sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1 | |
| 2 | /*--------------------------------------------------------------------*/ |
sewardj | 752f906 | 2010-05-03 21:38:49 +0000 | [diff] [blame] | 3 | /*--- begin guest_arm_toIR.c ---*/ |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 4 | /*--------------------------------------------------------------------*/ |
| 5 | |
| 6 | /* |
sewardj | 752f906 | 2010-05-03 21:38:49 +0000 | [diff] [blame] | 7 | This file is part of Valgrind, a dynamic binary instrumentation |
| 8 | framework. |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 9 | |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 10 | Copyright (C) 2004-2017 OpenWorks LLP |
sewardj | 752f906 | 2010-05-03 21:38:49 +0000 | [diff] [blame] | 11 | info@open-works.net |
sewardj | 64733c4 | 2010-10-12 10:10:46 +0000 | [diff] [blame] | 12 | |
| 13 | NEON support is |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 14 | Copyright (C) 2010-2017 Samsung Electronics |
sewardj | 64733c4 | 2010-10-12 10:10:46 +0000 | [diff] [blame] | 15 | contributed by Dmitry Zhurikhin <zhur@ispras.ru> |
| 16 | and Kirill Batuzov <batuzovk@ispras.ru> |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 17 | |
sewardj | 752f906 | 2010-05-03 21:38:49 +0000 | [diff] [blame] | 18 | This program is free software; you can redistribute it and/or |
| 19 | modify it under the terms of the GNU General Public License as |
| 20 | published by the Free Software Foundation; either version 2 of the |
| 21 | License, or (at your option) any later version. |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 22 | |
sewardj | 752f906 | 2010-05-03 21:38:49 +0000 | [diff] [blame] | 23 | This program is distributed in the hope that it will be useful, but |
| 24 | WITHOUT ANY WARRANTY; without even the implied warranty of |
| 25 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 26 | General Public License for more details. |
| 27 | |
| 28 | You should have received a copy of the GNU General Public License |
| 29 | along with this program; if not, write to the Free Software |
| 30 | Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA |
sewardj | 7bd6ffe | 2005-08-03 16:07:36 +0000 | [diff] [blame] | 31 | 02110-1301, USA. |
| 32 | |
sewardj | 752f906 | 2010-05-03 21:38:49 +0000 | [diff] [blame] | 33 | The GNU General Public License is contained in the file COPYING. |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 34 | */ |
| 35 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 36 | /* XXXX thumb to check: |
| 37 | that all cases where putIRegT writes r15, we generate a jump. |
| 38 | |
| 39 | All uses of newTemp assign to an IRTemp and not a UInt |
| 40 | |
| 41 | For all thumb loads and stores, including VFP ones, new-ITSTATE is |
| 42 | backed out before the memory op, and restored afterwards. This |
| 43 | needs to happen even after we go uncond. (and for sure it doesn't |
| 44 | happen for VFP loads/stores right now). |
| 45 | |
| 46 | VFP on thumb: check that we exclude all r13/r15 cases that we |
| 47 | should. |
| 48 | |
| 49 | XXXX thumb to do: improve the ITSTATE-zeroing optimisation by |
| 50 | taking into account the number of insns guarded by an IT. |
| 51 | |
| 52 | remove the nasty hack, in the spechelper, of looking for Or32(..., |
| 53 | 0xE0) in as the first arg to armg_calculate_condition, and instead |
| 54 | use Slice44 as specified in comments in the spechelper. |
| 55 | |
| 56 | add specialisations for armg_calculate_flag_c and _v, as they |
| 57 | are moderately often needed in Thumb code. |
| 58 | |
| 59 | Correctness: ITSTATE handling in Thumb SVCs is wrong. |
| 60 | |
| 61 | Correctness (obscure): in m_transtab, when invalidating code |
| 62 | address ranges, invalidate up to 18 bytes after the end of the |
| 63 | range. This is because the ITSTATE optimisation at the top of |
| 64 | _THUMB_WRK below analyses up to 18 bytes before the start of any |
| 65 | given instruction, and so might depend on the invalidated area. |
| 66 | */ |
| 67 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 68 | /* Limitations, etc |
| 69 | |
sewardj | 92001d4 | 2013-04-26 10:24:31 +0000 | [diff] [blame] | 70 | - pretty dodgy exception semantics for {LD,ST}Mxx and {LD,ST}RD. |
| 71 | These instructions are non-restartable in the case where the |
| 72 | transfer(s) fault. |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 73 | |
| 74 | - SWP: the restart jump back is Ijk_Boring; it should be |
| 75 | Ijk_NoRedir but that's expensive. See comments on casLE() in |
| 76 | guest_x86_toIR.c. |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 77 | */ |
| 78 | |
| 79 | /* "Special" instructions. |
| 80 | |
| 81 | This instruction decoder can decode four special instructions |
| 82 | which mean nothing natively (are no-ops as far as regs/mem are |
| 83 | concerned) but have meaning for supporting Valgrind. A special |
| 84 | instruction is flagged by a 16-byte preamble: |
| 85 | |
| 86 | E1A0C1EC E1A0C6EC E1A0CEEC E1A0C9EC |
| 87 | (mov r12, r12, ROR #3; mov r12, r12, ROR #13; |
| 88 | mov r12, r12, ROR #29; mov r12, r12, ROR #19) |
| 89 | |
| 90 | Following that, one of the following 3 are allowed |
| 91 | (standard interpretation in parentheses): |
| 92 | |
| 93 | E18AA00A (orr r10,r10,r10) R3 = client_request ( R4 ) |
| 94 | E18BB00B (orr r11,r11,r11) R3 = guest_NRADDR |
| 95 | E18CC00C (orr r12,r12,r12) branch-and-link-to-noredir R4 |
sewardj | a0d8eb8 | 2012-09-03 21:48:42 +0000 | [diff] [blame] | 96 | E1899009 (orr r9,r9,r9) IR injection |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 97 | |
| 98 | Any other bytes following the 16-byte preamble are illegal and |
| 99 | constitute a failure in instruction decoding. This all assumes |
| 100 | that the preamble will never occur except in specific code |
| 101 | fragments designed for Valgrind to catch. |
| 102 | */ |
| 103 | |
| 104 | /* Translates ARM(v5) code to IR. */ |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 105 | |
| 106 | #include "libvex_basictypes.h" |
| 107 | #include "libvex_ir.h" |
| 108 | #include "libvex.h" |
| 109 | #include "libvex_guest_arm.h" |
| 110 | |
sewardj | cef7d3e | 2009-07-02 12:21:59 +0000 | [diff] [blame] | 111 | #include "main_util.h" |
| 112 | #include "main_globals.h" |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 113 | #include "guest_generic_bb_to_IR.h" |
sewardj | cef7d3e | 2009-07-02 12:21:59 +0000 | [diff] [blame] | 114 | #include "guest_arm_defs.h" |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 115 | |
| 116 | |
| 117 | /*------------------------------------------------------------*/ |
| 118 | /*--- Globals ---*/ |
| 119 | /*------------------------------------------------------------*/ |
| 120 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 121 | /* These are set at the start of the translation of a instruction, so |
| 122 | that we don't have to pass them around endlessly. CONST means does |
| 123 | not change during translation of the instruction. |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 124 | */ |
| 125 | |
sewardj | 9b76916 | 2014-07-24 12:42:03 +0000 | [diff] [blame] | 126 | /* CONST: what is the host's endianness? This has to do with float vs |
| 127 | double register accesses on VFP, but it's complex and not properly |
| 128 | thought out. */ |
| 129 | static VexEndness host_endness; |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 130 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 131 | /* CONST: The guest address for the instruction currently being |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 132 | translated. This is the real, "decoded" address (not subject |
| 133 | to the CPSR.T kludge). */ |
| 134 | static Addr32 guest_R15_curr_instr_notENC; |
| 135 | |
| 136 | /* CONST, FOR ASSERTIONS ONLY. Indicates whether currently processed |
| 137 | insn is Thumb (True) or ARM (False). */ |
| 138 | static Bool __curr_is_Thumb; |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 139 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 140 | /* MOD: The IRSB* into which we're generating code. */ |
sewardj | dd40fdf | 2006-12-24 02:20:24 +0000 | [diff] [blame] | 141 | static IRSB* irsb; |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 142 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 143 | /* These are to do with handling writes to r15. They are initially |
| 144 | set at the start of disInstr_ARM_WRK to indicate no update, |
| 145 | possibly updated during the routine, and examined again at the end. |
| 146 | If they have been set to indicate a r15 update then a jump is |
| 147 | generated. Note, "explicit" jumps (b, bx, etc) are generated |
| 148 | directly, not using this mechanism -- this is intended to handle |
| 149 | the implicit-style jumps resulting from (eg) assigning to r15 as |
| 150 | the result of insns we wouldn't normally consider branchy. */ |
| 151 | |
| 152 | /* MOD. Initially False; set to True iff abovementioned handling is |
| 153 | required. */ |
| 154 | static Bool r15written; |
| 155 | |
| 156 | /* MOD. Initially IRTemp_INVALID. If the r15 branch to be generated |
| 157 | is conditional, this holds the gating IRTemp :: Ity_I32. If the |
| 158 | branch to be generated is unconditional, this remains |
| 159 | IRTemp_INVALID. */ |
| 160 | static IRTemp r15guard; /* :: Ity_I32, 0 or 1 */ |
| 161 | |
| 162 | /* MOD. Initially Ijk_Boring. If an r15 branch is to be generated, |
| 163 | this holds the jump kind. */ |
| 164 | static IRTemp r15kind; |
| 165 | |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 166 | |
| 167 | /*------------------------------------------------------------*/ |
| 168 | /*--- Debugging output ---*/ |
| 169 | /*------------------------------------------------------------*/ |
| 170 | |
| 171 | #define DIP(format, args...) \ |
| 172 | if (vex_traceflags & VEX_TRACE_FE) \ |
| 173 | vex_printf(format, ## args) |
| 174 | |
| 175 | #define DIS(buf, format, args...) \ |
| 176 | if (vex_traceflags & VEX_TRACE_FE) \ |
| 177 | vex_sprintf(buf, format, ## args) |
| 178 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 179 | #define ASSERT_IS_THUMB \ |
| 180 | do { vassert(__curr_is_Thumb); } while (0) |
| 181 | |
| 182 | #define ASSERT_IS_ARM \ |
| 183 | do { vassert(! __curr_is_Thumb); } while (0) |
| 184 | |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 185 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 186 | /*------------------------------------------------------------*/ |
| 187 | /*--- Helper bits and pieces for deconstructing the ---*/ |
| 188 | /*--- arm insn stream. ---*/ |
| 189 | /*------------------------------------------------------------*/ |
| 190 | |
| 191 | /* Do a little-endian load of a 32-bit word, regardless of the |
| 192 | endianness of the underlying host. */ |
florian | 8462d11 | 2014-09-24 15:18:09 +0000 | [diff] [blame] | 193 | static inline UInt getUIntLittleEndianly ( const UChar* p ) |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 194 | { |
| 195 | UInt w = 0; |
| 196 | w = (w << 8) | p[3]; |
| 197 | w = (w << 8) | p[2]; |
| 198 | w = (w << 8) | p[1]; |
| 199 | w = (w << 8) | p[0]; |
| 200 | return w; |
| 201 | } |
| 202 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 203 | /* Do a little-endian load of a 16-bit word, regardless of the |
| 204 | endianness of the underlying host. */ |
florian | 8462d11 | 2014-09-24 15:18:09 +0000 | [diff] [blame] | 205 | static inline UShort getUShortLittleEndianly ( const UChar* p ) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 206 | { |
| 207 | UShort w = 0; |
| 208 | w = (w << 8) | p[1]; |
| 209 | w = (w << 8) | p[0]; |
| 210 | return w; |
| 211 | } |
| 212 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 213 | static UInt ROR32 ( UInt x, UInt sh ) { |
| 214 | vassert(sh >= 0 && sh < 32); |
| 215 | if (sh == 0) |
| 216 | return x; |
| 217 | else |
| 218 | return (x << (32-sh)) | (x >> sh); |
| 219 | } |
| 220 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 221 | static Int popcount32 ( UInt x ) |
| 222 | { |
| 223 | Int res = 0, i; |
| 224 | for (i = 0; i < 32; i++) { |
| 225 | res += (x & 1); |
| 226 | x >>= 1; |
| 227 | } |
| 228 | return res; |
| 229 | } |
| 230 | |
| 231 | static UInt setbit32 ( UInt x, Int ix, UInt b ) |
| 232 | { |
| 233 | UInt mask = 1 << ix; |
| 234 | x &= ~mask; |
| 235 | x |= ((b << ix) & mask); |
| 236 | return x; |
| 237 | } |
| 238 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 239 | #define BITS2(_b1,_b0) \ |
| 240 | (((_b1) << 1) | (_b0)) |
| 241 | |
| 242 | #define BITS3(_b2,_b1,_b0) \ |
| 243 | (((_b2) << 2) | ((_b1) << 1) | (_b0)) |
| 244 | |
| 245 | #define BITS4(_b3,_b2,_b1,_b0) \ |
| 246 | (((_b3) << 3) | ((_b2) << 2) | ((_b1) << 1) | (_b0)) |
| 247 | |
| 248 | #define BITS8(_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 249 | ((BITS4((_b7),(_b6),(_b5),(_b4)) << 4) \ |
| 250 | | BITS4((_b3),(_b2),(_b1),(_b0))) |
cerion | a70a37b | 2004-12-03 18:54:08 +0000 | [diff] [blame] | 251 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 252 | #define BITS5(_b4,_b3,_b2,_b1,_b0) \ |
| 253 | (BITS8(0,0,0,(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 254 | #define BITS6(_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 255 | (BITS8(0,0,(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 256 | #define BITS7(_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 257 | (BITS8(0,(_b6),(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 258 | |
| 259 | #define BITS9(_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 260 | (((_b8) << 8) \ |
| 261 | | BITS8((_b7),(_b6),(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 262 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 263 | #define BITS10(_b9,_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 264 | (((_b9) << 9) | ((_b8) << 8) \ |
| 265 | | BITS8((_b7),(_b6),(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 266 | |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 267 | #define BITS11(_b10,_b9,_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 268 | ( ((_b10) << 10) | ((_b9) << 9) | ((_b8) << 8) \ |
| 269 | | BITS8((_b7),(_b6),(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 270 | |
| 271 | #define BITS12(_b11,_b10,_b9,_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 272 | ( ((_b11) << 11) | ((_b10) << 10) | ((_b9) << 9) | ((_b8) << 8) \ |
| 273 | | BITS8((_b7),(_b6),(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 274 | |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 275 | /* produces _uint[_bMax:_bMin] */ |
| 276 | #define SLICE_UInt(_uint,_bMax,_bMin) \ |
| 277 | (( ((UInt)(_uint)) >> (_bMin)) \ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 278 | & (UInt)((1ULL << ((_bMax) - (_bMin) + 1)) - 1ULL)) |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 279 | |
cerion | a70a37b | 2004-12-03 18:54:08 +0000 | [diff] [blame] | 280 | |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 281 | /*------------------------------------------------------------*/ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 282 | /*--- Helper bits and pieces for creating IR fragments. ---*/ |
| 283 | /*------------------------------------------------------------*/ |
| 284 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 285 | static IRExpr* mkU64 ( ULong i ) |
| 286 | { |
| 287 | return IRExpr_Const(IRConst_U64(i)); |
| 288 | } |
| 289 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 290 | static IRExpr* mkU32 ( UInt i ) |
| 291 | { |
| 292 | return IRExpr_Const(IRConst_U32(i)); |
| 293 | } |
| 294 | |
| 295 | static IRExpr* mkU8 ( UInt i ) |
| 296 | { |
| 297 | vassert(i < 256); |
| 298 | return IRExpr_Const(IRConst_U8( (UChar)i )); |
| 299 | } |
| 300 | |
| 301 | static IRExpr* mkexpr ( IRTemp tmp ) |
| 302 | { |
| 303 | return IRExpr_RdTmp(tmp); |
| 304 | } |
| 305 | |
| 306 | static IRExpr* unop ( IROp op, IRExpr* a ) |
| 307 | { |
| 308 | return IRExpr_Unop(op, a); |
| 309 | } |
| 310 | |
| 311 | static IRExpr* binop ( IROp op, IRExpr* a1, IRExpr* a2 ) |
| 312 | { |
| 313 | return IRExpr_Binop(op, a1, a2); |
| 314 | } |
| 315 | |
| 316 | static IRExpr* triop ( IROp op, IRExpr* a1, IRExpr* a2, IRExpr* a3 ) |
| 317 | { |
| 318 | return IRExpr_Triop(op, a1, a2, a3); |
| 319 | } |
| 320 | |
| 321 | static IRExpr* loadLE ( IRType ty, IRExpr* addr ) |
| 322 | { |
| 323 | return IRExpr_Load(Iend_LE, ty, addr); |
| 324 | } |
| 325 | |
| 326 | /* Add a statement to the list held by "irbb". */ |
| 327 | static void stmt ( IRStmt* st ) |
| 328 | { |
| 329 | addStmtToIRSB( irsb, st ); |
| 330 | } |
| 331 | |
| 332 | static void assign ( IRTemp dst, IRExpr* e ) |
| 333 | { |
| 334 | stmt( IRStmt_WrTmp(dst, e) ); |
| 335 | } |
| 336 | |
| 337 | static void storeLE ( IRExpr* addr, IRExpr* data ) |
| 338 | { |
| 339 | stmt( IRStmt_Store(Iend_LE, addr, data) ); |
| 340 | } |
| 341 | |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 342 | static void storeGuardedLE ( IRExpr* addr, IRExpr* data, IRTemp guardT ) |
| 343 | { |
| 344 | if (guardT == IRTemp_INVALID) { |
| 345 | /* unconditional */ |
| 346 | storeLE(addr, data); |
| 347 | } else { |
| 348 | stmt( IRStmt_StoreG(Iend_LE, addr, data, |
| 349 | binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0))) ); |
| 350 | } |
| 351 | } |
| 352 | |
| 353 | static void loadGuardedLE ( IRTemp dst, IRLoadGOp cvt, |
| 354 | IRExpr* addr, IRExpr* alt, |
| 355 | IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
| 356 | { |
| 357 | if (guardT == IRTemp_INVALID) { |
| 358 | /* unconditional */ |
| 359 | IRExpr* loaded = NULL; |
| 360 | switch (cvt) { |
| 361 | case ILGop_Ident32: |
| 362 | loaded = loadLE(Ity_I32, addr); break; |
| 363 | case ILGop_8Uto32: |
| 364 | loaded = unop(Iop_8Uto32, loadLE(Ity_I8, addr)); break; |
| 365 | case ILGop_8Sto32: |
| 366 | loaded = unop(Iop_8Sto32, loadLE(Ity_I8, addr)); break; |
| 367 | case ILGop_16Uto32: |
| 368 | loaded = unop(Iop_16Uto32, loadLE(Ity_I16, addr)); break; |
| 369 | case ILGop_16Sto32: |
| 370 | loaded = unop(Iop_16Sto32, loadLE(Ity_I16, addr)); break; |
| 371 | default: |
| 372 | vassert(0); |
| 373 | } |
| 374 | vassert(loaded != NULL); |
| 375 | assign(dst, loaded); |
| 376 | } else { |
| 377 | /* Generate a guarded load into 'dst', but apply 'cvt' to the |
| 378 | loaded data before putting the data in 'dst'. If the load |
| 379 | does not take place, 'alt' is placed directly in 'dst'. */ |
| 380 | stmt( IRStmt_LoadG(Iend_LE, cvt, dst, addr, alt, |
| 381 | binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0))) ); |
| 382 | } |
| 383 | } |
| 384 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 385 | /* Generate a new temporary of the given type. */ |
| 386 | static IRTemp newTemp ( IRType ty ) |
| 387 | { |
| 388 | vassert(isPlausibleIRType(ty)); |
| 389 | return newIRTemp( irsb->tyenv, ty ); |
| 390 | } |
| 391 | |
| 392 | /* Produces a value in 0 .. 3, which is encoded as per the type |
| 393 | IRRoundingMode. */ |
| 394 | static IRExpr* /* :: Ity_I32 */ get_FAKE_roundingmode ( void ) |
| 395 | { |
| 396 | return mkU32(Irrm_NEAREST); |
| 397 | } |
| 398 | |
| 399 | /* Generate an expression for SRC rotated right by ROT. */ |
| 400 | static IRExpr* genROR32( IRTemp src, Int rot ) |
| 401 | { |
| 402 | vassert(rot >= 0 && rot < 32); |
| 403 | if (rot == 0) |
| 404 | return mkexpr(src); |
| 405 | return |
| 406 | binop(Iop_Or32, |
| 407 | binop(Iop_Shl32, mkexpr(src), mkU8(32 - rot)), |
| 408 | binop(Iop_Shr32, mkexpr(src), mkU8(rot))); |
| 409 | } |
| 410 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 411 | static IRExpr* mkU128 ( ULong i ) |
| 412 | { |
| 413 | return binop(Iop_64HLtoV128, mkU64(i), mkU64(i)); |
| 414 | } |
| 415 | |
| 416 | /* Generate a 4-aligned version of the given expression if |
| 417 | the given condition is true. Else return it unchanged. */ |
| 418 | static IRExpr* align4if ( IRExpr* e, Bool b ) |
| 419 | { |
| 420 | if (b) |
| 421 | return binop(Iop_And32, e, mkU32(~3)); |
| 422 | else |
| 423 | return e; |
| 424 | } |
| 425 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 426 | |
| 427 | /*------------------------------------------------------------*/ |
| 428 | /*--- Helpers for accessing guest registers. ---*/ |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 429 | /*------------------------------------------------------------*/ |
| 430 | |
| 431 | #define OFFB_R0 offsetof(VexGuestARMState,guest_R0) |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 432 | #define OFFB_R1 offsetof(VexGuestARMState,guest_R1) |
| 433 | #define OFFB_R2 offsetof(VexGuestARMState,guest_R2) |
| 434 | #define OFFB_R3 offsetof(VexGuestARMState,guest_R3) |
| 435 | #define OFFB_R4 offsetof(VexGuestARMState,guest_R4) |
| 436 | #define OFFB_R5 offsetof(VexGuestARMState,guest_R5) |
| 437 | #define OFFB_R6 offsetof(VexGuestARMState,guest_R6) |
| 438 | #define OFFB_R7 offsetof(VexGuestARMState,guest_R7) |
| 439 | #define OFFB_R8 offsetof(VexGuestARMState,guest_R8) |
| 440 | #define OFFB_R9 offsetof(VexGuestARMState,guest_R9) |
| 441 | #define OFFB_R10 offsetof(VexGuestARMState,guest_R10) |
| 442 | #define OFFB_R11 offsetof(VexGuestARMState,guest_R11) |
| 443 | #define OFFB_R12 offsetof(VexGuestARMState,guest_R12) |
| 444 | #define OFFB_R13 offsetof(VexGuestARMState,guest_R13) |
| 445 | #define OFFB_R14 offsetof(VexGuestARMState,guest_R14) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 446 | #define OFFB_R15T offsetof(VexGuestARMState,guest_R15T) |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 447 | |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 448 | #define OFFB_CC_OP offsetof(VexGuestARMState,guest_CC_OP) |
| 449 | #define OFFB_CC_DEP1 offsetof(VexGuestARMState,guest_CC_DEP1) |
| 450 | #define OFFB_CC_DEP2 offsetof(VexGuestARMState,guest_CC_DEP2) |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 451 | #define OFFB_CC_NDEP offsetof(VexGuestARMState,guest_CC_NDEP) |
| 452 | #define OFFB_NRADDR offsetof(VexGuestARMState,guest_NRADDR) |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 453 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 454 | #define OFFB_D0 offsetof(VexGuestARMState,guest_D0) |
| 455 | #define OFFB_D1 offsetof(VexGuestARMState,guest_D1) |
| 456 | #define OFFB_D2 offsetof(VexGuestARMState,guest_D2) |
| 457 | #define OFFB_D3 offsetof(VexGuestARMState,guest_D3) |
| 458 | #define OFFB_D4 offsetof(VexGuestARMState,guest_D4) |
| 459 | #define OFFB_D5 offsetof(VexGuestARMState,guest_D5) |
| 460 | #define OFFB_D6 offsetof(VexGuestARMState,guest_D6) |
| 461 | #define OFFB_D7 offsetof(VexGuestARMState,guest_D7) |
| 462 | #define OFFB_D8 offsetof(VexGuestARMState,guest_D8) |
| 463 | #define OFFB_D9 offsetof(VexGuestARMState,guest_D9) |
| 464 | #define OFFB_D10 offsetof(VexGuestARMState,guest_D10) |
| 465 | #define OFFB_D11 offsetof(VexGuestARMState,guest_D11) |
| 466 | #define OFFB_D12 offsetof(VexGuestARMState,guest_D12) |
| 467 | #define OFFB_D13 offsetof(VexGuestARMState,guest_D13) |
| 468 | #define OFFB_D14 offsetof(VexGuestARMState,guest_D14) |
| 469 | #define OFFB_D15 offsetof(VexGuestARMState,guest_D15) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 470 | #define OFFB_D16 offsetof(VexGuestARMState,guest_D16) |
| 471 | #define OFFB_D17 offsetof(VexGuestARMState,guest_D17) |
| 472 | #define OFFB_D18 offsetof(VexGuestARMState,guest_D18) |
| 473 | #define OFFB_D19 offsetof(VexGuestARMState,guest_D19) |
| 474 | #define OFFB_D20 offsetof(VexGuestARMState,guest_D20) |
| 475 | #define OFFB_D21 offsetof(VexGuestARMState,guest_D21) |
| 476 | #define OFFB_D22 offsetof(VexGuestARMState,guest_D22) |
| 477 | #define OFFB_D23 offsetof(VexGuestARMState,guest_D23) |
| 478 | #define OFFB_D24 offsetof(VexGuestARMState,guest_D24) |
| 479 | #define OFFB_D25 offsetof(VexGuestARMState,guest_D25) |
| 480 | #define OFFB_D26 offsetof(VexGuestARMState,guest_D26) |
| 481 | #define OFFB_D27 offsetof(VexGuestARMState,guest_D27) |
| 482 | #define OFFB_D28 offsetof(VexGuestARMState,guest_D28) |
| 483 | #define OFFB_D29 offsetof(VexGuestARMState,guest_D29) |
| 484 | #define OFFB_D30 offsetof(VexGuestARMState,guest_D30) |
| 485 | #define OFFB_D31 offsetof(VexGuestARMState,guest_D31) |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 486 | |
| 487 | #define OFFB_FPSCR offsetof(VexGuestARMState,guest_FPSCR) |
| 488 | #define OFFB_TPIDRURO offsetof(VexGuestARMState,guest_TPIDRURO) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 489 | #define OFFB_ITSTATE offsetof(VexGuestARMState,guest_ITSTATE) |
| 490 | #define OFFB_QFLAG32 offsetof(VexGuestARMState,guest_QFLAG32) |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 491 | #define OFFB_GEFLAG0 offsetof(VexGuestARMState,guest_GEFLAG0) |
| 492 | #define OFFB_GEFLAG1 offsetof(VexGuestARMState,guest_GEFLAG1) |
| 493 | #define OFFB_GEFLAG2 offsetof(VexGuestARMState,guest_GEFLAG2) |
| 494 | #define OFFB_GEFLAG3 offsetof(VexGuestARMState,guest_GEFLAG3) |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 495 | |
sewardj | 05f5e01 | 2014-05-04 10:52:11 +0000 | [diff] [blame] | 496 | #define OFFB_CMSTART offsetof(VexGuestARMState,guest_CMSTART) |
| 497 | #define OFFB_CMLEN offsetof(VexGuestARMState,guest_CMLEN) |
sewardj | a0d8eb8 | 2012-09-03 21:48:42 +0000 | [diff] [blame] | 498 | |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 499 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 500 | /* ---------------- Integer registers ---------------- */ |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 501 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 502 | static Int integerGuestRegOffset ( UInt iregNo ) |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 503 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 504 | /* Do we care about endianness here? We do if sub-parts of integer |
| 505 | registers are accessed, but I don't think that ever happens on |
| 506 | ARM. */ |
| 507 | switch (iregNo) { |
| 508 | case 0: return OFFB_R0; |
| 509 | case 1: return OFFB_R1; |
| 510 | case 2: return OFFB_R2; |
| 511 | case 3: return OFFB_R3; |
| 512 | case 4: return OFFB_R4; |
| 513 | case 5: return OFFB_R5; |
| 514 | case 6: return OFFB_R6; |
| 515 | case 7: return OFFB_R7; |
| 516 | case 8: return OFFB_R8; |
| 517 | case 9: return OFFB_R9; |
| 518 | case 10: return OFFB_R10; |
| 519 | case 11: return OFFB_R11; |
| 520 | case 12: return OFFB_R12; |
| 521 | case 13: return OFFB_R13; |
| 522 | case 14: return OFFB_R14; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 523 | case 15: return OFFB_R15T; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 524 | default: vassert(0); |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 525 | } |
| 526 | } |
| 527 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 528 | /* Plain ("low level") read from a reg; no +8 offset magic for r15. */ |
| 529 | static IRExpr* llGetIReg ( UInt iregNo ) |
| 530 | { |
| 531 | vassert(iregNo < 16); |
| 532 | return IRExpr_Get( integerGuestRegOffset(iregNo), Ity_I32 ); |
| 533 | } |
| 534 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 535 | /* Architected read from a reg in ARM mode. This automagically adds 8 |
| 536 | to all reads of r15. */ |
| 537 | static IRExpr* getIRegA ( UInt iregNo ) |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 538 | { |
| 539 | IRExpr* e; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 540 | ASSERT_IS_ARM; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 541 | vassert(iregNo < 16); |
| 542 | if (iregNo == 15) { |
| 543 | /* If asked for r15, don't read the guest state value, as that |
| 544 | may not be up to date in the case where loop unrolling has |
| 545 | happened, because the first insn's write to the block is |
| 546 | omitted; hence in the 2nd and subsequent unrollings we don't |
| 547 | have a correct value in guest r15. Instead produce the |
| 548 | constant that we know would be produced at this point. */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 549 | vassert(0 == (guest_R15_curr_instr_notENC & 3)); |
| 550 | e = mkU32(guest_R15_curr_instr_notENC + 8); |
| 551 | } else { |
| 552 | e = IRExpr_Get( integerGuestRegOffset(iregNo), Ity_I32 ); |
| 553 | } |
| 554 | return e; |
| 555 | } |
| 556 | |
| 557 | /* Architected read from a reg in Thumb mode. This automagically adds |
| 558 | 4 to all reads of r15. */ |
| 559 | static IRExpr* getIRegT ( UInt iregNo ) |
| 560 | { |
| 561 | IRExpr* e; |
| 562 | ASSERT_IS_THUMB; |
| 563 | vassert(iregNo < 16); |
| 564 | if (iregNo == 15) { |
| 565 | /* Ditto comment in getIReg. */ |
| 566 | vassert(0 == (guest_R15_curr_instr_notENC & 1)); |
| 567 | e = mkU32(guest_R15_curr_instr_notENC + 4); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 568 | } else { |
| 569 | e = IRExpr_Get( integerGuestRegOffset(iregNo), Ity_I32 ); |
| 570 | } |
| 571 | return e; |
| 572 | } |
| 573 | |
| 574 | /* Plain ("low level") write to a reg; no jump or alignment magic for |
| 575 | r15. */ |
| 576 | static void llPutIReg ( UInt iregNo, IRExpr* e ) |
| 577 | { |
| 578 | vassert(iregNo < 16); |
| 579 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I32); |
| 580 | stmt( IRStmt_Put(integerGuestRegOffset(iregNo), e) ); |
| 581 | } |
| 582 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 583 | /* Architected write to an integer register in ARM mode. If it is to |
| 584 | r15, record info so at the end of this insn's translation, a branch |
| 585 | to it can be made. Also handles conditional writes to the |
| 586 | register: if guardT == IRTemp_INVALID then the write is |
| 587 | unconditional. If writing r15, also 4-align it. */ |
| 588 | static void putIRegA ( UInt iregNo, |
| 589 | IRExpr* e, |
| 590 | IRTemp guardT /* :: Ity_I32, 0 or 1 */, |
| 591 | IRJumpKind jk /* if a jump is generated */ ) |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 592 | { |
| 593 | /* if writing r15, force e to be 4-aligned. */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 594 | // INTERWORKING FIXME. this needs to be relaxed so that |
| 595 | // puts caused by LDMxx which load r15 interwork right. |
| 596 | // but is no aligned too relaxed? |
| 597 | //if (iregNo == 15) |
| 598 | // e = binop(Iop_And32, e, mkU32(~3)); |
| 599 | ASSERT_IS_ARM; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 600 | /* So, generate either an unconditional or a conditional write to |
| 601 | the reg. */ |
| 602 | if (guardT == IRTemp_INVALID) { |
| 603 | /* unconditional write */ |
| 604 | llPutIReg( iregNo, e ); |
| 605 | } else { |
| 606 | llPutIReg( iregNo, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 607 | IRExpr_ITE( binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0)), |
| 608 | e, llGetIReg(iregNo) )); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 609 | } |
| 610 | if (iregNo == 15) { |
| 611 | // assert against competing r15 updates. Shouldn't |
| 612 | // happen; should be ruled out by the instr matching |
| 613 | // logic. |
| 614 | vassert(r15written == False); |
| 615 | vassert(r15guard == IRTemp_INVALID); |
| 616 | vassert(r15kind == Ijk_Boring); |
| 617 | r15written = True; |
| 618 | r15guard = guardT; |
| 619 | r15kind = jk; |
| 620 | } |
| 621 | } |
| 622 | |
| 623 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 624 | /* Architected write to an integer register in Thumb mode. Writes to |
| 625 | r15 are not allowed. Handles conditional writes to the register: |
| 626 | if guardT == IRTemp_INVALID then the write is unconditional. */ |
| 627 | static void putIRegT ( UInt iregNo, |
| 628 | IRExpr* e, |
| 629 | IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
| 630 | { |
| 631 | /* So, generate either an unconditional or a conditional write to |
| 632 | the reg. */ |
| 633 | ASSERT_IS_THUMB; |
| 634 | vassert(iregNo >= 0 && iregNo <= 14); |
| 635 | if (guardT == IRTemp_INVALID) { |
| 636 | /* unconditional write */ |
| 637 | llPutIReg( iregNo, e ); |
| 638 | } else { |
| 639 | llPutIReg( iregNo, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 640 | IRExpr_ITE( binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0)), |
| 641 | e, llGetIReg(iregNo) )); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 642 | } |
| 643 | } |
| 644 | |
| 645 | |
| 646 | /* Thumb16 and Thumb32 only. |
| 647 | Returns true if reg is 13 or 15. Implements the BadReg |
| 648 | predicate in the ARM ARM. */ |
| 649 | static Bool isBadRegT ( UInt r ) |
| 650 | { |
| 651 | vassert(r <= 15); |
| 652 | ASSERT_IS_THUMB; |
| 653 | return r == 13 || r == 15; |
| 654 | } |
| 655 | |
| 656 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 657 | /* ---------------- Double registers ---------------- */ |
| 658 | |
| 659 | static Int doubleGuestRegOffset ( UInt dregNo ) |
| 660 | { |
| 661 | /* Do we care about endianness here? Probably do if we ever get |
| 662 | into the situation of dealing with the single-precision VFP |
| 663 | registers. */ |
| 664 | switch (dregNo) { |
| 665 | case 0: return OFFB_D0; |
| 666 | case 1: return OFFB_D1; |
| 667 | case 2: return OFFB_D2; |
| 668 | case 3: return OFFB_D3; |
| 669 | case 4: return OFFB_D4; |
| 670 | case 5: return OFFB_D5; |
| 671 | case 6: return OFFB_D6; |
| 672 | case 7: return OFFB_D7; |
| 673 | case 8: return OFFB_D8; |
| 674 | case 9: return OFFB_D9; |
| 675 | case 10: return OFFB_D10; |
| 676 | case 11: return OFFB_D11; |
| 677 | case 12: return OFFB_D12; |
| 678 | case 13: return OFFB_D13; |
| 679 | case 14: return OFFB_D14; |
| 680 | case 15: return OFFB_D15; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 681 | case 16: return OFFB_D16; |
| 682 | case 17: return OFFB_D17; |
| 683 | case 18: return OFFB_D18; |
| 684 | case 19: return OFFB_D19; |
| 685 | case 20: return OFFB_D20; |
| 686 | case 21: return OFFB_D21; |
| 687 | case 22: return OFFB_D22; |
| 688 | case 23: return OFFB_D23; |
| 689 | case 24: return OFFB_D24; |
| 690 | case 25: return OFFB_D25; |
| 691 | case 26: return OFFB_D26; |
| 692 | case 27: return OFFB_D27; |
| 693 | case 28: return OFFB_D28; |
| 694 | case 29: return OFFB_D29; |
| 695 | case 30: return OFFB_D30; |
| 696 | case 31: return OFFB_D31; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 697 | default: vassert(0); |
| 698 | } |
| 699 | } |
| 700 | |
| 701 | /* Plain ("low level") read from a VFP Dreg. */ |
| 702 | static IRExpr* llGetDReg ( UInt dregNo ) |
| 703 | { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 704 | vassert(dregNo < 32); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 705 | return IRExpr_Get( doubleGuestRegOffset(dregNo), Ity_F64 ); |
| 706 | } |
| 707 | |
| 708 | /* Architected read from a VFP Dreg. */ |
| 709 | static IRExpr* getDReg ( UInt dregNo ) { |
| 710 | return llGetDReg( dregNo ); |
| 711 | } |
| 712 | |
| 713 | /* Plain ("low level") write to a VFP Dreg. */ |
| 714 | static void llPutDReg ( UInt dregNo, IRExpr* e ) |
| 715 | { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 716 | vassert(dregNo < 32); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 717 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_F64); |
| 718 | stmt( IRStmt_Put(doubleGuestRegOffset(dregNo), e) ); |
| 719 | } |
| 720 | |
| 721 | /* Architected write to a VFP Dreg. Handles conditional writes to the |
| 722 | register: if guardT == IRTemp_INVALID then the write is |
| 723 | unconditional. */ |
| 724 | static void putDReg ( UInt dregNo, |
| 725 | IRExpr* e, |
| 726 | IRTemp guardT /* :: Ity_I32, 0 or 1 */) |
| 727 | { |
| 728 | /* So, generate either an unconditional or a conditional write to |
| 729 | the reg. */ |
| 730 | if (guardT == IRTemp_INVALID) { |
| 731 | /* unconditional write */ |
| 732 | llPutDReg( dregNo, e ); |
| 733 | } else { |
| 734 | llPutDReg( dregNo, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 735 | IRExpr_ITE( binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0)), |
| 736 | e, llGetDReg(dregNo) )); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 737 | } |
| 738 | } |
| 739 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 740 | /* And now exactly the same stuff all over again, but this time |
| 741 | taking/returning I64 rather than F64, to support 64-bit Neon |
| 742 | ops. */ |
| 743 | |
| 744 | /* Plain ("low level") read from a Neon Integer Dreg. */ |
| 745 | static IRExpr* llGetDRegI64 ( UInt dregNo ) |
| 746 | { |
| 747 | vassert(dregNo < 32); |
| 748 | return IRExpr_Get( doubleGuestRegOffset(dregNo), Ity_I64 ); |
| 749 | } |
| 750 | |
| 751 | /* Architected read from a Neon Integer Dreg. */ |
| 752 | static IRExpr* getDRegI64 ( UInt dregNo ) { |
| 753 | return llGetDRegI64( dregNo ); |
| 754 | } |
| 755 | |
| 756 | /* Plain ("low level") write to a Neon Integer Dreg. */ |
| 757 | static void llPutDRegI64 ( UInt dregNo, IRExpr* e ) |
| 758 | { |
| 759 | vassert(dregNo < 32); |
| 760 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I64); |
| 761 | stmt( IRStmt_Put(doubleGuestRegOffset(dregNo), e) ); |
| 762 | } |
| 763 | |
| 764 | /* Architected write to a Neon Integer Dreg. Handles conditional |
| 765 | writes to the register: if guardT == IRTemp_INVALID then the write |
| 766 | is unconditional. */ |
| 767 | static void putDRegI64 ( UInt dregNo, |
| 768 | IRExpr* e, |
| 769 | IRTemp guardT /* :: Ity_I32, 0 or 1 */) |
| 770 | { |
| 771 | /* So, generate either an unconditional or a conditional write to |
| 772 | the reg. */ |
| 773 | if (guardT == IRTemp_INVALID) { |
| 774 | /* unconditional write */ |
| 775 | llPutDRegI64( dregNo, e ); |
| 776 | } else { |
| 777 | llPutDRegI64( dregNo, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 778 | IRExpr_ITE( binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0)), |
| 779 | e, llGetDRegI64(dregNo) )); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 780 | } |
| 781 | } |
| 782 | |
| 783 | /* ---------------- Quad registers ---------------- */ |
| 784 | |
| 785 | static Int quadGuestRegOffset ( UInt qregNo ) |
| 786 | { |
| 787 | /* Do we care about endianness here? Probably do if we ever get |
| 788 | into the situation of dealing with the 64 bit Neon registers. */ |
| 789 | switch (qregNo) { |
| 790 | case 0: return OFFB_D0; |
| 791 | case 1: return OFFB_D2; |
| 792 | case 2: return OFFB_D4; |
| 793 | case 3: return OFFB_D6; |
| 794 | case 4: return OFFB_D8; |
| 795 | case 5: return OFFB_D10; |
| 796 | case 6: return OFFB_D12; |
| 797 | case 7: return OFFB_D14; |
| 798 | case 8: return OFFB_D16; |
| 799 | case 9: return OFFB_D18; |
| 800 | case 10: return OFFB_D20; |
| 801 | case 11: return OFFB_D22; |
| 802 | case 12: return OFFB_D24; |
| 803 | case 13: return OFFB_D26; |
| 804 | case 14: return OFFB_D28; |
| 805 | case 15: return OFFB_D30; |
| 806 | default: vassert(0); |
| 807 | } |
| 808 | } |
| 809 | |
| 810 | /* Plain ("low level") read from a Neon Qreg. */ |
| 811 | static IRExpr* llGetQReg ( UInt qregNo ) |
| 812 | { |
| 813 | vassert(qregNo < 16); |
| 814 | return IRExpr_Get( quadGuestRegOffset(qregNo), Ity_V128 ); |
| 815 | } |
| 816 | |
| 817 | /* Architected read from a Neon Qreg. */ |
| 818 | static IRExpr* getQReg ( UInt qregNo ) { |
| 819 | return llGetQReg( qregNo ); |
| 820 | } |
| 821 | |
| 822 | /* Plain ("low level") write to a Neon Qreg. */ |
| 823 | static void llPutQReg ( UInt qregNo, IRExpr* e ) |
| 824 | { |
| 825 | vassert(qregNo < 16); |
| 826 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_V128); |
| 827 | stmt( IRStmt_Put(quadGuestRegOffset(qregNo), e) ); |
| 828 | } |
| 829 | |
| 830 | /* Architected write to a Neon Qreg. Handles conditional writes to the |
| 831 | register: if guardT == IRTemp_INVALID then the write is |
| 832 | unconditional. */ |
| 833 | static void putQReg ( UInt qregNo, |
| 834 | IRExpr* e, |
| 835 | IRTemp guardT /* :: Ity_I32, 0 or 1 */) |
| 836 | { |
| 837 | /* So, generate either an unconditional or a conditional write to |
| 838 | the reg. */ |
| 839 | if (guardT == IRTemp_INVALID) { |
| 840 | /* unconditional write */ |
| 841 | llPutQReg( qregNo, e ); |
| 842 | } else { |
| 843 | llPutQReg( qregNo, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 844 | IRExpr_ITE( binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0)), |
| 845 | e, llGetQReg(qregNo) )); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 846 | } |
| 847 | } |
| 848 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 849 | |
| 850 | /* ---------------- Float registers ---------------- */ |
| 851 | |
| 852 | static Int floatGuestRegOffset ( UInt fregNo ) |
| 853 | { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 854 | /* Start with the offset of the containing double, and then correct |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 855 | for endianness. Actually this is completely bogus and needs |
| 856 | careful thought. */ |
| 857 | Int off; |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 858 | /* NB! Limit is 64, not 32, because we might be pulling F32 bits |
| 859 | out of SIMD registers, and there are 16 SIMD registers each of |
| 860 | 128 bits (4 x F32). */ |
| 861 | vassert(fregNo < 64); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 862 | off = doubleGuestRegOffset(fregNo >> 1); |
sewardj | 9b76916 | 2014-07-24 12:42:03 +0000 | [diff] [blame] | 863 | if (host_endness == VexEndnessLE) { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 864 | if (fregNo & 1) |
| 865 | off += 4; |
sewardj | 9b76916 | 2014-07-24 12:42:03 +0000 | [diff] [blame] | 866 | } else { |
| 867 | vassert(0); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 868 | } |
| 869 | return off; |
| 870 | } |
| 871 | |
| 872 | /* Plain ("low level") read from a VFP Freg. */ |
| 873 | static IRExpr* llGetFReg ( UInt fregNo ) |
| 874 | { |
| 875 | vassert(fregNo < 32); |
| 876 | return IRExpr_Get( floatGuestRegOffset(fregNo), Ity_F32 ); |
| 877 | } |
| 878 | |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 879 | static IRExpr* llGetFReg_up_to_64 ( UInt fregNo ) |
| 880 | { |
| 881 | vassert(fregNo < 64); |
| 882 | return IRExpr_Get( floatGuestRegOffset(fregNo), Ity_F32 ); |
| 883 | } |
| 884 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 885 | /* Architected read from a VFP Freg. */ |
| 886 | static IRExpr* getFReg ( UInt fregNo ) { |
| 887 | return llGetFReg( fregNo ); |
| 888 | } |
| 889 | |
| 890 | /* Plain ("low level") write to a VFP Freg. */ |
| 891 | static void llPutFReg ( UInt fregNo, IRExpr* e ) |
| 892 | { |
| 893 | vassert(fregNo < 32); |
| 894 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_F32); |
| 895 | stmt( IRStmt_Put(floatGuestRegOffset(fregNo), e) ); |
| 896 | } |
| 897 | |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 898 | static void llPutFReg_up_to_64 ( UInt fregNo, IRExpr* e ) |
| 899 | { |
| 900 | vassert(fregNo < 64); |
| 901 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_F32); |
| 902 | stmt( IRStmt_Put(floatGuestRegOffset(fregNo), e) ); |
| 903 | } |
| 904 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 905 | /* Architected write to a VFP Freg. Handles conditional writes to the |
| 906 | register: if guardT == IRTemp_INVALID then the write is |
| 907 | unconditional. */ |
| 908 | static void putFReg ( UInt fregNo, |
| 909 | IRExpr* e, |
| 910 | IRTemp guardT /* :: Ity_I32, 0 or 1 */) |
| 911 | { |
| 912 | /* So, generate either an unconditional or a conditional write to |
| 913 | the reg. */ |
| 914 | if (guardT == IRTemp_INVALID) { |
| 915 | /* unconditional write */ |
| 916 | llPutFReg( fregNo, e ); |
| 917 | } else { |
| 918 | llPutFReg( fregNo, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 919 | IRExpr_ITE( binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0)), |
| 920 | e, llGetFReg(fregNo) )); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 921 | } |
| 922 | } |
| 923 | |
| 924 | |
| 925 | /* ---------------- Misc registers ---------------- */ |
| 926 | |
| 927 | static void putMiscReg32 ( UInt gsoffset, |
| 928 | IRExpr* e, /* :: Ity_I32 */ |
| 929 | IRTemp guardT /* :: Ity_I32, 0 or 1 */) |
| 930 | { |
| 931 | switch (gsoffset) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 932 | case OFFB_FPSCR: break; |
| 933 | case OFFB_QFLAG32: break; |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 934 | case OFFB_GEFLAG0: break; |
| 935 | case OFFB_GEFLAG1: break; |
| 936 | case OFFB_GEFLAG2: break; |
| 937 | case OFFB_GEFLAG3: break; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 938 | default: vassert(0); /* awaiting more cases */ |
| 939 | } |
| 940 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I32); |
| 941 | |
| 942 | if (guardT == IRTemp_INVALID) { |
| 943 | /* unconditional write */ |
| 944 | stmt(IRStmt_Put(gsoffset, e)); |
| 945 | } else { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 946 | stmt(IRStmt_Put( |
| 947 | gsoffset, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 948 | IRExpr_ITE( binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0)), |
| 949 | e, IRExpr_Get(gsoffset, Ity_I32) ) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 950 | )); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 951 | } |
| 952 | } |
| 953 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 954 | static IRTemp get_ITSTATE ( void ) |
| 955 | { |
| 956 | ASSERT_IS_THUMB; |
| 957 | IRTemp t = newTemp(Ity_I32); |
| 958 | assign(t, IRExpr_Get( OFFB_ITSTATE, Ity_I32)); |
| 959 | return t; |
| 960 | } |
| 961 | |
| 962 | static void put_ITSTATE ( IRTemp t ) |
| 963 | { |
| 964 | ASSERT_IS_THUMB; |
| 965 | stmt( IRStmt_Put( OFFB_ITSTATE, mkexpr(t)) ); |
| 966 | } |
| 967 | |
| 968 | static IRTemp get_QFLAG32 ( void ) |
| 969 | { |
| 970 | IRTemp t = newTemp(Ity_I32); |
| 971 | assign(t, IRExpr_Get( OFFB_QFLAG32, Ity_I32)); |
| 972 | return t; |
| 973 | } |
| 974 | |
| 975 | static void put_QFLAG32 ( IRTemp t, IRTemp condT ) |
| 976 | { |
| 977 | putMiscReg32( OFFB_QFLAG32, mkexpr(t), condT ); |
| 978 | } |
| 979 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 980 | /* Stickily set the 'Q' flag (APSR bit 27) of the APSR (Application Program |
| 981 | Status Register) to indicate that overflow or saturation occurred. |
| 982 | Nb: t must be zero to denote no saturation, and any nonzero |
| 983 | value to indicate saturation. */ |
| 984 | static void or_into_QFLAG32 ( IRExpr* e, IRTemp condT ) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 985 | { |
| 986 | IRTemp old = get_QFLAG32(); |
| 987 | IRTemp nyu = newTemp(Ity_I32); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 988 | assign(nyu, binop(Iop_Or32, mkexpr(old), e) ); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 989 | put_QFLAG32(nyu, condT); |
| 990 | } |
| 991 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 992 | /* Generate code to set APSR.GE[flagNo]. Each fn call sets 1 bit. |
| 993 | flagNo: which flag bit to set [3...0] |
| 994 | lowbits_to_ignore: 0 = look at all 32 bits |
| 995 | 8 = look at top 24 bits only |
| 996 | 16 = look at top 16 bits only |
| 997 | 31 = look at the top bit only |
| 998 | e: input value to be evaluated. |
| 999 | The new value is taken from 'e' with the lowest 'lowbits_to_ignore' |
| 1000 | masked out. If the resulting value is zero then the GE flag is |
| 1001 | set to 0; any other value sets the flag to 1. */ |
| 1002 | static void put_GEFLAG32 ( Int flagNo, /* 0, 1, 2 or 3 */ |
| 1003 | Int lowbits_to_ignore, /* 0, 8, 16 or 31 */ |
| 1004 | IRExpr* e, /* Ity_I32 */ |
| 1005 | IRTemp condT ) |
| 1006 | { |
| 1007 | vassert( flagNo >= 0 && flagNo <= 3 ); |
| 1008 | vassert( lowbits_to_ignore == 0 || |
| 1009 | lowbits_to_ignore == 8 || |
| 1010 | lowbits_to_ignore == 16 || |
| 1011 | lowbits_to_ignore == 31 ); |
| 1012 | IRTemp masked = newTemp(Ity_I32); |
| 1013 | assign(masked, binop(Iop_Shr32, e, mkU8(lowbits_to_ignore))); |
florian | 24e0c13 | 2015-06-05 19:48:24 +0000 | [diff] [blame] | 1014 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 1015 | switch (flagNo) { |
| 1016 | case 0: putMiscReg32(OFFB_GEFLAG0, mkexpr(masked), condT); break; |
| 1017 | case 1: putMiscReg32(OFFB_GEFLAG1, mkexpr(masked), condT); break; |
| 1018 | case 2: putMiscReg32(OFFB_GEFLAG2, mkexpr(masked), condT); break; |
| 1019 | case 3: putMiscReg32(OFFB_GEFLAG3, mkexpr(masked), condT); break; |
| 1020 | default: vassert(0); |
| 1021 | } |
| 1022 | } |
| 1023 | |
| 1024 | /* Return the (32-bit, zero-or-nonzero representation scheme) of |
| 1025 | the specified GE flag. */ |
| 1026 | static IRExpr* get_GEFLAG32( Int flagNo /* 0, 1, 2, 3 */ ) |
| 1027 | { |
| 1028 | switch (flagNo) { |
| 1029 | case 0: return IRExpr_Get( OFFB_GEFLAG0, Ity_I32 ); |
| 1030 | case 1: return IRExpr_Get( OFFB_GEFLAG1, Ity_I32 ); |
| 1031 | case 2: return IRExpr_Get( OFFB_GEFLAG2, Ity_I32 ); |
| 1032 | case 3: return IRExpr_Get( OFFB_GEFLAG3, Ity_I32 ); |
| 1033 | default: vassert(0); |
| 1034 | } |
| 1035 | } |
| 1036 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 1037 | /* Set all 4 GE flags from the given 32-bit value as follows: GE 3 and |
| 1038 | 2 are set from bit 31 of the value, and GE 1 and 0 are set from bit |
| 1039 | 15 of the value. All other bits are ignored. */ |
| 1040 | static void set_GE_32_10_from_bits_31_15 ( IRTemp t32, IRTemp condT ) |
| 1041 | { |
| 1042 | IRTemp ge10 = newTemp(Ity_I32); |
| 1043 | IRTemp ge32 = newTemp(Ity_I32); |
| 1044 | assign(ge10, binop(Iop_And32, mkexpr(t32), mkU32(0x00008000))); |
| 1045 | assign(ge32, binop(Iop_And32, mkexpr(t32), mkU32(0x80000000))); |
| 1046 | put_GEFLAG32( 0, 0, mkexpr(ge10), condT ); |
| 1047 | put_GEFLAG32( 1, 0, mkexpr(ge10), condT ); |
| 1048 | put_GEFLAG32( 2, 0, mkexpr(ge32), condT ); |
| 1049 | put_GEFLAG32( 3, 0, mkexpr(ge32), condT ); |
| 1050 | } |
| 1051 | |
| 1052 | |
| 1053 | /* Set all 4 GE flags from the given 32-bit value as follows: GE 3 |
| 1054 | from bit 31, GE 2 from bit 23, GE 1 from bit 15, and GE0 from |
| 1055 | bit 7. All other bits are ignored. */ |
| 1056 | static void set_GE_3_2_1_0_from_bits_31_23_15_7 ( IRTemp t32, IRTemp condT ) |
| 1057 | { |
| 1058 | IRTemp ge0 = newTemp(Ity_I32); |
| 1059 | IRTemp ge1 = newTemp(Ity_I32); |
| 1060 | IRTemp ge2 = newTemp(Ity_I32); |
| 1061 | IRTemp ge3 = newTemp(Ity_I32); |
| 1062 | assign(ge0, binop(Iop_And32, mkexpr(t32), mkU32(0x00000080))); |
| 1063 | assign(ge1, binop(Iop_And32, mkexpr(t32), mkU32(0x00008000))); |
| 1064 | assign(ge2, binop(Iop_And32, mkexpr(t32), mkU32(0x00800000))); |
| 1065 | assign(ge3, binop(Iop_And32, mkexpr(t32), mkU32(0x80000000))); |
| 1066 | put_GEFLAG32( 0, 0, mkexpr(ge0), condT ); |
| 1067 | put_GEFLAG32( 1, 0, mkexpr(ge1), condT ); |
| 1068 | put_GEFLAG32( 2, 0, mkexpr(ge2), condT ); |
| 1069 | put_GEFLAG32( 3, 0, mkexpr(ge3), condT ); |
| 1070 | } |
| 1071 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1072 | |
| 1073 | /* ---------------- FPSCR stuff ---------------- */ |
| 1074 | |
| 1075 | /* Generate IR to get hold of the rounding mode bits in FPSCR, and |
| 1076 | convert them to IR format. Bind the final result to the |
| 1077 | returned temp. */ |
| 1078 | static IRTemp /* :: Ity_I32 */ mk_get_IR_rounding_mode ( void ) |
| 1079 | { |
| 1080 | /* The ARMvfp encoding for rounding mode bits is: |
| 1081 | 00 to nearest |
| 1082 | 01 to +infinity |
| 1083 | 10 to -infinity |
| 1084 | 11 to zero |
| 1085 | We need to convert that to the IR encoding: |
| 1086 | 00 to nearest (the default) |
| 1087 | 10 to +infinity |
| 1088 | 01 to -infinity |
| 1089 | 11 to zero |
| 1090 | Which can be done by swapping bits 0 and 1. |
| 1091 | The rmode bits are at 23:22 in FPSCR. |
| 1092 | */ |
| 1093 | IRTemp armEncd = newTemp(Ity_I32); |
| 1094 | IRTemp swapped = newTemp(Ity_I32); |
| 1095 | /* Fish FPSCR[23:22] out, and slide to bottom. Doesn't matter that |
| 1096 | we don't zero out bits 24 and above, since the assignment to |
| 1097 | 'swapped' will mask them out anyway. */ |
| 1098 | assign(armEncd, |
| 1099 | binop(Iop_Shr32, IRExpr_Get(OFFB_FPSCR, Ity_I32), mkU8(22))); |
| 1100 | /* Now swap them. */ |
| 1101 | assign(swapped, |
| 1102 | binop(Iop_Or32, |
| 1103 | binop(Iop_And32, |
| 1104 | binop(Iop_Shl32, mkexpr(armEncd), mkU8(1)), |
| 1105 | mkU32(2)), |
| 1106 | binop(Iop_And32, |
| 1107 | binop(Iop_Shr32, mkexpr(armEncd), mkU8(1)), |
| 1108 | mkU32(1)) |
| 1109 | )); |
| 1110 | return swapped; |
| 1111 | } |
| 1112 | |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1113 | |
| 1114 | /*------------------------------------------------------------*/ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1115 | /*--- Helpers for flag handling and conditional insns ---*/ |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1116 | /*------------------------------------------------------------*/ |
| 1117 | |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 1118 | static const HChar* name_ARMCondcode ( ARMCondcode cond ) |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1119 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1120 | switch (cond) { |
| 1121 | case ARMCondEQ: return "{eq}"; |
| 1122 | case ARMCondNE: return "{ne}"; |
| 1123 | case ARMCondHS: return "{hs}"; // or 'cs' |
| 1124 | case ARMCondLO: return "{lo}"; // or 'cc' |
| 1125 | case ARMCondMI: return "{mi}"; |
| 1126 | case ARMCondPL: return "{pl}"; |
| 1127 | case ARMCondVS: return "{vs}"; |
| 1128 | case ARMCondVC: return "{vc}"; |
| 1129 | case ARMCondHI: return "{hi}"; |
| 1130 | case ARMCondLS: return "{ls}"; |
| 1131 | case ARMCondGE: return "{ge}"; |
| 1132 | case ARMCondLT: return "{lt}"; |
| 1133 | case ARMCondGT: return "{gt}"; |
| 1134 | case ARMCondLE: return "{le}"; |
| 1135 | case ARMCondAL: return ""; // {al}: is the default |
| 1136 | case ARMCondNV: return "{nv}"; |
| 1137 | default: vpanic("name_ARMCondcode"); |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1138 | } |
| 1139 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1140 | /* and a handy shorthand for it */ |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 1141 | static const HChar* nCC ( ARMCondcode cond ) { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1142 | return name_ARMCondcode(cond); |
cerion | f7da63d | 2004-12-09 19:04:57 +0000 | [diff] [blame] | 1143 | } |
| 1144 | |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 1145 | |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1146 | /* Build IR to calculate some particular condition from stored |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1147 | CC_OP/CC_DEP1/CC_DEP2/CC_NDEP. Returns an expression of type |
| 1148 | Ity_I32, suitable for narrowing. Although the return type is |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1149 | Ity_I32, the returned value is either 0 or 1. 'cond' must be |
| 1150 | :: Ity_I32 and must denote the condition to compute in |
| 1151 | bits 7:4, and be zero everywhere else. |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 1152 | */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1153 | static IRExpr* mk_armg_calculate_condition_dyn ( IRExpr* cond ) |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1154 | { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1155 | vassert(typeOfIRExpr(irsb->tyenv, cond) == Ity_I32); |
sewardj | bb8b394 | 2011-05-01 18:47:10 +0000 | [diff] [blame] | 1156 | /* And 'cond' had better produce a value in which only bits 7:4 are |
| 1157 | nonzero. However, obviously we can't assert for that. */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1158 | |
| 1159 | /* So what we're constructing for the first argument is |
sewardj | bb8b394 | 2011-05-01 18:47:10 +0000 | [diff] [blame] | 1160 | "(cond << 4) | stored-operation". |
| 1161 | However, as per comments above, 'cond' must be supplied |
| 1162 | pre-shifted to this function. |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1163 | |
| 1164 | This pairing scheme requires that the ARM_CC_OP_ values all fit |
| 1165 | in 4 bits. Hence we are passing a (COND, OP) pair in the lowest |
| 1166 | 8 bits of the first argument. */ |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1167 | IRExpr** args |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1168 | = mkIRExprVec_4( |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1169 | binop(Iop_Or32, IRExpr_Get(OFFB_CC_OP, Ity_I32), cond), |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1170 | IRExpr_Get(OFFB_CC_DEP1, Ity_I32), |
| 1171 | IRExpr_Get(OFFB_CC_DEP2, Ity_I32), |
| 1172 | IRExpr_Get(OFFB_CC_NDEP, Ity_I32) |
| 1173 | ); |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1174 | IRExpr* call |
| 1175 | = mkIRExprCCall( |
| 1176 | Ity_I32, |
| 1177 | 0/*regparm*/, |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 1178 | "armg_calculate_condition", &armg_calculate_condition, |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1179 | args |
| 1180 | ); |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 1181 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1182 | /* Exclude the requested condition, OP and NDEP from definedness |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1183 | checking. We're only interested in DEP1 and DEP2. */ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1184 | call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3); |
| 1185 | return call; |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1186 | } |
| 1187 | |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 1188 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1189 | /* Build IR to calculate some particular condition from stored |
| 1190 | CC_OP/CC_DEP1/CC_DEP2/CC_NDEP. Returns an expression of type |
| 1191 | Ity_I32, suitable for narrowing. Although the return type is |
| 1192 | Ity_I32, the returned value is either 0 or 1. |
| 1193 | */ |
| 1194 | static IRExpr* mk_armg_calculate_condition ( ARMCondcode cond ) |
| 1195 | { |
| 1196 | /* First arg is "(cond << 4) | condition". This requires that the |
| 1197 | ARM_CC_OP_ values all fit in 4 bits. Hence we are passing a |
| 1198 | (COND, OP) pair in the lowest 8 bits of the first argument. */ |
| 1199 | vassert(cond >= 0 && cond <= 15); |
| 1200 | return mk_armg_calculate_condition_dyn( mkU32(cond << 4) ); |
| 1201 | } |
| 1202 | |
| 1203 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1204 | /* Build IR to calculate just the carry flag from stored |
| 1205 | CC_OP/CC_DEP1/CC_DEP2/CC_NDEP. Returns an expression :: |
| 1206 | Ity_I32. */ |
| 1207 | static IRExpr* mk_armg_calculate_flag_c ( void ) |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1208 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1209 | IRExpr** args |
| 1210 | = mkIRExprVec_4( IRExpr_Get(OFFB_CC_OP, Ity_I32), |
| 1211 | IRExpr_Get(OFFB_CC_DEP1, Ity_I32), |
| 1212 | IRExpr_Get(OFFB_CC_DEP2, Ity_I32), |
| 1213 | IRExpr_Get(OFFB_CC_NDEP, Ity_I32) ); |
| 1214 | IRExpr* call |
| 1215 | = mkIRExprCCall( |
| 1216 | Ity_I32, |
| 1217 | 0/*regparm*/, |
| 1218 | "armg_calculate_flag_c", &armg_calculate_flag_c, |
| 1219 | args |
| 1220 | ); |
| 1221 | /* Exclude OP and NDEP from definedness checking. We're only |
| 1222 | interested in DEP1 and DEP2. */ |
| 1223 | call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3); |
| 1224 | return call; |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1225 | } |
| 1226 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1227 | |
| 1228 | /* Build IR to calculate just the overflow flag from stored |
| 1229 | CC_OP/CC_DEP1/CC_DEP2/CC_NDEP. Returns an expression :: |
| 1230 | Ity_I32. */ |
| 1231 | static IRExpr* mk_armg_calculate_flag_v ( void ) |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1232 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1233 | IRExpr** args |
| 1234 | = mkIRExprVec_4( IRExpr_Get(OFFB_CC_OP, Ity_I32), |
| 1235 | IRExpr_Get(OFFB_CC_DEP1, Ity_I32), |
| 1236 | IRExpr_Get(OFFB_CC_DEP2, Ity_I32), |
| 1237 | IRExpr_Get(OFFB_CC_NDEP, Ity_I32) ); |
| 1238 | IRExpr* call |
| 1239 | = mkIRExprCCall( |
| 1240 | Ity_I32, |
| 1241 | 0/*regparm*/, |
| 1242 | "armg_calculate_flag_v", &armg_calculate_flag_v, |
| 1243 | args |
| 1244 | ); |
| 1245 | /* Exclude OP and NDEP from definedness checking. We're only |
| 1246 | interested in DEP1 and DEP2. */ |
| 1247 | call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3); |
| 1248 | return call; |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1249 | } |
| 1250 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1251 | |
| 1252 | /* Build IR to calculate N Z C V in bits 31:28 of the |
| 1253 | returned word. */ |
| 1254 | static IRExpr* mk_armg_calculate_flags_nzcv ( void ) |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1255 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1256 | IRExpr** args |
| 1257 | = mkIRExprVec_4( IRExpr_Get(OFFB_CC_OP, Ity_I32), |
| 1258 | IRExpr_Get(OFFB_CC_DEP1, Ity_I32), |
| 1259 | IRExpr_Get(OFFB_CC_DEP2, Ity_I32), |
| 1260 | IRExpr_Get(OFFB_CC_NDEP, Ity_I32) ); |
| 1261 | IRExpr* call |
| 1262 | = mkIRExprCCall( |
| 1263 | Ity_I32, |
| 1264 | 0/*regparm*/, |
| 1265 | "armg_calculate_flags_nzcv", &armg_calculate_flags_nzcv, |
| 1266 | args |
| 1267 | ); |
| 1268 | /* Exclude OP and NDEP from definedness checking. We're only |
| 1269 | interested in DEP1 and DEP2. */ |
| 1270 | call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3); |
| 1271 | return call; |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1272 | } |
| 1273 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1274 | static IRExpr* mk_armg_calculate_flag_qc ( IRExpr* resL, IRExpr* resR, Bool Q ) |
| 1275 | { |
| 1276 | IRExpr** args1; |
| 1277 | IRExpr** args2; |
| 1278 | IRExpr *call1, *call2, *res; |
| 1279 | |
| 1280 | if (Q) { |
| 1281 | args1 = mkIRExprVec_4 ( binop(Iop_GetElem32x4, resL, mkU8(0)), |
| 1282 | binop(Iop_GetElem32x4, resL, mkU8(1)), |
| 1283 | binop(Iop_GetElem32x4, resR, mkU8(0)), |
| 1284 | binop(Iop_GetElem32x4, resR, mkU8(1)) ); |
| 1285 | args2 = mkIRExprVec_4 ( binop(Iop_GetElem32x4, resL, mkU8(2)), |
| 1286 | binop(Iop_GetElem32x4, resL, mkU8(3)), |
| 1287 | binop(Iop_GetElem32x4, resR, mkU8(2)), |
| 1288 | binop(Iop_GetElem32x4, resR, mkU8(3)) ); |
| 1289 | } else { |
| 1290 | args1 = mkIRExprVec_4 ( binop(Iop_GetElem32x2, resL, mkU8(0)), |
| 1291 | binop(Iop_GetElem32x2, resL, mkU8(1)), |
| 1292 | binop(Iop_GetElem32x2, resR, mkU8(0)), |
| 1293 | binop(Iop_GetElem32x2, resR, mkU8(1)) ); |
| 1294 | } |
| 1295 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1296 | call1 = mkIRExprCCall( |
| 1297 | Ity_I32, |
| 1298 | 0/*regparm*/, |
| 1299 | "armg_calculate_flag_qc", &armg_calculate_flag_qc, |
| 1300 | args1 |
| 1301 | ); |
| 1302 | if (Q) { |
| 1303 | call2 = mkIRExprCCall( |
| 1304 | Ity_I32, |
| 1305 | 0/*regparm*/, |
| 1306 | "armg_calculate_flag_qc", &armg_calculate_flag_qc, |
| 1307 | args2 |
| 1308 | ); |
| 1309 | } |
| 1310 | if (Q) { |
| 1311 | res = binop(Iop_Or32, call1, call2); |
| 1312 | } else { |
| 1313 | res = call1; |
| 1314 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1315 | return res; |
| 1316 | } |
| 1317 | |
sewardj | 9dbbd7b | 2010-08-22 18:24:51 +0000 | [diff] [blame] | 1318 | // FIXME: this is named wrongly .. looks like a sticky set of |
| 1319 | // QC, not a write to it. |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1320 | static void setFlag_QC ( IRExpr* resL, IRExpr* resR, Bool Q, |
| 1321 | IRTemp condT ) |
| 1322 | { |
| 1323 | putMiscReg32 (OFFB_FPSCR, |
| 1324 | binop(Iop_Or32, |
| 1325 | IRExpr_Get(OFFB_FPSCR, Ity_I32), |
| 1326 | binop(Iop_Shl32, |
| 1327 | mk_armg_calculate_flag_qc(resL, resR, Q), |
| 1328 | mkU8(27))), |
| 1329 | condT); |
| 1330 | } |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1331 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1332 | /* Build IR to conditionally set the flags thunk. As with putIReg, if |
| 1333 | guard is IRTemp_INVALID then it's unconditional, else it holds a |
| 1334 | condition :: Ity_I32. */ |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1335 | static |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1336 | void setFlags_D1_D2_ND ( UInt cc_op, IRTemp t_dep1, |
| 1337 | IRTemp t_dep2, IRTemp t_ndep, |
| 1338 | IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1339 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1340 | vassert(typeOfIRTemp(irsb->tyenv, t_dep1 == Ity_I32)); |
| 1341 | vassert(typeOfIRTemp(irsb->tyenv, t_dep2 == Ity_I32)); |
| 1342 | vassert(typeOfIRTemp(irsb->tyenv, t_ndep == Ity_I32)); |
| 1343 | vassert(cc_op >= ARMG_CC_OP_COPY && cc_op < ARMG_CC_OP_NUMBER); |
| 1344 | if (guardT == IRTemp_INVALID) { |
| 1345 | /* unconditional */ |
| 1346 | stmt( IRStmt_Put( OFFB_CC_OP, mkU32(cc_op) )); |
| 1347 | stmt( IRStmt_Put( OFFB_CC_DEP1, mkexpr(t_dep1) )); |
| 1348 | stmt( IRStmt_Put( OFFB_CC_DEP2, mkexpr(t_dep2) )); |
| 1349 | stmt( IRStmt_Put( OFFB_CC_NDEP, mkexpr(t_ndep) )); |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1350 | } else { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1351 | /* conditional */ |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 1352 | IRTemp c1 = newTemp(Ity_I1); |
| 1353 | assign( c1, binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0)) ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1354 | stmt( IRStmt_Put( |
| 1355 | OFFB_CC_OP, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 1356 | IRExpr_ITE( mkexpr(c1), |
| 1357 | mkU32(cc_op), |
| 1358 | IRExpr_Get(OFFB_CC_OP, Ity_I32) ) )); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1359 | stmt( IRStmt_Put( |
| 1360 | OFFB_CC_DEP1, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 1361 | IRExpr_ITE( mkexpr(c1), |
| 1362 | mkexpr(t_dep1), |
| 1363 | IRExpr_Get(OFFB_CC_DEP1, Ity_I32) ) )); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1364 | stmt( IRStmt_Put( |
| 1365 | OFFB_CC_DEP2, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 1366 | IRExpr_ITE( mkexpr(c1), |
| 1367 | mkexpr(t_dep2), |
| 1368 | IRExpr_Get(OFFB_CC_DEP2, Ity_I32) ) )); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1369 | stmt( IRStmt_Put( |
| 1370 | OFFB_CC_NDEP, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 1371 | IRExpr_ITE( mkexpr(c1), |
| 1372 | mkexpr(t_ndep), |
| 1373 | IRExpr_Get(OFFB_CC_NDEP, Ity_I32) ) )); |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 1374 | } |
| 1375 | } |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 1376 | |
| 1377 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1378 | /* Minor variant of the above that sets NDEP to zero (if it |
| 1379 | sets it at all) */ |
| 1380 | static void setFlags_D1_D2 ( UInt cc_op, IRTemp t_dep1, |
| 1381 | IRTemp t_dep2, |
| 1382 | IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
cerion | 397f9e5 | 2004-12-15 13:04:06 +0000 | [diff] [blame] | 1383 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1384 | IRTemp z32 = newTemp(Ity_I32); |
| 1385 | assign( z32, mkU32(0) ); |
| 1386 | setFlags_D1_D2_ND( cc_op, t_dep1, t_dep2, z32, guardT ); |
cerion | 397f9e5 | 2004-12-15 13:04:06 +0000 | [diff] [blame] | 1387 | } |
| 1388 | |
| 1389 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1390 | /* Minor variant of the above that sets DEP2 to zero (if it |
| 1391 | sets it at all) */ |
| 1392 | static void setFlags_D1_ND ( UInt cc_op, IRTemp t_dep1, |
| 1393 | IRTemp t_ndep, |
| 1394 | IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
cerion | 397f9e5 | 2004-12-15 13:04:06 +0000 | [diff] [blame] | 1395 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1396 | IRTemp z32 = newTemp(Ity_I32); |
| 1397 | assign( z32, mkU32(0) ); |
| 1398 | setFlags_D1_D2_ND( cc_op, t_dep1, z32, t_ndep, guardT ); |
cerion | 397f9e5 | 2004-12-15 13:04:06 +0000 | [diff] [blame] | 1399 | } |
| 1400 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1401 | |
| 1402 | /* Minor variant of the above that sets DEP2 and NDEP to zero (if it |
| 1403 | sets them at all) */ |
| 1404 | static void setFlags_D1 ( UInt cc_op, IRTemp t_dep1, |
| 1405 | IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
cerion | 397f9e5 | 2004-12-15 13:04:06 +0000 | [diff] [blame] | 1406 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1407 | IRTemp z32 = newTemp(Ity_I32); |
| 1408 | assign( z32, mkU32(0) ); |
| 1409 | setFlags_D1_D2_ND( cc_op, t_dep1, z32, z32, guardT ); |
cerion | 397f9e5 | 2004-12-15 13:04:06 +0000 | [diff] [blame] | 1410 | } |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 1411 | |
| 1412 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1413 | /* ARM only */ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1414 | /* Generate a side-exit to the next instruction, if the given guard |
| 1415 | expression :: Ity_I32 is 0 (note! the side exit is taken if the |
| 1416 | condition is false!) This is used to skip over conditional |
| 1417 | instructions which we can't generate straight-line code for, either |
| 1418 | because they are too complex or (more likely) they potentially |
| 1419 | generate exceptions. |
cerion | f7da63d | 2004-12-09 19:04:57 +0000 | [diff] [blame] | 1420 | */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1421 | static void mk_skip_over_A32_if_cond_is_false ( |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1422 | IRTemp guardT /* :: Ity_I32, 0 or 1 */ |
| 1423 | ) |
cerion | f7da63d | 2004-12-09 19:04:57 +0000 | [diff] [blame] | 1424 | { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1425 | ASSERT_IS_ARM; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1426 | vassert(guardT != IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1427 | vassert(0 == (guest_R15_curr_instr_notENC & 3)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1428 | stmt( IRStmt_Exit( |
| 1429 | unop(Iop_Not1, unop(Iop_32to1, mkexpr(guardT))), |
| 1430 | Ijk_Boring, |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 1431 | IRConst_U32(toUInt(guest_R15_curr_instr_notENC + 4)), |
| 1432 | OFFB_R15T |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1433 | )); |
| 1434 | } |
| 1435 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1436 | /* Thumb16 only */ |
| 1437 | /* ditto, but jump over a 16-bit thumb insn */ |
| 1438 | static void mk_skip_over_T16_if_cond_is_false ( |
| 1439 | IRTemp guardT /* :: Ity_I32, 0 or 1 */ |
| 1440 | ) |
| 1441 | { |
| 1442 | ASSERT_IS_THUMB; |
| 1443 | vassert(guardT != IRTemp_INVALID); |
| 1444 | vassert(0 == (guest_R15_curr_instr_notENC & 1)); |
| 1445 | stmt( IRStmt_Exit( |
| 1446 | unop(Iop_Not1, unop(Iop_32to1, mkexpr(guardT))), |
| 1447 | Ijk_Boring, |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 1448 | IRConst_U32(toUInt((guest_R15_curr_instr_notENC + 2) | 1)), |
| 1449 | OFFB_R15T |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1450 | )); |
| 1451 | } |
| 1452 | |
| 1453 | |
| 1454 | /* Thumb32 only */ |
| 1455 | /* ditto, but jump over a 32-bit thumb insn */ |
| 1456 | static void mk_skip_over_T32_if_cond_is_false ( |
| 1457 | IRTemp guardT /* :: Ity_I32, 0 or 1 */ |
| 1458 | ) |
| 1459 | { |
| 1460 | ASSERT_IS_THUMB; |
| 1461 | vassert(guardT != IRTemp_INVALID); |
| 1462 | vassert(0 == (guest_R15_curr_instr_notENC & 1)); |
| 1463 | stmt( IRStmt_Exit( |
| 1464 | unop(Iop_Not1, unop(Iop_32to1, mkexpr(guardT))), |
| 1465 | Ijk_Boring, |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 1466 | IRConst_U32(toUInt((guest_R15_curr_instr_notENC + 4) | 1)), |
| 1467 | OFFB_R15T |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1468 | )); |
| 1469 | } |
| 1470 | |
| 1471 | |
| 1472 | /* Thumb16 and Thumb32 only |
| 1473 | Generate a SIGILL followed by a restart of the current instruction |
| 1474 | if the given temp is nonzero. */ |
| 1475 | static void gen_SIGILL_T_if_nonzero ( IRTemp t /* :: Ity_I32 */ ) |
| 1476 | { |
| 1477 | ASSERT_IS_THUMB; |
| 1478 | vassert(t != IRTemp_INVALID); |
| 1479 | vassert(0 == (guest_R15_curr_instr_notENC & 1)); |
| 1480 | stmt( |
| 1481 | IRStmt_Exit( |
| 1482 | binop(Iop_CmpNE32, mkexpr(t), mkU32(0)), |
| 1483 | Ijk_NoDecode, |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 1484 | IRConst_U32(toUInt(guest_R15_curr_instr_notENC | 1)), |
| 1485 | OFFB_R15T |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1486 | ) |
| 1487 | ); |
| 1488 | } |
| 1489 | |
| 1490 | |
| 1491 | /* Inspect the old_itstate, and generate a SIGILL if it indicates that |
| 1492 | we are currently in an IT block and are not the last in the block. |
| 1493 | This also rolls back guest_ITSTATE to its old value before the exit |
| 1494 | and restores it to its new value afterwards. This is so that if |
| 1495 | the exit is taken, we have an up to date version of ITSTATE |
| 1496 | available. Without doing that, we have no hope of making precise |
| 1497 | exceptions work. */ |
| 1498 | static void gen_SIGILL_T_if_in_but_NLI_ITBlock ( |
| 1499 | IRTemp old_itstate /* :: Ity_I32 */, |
| 1500 | IRTemp new_itstate /* :: Ity_I32 */ |
| 1501 | ) |
| 1502 | { |
| 1503 | ASSERT_IS_THUMB; |
| 1504 | put_ITSTATE(old_itstate); // backout |
| 1505 | IRTemp guards_for_next3 = newTemp(Ity_I32); |
| 1506 | assign(guards_for_next3, |
| 1507 | binop(Iop_Shr32, mkexpr(old_itstate), mkU8(8))); |
| 1508 | gen_SIGILL_T_if_nonzero(guards_for_next3); |
| 1509 | put_ITSTATE(new_itstate); //restore |
| 1510 | } |
| 1511 | |
| 1512 | |
| 1513 | /* Simpler version of the above, which generates a SIGILL if |
| 1514 | we're anywhere within an IT block. */ |
| 1515 | static void gen_SIGILL_T_if_in_ITBlock ( |
| 1516 | IRTemp old_itstate /* :: Ity_I32 */, |
| 1517 | IRTemp new_itstate /* :: Ity_I32 */ |
| 1518 | ) |
| 1519 | { |
| 1520 | put_ITSTATE(old_itstate); // backout |
| 1521 | gen_SIGILL_T_if_nonzero(old_itstate); |
| 1522 | put_ITSTATE(new_itstate); //restore |
| 1523 | } |
| 1524 | |
| 1525 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 1526 | /* Generate an APSR value, from the NZCV thunk, and |
| 1527 | from QFLAG32 and GEFLAG0 .. GEFLAG3. */ |
| 1528 | static IRTemp synthesise_APSR ( void ) |
| 1529 | { |
| 1530 | IRTemp res1 = newTemp(Ity_I32); |
| 1531 | // Get NZCV |
| 1532 | assign( res1, mk_armg_calculate_flags_nzcv() ); |
| 1533 | // OR in the Q value |
| 1534 | IRTemp res2 = newTemp(Ity_I32); |
| 1535 | assign( |
| 1536 | res2, |
| 1537 | binop(Iop_Or32, |
| 1538 | mkexpr(res1), |
| 1539 | binop(Iop_Shl32, |
| 1540 | unop(Iop_1Uto32, |
| 1541 | binop(Iop_CmpNE32, |
| 1542 | mkexpr(get_QFLAG32()), |
| 1543 | mkU32(0))), |
| 1544 | mkU8(ARMG_CC_SHIFT_Q))) |
| 1545 | ); |
| 1546 | // OR in GE0 .. GE3 |
| 1547 | IRExpr* ge0 |
| 1548 | = unop(Iop_1Uto32, binop(Iop_CmpNE32, get_GEFLAG32(0), mkU32(0))); |
| 1549 | IRExpr* ge1 |
| 1550 | = unop(Iop_1Uto32, binop(Iop_CmpNE32, get_GEFLAG32(1), mkU32(0))); |
| 1551 | IRExpr* ge2 |
| 1552 | = unop(Iop_1Uto32, binop(Iop_CmpNE32, get_GEFLAG32(2), mkU32(0))); |
| 1553 | IRExpr* ge3 |
| 1554 | = unop(Iop_1Uto32, binop(Iop_CmpNE32, get_GEFLAG32(3), mkU32(0))); |
| 1555 | IRTemp res3 = newTemp(Ity_I32); |
| 1556 | assign(res3, |
| 1557 | binop(Iop_Or32, |
| 1558 | mkexpr(res2), |
| 1559 | binop(Iop_Or32, |
| 1560 | binop(Iop_Or32, |
| 1561 | binop(Iop_Shl32, ge0, mkU8(16)), |
| 1562 | binop(Iop_Shl32, ge1, mkU8(17))), |
| 1563 | binop(Iop_Or32, |
| 1564 | binop(Iop_Shl32, ge2, mkU8(18)), |
| 1565 | binop(Iop_Shl32, ge3, mkU8(19))) ))); |
| 1566 | return res3; |
| 1567 | } |
| 1568 | |
| 1569 | |
| 1570 | /* and the inverse transformation: given an APSR value, |
| 1571 | set the NZCV thunk, the Q flag, and the GE flags. */ |
| 1572 | static void desynthesise_APSR ( Bool write_nzcvq, Bool write_ge, |
| 1573 | IRTemp apsrT, IRTemp condT ) |
| 1574 | { |
| 1575 | vassert(write_nzcvq || write_ge); |
| 1576 | if (write_nzcvq) { |
| 1577 | // Do NZCV |
| 1578 | IRTemp immT = newTemp(Ity_I32); |
| 1579 | assign(immT, binop(Iop_And32, mkexpr(apsrT), mkU32(0xF0000000)) ); |
| 1580 | setFlags_D1(ARMG_CC_OP_COPY, immT, condT); |
| 1581 | // Do Q |
| 1582 | IRTemp qnewT = newTemp(Ity_I32); |
| 1583 | assign(qnewT, binop(Iop_And32, mkexpr(apsrT), mkU32(ARMG_CC_MASK_Q))); |
| 1584 | put_QFLAG32(qnewT, condT); |
| 1585 | } |
| 1586 | if (write_ge) { |
| 1587 | // Do GE3..0 |
| 1588 | put_GEFLAG32(0, 0, binop(Iop_And32, mkexpr(apsrT), mkU32(1<<16)), |
| 1589 | condT); |
| 1590 | put_GEFLAG32(1, 0, binop(Iop_And32, mkexpr(apsrT), mkU32(1<<17)), |
| 1591 | condT); |
| 1592 | put_GEFLAG32(2, 0, binop(Iop_And32, mkexpr(apsrT), mkU32(1<<18)), |
| 1593 | condT); |
| 1594 | put_GEFLAG32(3, 0, binop(Iop_And32, mkexpr(apsrT), mkU32(1<<19)), |
| 1595 | condT); |
| 1596 | } |
| 1597 | } |
| 1598 | |
| 1599 | |
| 1600 | /*------------------------------------------------------------*/ |
| 1601 | /*--- Helpers for saturation ---*/ |
| 1602 | /*------------------------------------------------------------*/ |
| 1603 | |
| 1604 | /* FIXME: absolutely the only diff. between (a) armUnsignedSatQ and |
| 1605 | (b) armSignedSatQ is that in (a) the floor is set to 0, whereas in |
| 1606 | (b) the floor is computed from the value of imm5. these two fnsn |
| 1607 | should be commoned up. */ |
| 1608 | |
| 1609 | /* UnsignedSatQ(): 'clamp' each value so it lies between 0 <= x <= (2^N)-1 |
| 1610 | Optionally return flag resQ saying whether saturation occurred. |
| 1611 | See definition in manual, section A2.2.1, page 41 |
| 1612 | (bits(N), boolean) UnsignedSatQ( integer i, integer N ) |
| 1613 | { |
| 1614 | if ( i > (2^N)-1 ) { result = (2^N)-1; saturated = TRUE; } |
| 1615 | elsif ( i < 0 ) { result = 0; saturated = TRUE; } |
| 1616 | else { result = i; saturated = FALSE; } |
| 1617 | return ( result<N-1:0>, saturated ); |
| 1618 | } |
| 1619 | */ |
| 1620 | static void armUnsignedSatQ( IRTemp* res, /* OUT - Ity_I32 */ |
| 1621 | IRTemp* resQ, /* OUT - Ity_I32 */ |
| 1622 | IRTemp regT, /* value to clamp - Ity_I32 */ |
| 1623 | UInt imm5 ) /* saturation ceiling */ |
| 1624 | { |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 1625 | ULong ceil64 = (1ULL << imm5) - 1; // (2^imm5)-1 |
| 1626 | UInt ceil = (UInt)ceil64; |
| 1627 | UInt floor = 0; |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 1628 | |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 1629 | IRTemp nd0 = newTemp(Ity_I32); |
| 1630 | IRTemp nd1 = newTemp(Ity_I32); |
| 1631 | IRTemp nd2 = newTemp(Ity_I1); |
| 1632 | IRTemp nd3 = newTemp(Ity_I32); |
| 1633 | IRTemp nd4 = newTemp(Ity_I32); |
| 1634 | IRTemp nd5 = newTemp(Ity_I1); |
| 1635 | IRTemp nd6 = newTemp(Ity_I32); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 1636 | |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 1637 | assign( nd0, mkexpr(regT) ); |
| 1638 | assign( nd1, mkU32(ceil) ); |
| 1639 | assign( nd2, binop( Iop_CmpLT32S, mkexpr(nd1), mkexpr(nd0) ) ); |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 1640 | assign( nd3, IRExpr_ITE(mkexpr(nd2), mkexpr(nd1), mkexpr(nd0)) ); |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 1641 | assign( nd4, mkU32(floor) ); |
| 1642 | assign( nd5, binop( Iop_CmpLT32S, mkexpr(nd3), mkexpr(nd4) ) ); |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 1643 | assign( nd6, IRExpr_ITE(mkexpr(nd5), mkexpr(nd4), mkexpr(nd3)) ); |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 1644 | assign( *res, mkexpr(nd6) ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 1645 | |
| 1646 | /* if saturation occurred, then resQ is set to some nonzero value |
| 1647 | if sat did not occur, resQ is guaranteed to be zero. */ |
| 1648 | if (resQ) { |
| 1649 | assign( *resQ, binop(Iop_Xor32, mkexpr(*res), mkexpr(regT)) ); |
| 1650 | } |
| 1651 | } |
| 1652 | |
| 1653 | |
| 1654 | /* SignedSatQ(): 'clamp' each value so it lies between -2^N <= x <= (2^N) - 1 |
| 1655 | Optionally return flag resQ saying whether saturation occurred. |
| 1656 | - see definition in manual, section A2.2.1, page 41 |
| 1657 | (bits(N), boolean ) SignedSatQ( integer i, integer N ) |
| 1658 | { |
| 1659 | if ( i > 2^(N-1) - 1 ) { result = 2^(N-1) - 1; saturated = TRUE; } |
| 1660 | elsif ( i < -(2^(N-1)) ) { result = -(2^(N-1)); saturated = FALSE; } |
| 1661 | else { result = i; saturated = FALSE; } |
| 1662 | return ( result[N-1:0], saturated ); |
| 1663 | } |
| 1664 | */ |
| 1665 | static void armSignedSatQ( IRTemp regT, /* value to clamp - Ity_I32 */ |
| 1666 | UInt imm5, /* saturation ceiling */ |
| 1667 | IRTemp* res, /* OUT - Ity_I32 */ |
| 1668 | IRTemp* resQ ) /* OUT - Ity_I32 */ |
| 1669 | { |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 1670 | Long ceil64 = (1LL << (imm5-1)) - 1; // (2^(imm5-1))-1 |
| 1671 | Long floor64 = -(1LL << (imm5-1)); // -(2^(imm5-1)) |
| 1672 | Int ceil = (Int)ceil64; |
| 1673 | Int floor = (Int)floor64; |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 1674 | |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 1675 | IRTemp nd0 = newTemp(Ity_I32); |
| 1676 | IRTemp nd1 = newTemp(Ity_I32); |
| 1677 | IRTemp nd2 = newTemp(Ity_I1); |
| 1678 | IRTemp nd3 = newTemp(Ity_I32); |
| 1679 | IRTemp nd4 = newTemp(Ity_I32); |
| 1680 | IRTemp nd5 = newTemp(Ity_I1); |
| 1681 | IRTemp nd6 = newTemp(Ity_I32); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 1682 | |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 1683 | assign( nd0, mkexpr(regT) ); |
| 1684 | assign( nd1, mkU32(ceil) ); |
| 1685 | assign( nd2, binop( Iop_CmpLT32S, mkexpr(nd1), mkexpr(nd0) ) ); |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 1686 | assign( nd3, IRExpr_ITE( mkexpr(nd2), mkexpr(nd1), mkexpr(nd0) ) ); |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 1687 | assign( nd4, mkU32(floor) ); |
| 1688 | assign( nd5, binop( Iop_CmpLT32S, mkexpr(nd3), mkexpr(nd4) ) ); |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 1689 | assign( nd6, IRExpr_ITE( mkexpr(nd5), mkexpr(nd4), mkexpr(nd3) ) ); |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 1690 | assign( *res, mkexpr(nd6) ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 1691 | |
| 1692 | /* if saturation occurred, then resQ is set to some nonzero value |
| 1693 | if sat did not occur, resQ is guaranteed to be zero. */ |
| 1694 | if (resQ) { |
| 1695 | assign( *resQ, binop(Iop_Xor32, mkexpr(*res), mkexpr(regT)) ); |
| 1696 | } |
| 1697 | } |
| 1698 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1699 | |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 1700 | /* Compute a value 0 :: I32 or 1 :: I32, indicating whether signed |
| 1701 | overflow occurred for 32-bit addition. Needs both args and the |
| 1702 | result. HD p27. */ |
| 1703 | static |
| 1704 | IRExpr* signed_overflow_after_Add32 ( IRExpr* resE, |
| 1705 | IRTemp argL, IRTemp argR ) |
| 1706 | { |
| 1707 | IRTemp res = newTemp(Ity_I32); |
| 1708 | assign(res, resE); |
| 1709 | return |
| 1710 | binop( Iop_Shr32, |
| 1711 | binop( Iop_And32, |
| 1712 | binop( Iop_Xor32, mkexpr(res), mkexpr(argL) ), |
| 1713 | binop( Iop_Xor32, mkexpr(res), mkexpr(argR) )), |
| 1714 | mkU8(31) ); |
| 1715 | } |
| 1716 | |
sewardj | 44ce46d | 2012-07-11 13:19:10 +0000 | [diff] [blame] | 1717 | /* Similarly .. also from HD p27 .. */ |
| 1718 | static |
| 1719 | IRExpr* signed_overflow_after_Sub32 ( IRExpr* resE, |
| 1720 | IRTemp argL, IRTemp argR ) |
| 1721 | { |
| 1722 | IRTemp res = newTemp(Ity_I32); |
| 1723 | assign(res, resE); |
| 1724 | return |
| 1725 | binop( Iop_Shr32, |
| 1726 | binop( Iop_And32, |
| 1727 | binop( Iop_Xor32, mkexpr(argL), mkexpr(argR) ), |
| 1728 | binop( Iop_Xor32, mkexpr(res), mkexpr(argL) )), |
| 1729 | mkU8(31) ); |
| 1730 | } |
| 1731 | |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 1732 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 1733 | /*------------------------------------------------------------*/ |
| 1734 | /*--- Larger helpers ---*/ |
| 1735 | /*------------------------------------------------------------*/ |
| 1736 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1737 | /* Compute both the result and new C flag value for a LSL by an imm5 |
| 1738 | or by a register operand. May generate reads of the old C value |
| 1739 | (hence only safe to use before any writes to guest state happen). |
| 1740 | Are factored out so can be used by both ARM and Thumb. |
| 1741 | |
| 1742 | Note that in compute_result_and_C_after_{LSL,LSR,ASR}_by{imm5,reg}, |
| 1743 | "res" (the result) is a.k.a. "shop", shifter operand |
| 1744 | "newC" (the new C) is a.k.a. "shco", shifter carry out |
| 1745 | |
| 1746 | The calling convention for res and newC is a bit funny. They could |
| 1747 | be passed by value, but instead are passed by ref. |
sewardj | bb8b394 | 2011-05-01 18:47:10 +0000 | [diff] [blame] | 1748 | |
| 1749 | The C (shco) value computed must be zero in bits 31:1, as the IR |
| 1750 | optimisations for flag handling (guest_arm_spechelper) rely on |
| 1751 | that, and the slow-path handlers (armg_calculate_flags_nzcv) assert |
| 1752 | for it. Same applies to all these functions that compute shco |
| 1753 | after a shift or rotate, not just this one. |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1754 | */ |
| 1755 | |
| 1756 | static void compute_result_and_C_after_LSL_by_imm5 ( |
| 1757 | /*OUT*/HChar* buf, |
| 1758 | IRTemp* res, |
| 1759 | IRTemp* newC, |
| 1760 | IRTemp rMt, UInt shift_amt, /* operands */ |
| 1761 | UInt rM /* only for debug printing */ |
| 1762 | ) |
| 1763 | { |
| 1764 | if (shift_amt == 0) { |
| 1765 | if (newC) { |
| 1766 | assign( *newC, mk_armg_calculate_flag_c() ); |
| 1767 | } |
| 1768 | assign( *res, mkexpr(rMt) ); |
| 1769 | DIS(buf, "r%u", rM); |
| 1770 | } else { |
| 1771 | vassert(shift_amt >= 1 && shift_amt <= 31); |
| 1772 | if (newC) { |
| 1773 | assign( *newC, |
| 1774 | binop(Iop_And32, |
| 1775 | binop(Iop_Shr32, mkexpr(rMt), |
| 1776 | mkU8(32 - shift_amt)), |
| 1777 | mkU32(1))); |
| 1778 | } |
| 1779 | assign( *res, |
| 1780 | binop(Iop_Shl32, mkexpr(rMt), mkU8(shift_amt)) ); |
| 1781 | DIS(buf, "r%u, LSL #%u", rM, shift_amt); |
| 1782 | } |
| 1783 | } |
| 1784 | |
| 1785 | |
| 1786 | static void compute_result_and_C_after_LSL_by_reg ( |
| 1787 | /*OUT*/HChar* buf, |
| 1788 | IRTemp* res, |
| 1789 | IRTemp* newC, |
| 1790 | IRTemp rMt, IRTemp rSt, /* operands */ |
| 1791 | UInt rM, UInt rS /* only for debug printing */ |
| 1792 | ) |
| 1793 | { |
| 1794 | // shift left in range 0 .. 255 |
| 1795 | // amt = rS & 255 |
| 1796 | // res = amt < 32 ? Rm << amt : 0 |
| 1797 | // newC = amt == 0 ? oldC : |
| 1798 | // amt in 1..32 ? Rm[32-amt] : 0 |
| 1799 | IRTemp amtT = newTemp(Ity_I32); |
| 1800 | assign( amtT, binop(Iop_And32, mkexpr(rSt), mkU32(255)) ); |
| 1801 | if (newC) { |
| 1802 | /* mux0X(amt == 0, |
| 1803 | mux0X(amt < 32, |
| 1804 | 0, |
sewardj | bb8b394 | 2011-05-01 18:47:10 +0000 | [diff] [blame] | 1805 | Rm[(32-amt) & 31]), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1806 | oldC) |
| 1807 | */ |
| 1808 | /* About the best you can do is pray that iropt is able |
| 1809 | to nuke most or all of the following junk. */ |
| 1810 | IRTemp oldC = newTemp(Ity_I32); |
| 1811 | assign(oldC, mk_armg_calculate_flag_c() ); |
| 1812 | assign( |
| 1813 | *newC, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 1814 | IRExpr_ITE( |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 1815 | binop(Iop_CmpEQ32, mkexpr(amtT), mkU32(0)), |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 1816 | mkexpr(oldC), |
| 1817 | IRExpr_ITE( |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 1818 | binop(Iop_CmpLE32U, mkexpr(amtT), mkU32(32)), |
sewardj | bb8b394 | 2011-05-01 18:47:10 +0000 | [diff] [blame] | 1819 | binop(Iop_And32, |
| 1820 | binop(Iop_Shr32, |
| 1821 | mkexpr(rMt), |
| 1822 | unop(Iop_32to8, |
| 1823 | binop(Iop_And32, |
| 1824 | binop(Iop_Sub32, |
| 1825 | mkU32(32), |
| 1826 | mkexpr(amtT)), |
| 1827 | mkU32(31) |
| 1828 | ) |
| 1829 | ) |
| 1830 | ), |
| 1831 | mkU32(1) |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 1832 | ), |
| 1833 | mkU32(0) |
| 1834 | ) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1835 | ) |
| 1836 | ); |
| 1837 | } |
| 1838 | // (Rm << (Rs & 31)) & (((Rs & 255) - 32) >>s 31) |
| 1839 | // Lhs of the & limits the shift to 31 bits, so as to |
| 1840 | // give known IR semantics. Rhs of the & is all 1s for |
| 1841 | // Rs <= 31 and all 0s for Rs >= 32. |
| 1842 | assign( |
| 1843 | *res, |
| 1844 | binop( |
| 1845 | Iop_And32, |
| 1846 | binop(Iop_Shl32, |
| 1847 | mkexpr(rMt), |
| 1848 | unop(Iop_32to8, |
| 1849 | binop(Iop_And32, mkexpr(rSt), mkU32(31)))), |
| 1850 | binop(Iop_Sar32, |
| 1851 | binop(Iop_Sub32, |
| 1852 | mkexpr(amtT), |
| 1853 | mkU32(32)), |
| 1854 | mkU8(31)))); |
| 1855 | DIS(buf, "r%u, LSL r%u", rM, rS); |
| 1856 | } |
| 1857 | |
| 1858 | |
| 1859 | static void compute_result_and_C_after_LSR_by_imm5 ( |
| 1860 | /*OUT*/HChar* buf, |
| 1861 | IRTemp* res, |
| 1862 | IRTemp* newC, |
| 1863 | IRTemp rMt, UInt shift_amt, /* operands */ |
| 1864 | UInt rM /* only for debug printing */ |
| 1865 | ) |
| 1866 | { |
| 1867 | if (shift_amt == 0) { |
| 1868 | // conceptually a 32-bit shift, however: |
| 1869 | // res = 0 |
| 1870 | // newC = Rm[31] |
| 1871 | if (newC) { |
| 1872 | assign( *newC, |
| 1873 | binop(Iop_And32, |
| 1874 | binop(Iop_Shr32, mkexpr(rMt), mkU8(31)), |
| 1875 | mkU32(1))); |
| 1876 | } |
| 1877 | assign( *res, mkU32(0) ); |
| 1878 | DIS(buf, "r%u, LSR #0(a.k.a. 32)", rM); |
| 1879 | } else { |
| 1880 | // shift in range 1..31 |
| 1881 | // res = Rm >>u shift_amt |
| 1882 | // newC = Rm[shift_amt - 1] |
| 1883 | vassert(shift_amt >= 1 && shift_amt <= 31); |
| 1884 | if (newC) { |
| 1885 | assign( *newC, |
| 1886 | binop(Iop_And32, |
| 1887 | binop(Iop_Shr32, mkexpr(rMt), |
| 1888 | mkU8(shift_amt - 1)), |
| 1889 | mkU32(1))); |
| 1890 | } |
| 1891 | assign( *res, |
| 1892 | binop(Iop_Shr32, mkexpr(rMt), mkU8(shift_amt)) ); |
| 1893 | DIS(buf, "r%u, LSR #%u", rM, shift_amt); |
| 1894 | } |
| 1895 | } |
| 1896 | |
| 1897 | |
| 1898 | static void compute_result_and_C_after_LSR_by_reg ( |
| 1899 | /*OUT*/HChar* buf, |
| 1900 | IRTemp* res, |
| 1901 | IRTemp* newC, |
| 1902 | IRTemp rMt, IRTemp rSt, /* operands */ |
| 1903 | UInt rM, UInt rS /* only for debug printing */ |
| 1904 | ) |
| 1905 | { |
| 1906 | // shift right in range 0 .. 255 |
| 1907 | // amt = rS & 255 |
| 1908 | // res = amt < 32 ? Rm >>u amt : 0 |
| 1909 | // newC = amt == 0 ? oldC : |
| 1910 | // amt in 1..32 ? Rm[amt-1] : 0 |
| 1911 | IRTemp amtT = newTemp(Ity_I32); |
| 1912 | assign( amtT, binop(Iop_And32, mkexpr(rSt), mkU32(255)) ); |
| 1913 | if (newC) { |
| 1914 | /* mux0X(amt == 0, |
| 1915 | mux0X(amt < 32, |
| 1916 | 0, |
sewardj | bb8b394 | 2011-05-01 18:47:10 +0000 | [diff] [blame] | 1917 | Rm[(amt-1) & 31]), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1918 | oldC) |
| 1919 | */ |
| 1920 | IRTemp oldC = newTemp(Ity_I32); |
| 1921 | assign(oldC, mk_armg_calculate_flag_c() ); |
| 1922 | assign( |
| 1923 | *newC, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 1924 | IRExpr_ITE( |
| 1925 | binop(Iop_CmpEQ32, mkexpr(amtT), mkU32(0)), |
| 1926 | mkexpr(oldC), |
| 1927 | IRExpr_ITE( |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 1928 | binop(Iop_CmpLE32U, mkexpr(amtT), mkU32(32)), |
sewardj | bb8b394 | 2011-05-01 18:47:10 +0000 | [diff] [blame] | 1929 | binop(Iop_And32, |
| 1930 | binop(Iop_Shr32, |
| 1931 | mkexpr(rMt), |
| 1932 | unop(Iop_32to8, |
| 1933 | binop(Iop_And32, |
| 1934 | binop(Iop_Sub32, |
| 1935 | mkexpr(amtT), |
| 1936 | mkU32(1)), |
| 1937 | mkU32(31) |
| 1938 | ) |
| 1939 | ) |
| 1940 | ), |
| 1941 | mkU32(1) |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 1942 | ), |
| 1943 | mkU32(0) |
| 1944 | ) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 1945 | ) |
| 1946 | ); |
| 1947 | } |
| 1948 | // (Rm >>u (Rs & 31)) & (((Rs & 255) - 32) >>s 31) |
| 1949 | // Lhs of the & limits the shift to 31 bits, so as to |
| 1950 | // give known IR semantics. Rhs of the & is all 1s for |
| 1951 | // Rs <= 31 and all 0s for Rs >= 32. |
| 1952 | assign( |
| 1953 | *res, |
| 1954 | binop( |
| 1955 | Iop_And32, |
| 1956 | binop(Iop_Shr32, |
| 1957 | mkexpr(rMt), |
| 1958 | unop(Iop_32to8, |
| 1959 | binop(Iop_And32, mkexpr(rSt), mkU32(31)))), |
| 1960 | binop(Iop_Sar32, |
| 1961 | binop(Iop_Sub32, |
| 1962 | mkexpr(amtT), |
| 1963 | mkU32(32)), |
| 1964 | mkU8(31)))); |
| 1965 | DIS(buf, "r%u, LSR r%u", rM, rS); |
| 1966 | } |
| 1967 | |
| 1968 | |
| 1969 | static void compute_result_and_C_after_ASR_by_imm5 ( |
| 1970 | /*OUT*/HChar* buf, |
| 1971 | IRTemp* res, |
| 1972 | IRTemp* newC, |
| 1973 | IRTemp rMt, UInt shift_amt, /* operands */ |
| 1974 | UInt rM /* only for debug printing */ |
| 1975 | ) |
| 1976 | { |
| 1977 | if (shift_amt == 0) { |
| 1978 | // conceptually a 32-bit shift, however: |
| 1979 | // res = Rm >>s 31 |
| 1980 | // newC = Rm[31] |
| 1981 | if (newC) { |
| 1982 | assign( *newC, |
| 1983 | binop(Iop_And32, |
| 1984 | binop(Iop_Shr32, mkexpr(rMt), mkU8(31)), |
| 1985 | mkU32(1))); |
| 1986 | } |
| 1987 | assign( *res, binop(Iop_Sar32, mkexpr(rMt), mkU8(31)) ); |
| 1988 | DIS(buf, "r%u, ASR #0(a.k.a. 32)", rM); |
| 1989 | } else { |
| 1990 | // shift in range 1..31 |
| 1991 | // res = Rm >>s shift_amt |
| 1992 | // newC = Rm[shift_amt - 1] |
| 1993 | vassert(shift_amt >= 1 && shift_amt <= 31); |
| 1994 | if (newC) { |
| 1995 | assign( *newC, |
| 1996 | binop(Iop_And32, |
| 1997 | binop(Iop_Shr32, mkexpr(rMt), |
| 1998 | mkU8(shift_amt - 1)), |
| 1999 | mkU32(1))); |
| 2000 | } |
| 2001 | assign( *res, |
| 2002 | binop(Iop_Sar32, mkexpr(rMt), mkU8(shift_amt)) ); |
| 2003 | DIS(buf, "r%u, ASR #%u", rM, shift_amt); |
| 2004 | } |
| 2005 | } |
| 2006 | |
| 2007 | |
| 2008 | static void compute_result_and_C_after_ASR_by_reg ( |
| 2009 | /*OUT*/HChar* buf, |
| 2010 | IRTemp* res, |
| 2011 | IRTemp* newC, |
| 2012 | IRTemp rMt, IRTemp rSt, /* operands */ |
| 2013 | UInt rM, UInt rS /* only for debug printing */ |
| 2014 | ) |
| 2015 | { |
| 2016 | // arithmetic shift right in range 0 .. 255 |
| 2017 | // amt = rS & 255 |
| 2018 | // res = amt < 32 ? Rm >>s amt : Rm >>s 31 |
| 2019 | // newC = amt == 0 ? oldC : |
| 2020 | // amt in 1..32 ? Rm[amt-1] : Rm[31] |
| 2021 | IRTemp amtT = newTemp(Ity_I32); |
| 2022 | assign( amtT, binop(Iop_And32, mkexpr(rSt), mkU32(255)) ); |
| 2023 | if (newC) { |
| 2024 | /* mux0X(amt == 0, |
| 2025 | mux0X(amt < 32, |
| 2026 | Rm[31], |
| 2027 | Rm[(amt-1) & 31]) |
| 2028 | oldC) |
| 2029 | */ |
| 2030 | IRTemp oldC = newTemp(Ity_I32); |
| 2031 | assign(oldC, mk_armg_calculate_flag_c() ); |
| 2032 | assign( |
| 2033 | *newC, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 2034 | IRExpr_ITE( |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 2035 | binop(Iop_CmpEQ32, mkexpr(amtT), mkU32(0)), |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 2036 | mkexpr(oldC), |
| 2037 | IRExpr_ITE( |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 2038 | binop(Iop_CmpLE32U, mkexpr(amtT), mkU32(32)), |
sewardj | bb8b394 | 2011-05-01 18:47:10 +0000 | [diff] [blame] | 2039 | binop(Iop_And32, |
| 2040 | binop(Iop_Shr32, |
| 2041 | mkexpr(rMt), |
sewardj | bb8b394 | 2011-05-01 18:47:10 +0000 | [diff] [blame] | 2042 | unop(Iop_32to8, |
| 2043 | binop(Iop_And32, |
| 2044 | binop(Iop_Sub32, |
| 2045 | mkexpr(amtT), |
| 2046 | mkU32(1)), |
| 2047 | mkU32(31) |
| 2048 | ) |
| 2049 | ) |
| 2050 | ), |
| 2051 | mkU32(1) |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 2052 | ), |
| 2053 | binop(Iop_And32, |
| 2054 | binop(Iop_Shr32, |
| 2055 | mkexpr(rMt), |
| 2056 | mkU8(31) |
| 2057 | ), |
| 2058 | mkU32(1) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2059 | ) |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 2060 | ) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2061 | ) |
| 2062 | ); |
| 2063 | } |
| 2064 | // (Rm >>s (amt <u 32 ? amt : 31)) |
| 2065 | assign( |
| 2066 | *res, |
| 2067 | binop( |
| 2068 | Iop_Sar32, |
| 2069 | mkexpr(rMt), |
| 2070 | unop( |
| 2071 | Iop_32to8, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 2072 | IRExpr_ITE( |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 2073 | binop(Iop_CmpLT32U, mkexpr(amtT), mkU32(32)), |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 2074 | mkexpr(amtT), |
| 2075 | mkU32(31))))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2076 | DIS(buf, "r%u, ASR r%u", rM, rS); |
| 2077 | } |
| 2078 | |
| 2079 | |
| 2080 | static void compute_result_and_C_after_ROR_by_reg ( |
| 2081 | /*OUT*/HChar* buf, |
| 2082 | IRTemp* res, |
| 2083 | IRTemp* newC, |
| 2084 | IRTemp rMt, IRTemp rSt, /* operands */ |
| 2085 | UInt rM, UInt rS /* only for debug printing */ |
| 2086 | ) |
| 2087 | { |
| 2088 | // rotate right in range 0 .. 255 |
| 2089 | // amt = rS & 255 |
| 2090 | // shop = Rm `ror` (amt & 31) |
| 2091 | // shco = amt == 0 ? oldC : Rm[(amt-1) & 31] |
| 2092 | IRTemp amtT = newTemp(Ity_I32); |
| 2093 | assign( amtT, binop(Iop_And32, mkexpr(rSt), mkU32(255)) ); |
| 2094 | IRTemp amt5T = newTemp(Ity_I32); |
| 2095 | assign( amt5T, binop(Iop_And32, mkexpr(rSt), mkU32(31)) ); |
| 2096 | IRTemp oldC = newTemp(Ity_I32); |
| 2097 | assign(oldC, mk_armg_calculate_flag_c() ); |
| 2098 | if (newC) { |
| 2099 | assign( |
| 2100 | *newC, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 2101 | IRExpr_ITE( |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 2102 | binop(Iop_CmpNE32, mkexpr(amtT), mkU32(0)), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2103 | binop(Iop_And32, |
| 2104 | binop(Iop_Shr32, |
| 2105 | mkexpr(rMt), |
| 2106 | unop(Iop_32to8, |
| 2107 | binop(Iop_And32, |
| 2108 | binop(Iop_Sub32, |
| 2109 | mkexpr(amtT), |
| 2110 | mkU32(1) |
| 2111 | ), |
| 2112 | mkU32(31) |
| 2113 | ) |
| 2114 | ) |
| 2115 | ), |
| 2116 | mkU32(1) |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 2117 | ), |
| 2118 | mkexpr(oldC) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2119 | ) |
| 2120 | ); |
| 2121 | } |
| 2122 | assign( |
| 2123 | *res, |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 2124 | IRExpr_ITE( |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 2125 | binop(Iop_CmpNE32, mkexpr(amt5T), mkU32(0)), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2126 | binop(Iop_Or32, |
| 2127 | binop(Iop_Shr32, |
| 2128 | mkexpr(rMt), |
| 2129 | unop(Iop_32to8, mkexpr(amt5T)) |
| 2130 | ), |
| 2131 | binop(Iop_Shl32, |
| 2132 | mkexpr(rMt), |
| 2133 | unop(Iop_32to8, |
| 2134 | binop(Iop_Sub32, mkU32(32), mkexpr(amt5T)) |
| 2135 | ) |
| 2136 | ) |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 2137 | ), |
| 2138 | mkexpr(rMt) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2139 | ) |
| 2140 | ); |
| 2141 | DIS(buf, "r%u, ROR r#%u", rM, rS); |
| 2142 | } |
| 2143 | |
| 2144 | |
| 2145 | /* Generate an expression corresponding to the immediate-shift case of |
| 2146 | a shifter operand. This is used both for ARM and Thumb2. |
| 2147 | |
| 2148 | Bind it to a temporary, and return that via *res. If newC is |
| 2149 | non-NULL, also compute a value for the shifter's carry out (in the |
| 2150 | LSB of a word), bind it to a temporary, and return that via *shco. |
| 2151 | |
| 2152 | Generates GETs from the guest state and is therefore not safe to |
| 2153 | use once we start doing PUTs to it, for any given instruction. |
| 2154 | |
| 2155 | 'how' is encoded thusly: |
| 2156 | 00b LSL, 01b LSR, 10b ASR, 11b ROR |
| 2157 | Most but not all ARM and Thumb integer insns use this encoding. |
| 2158 | Be careful to ensure the right value is passed here. |
| 2159 | */ |
| 2160 | static void compute_result_and_C_after_shift_by_imm5 ( |
| 2161 | /*OUT*/HChar* buf, |
| 2162 | /*OUT*/IRTemp* res, |
| 2163 | /*OUT*/IRTemp* newC, |
| 2164 | IRTemp rMt, /* reg to shift */ |
| 2165 | UInt how, /* what kind of shift */ |
| 2166 | UInt shift_amt, /* shift amount (0..31) */ |
| 2167 | UInt rM /* only for debug printing */ |
| 2168 | ) |
| 2169 | { |
| 2170 | vassert(shift_amt < 32); |
| 2171 | vassert(how < 4); |
| 2172 | |
| 2173 | switch (how) { |
| 2174 | |
| 2175 | case 0: |
| 2176 | compute_result_and_C_after_LSL_by_imm5( |
| 2177 | buf, res, newC, rMt, shift_amt, rM |
| 2178 | ); |
| 2179 | break; |
| 2180 | |
| 2181 | case 1: |
| 2182 | compute_result_and_C_after_LSR_by_imm5( |
| 2183 | buf, res, newC, rMt, shift_amt, rM |
| 2184 | ); |
| 2185 | break; |
| 2186 | |
| 2187 | case 2: |
| 2188 | compute_result_and_C_after_ASR_by_imm5( |
| 2189 | buf, res, newC, rMt, shift_amt, rM |
| 2190 | ); |
| 2191 | break; |
| 2192 | |
| 2193 | case 3: |
| 2194 | if (shift_amt == 0) { |
| 2195 | IRTemp oldcT = newTemp(Ity_I32); |
| 2196 | // rotate right 1 bit through carry (?) |
| 2197 | // RRX -- described at ARM ARM A5-17 |
| 2198 | // res = (oldC << 31) | (Rm >>u 1) |
| 2199 | // newC = Rm[0] |
| 2200 | if (newC) { |
| 2201 | assign( *newC, |
| 2202 | binop(Iop_And32, mkexpr(rMt), mkU32(1))); |
| 2203 | } |
| 2204 | assign( oldcT, mk_armg_calculate_flag_c() ); |
| 2205 | assign( *res, |
| 2206 | binop(Iop_Or32, |
| 2207 | binop(Iop_Shl32, mkexpr(oldcT), mkU8(31)), |
| 2208 | binop(Iop_Shr32, mkexpr(rMt), mkU8(1))) ); |
| 2209 | DIS(buf, "r%u, RRX", rM); |
| 2210 | } else { |
| 2211 | // rotate right in range 1..31 |
| 2212 | // res = Rm `ror` shift_amt |
| 2213 | // newC = Rm[shift_amt - 1] |
| 2214 | vassert(shift_amt >= 1 && shift_amt <= 31); |
| 2215 | if (newC) { |
| 2216 | assign( *newC, |
| 2217 | binop(Iop_And32, |
| 2218 | binop(Iop_Shr32, mkexpr(rMt), |
| 2219 | mkU8(shift_amt - 1)), |
| 2220 | mkU32(1))); |
| 2221 | } |
| 2222 | assign( *res, |
| 2223 | binop(Iop_Or32, |
| 2224 | binop(Iop_Shr32, mkexpr(rMt), mkU8(shift_amt)), |
| 2225 | binop(Iop_Shl32, mkexpr(rMt), |
| 2226 | mkU8(32-shift_amt)))); |
| 2227 | DIS(buf, "r%u, ROR #%u", rM, shift_amt); |
| 2228 | } |
| 2229 | break; |
| 2230 | |
| 2231 | default: |
| 2232 | /*NOTREACHED*/ |
| 2233 | vassert(0); |
| 2234 | } |
| 2235 | } |
| 2236 | |
| 2237 | |
| 2238 | /* Generate an expression corresponding to the register-shift case of |
| 2239 | a shifter operand. This is used both for ARM and Thumb2. |
| 2240 | |
| 2241 | Bind it to a temporary, and return that via *res. If newC is |
| 2242 | non-NULL, also compute a value for the shifter's carry out (in the |
| 2243 | LSB of a word), bind it to a temporary, and return that via *shco. |
| 2244 | |
| 2245 | Generates GETs from the guest state and is therefore not safe to |
| 2246 | use once we start doing PUTs to it, for any given instruction. |
| 2247 | |
| 2248 | 'how' is encoded thusly: |
| 2249 | 00b LSL, 01b LSR, 10b ASR, 11b ROR |
| 2250 | Most but not all ARM and Thumb integer insns use this encoding. |
| 2251 | Be careful to ensure the right value is passed here. |
| 2252 | */ |
| 2253 | static void compute_result_and_C_after_shift_by_reg ( |
| 2254 | /*OUT*/HChar* buf, |
| 2255 | /*OUT*/IRTemp* res, |
| 2256 | /*OUT*/IRTemp* newC, |
| 2257 | IRTemp rMt, /* reg to shift */ |
| 2258 | UInt how, /* what kind of shift */ |
| 2259 | IRTemp rSt, /* shift amount */ |
| 2260 | UInt rM, /* only for debug printing */ |
| 2261 | UInt rS /* only for debug printing */ |
| 2262 | ) |
| 2263 | { |
| 2264 | vassert(how < 4); |
| 2265 | switch (how) { |
| 2266 | case 0: { /* LSL */ |
| 2267 | compute_result_and_C_after_LSL_by_reg( |
| 2268 | buf, res, newC, rMt, rSt, rM, rS |
| 2269 | ); |
| 2270 | break; |
| 2271 | } |
| 2272 | case 1: { /* LSR */ |
| 2273 | compute_result_and_C_after_LSR_by_reg( |
| 2274 | buf, res, newC, rMt, rSt, rM, rS |
| 2275 | ); |
| 2276 | break; |
| 2277 | } |
| 2278 | case 2: { /* ASR */ |
| 2279 | compute_result_and_C_after_ASR_by_reg( |
| 2280 | buf, res, newC, rMt, rSt, rM, rS |
| 2281 | ); |
| 2282 | break; |
| 2283 | } |
| 2284 | case 3: { /* ROR */ |
| 2285 | compute_result_and_C_after_ROR_by_reg( |
| 2286 | buf, res, newC, rMt, rSt, rM, rS |
| 2287 | ); |
| 2288 | break; |
| 2289 | } |
| 2290 | default: |
| 2291 | /*NOTREACHED*/ |
| 2292 | vassert(0); |
| 2293 | } |
| 2294 | } |
| 2295 | |
| 2296 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2297 | /* Generate an expression corresponding to a shifter_operand, bind it |
| 2298 | to a temporary, and return that via *shop. If shco is non-NULL, |
| 2299 | also compute a value for the shifter's carry out (in the LSB of a |
| 2300 | word), bind it to a temporary, and return that via *shco. |
| 2301 | |
| 2302 | If for some reason we can't come up with a shifter operand (missing |
| 2303 | case? not really a shifter operand?) return False. |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2304 | |
| 2305 | Generates GETs from the guest state and is therefore not safe to |
| 2306 | use once we start doing PUTs to it, for any given instruction. |
| 2307 | |
| 2308 | For ARM insns only; not for Thumb. |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2309 | */ |
| 2310 | static Bool mk_shifter_operand ( UInt insn_25, UInt insn_11_0, |
| 2311 | /*OUT*/IRTemp* shop, |
| 2312 | /*OUT*/IRTemp* shco, |
| 2313 | /*OUT*/HChar* buf ) |
| 2314 | { |
| 2315 | UInt insn_4 = (insn_11_0 >> 4) & 1; |
| 2316 | UInt insn_7 = (insn_11_0 >> 7) & 1; |
| 2317 | vassert(insn_25 <= 0x1); |
| 2318 | vassert(insn_11_0 <= 0xFFF); |
| 2319 | |
| 2320 | vassert(shop && *shop == IRTemp_INVALID); |
| 2321 | *shop = newTemp(Ity_I32); |
| 2322 | |
| 2323 | if (shco) { |
| 2324 | vassert(*shco == IRTemp_INVALID); |
| 2325 | *shco = newTemp(Ity_I32); |
cerion | b85e8bb | 2005-02-16 08:54:33 +0000 | [diff] [blame] | 2326 | } |
cerion | 060c542 | 2004-12-10 15:28:43 +0000 | [diff] [blame] | 2327 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2328 | /* 32-bit immediate */ |
| 2329 | |
| 2330 | if (insn_25 == 1) { |
| 2331 | /* immediate: (7:0) rotated right by 2 * (11:8) */ |
| 2332 | UInt imm = (insn_11_0 >> 0) & 0xFF; |
| 2333 | UInt rot = 2 * ((insn_11_0 >> 8) & 0xF); |
| 2334 | vassert(rot <= 30); |
| 2335 | imm = ROR32(imm, rot); |
| 2336 | if (shco) { |
| 2337 | if (rot == 0) { |
| 2338 | assign( *shco, mk_armg_calculate_flag_c() ); |
| 2339 | } else { |
| 2340 | assign( *shco, mkU32( (imm >> 31) & 1 ) ); |
| 2341 | } |
cerion | b85e8bb | 2005-02-16 08:54:33 +0000 | [diff] [blame] | 2342 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2343 | DIS(buf, "#0x%x", imm); |
| 2344 | assign( *shop, mkU32(imm) ); |
| 2345 | return True; |
cerion | b85e8bb | 2005-02-16 08:54:33 +0000 | [diff] [blame] | 2346 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2347 | |
| 2348 | /* Shift/rotate by immediate */ |
| 2349 | |
| 2350 | if (insn_25 == 0 && insn_4 == 0) { |
| 2351 | /* Rm (3:0) shifted (6:5) by immediate (11:7) */ |
| 2352 | UInt shift_amt = (insn_11_0 >> 7) & 0x1F; |
| 2353 | UInt rM = (insn_11_0 >> 0) & 0xF; |
| 2354 | UInt how = (insn_11_0 >> 5) & 3; |
| 2355 | /* how: 00 = Shl, 01 = Shr, 10 = Sar, 11 = Ror */ |
| 2356 | IRTemp rMt = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2357 | assign(rMt, getIRegA(rM)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2358 | |
| 2359 | vassert(shift_amt <= 31); |
| 2360 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2361 | compute_result_and_C_after_shift_by_imm5( |
| 2362 | buf, shop, shco, rMt, how, shift_amt, rM |
| 2363 | ); |
| 2364 | return True; |
cerion | b85e8bb | 2005-02-16 08:54:33 +0000 | [diff] [blame] | 2365 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2366 | |
| 2367 | /* Shift/rotate by register */ |
| 2368 | if (insn_25 == 0 && insn_4 == 1) { |
| 2369 | /* Rm (3:0) shifted (6:5) by Rs (11:8) */ |
| 2370 | UInt rM = (insn_11_0 >> 0) & 0xF; |
| 2371 | UInt rS = (insn_11_0 >> 8) & 0xF; |
| 2372 | UInt how = (insn_11_0 >> 5) & 3; |
| 2373 | /* how: 00 = Shl, 01 = Shr, 10 = Sar, 11 = Ror */ |
| 2374 | IRTemp rMt = newTemp(Ity_I32); |
| 2375 | IRTemp rSt = newTemp(Ity_I32); |
| 2376 | |
| 2377 | if (insn_7 == 1) |
| 2378 | return False; /* not really a shifter operand */ |
| 2379 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2380 | assign(rMt, getIRegA(rM)); |
| 2381 | assign(rSt, getIRegA(rS)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2382 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2383 | compute_result_and_C_after_shift_by_reg( |
| 2384 | buf, shop, shco, rMt, how, rSt, rM, rS |
| 2385 | ); |
| 2386 | return True; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2387 | } |
| 2388 | |
| 2389 | vex_printf("mk_shifter_operand(0x%x,0x%x)\n", insn_25, insn_11_0 ); |
| 2390 | return False; |
cerion | 060c542 | 2004-12-10 15:28:43 +0000 | [diff] [blame] | 2391 | } |
cerion | f7da63d | 2004-12-09 19:04:57 +0000 | [diff] [blame] | 2392 | |
| 2393 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2394 | /* ARM only */ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2395 | static |
| 2396 | IRExpr* mk_EA_reg_plusminus_imm12 ( UInt rN, UInt bU, UInt imm12, |
| 2397 | /*OUT*/HChar* buf ) |
cerion | 060c542 | 2004-12-10 15:28:43 +0000 | [diff] [blame] | 2398 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2399 | vassert(rN < 16); |
| 2400 | vassert(bU < 2); |
| 2401 | vassert(imm12 < 0x1000); |
florian | 5df8ab0 | 2012-10-13 19:34:19 +0000 | [diff] [blame] | 2402 | HChar opChar = bU == 1 ? '+' : '-'; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2403 | DIS(buf, "[r%u, #%c%u]", rN, opChar, imm12); |
| 2404 | return |
| 2405 | binop( (bU == 1 ? Iop_Add32 : Iop_Sub32), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2406 | getIRegA(rN), |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2407 | mkU32(imm12) ); |
cerion | f7da63d | 2004-12-09 19:04:57 +0000 | [diff] [blame] | 2408 | } |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 2409 | |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 2410 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2411 | /* ARM only. |
| 2412 | NB: This is "DecodeImmShift" in newer versions of the the ARM ARM. |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 2413 | */ |
cerion | a70a37b | 2004-12-03 18:54:08 +0000 | [diff] [blame] | 2414 | static |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2415 | IRExpr* mk_EA_reg_plusminus_shifted_reg ( UInt rN, UInt bU, UInt rM, |
| 2416 | UInt sh2, UInt imm5, |
| 2417 | /*OUT*/HChar* buf ) |
cerion | a70a37b | 2004-12-03 18:54:08 +0000 | [diff] [blame] | 2418 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2419 | vassert(rN < 16); |
| 2420 | vassert(bU < 2); |
| 2421 | vassert(rM < 16); |
| 2422 | vassert(sh2 < 4); |
| 2423 | vassert(imm5 < 32); |
florian | 5df8ab0 | 2012-10-13 19:34:19 +0000 | [diff] [blame] | 2424 | HChar opChar = bU == 1 ? '+' : '-'; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2425 | IRExpr* index = NULL; |
| 2426 | switch (sh2) { |
| 2427 | case 0: /* LSL */ |
| 2428 | /* imm5 can be in the range 0 .. 31 inclusive. */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2429 | index = binop(Iop_Shl32, getIRegA(rM), mkU8(imm5)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2430 | DIS(buf, "[r%u, %c r%u LSL #%u]", rN, opChar, rM, imm5); |
cerion | b85e8bb | 2005-02-16 08:54:33 +0000 | [diff] [blame] | 2431 | break; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2432 | case 1: /* LSR */ |
| 2433 | if (imm5 == 0) { |
| 2434 | index = mkU32(0); |
| 2435 | vassert(0); // ATC |
| 2436 | } else { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2437 | index = binop(Iop_Shr32, getIRegA(rM), mkU8(imm5)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2438 | } |
| 2439 | DIS(buf, "[r%u, %cr%u, LSR #%u]", |
| 2440 | rN, opChar, rM, imm5 == 0 ? 32 : imm5); |
cerion | b85e8bb | 2005-02-16 08:54:33 +0000 | [diff] [blame] | 2441 | break; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2442 | case 2: /* ASR */ |
| 2443 | /* Doesn't this just mean that the behaviour with imm5 == 0 |
| 2444 | is the same as if it had been 31 ? */ |
| 2445 | if (imm5 == 0) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2446 | index = binop(Iop_Sar32, getIRegA(rM), mkU8(31)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2447 | vassert(0); // ATC |
| 2448 | } else { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2449 | index = binop(Iop_Sar32, getIRegA(rM), mkU8(imm5)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2450 | } |
| 2451 | DIS(buf, "[r%u, %cr%u, ASR #%u]", |
| 2452 | rN, opChar, rM, imm5 == 0 ? 32 : imm5); |
cerion | b85e8bb | 2005-02-16 08:54:33 +0000 | [diff] [blame] | 2453 | break; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2454 | case 3: /* ROR or RRX */ |
| 2455 | if (imm5 == 0) { |
| 2456 | IRTemp rmT = newTemp(Ity_I32); |
| 2457 | IRTemp cflagT = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2458 | assign(rmT, getIRegA(rM)); |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 2459 | assign(cflagT, mk_armg_calculate_flag_c()); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2460 | index = binop(Iop_Or32, |
| 2461 | binop(Iop_Shl32, mkexpr(cflagT), mkU8(31)), |
| 2462 | binop(Iop_Shr32, mkexpr(rmT), mkU8(1))); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2463 | DIS(buf, "[r%u, %cr%u, RRX]", rN, opChar, rM); |
| 2464 | } else { |
| 2465 | IRTemp rmT = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2466 | assign(rmT, getIRegA(rM)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2467 | vassert(imm5 >= 1 && imm5 <= 31); |
| 2468 | index = binop(Iop_Or32, |
| 2469 | binop(Iop_Shl32, mkexpr(rmT), mkU8(32-imm5)), |
| 2470 | binop(Iop_Shr32, mkexpr(rmT), mkU8(imm5))); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2471 | DIS(buf, "[r%u, %cr%u, ROR #%u]", rN, opChar, rM, imm5); |
| 2472 | } |
| 2473 | break; |
cerion | b85e8bb | 2005-02-16 08:54:33 +0000 | [diff] [blame] | 2474 | default: |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2475 | vassert(0); |
cerion | b85e8bb | 2005-02-16 08:54:33 +0000 | [diff] [blame] | 2476 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2477 | vassert(index); |
| 2478 | return binop(bU == 1 ? Iop_Add32 : Iop_Sub32, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2479 | getIRegA(rN), index); |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 2480 | } |
| 2481 | |
| 2482 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2483 | /* ARM only */ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2484 | static |
| 2485 | IRExpr* mk_EA_reg_plusminus_imm8 ( UInt rN, UInt bU, UInt imm8, |
| 2486 | /*OUT*/HChar* buf ) |
cerion | f7da63d | 2004-12-09 19:04:57 +0000 | [diff] [blame] | 2487 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2488 | vassert(rN < 16); |
| 2489 | vassert(bU < 2); |
| 2490 | vassert(imm8 < 0x100); |
florian | 5df8ab0 | 2012-10-13 19:34:19 +0000 | [diff] [blame] | 2491 | HChar opChar = bU == 1 ? '+' : '-'; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2492 | DIS(buf, "[r%u, #%c%u]", rN, opChar, imm8); |
| 2493 | return |
| 2494 | binop( (bU == 1 ? Iop_Add32 : Iop_Sub32), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2495 | getIRegA(rN), |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2496 | mkU32(imm8) ); |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 2497 | } |
| 2498 | |
| 2499 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2500 | /* ARM only */ |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 2501 | static |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2502 | IRExpr* mk_EA_reg_plusminus_reg ( UInt rN, UInt bU, UInt rM, |
| 2503 | /*OUT*/HChar* buf ) |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 2504 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2505 | vassert(rN < 16); |
| 2506 | vassert(bU < 2); |
| 2507 | vassert(rM < 16); |
florian | 5df8ab0 | 2012-10-13 19:34:19 +0000 | [diff] [blame] | 2508 | HChar opChar = bU == 1 ? '+' : '-'; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2509 | IRExpr* index = getIRegA(rM); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2510 | DIS(buf, "[r%u, %c r%u]", rN, opChar, rM); |
| 2511 | return binop(bU == 1 ? Iop_Add32 : Iop_Sub32, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2512 | getIRegA(rN), index); |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 2513 | } |
| 2514 | |
| 2515 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2516 | /* irRes :: Ity_I32 holds a floating point comparison result encoded |
| 2517 | as an IRCmpF64Result. Generate code to convert it to an |
| 2518 | ARM-encoded (N,Z,C,V) group in the lowest 4 bits of an I32 value. |
| 2519 | Assign a new temp to hold that value, and return the temp. */ |
cerion | a70a37b | 2004-12-03 18:54:08 +0000 | [diff] [blame] | 2520 | static |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2521 | IRTemp mk_convert_IRCmpF64Result_to_NZCV ( IRTemp irRes ) |
cerion | a70a37b | 2004-12-03 18:54:08 +0000 | [diff] [blame] | 2522 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2523 | IRTemp ix = newTemp(Ity_I32); |
| 2524 | IRTemp termL = newTemp(Ity_I32); |
| 2525 | IRTemp termR = newTemp(Ity_I32); |
| 2526 | IRTemp nzcv = newTemp(Ity_I32); |
cerion | a70a37b | 2004-12-03 18:54:08 +0000 | [diff] [blame] | 2527 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2528 | /* This is where the fun starts. We have to convert 'irRes' from |
| 2529 | an IR-convention return result (IRCmpF64Result) to an |
| 2530 | ARM-encoded (N,Z,C,V) group. The final result is in the bottom |
| 2531 | 4 bits of 'nzcv'. */ |
| 2532 | /* Map compare result from IR to ARM(nzcv) */ |
| 2533 | /* |
| 2534 | FP cmp result | IR | ARM(nzcv) |
| 2535 | -------------------------------- |
| 2536 | UN 0x45 0011 |
| 2537 | LT 0x01 1000 |
| 2538 | GT 0x00 0010 |
| 2539 | EQ 0x40 0110 |
| 2540 | */ |
| 2541 | /* Now since you're probably wondering WTF .. |
cerion | 397f9e5 | 2004-12-15 13:04:06 +0000 | [diff] [blame] | 2542 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2543 | ix fishes the useful bits out of the IR value, bits 6 and 0, and |
| 2544 | places them side by side, giving a number which is 0, 1, 2 or 3. |
cerion | 397f9e5 | 2004-12-15 13:04:06 +0000 | [diff] [blame] | 2545 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2546 | termL is a sequence cooked up by GNU superopt. It converts ix |
| 2547 | into an almost correct value NZCV value (incredibly), except |
| 2548 | for the case of UN, where it produces 0100 instead of the |
| 2549 | required 0011. |
cerion | 397f9e5 | 2004-12-15 13:04:06 +0000 | [diff] [blame] | 2550 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2551 | termR is therefore a correction term, also computed from ix. It |
| 2552 | is 1 in the UN case and 0 for LT, GT and UN. Hence, to get |
| 2553 | the final correct value, we subtract termR from termL. |
cerion | a70a37b | 2004-12-03 18:54:08 +0000 | [diff] [blame] | 2554 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2555 | Don't take my word for it. There's a test program at the bottom |
| 2556 | of this file, to try this out with. |
| 2557 | */ |
| 2558 | assign( |
| 2559 | ix, |
| 2560 | binop(Iop_Or32, |
| 2561 | binop(Iop_And32, |
| 2562 | binop(Iop_Shr32, mkexpr(irRes), mkU8(5)), |
| 2563 | mkU32(3)), |
| 2564 | binop(Iop_And32, mkexpr(irRes), mkU32(1)))); |
cerion | a70a37b | 2004-12-03 18:54:08 +0000 | [diff] [blame] | 2565 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2566 | assign( |
| 2567 | termL, |
| 2568 | binop(Iop_Add32, |
| 2569 | binop(Iop_Shr32, |
| 2570 | binop(Iop_Sub32, |
| 2571 | binop(Iop_Shl32, |
| 2572 | binop(Iop_Xor32, mkexpr(ix), mkU32(1)), |
| 2573 | mkU8(30)), |
| 2574 | mkU32(1)), |
| 2575 | mkU8(29)), |
| 2576 | mkU32(1))); |
cerion | a70a37b | 2004-12-03 18:54:08 +0000 | [diff] [blame] | 2577 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2578 | assign( |
| 2579 | termR, |
| 2580 | binop(Iop_And32, |
| 2581 | binop(Iop_And32, |
| 2582 | mkexpr(ix), |
| 2583 | binop(Iop_Shr32, mkexpr(ix), mkU8(1))), |
| 2584 | mkU32(1))); |
cerion | f7da63d | 2004-12-09 19:04:57 +0000 | [diff] [blame] | 2585 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 2586 | assign(nzcv, binop(Iop_Sub32, mkexpr(termL), mkexpr(termR))); |
| 2587 | return nzcv; |
cerion | a70a37b | 2004-12-03 18:54:08 +0000 | [diff] [blame] | 2588 | } |
| 2589 | |
| 2590 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2591 | /* Thumb32 only. This is "ThumbExpandImm" in the ARM ARM. If |
| 2592 | updatesC is non-NULL, a boolean is written to it indicating whether |
| 2593 | or not the C flag is updated, as per ARM ARM "ThumbExpandImm_C". |
| 2594 | */ |
| 2595 | static UInt thumbExpandImm ( Bool* updatesC, |
| 2596 | UInt imm1, UInt imm3, UInt imm8 ) |
| 2597 | { |
| 2598 | vassert(imm1 < (1<<1)); |
| 2599 | vassert(imm3 < (1<<3)); |
| 2600 | vassert(imm8 < (1<<8)); |
| 2601 | UInt i_imm3_a = (imm1 << 4) | (imm3 << 1) | ((imm8 >> 7) & 1); |
| 2602 | UInt abcdefgh = imm8; |
| 2603 | UInt lbcdefgh = imm8 | 0x80; |
| 2604 | if (updatesC) { |
| 2605 | *updatesC = i_imm3_a >= 8; |
| 2606 | } |
| 2607 | switch (i_imm3_a) { |
| 2608 | case 0: case 1: |
| 2609 | return abcdefgh; |
| 2610 | case 2: case 3: |
| 2611 | return (abcdefgh << 16) | abcdefgh; |
| 2612 | case 4: case 5: |
| 2613 | return (abcdefgh << 24) | (abcdefgh << 8); |
| 2614 | case 6: case 7: |
| 2615 | return (abcdefgh << 24) | (abcdefgh << 16) |
| 2616 | | (abcdefgh << 8) | abcdefgh; |
| 2617 | case 8 ... 31: |
| 2618 | return lbcdefgh << (32 - i_imm3_a); |
| 2619 | default: |
| 2620 | break; |
| 2621 | } |
| 2622 | /*NOTREACHED*/vassert(0); |
| 2623 | } |
| 2624 | |
| 2625 | |
| 2626 | /* Version of thumbExpandImm where we simply feed it the |
| 2627 | instruction halfwords (the lowest addressed one is I0). */ |
| 2628 | static UInt thumbExpandImm_from_I0_I1 ( Bool* updatesC, |
| 2629 | UShort i0s, UShort i1s ) |
| 2630 | { |
| 2631 | UInt i0 = (UInt)i0s; |
| 2632 | UInt i1 = (UInt)i1s; |
| 2633 | UInt imm1 = SLICE_UInt(i0,10,10); |
| 2634 | UInt imm3 = SLICE_UInt(i1,14,12); |
| 2635 | UInt imm8 = SLICE_UInt(i1,7,0); |
| 2636 | return thumbExpandImm(updatesC, imm1, imm3, imm8); |
| 2637 | } |
| 2638 | |
| 2639 | |
| 2640 | /* Thumb16 only. Given the firstcond and mask fields from an IT |
| 2641 | instruction, compute the 32-bit ITSTATE value implied, as described |
| 2642 | in libvex_guest_arm.h. This is not the ARM ARM representation. |
| 2643 | Also produce the t/e chars for the 2nd, 3rd, 4th insns, for |
| 2644 | disassembly printing. Returns False if firstcond or mask |
| 2645 | denote something invalid. |
| 2646 | |
| 2647 | The number and conditions for the instructions to be |
| 2648 | conditionalised depend on firstcond and mask: |
| 2649 | |
| 2650 | mask cond 1 cond 2 cond 3 cond 4 |
| 2651 | |
| 2652 | 1000 fc[3:0] |
| 2653 | x100 fc[3:0] fc[3:1]:x |
| 2654 | xy10 fc[3:0] fc[3:1]:x fc[3:1]:y |
| 2655 | xyz1 fc[3:0] fc[3:1]:x fc[3:1]:y fc[3:1]:z |
| 2656 | |
| 2657 | The condition fields are assembled in *itstate backwards (cond 4 at |
| 2658 | the top, cond 1 at the bottom). Conditions are << 4'd and then |
| 2659 | ^0xE'd, and those fields that correspond to instructions in the IT |
| 2660 | block are tagged with a 1 bit. |
| 2661 | */ |
| 2662 | static Bool compute_ITSTATE ( /*OUT*/UInt* itstate, |
florian | 5df8ab0 | 2012-10-13 19:34:19 +0000 | [diff] [blame] | 2663 | /*OUT*/HChar* ch1, |
| 2664 | /*OUT*/HChar* ch2, |
| 2665 | /*OUT*/HChar* ch3, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2666 | UInt firstcond, UInt mask ) |
| 2667 | { |
| 2668 | vassert(firstcond <= 0xF); |
| 2669 | vassert(mask <= 0xF); |
| 2670 | *itstate = 0; |
| 2671 | *ch1 = *ch2 = *ch3 = '.'; |
| 2672 | if (mask == 0) |
| 2673 | return False; /* the logic below actually ensures this anyway, |
| 2674 | but clearer to make it explicit. */ |
| 2675 | if (firstcond == 0xF) |
| 2676 | return False; /* NV is not allowed */ |
| 2677 | if (firstcond == 0xE && popcount32(mask) != 1) |
| 2678 | return False; /* if firstcond is AL then all the rest must be too */ |
| 2679 | |
| 2680 | UInt m3 = (mask >> 3) & 1; |
| 2681 | UInt m2 = (mask >> 2) & 1; |
| 2682 | UInt m1 = (mask >> 1) & 1; |
| 2683 | UInt m0 = (mask >> 0) & 1; |
| 2684 | |
| 2685 | UInt fc = (firstcond << 4) | 1/*in-IT-block*/; |
| 2686 | UInt ni = (0xE/*AL*/ << 4) | 0/*not-in-IT-block*/; |
| 2687 | |
| 2688 | if (m3 == 1 && (m2|m1|m0) == 0) { |
| 2689 | *itstate = (ni << 24) | (ni << 16) | (ni << 8) | fc; |
| 2690 | *itstate ^= 0xE0E0E0E0; |
| 2691 | return True; |
| 2692 | } |
| 2693 | |
| 2694 | if (m2 == 1 && (m1|m0) == 0) { |
| 2695 | *itstate = (ni << 24) | (ni << 16) | (setbit32(fc, 4, m3) << 8) | fc; |
| 2696 | *itstate ^= 0xE0E0E0E0; |
| 2697 | *ch1 = m3 == (firstcond & 1) ? 't' : 'e'; |
| 2698 | return True; |
| 2699 | } |
| 2700 | |
| 2701 | if (m1 == 1 && m0 == 0) { |
| 2702 | *itstate = (ni << 24) |
| 2703 | | (setbit32(fc, 4, m2) << 16) |
| 2704 | | (setbit32(fc, 4, m3) << 8) | fc; |
| 2705 | *itstate ^= 0xE0E0E0E0; |
| 2706 | *ch1 = m3 == (firstcond & 1) ? 't' : 'e'; |
| 2707 | *ch2 = m2 == (firstcond & 1) ? 't' : 'e'; |
| 2708 | return True; |
| 2709 | } |
| 2710 | |
| 2711 | if (m0 == 1) { |
| 2712 | *itstate = (setbit32(fc, 4, m1) << 24) |
| 2713 | | (setbit32(fc, 4, m2) << 16) |
| 2714 | | (setbit32(fc, 4, m3) << 8) | fc; |
| 2715 | *itstate ^= 0xE0E0E0E0; |
| 2716 | *ch1 = m3 == (firstcond & 1) ? 't' : 'e'; |
| 2717 | *ch2 = m2 == (firstcond & 1) ? 't' : 'e'; |
| 2718 | *ch3 = m1 == (firstcond & 1) ? 't' : 'e'; |
| 2719 | return True; |
| 2720 | } |
| 2721 | |
| 2722 | return False; |
| 2723 | } |
| 2724 | |
sewardj | 677e5af | 2010-09-02 21:02:47 +0000 | [diff] [blame] | 2725 | |
| 2726 | /* Generate IR to do 32-bit bit reversal, a la Hacker's Delight |
| 2727 | Chapter 7 Section 1. */ |
| 2728 | static IRTemp gen_BITREV ( IRTemp x0 ) |
| 2729 | { |
| 2730 | IRTemp x1 = newTemp(Ity_I32); |
| 2731 | IRTemp x2 = newTemp(Ity_I32); |
| 2732 | IRTemp x3 = newTemp(Ity_I32); |
| 2733 | IRTemp x4 = newTemp(Ity_I32); |
| 2734 | IRTemp x5 = newTemp(Ity_I32); |
| 2735 | UInt c1 = 0x55555555; |
| 2736 | UInt c2 = 0x33333333; |
| 2737 | UInt c3 = 0x0F0F0F0F; |
| 2738 | UInt c4 = 0x00FF00FF; |
| 2739 | UInt c5 = 0x0000FFFF; |
| 2740 | assign(x1, |
| 2741 | binop(Iop_Or32, |
| 2742 | binop(Iop_Shl32, |
| 2743 | binop(Iop_And32, mkexpr(x0), mkU32(c1)), |
| 2744 | mkU8(1)), |
| 2745 | binop(Iop_Shr32, |
| 2746 | binop(Iop_And32, mkexpr(x0), mkU32(~c1)), |
| 2747 | mkU8(1)) |
| 2748 | )); |
| 2749 | assign(x2, |
| 2750 | binop(Iop_Or32, |
| 2751 | binop(Iop_Shl32, |
| 2752 | binop(Iop_And32, mkexpr(x1), mkU32(c2)), |
| 2753 | mkU8(2)), |
| 2754 | binop(Iop_Shr32, |
| 2755 | binop(Iop_And32, mkexpr(x1), mkU32(~c2)), |
| 2756 | mkU8(2)) |
| 2757 | )); |
| 2758 | assign(x3, |
| 2759 | binop(Iop_Or32, |
| 2760 | binop(Iop_Shl32, |
| 2761 | binop(Iop_And32, mkexpr(x2), mkU32(c3)), |
| 2762 | mkU8(4)), |
| 2763 | binop(Iop_Shr32, |
| 2764 | binop(Iop_And32, mkexpr(x2), mkU32(~c3)), |
| 2765 | mkU8(4)) |
| 2766 | )); |
| 2767 | assign(x4, |
| 2768 | binop(Iop_Or32, |
| 2769 | binop(Iop_Shl32, |
| 2770 | binop(Iop_And32, mkexpr(x3), mkU32(c4)), |
| 2771 | mkU8(8)), |
| 2772 | binop(Iop_Shr32, |
| 2773 | binop(Iop_And32, mkexpr(x3), mkU32(~c4)), |
| 2774 | mkU8(8)) |
| 2775 | )); |
| 2776 | assign(x5, |
| 2777 | binop(Iop_Or32, |
| 2778 | binop(Iop_Shl32, |
| 2779 | binop(Iop_And32, mkexpr(x4), mkU32(c5)), |
| 2780 | mkU8(16)), |
| 2781 | binop(Iop_Shr32, |
| 2782 | binop(Iop_And32, mkexpr(x4), mkU32(~c5)), |
| 2783 | mkU8(16)) |
| 2784 | )); |
| 2785 | return x5; |
| 2786 | } |
| 2787 | |
| 2788 | |
sewardj | 27312d3 | 2010-09-26 00:48:41 +0000 | [diff] [blame] | 2789 | /* Generate IR to do rearrange bytes 3:2:1:0 in a word in to the order |
| 2790 | 0:1:2:3 (aka byte-swap). */ |
| 2791 | static IRTemp gen_REV ( IRTemp arg ) |
| 2792 | { |
| 2793 | IRTemp res = newTemp(Ity_I32); |
| 2794 | assign(res, |
| 2795 | binop(Iop_Or32, |
| 2796 | binop(Iop_Shl32, mkexpr(arg), mkU8(24)), |
| 2797 | binop(Iop_Or32, |
| 2798 | binop(Iop_And32, binop(Iop_Shl32, mkexpr(arg), mkU8(8)), |
| 2799 | mkU32(0x00FF0000)), |
| 2800 | binop(Iop_Or32, |
| 2801 | binop(Iop_And32, binop(Iop_Shr32, mkexpr(arg), mkU8(8)), |
| 2802 | mkU32(0x0000FF00)), |
| 2803 | binop(Iop_And32, binop(Iop_Shr32, mkexpr(arg), mkU8(24)), |
| 2804 | mkU32(0x000000FF) ) |
| 2805 | )))); |
| 2806 | return res; |
| 2807 | } |
| 2808 | |
| 2809 | |
| 2810 | /* Generate IR to do rearrange bytes 3:2:1:0 in a word in to the order |
| 2811 | 2:3:0:1 (swap within lo and hi halves). */ |
| 2812 | static IRTemp gen_REV16 ( IRTemp arg ) |
| 2813 | { |
| 2814 | IRTemp res = newTemp(Ity_I32); |
| 2815 | assign(res, |
| 2816 | binop(Iop_Or32, |
| 2817 | binop(Iop_And32, |
| 2818 | binop(Iop_Shl32, mkexpr(arg), mkU8(8)), |
| 2819 | mkU32(0xFF00FF00)), |
| 2820 | binop(Iop_And32, |
| 2821 | binop(Iop_Shr32, mkexpr(arg), mkU8(8)), |
| 2822 | mkU32(0x00FF00FF)))); |
| 2823 | return res; |
| 2824 | } |
| 2825 | |
| 2826 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2827 | /*------------------------------------------------------------*/ |
| 2828 | /*--- Advanced SIMD (NEON) instructions ---*/ |
| 2829 | /*------------------------------------------------------------*/ |
| 2830 | |
| 2831 | /*------------------------------------------------------------*/ |
| 2832 | /*--- NEON data processing ---*/ |
| 2833 | /*------------------------------------------------------------*/ |
| 2834 | |
| 2835 | /* For all NEON DP ops, we use the normal scheme to handle conditional |
| 2836 | writes to registers -- pass in condT and hand that on to the |
| 2837 | put*Reg functions. In ARM mode condT is always IRTemp_INVALID |
| 2838 | since NEON is unconditional for ARM. In Thumb mode condT is |
| 2839 | derived from the ITSTATE shift register in the normal way. */ |
| 2840 | |
| 2841 | static |
| 2842 | UInt get_neon_d_regno(UInt theInstr) |
| 2843 | { |
| 2844 | UInt x = ((theInstr >> 18) & 0x10) | ((theInstr >> 12) & 0xF); |
| 2845 | if (theInstr & 0x40) { |
| 2846 | if (x & 1) { |
| 2847 | x = x + 0x100; |
| 2848 | } else { |
| 2849 | x = x >> 1; |
| 2850 | } |
| 2851 | } |
| 2852 | return x; |
| 2853 | } |
| 2854 | |
| 2855 | static |
| 2856 | UInt get_neon_n_regno(UInt theInstr) |
| 2857 | { |
| 2858 | UInt x = ((theInstr >> 3) & 0x10) | ((theInstr >> 16) & 0xF); |
| 2859 | if (theInstr & 0x40) { |
| 2860 | if (x & 1) { |
| 2861 | x = x + 0x100; |
| 2862 | } else { |
| 2863 | x = x >> 1; |
| 2864 | } |
| 2865 | } |
| 2866 | return x; |
| 2867 | } |
| 2868 | |
| 2869 | static |
| 2870 | UInt get_neon_m_regno(UInt theInstr) |
| 2871 | { |
| 2872 | UInt x = ((theInstr >> 1) & 0x10) | (theInstr & 0xF); |
| 2873 | if (theInstr & 0x40) { |
| 2874 | if (x & 1) { |
| 2875 | x = x + 0x100; |
| 2876 | } else { |
| 2877 | x = x >> 1; |
| 2878 | } |
| 2879 | } |
| 2880 | return x; |
| 2881 | } |
| 2882 | |
| 2883 | static |
| 2884 | Bool dis_neon_vext ( UInt theInstr, IRTemp condT ) |
| 2885 | { |
| 2886 | UInt dreg = get_neon_d_regno(theInstr); |
| 2887 | UInt mreg = get_neon_m_regno(theInstr); |
| 2888 | UInt nreg = get_neon_n_regno(theInstr); |
| 2889 | UInt imm4 = (theInstr >> 8) & 0xf; |
| 2890 | UInt Q = (theInstr >> 6) & 1; |
| 2891 | HChar reg_t = Q ? 'q' : 'd'; |
| 2892 | |
| 2893 | if (Q) { |
sewardj | e6b9bd9 | 2014-09-01 11:32:47 +0000 | [diff] [blame] | 2894 | putQReg(dreg, triop(Iop_SliceV128, /*hiV128*/getQReg(mreg), |
| 2895 | /*loV128*/getQReg(nreg), mkU8(imm4)), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2896 | } else { |
sewardj | e6b9bd9 | 2014-09-01 11:32:47 +0000 | [diff] [blame] | 2897 | putDRegI64(dreg, triop(Iop_Slice64, /*hiI64*/getDRegI64(mreg), |
| 2898 | /*loI64*/getDRegI64(nreg), mkU8(imm4)), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2899 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 2900 | DIP("vext.8 %c%u, %c%u, %c%u, #%u\n", reg_t, dreg, reg_t, nreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2901 | reg_t, mreg, imm4); |
| 2902 | return True; |
| 2903 | } |
| 2904 | |
sewardj | e84eeb4 | 2014-03-15 11:41:39 +0000 | [diff] [blame] | 2905 | /* Generate specific vector FP binary ops, possibly with a fake |
| 2906 | rounding mode as required by the primop. */ |
| 2907 | static |
| 2908 | IRExpr* binop_w_fake_RM ( IROp op, IRExpr* argL, IRExpr* argR ) |
| 2909 | { |
| 2910 | switch (op) { |
| 2911 | case Iop_Add32Fx4: |
| 2912 | case Iop_Sub32Fx4: |
| 2913 | case Iop_Mul32Fx4: |
| 2914 | return triop(op, get_FAKE_roundingmode(), argL, argR ); |
| 2915 | case Iop_Add32x4: case Iop_Add16x8: |
| 2916 | case Iop_Sub32x4: case Iop_Sub16x8: |
| 2917 | case Iop_Mul32x4: case Iop_Mul16x8: |
| 2918 | case Iop_Mul32x2: case Iop_Mul16x4: |
| 2919 | case Iop_Add32Fx2: |
| 2920 | case Iop_Sub32Fx2: |
| 2921 | case Iop_Mul32Fx2: |
| 2922 | case Iop_PwAdd32Fx2: |
| 2923 | return binop(op, argL, argR); |
| 2924 | default: |
| 2925 | ppIROp(op); |
| 2926 | vassert(0); |
| 2927 | } |
| 2928 | } |
| 2929 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 2930 | /* VTBL, VTBX */ |
| 2931 | static |
| 2932 | Bool dis_neon_vtb ( UInt theInstr, IRTemp condT ) |
| 2933 | { |
| 2934 | UInt op = (theInstr >> 6) & 1; |
| 2935 | UInt dreg = get_neon_d_regno(theInstr & ~(1 << 6)); |
| 2936 | UInt nreg = get_neon_n_regno(theInstr & ~(1 << 6)); |
| 2937 | UInt mreg = get_neon_m_regno(theInstr & ~(1 << 6)); |
| 2938 | UInt len = (theInstr >> 8) & 3; |
| 2939 | Int i; |
| 2940 | IROp cmp; |
| 2941 | ULong imm; |
| 2942 | IRTemp arg_l; |
| 2943 | IRTemp old_mask, new_mask, cur_mask; |
| 2944 | IRTemp old_res, new_res; |
| 2945 | IRTemp old_arg, new_arg; |
| 2946 | |
| 2947 | if (dreg >= 0x100 || mreg >= 0x100 || nreg >= 0x100) |
| 2948 | return False; |
| 2949 | if (nreg + len > 31) |
| 2950 | return False; |
| 2951 | |
| 2952 | cmp = Iop_CmpGT8Ux8; |
| 2953 | |
| 2954 | old_mask = newTemp(Ity_I64); |
| 2955 | old_res = newTemp(Ity_I64); |
| 2956 | old_arg = newTemp(Ity_I64); |
| 2957 | assign(old_mask, mkU64(0)); |
| 2958 | assign(old_res, mkU64(0)); |
| 2959 | assign(old_arg, getDRegI64(mreg)); |
| 2960 | imm = 8; |
| 2961 | imm = (imm << 8) | imm; |
| 2962 | imm = (imm << 16) | imm; |
| 2963 | imm = (imm << 32) | imm; |
| 2964 | |
| 2965 | for (i = 0; i <= len; i++) { |
| 2966 | arg_l = newTemp(Ity_I64); |
| 2967 | new_mask = newTemp(Ity_I64); |
| 2968 | cur_mask = newTemp(Ity_I64); |
| 2969 | new_res = newTemp(Ity_I64); |
| 2970 | new_arg = newTemp(Ity_I64); |
| 2971 | assign(arg_l, getDRegI64(nreg+i)); |
| 2972 | assign(new_arg, binop(Iop_Sub8x8, mkexpr(old_arg), mkU64(imm))); |
| 2973 | assign(cur_mask, binop(cmp, mkU64(imm), mkexpr(old_arg))); |
| 2974 | assign(new_mask, binop(Iop_Or64, mkexpr(old_mask), mkexpr(cur_mask))); |
| 2975 | assign(new_res, binop(Iop_Or64, |
| 2976 | mkexpr(old_res), |
| 2977 | binop(Iop_And64, |
| 2978 | binop(Iop_Perm8x8, |
| 2979 | mkexpr(arg_l), |
| 2980 | binop(Iop_And64, |
| 2981 | mkexpr(old_arg), |
| 2982 | mkexpr(cur_mask))), |
| 2983 | mkexpr(cur_mask)))); |
| 2984 | |
| 2985 | old_arg = new_arg; |
| 2986 | old_mask = new_mask; |
| 2987 | old_res = new_res; |
| 2988 | } |
| 2989 | if (op) { |
| 2990 | new_res = newTemp(Ity_I64); |
| 2991 | assign(new_res, binop(Iop_Or64, |
| 2992 | binop(Iop_And64, |
| 2993 | getDRegI64(dreg), |
| 2994 | unop(Iop_Not64, mkexpr(old_mask))), |
| 2995 | mkexpr(old_res))); |
| 2996 | old_res = new_res; |
| 2997 | } |
| 2998 | |
| 2999 | putDRegI64(dreg, mkexpr(old_res), condT); |
| 3000 | DIP("vtb%c.8 d%u, {", op ? 'x' : 'l', dreg); |
| 3001 | if (len > 0) { |
| 3002 | DIP("d%u-d%u", nreg, nreg + len); |
| 3003 | } else { |
| 3004 | DIP("d%u", nreg); |
| 3005 | } |
| 3006 | DIP("}, d%u\n", mreg); |
| 3007 | return True; |
| 3008 | } |
| 3009 | |
| 3010 | /* VDUP (scalar) */ |
| 3011 | static |
| 3012 | Bool dis_neon_vdup ( UInt theInstr, IRTemp condT ) |
| 3013 | { |
| 3014 | UInt Q = (theInstr >> 6) & 1; |
| 3015 | UInt dreg = ((theInstr >> 18) & 0x10) | ((theInstr >> 12) & 0xF); |
| 3016 | UInt mreg = ((theInstr >> 1) & 0x10) | (theInstr & 0xF); |
| 3017 | UInt imm4 = (theInstr >> 16) & 0xF; |
| 3018 | UInt index; |
| 3019 | UInt size; |
| 3020 | IRTemp arg_m; |
| 3021 | IRTemp res; |
| 3022 | IROp op, op2; |
| 3023 | |
| 3024 | if ((imm4 == 0) || (imm4 == 8)) |
| 3025 | return False; |
| 3026 | if ((Q == 1) && ((dreg & 1) == 1)) |
| 3027 | return False; |
| 3028 | if (Q) |
| 3029 | dreg >>= 1; |
| 3030 | arg_m = newTemp(Ity_I64); |
| 3031 | assign(arg_m, getDRegI64(mreg)); |
| 3032 | if (Q) |
| 3033 | res = newTemp(Ity_V128); |
| 3034 | else |
| 3035 | res = newTemp(Ity_I64); |
| 3036 | if ((imm4 & 1) == 1) { |
| 3037 | op = Q ? Iop_Dup8x16 : Iop_Dup8x8; |
| 3038 | op2 = Iop_GetElem8x8; |
| 3039 | index = imm4 >> 1; |
| 3040 | size = 8; |
| 3041 | } else if ((imm4 & 3) == 2) { |
| 3042 | op = Q ? Iop_Dup16x8 : Iop_Dup16x4; |
| 3043 | op2 = Iop_GetElem16x4; |
| 3044 | index = imm4 >> 2; |
| 3045 | size = 16; |
| 3046 | } else if ((imm4 & 7) == 4) { |
| 3047 | op = Q ? Iop_Dup32x4 : Iop_Dup32x2; |
| 3048 | op2 = Iop_GetElem32x2; |
| 3049 | index = imm4 >> 3; |
| 3050 | size = 32; |
sewardj | 9dbbd7b | 2010-08-22 18:24:51 +0000 | [diff] [blame] | 3051 | } else { |
| 3052 | return False; // can this ever happen? |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3053 | } |
| 3054 | assign(res, unop(op, binop(op2, mkexpr(arg_m), mkU8(index)))); |
| 3055 | if (Q) { |
| 3056 | putQReg(dreg, mkexpr(res), condT); |
| 3057 | } else { |
| 3058 | putDRegI64(dreg, mkexpr(res), condT); |
| 3059 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3060 | DIP("vdup.%u %c%u, d%u[%u]\n", size, Q ? 'q' : 'd', dreg, mreg, index); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3061 | return True; |
| 3062 | } |
| 3063 | |
| 3064 | /* A7.4.1 Three registers of the same length */ |
| 3065 | static |
| 3066 | Bool dis_neon_data_3same ( UInt theInstr, IRTemp condT ) |
| 3067 | { |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 3068 | /* In paths where this returns False, indicating a non-decodable |
| 3069 | instruction, there may still be some IR assignments to temporaries |
| 3070 | generated. This is inconvenient but harmless, and the post-front-end |
| 3071 | IR optimisation pass will just remove them anyway. So there's no |
| 3072 | effort made here to tidy it up. |
| 3073 | */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3074 | UInt Q = (theInstr >> 6) & 1; |
| 3075 | UInt dreg = get_neon_d_regno(theInstr); |
| 3076 | UInt nreg = get_neon_n_regno(theInstr); |
| 3077 | UInt mreg = get_neon_m_regno(theInstr); |
| 3078 | UInt A = (theInstr >> 8) & 0xF; |
| 3079 | UInt B = (theInstr >> 4) & 1; |
| 3080 | UInt C = (theInstr >> 20) & 0x3; |
| 3081 | UInt U = (theInstr >> 24) & 1; |
| 3082 | UInt size = C; |
| 3083 | |
| 3084 | IRTemp arg_n; |
| 3085 | IRTemp arg_m; |
| 3086 | IRTemp res; |
| 3087 | |
| 3088 | if (Q) { |
| 3089 | arg_n = newTemp(Ity_V128); |
| 3090 | arg_m = newTemp(Ity_V128); |
| 3091 | res = newTemp(Ity_V128); |
| 3092 | assign(arg_n, getQReg(nreg)); |
| 3093 | assign(arg_m, getQReg(mreg)); |
| 3094 | } else { |
| 3095 | arg_n = newTemp(Ity_I64); |
| 3096 | arg_m = newTemp(Ity_I64); |
| 3097 | res = newTemp(Ity_I64); |
| 3098 | assign(arg_n, getDRegI64(nreg)); |
| 3099 | assign(arg_m, getDRegI64(mreg)); |
| 3100 | } |
| 3101 | |
| 3102 | switch(A) { |
| 3103 | case 0: |
| 3104 | if (B == 0) { |
| 3105 | /* VHADD */ |
| 3106 | ULong imm = 0; |
| 3107 | IRExpr *imm_val; |
| 3108 | IROp addOp; |
| 3109 | IROp andOp; |
| 3110 | IROp shOp; |
florian | 5df8ab0 | 2012-10-13 19:34:19 +0000 | [diff] [blame] | 3111 | HChar regType = Q ? 'q' : 'd'; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3112 | |
| 3113 | if (size == 3) |
| 3114 | return False; |
| 3115 | switch(size) { |
| 3116 | case 0: imm = 0x101010101010101LL; break; |
| 3117 | case 1: imm = 0x1000100010001LL; break; |
| 3118 | case 2: imm = 0x100000001LL; break; |
| 3119 | default: vassert(0); |
| 3120 | } |
| 3121 | if (Q) { |
| 3122 | imm_val = binop(Iop_64HLtoV128, mkU64(imm), mkU64(imm)); |
| 3123 | andOp = Iop_AndV128; |
| 3124 | } else { |
| 3125 | imm_val = mkU64(imm); |
| 3126 | andOp = Iop_And64; |
| 3127 | } |
| 3128 | if (U) { |
| 3129 | switch(size) { |
| 3130 | case 0: |
| 3131 | addOp = Q ? Iop_Add8x16 : Iop_Add8x8; |
| 3132 | shOp = Q ? Iop_ShrN8x16 : Iop_ShrN8x8; |
| 3133 | break; |
| 3134 | case 1: |
| 3135 | addOp = Q ? Iop_Add16x8 : Iop_Add16x4; |
| 3136 | shOp = Q ? Iop_ShrN16x8 : Iop_ShrN16x4; |
| 3137 | break; |
| 3138 | case 2: |
| 3139 | addOp = Q ? Iop_Add32x4 : Iop_Add32x2; |
| 3140 | shOp = Q ? Iop_ShrN32x4 : Iop_ShrN32x2; |
| 3141 | break; |
| 3142 | default: |
| 3143 | vassert(0); |
| 3144 | } |
| 3145 | } else { |
| 3146 | switch(size) { |
| 3147 | case 0: |
| 3148 | addOp = Q ? Iop_Add8x16 : Iop_Add8x8; |
| 3149 | shOp = Q ? Iop_SarN8x16 : Iop_SarN8x8; |
| 3150 | break; |
| 3151 | case 1: |
| 3152 | addOp = Q ? Iop_Add16x8 : Iop_Add16x4; |
| 3153 | shOp = Q ? Iop_SarN16x8 : Iop_SarN16x4; |
| 3154 | break; |
| 3155 | case 2: |
| 3156 | addOp = Q ? Iop_Add32x4 : Iop_Add32x2; |
| 3157 | shOp = Q ? Iop_SarN32x4 : Iop_SarN32x2; |
| 3158 | break; |
| 3159 | default: |
| 3160 | vassert(0); |
| 3161 | } |
| 3162 | } |
| 3163 | assign(res, |
| 3164 | binop(addOp, |
| 3165 | binop(addOp, |
| 3166 | binop(shOp, mkexpr(arg_m), mkU8(1)), |
| 3167 | binop(shOp, mkexpr(arg_n), mkU8(1))), |
| 3168 | binop(shOp, |
| 3169 | binop(addOp, |
| 3170 | binop(andOp, mkexpr(arg_m), imm_val), |
| 3171 | binop(andOp, mkexpr(arg_n), imm_val)), |
| 3172 | mkU8(1)))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3173 | DIP("vhadd.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3174 | U ? 'u' : 's', 8 << size, regType, |
| 3175 | dreg, regType, nreg, regType, mreg); |
| 3176 | } else { |
| 3177 | /* VQADD */ |
| 3178 | IROp op, op2; |
| 3179 | IRTemp tmp; |
florian | 5df8ab0 | 2012-10-13 19:34:19 +0000 | [diff] [blame] | 3180 | HChar reg_t = Q ? 'q' : 'd'; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3181 | if (Q) { |
| 3182 | switch (size) { |
| 3183 | case 0: |
| 3184 | op = U ? Iop_QAdd8Ux16 : Iop_QAdd8Sx16; |
| 3185 | op2 = Iop_Add8x16; |
| 3186 | break; |
| 3187 | case 1: |
| 3188 | op = U ? Iop_QAdd16Ux8 : Iop_QAdd16Sx8; |
| 3189 | op2 = Iop_Add16x8; |
| 3190 | break; |
| 3191 | case 2: |
| 3192 | op = U ? Iop_QAdd32Ux4 : Iop_QAdd32Sx4; |
| 3193 | op2 = Iop_Add32x4; |
| 3194 | break; |
| 3195 | case 3: |
| 3196 | op = U ? Iop_QAdd64Ux2 : Iop_QAdd64Sx2; |
| 3197 | op2 = Iop_Add64x2; |
| 3198 | break; |
| 3199 | default: |
| 3200 | vassert(0); |
| 3201 | } |
| 3202 | } else { |
| 3203 | switch (size) { |
| 3204 | case 0: |
| 3205 | op = U ? Iop_QAdd8Ux8 : Iop_QAdd8Sx8; |
| 3206 | op2 = Iop_Add8x8; |
| 3207 | break; |
| 3208 | case 1: |
| 3209 | op = U ? Iop_QAdd16Ux4 : Iop_QAdd16Sx4; |
| 3210 | op2 = Iop_Add16x4; |
| 3211 | break; |
| 3212 | case 2: |
| 3213 | op = U ? Iop_QAdd32Ux2 : Iop_QAdd32Sx2; |
| 3214 | op2 = Iop_Add32x2; |
| 3215 | break; |
| 3216 | case 3: |
| 3217 | op = U ? Iop_QAdd64Ux1 : Iop_QAdd64Sx1; |
| 3218 | op2 = Iop_Add64; |
| 3219 | break; |
| 3220 | default: |
| 3221 | vassert(0); |
| 3222 | } |
| 3223 | } |
| 3224 | if (Q) { |
| 3225 | tmp = newTemp(Ity_V128); |
| 3226 | } else { |
| 3227 | tmp = newTemp(Ity_I64); |
| 3228 | } |
| 3229 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3230 | assign(tmp, binop(op2, mkexpr(arg_n), mkexpr(arg_m))); |
| 3231 | setFlag_QC(mkexpr(res), mkexpr(tmp), Q, condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3232 | DIP("vqadd.%c%d %c%u %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3233 | U ? 'u' : 's', |
| 3234 | 8 << size, reg_t, dreg, reg_t, nreg, reg_t, mreg); |
| 3235 | } |
| 3236 | break; |
| 3237 | case 1: |
| 3238 | if (B == 0) { |
| 3239 | /* VRHADD */ |
| 3240 | /* VRHADD C, A, B ::= |
| 3241 | C = (A >> 1) + (B >> 1) + (((A & 1) + (B & 1) + 1) >> 1) */ |
| 3242 | IROp shift_op, add_op; |
| 3243 | IRTemp cc; |
| 3244 | ULong one = 1; |
| 3245 | HChar reg_t = Q ? 'q' : 'd'; |
| 3246 | switch (size) { |
| 3247 | case 0: one = (one << 8) | one; /* fall through */ |
| 3248 | case 1: one = (one << 16) | one; /* fall through */ |
| 3249 | case 2: one = (one << 32) | one; break; |
| 3250 | case 3: return False; |
| 3251 | default: vassert(0); |
| 3252 | } |
| 3253 | if (Q) { |
| 3254 | switch (size) { |
| 3255 | case 0: |
| 3256 | shift_op = U ? Iop_ShrN8x16 : Iop_SarN8x16; |
| 3257 | add_op = Iop_Add8x16; |
| 3258 | break; |
| 3259 | case 1: |
| 3260 | shift_op = U ? Iop_ShrN16x8 : Iop_SarN16x8; |
| 3261 | add_op = Iop_Add16x8; |
| 3262 | break; |
| 3263 | case 2: |
| 3264 | shift_op = U ? Iop_ShrN32x4 : Iop_SarN32x4; |
| 3265 | add_op = Iop_Add32x4; |
| 3266 | break; |
| 3267 | case 3: |
| 3268 | return False; |
| 3269 | default: |
| 3270 | vassert(0); |
| 3271 | } |
| 3272 | } else { |
| 3273 | switch (size) { |
| 3274 | case 0: |
| 3275 | shift_op = U ? Iop_ShrN8x8 : Iop_SarN8x8; |
| 3276 | add_op = Iop_Add8x8; |
| 3277 | break; |
| 3278 | case 1: |
| 3279 | shift_op = U ? Iop_ShrN16x4 : Iop_SarN16x4; |
| 3280 | add_op = Iop_Add16x4; |
| 3281 | break; |
| 3282 | case 2: |
| 3283 | shift_op = U ? Iop_ShrN32x2 : Iop_SarN32x2; |
| 3284 | add_op = Iop_Add32x2; |
| 3285 | break; |
| 3286 | case 3: |
| 3287 | return False; |
| 3288 | default: |
| 3289 | vassert(0); |
| 3290 | } |
| 3291 | } |
| 3292 | if (Q) { |
| 3293 | cc = newTemp(Ity_V128); |
| 3294 | assign(cc, binop(shift_op, |
| 3295 | binop(add_op, |
| 3296 | binop(add_op, |
| 3297 | binop(Iop_AndV128, |
| 3298 | mkexpr(arg_n), |
| 3299 | binop(Iop_64HLtoV128, |
| 3300 | mkU64(one), |
| 3301 | mkU64(one))), |
| 3302 | binop(Iop_AndV128, |
| 3303 | mkexpr(arg_m), |
| 3304 | binop(Iop_64HLtoV128, |
| 3305 | mkU64(one), |
| 3306 | mkU64(one)))), |
| 3307 | binop(Iop_64HLtoV128, |
| 3308 | mkU64(one), |
| 3309 | mkU64(one))), |
| 3310 | mkU8(1))); |
| 3311 | assign(res, binop(add_op, |
| 3312 | binop(add_op, |
| 3313 | binop(shift_op, |
| 3314 | mkexpr(arg_n), |
| 3315 | mkU8(1)), |
| 3316 | binop(shift_op, |
| 3317 | mkexpr(arg_m), |
| 3318 | mkU8(1))), |
| 3319 | mkexpr(cc))); |
| 3320 | } else { |
| 3321 | cc = newTemp(Ity_I64); |
| 3322 | assign(cc, binop(shift_op, |
| 3323 | binop(add_op, |
| 3324 | binop(add_op, |
| 3325 | binop(Iop_And64, |
| 3326 | mkexpr(arg_n), |
| 3327 | mkU64(one)), |
| 3328 | binop(Iop_And64, |
| 3329 | mkexpr(arg_m), |
| 3330 | mkU64(one))), |
| 3331 | mkU64(one)), |
| 3332 | mkU8(1))); |
| 3333 | assign(res, binop(add_op, |
| 3334 | binop(add_op, |
| 3335 | binop(shift_op, |
| 3336 | mkexpr(arg_n), |
| 3337 | mkU8(1)), |
| 3338 | binop(shift_op, |
| 3339 | mkexpr(arg_m), |
| 3340 | mkU8(1))), |
| 3341 | mkexpr(cc))); |
| 3342 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3343 | DIP("vrhadd.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3344 | U ? 'u' : 's', |
| 3345 | 8 << size, reg_t, dreg, reg_t, nreg, reg_t, mreg); |
| 3346 | } else { |
| 3347 | if (U == 0) { |
| 3348 | switch(C) { |
| 3349 | case 0: { |
| 3350 | /* VAND */ |
| 3351 | HChar reg_t = Q ? 'q' : 'd'; |
| 3352 | if (Q) { |
| 3353 | assign(res, binop(Iop_AndV128, mkexpr(arg_n), |
| 3354 | mkexpr(arg_m))); |
| 3355 | } else { |
| 3356 | assign(res, binop(Iop_And64, mkexpr(arg_n), |
| 3357 | mkexpr(arg_m))); |
| 3358 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3359 | DIP("vand %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3360 | reg_t, dreg, reg_t, nreg, reg_t, mreg); |
| 3361 | break; |
| 3362 | } |
| 3363 | case 1: { |
| 3364 | /* VBIC */ |
| 3365 | HChar reg_t = Q ? 'q' : 'd'; |
| 3366 | if (Q) { |
| 3367 | assign(res, binop(Iop_AndV128,mkexpr(arg_n), |
| 3368 | unop(Iop_NotV128, mkexpr(arg_m)))); |
| 3369 | } else { |
| 3370 | assign(res, binop(Iop_And64, mkexpr(arg_n), |
| 3371 | unop(Iop_Not64, mkexpr(arg_m)))); |
| 3372 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3373 | DIP("vbic %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3374 | reg_t, dreg, reg_t, nreg, reg_t, mreg); |
| 3375 | break; |
| 3376 | } |
| 3377 | case 2: |
| 3378 | if ( nreg != mreg) { |
| 3379 | /* VORR */ |
| 3380 | HChar reg_t = Q ? 'q' : 'd'; |
| 3381 | if (Q) { |
| 3382 | assign(res, binop(Iop_OrV128, mkexpr(arg_n), |
| 3383 | mkexpr(arg_m))); |
| 3384 | } else { |
| 3385 | assign(res, binop(Iop_Or64, mkexpr(arg_n), |
| 3386 | mkexpr(arg_m))); |
| 3387 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3388 | DIP("vorr %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3389 | reg_t, dreg, reg_t, nreg, reg_t, mreg); |
| 3390 | } else { |
| 3391 | /* VMOV */ |
| 3392 | HChar reg_t = Q ? 'q' : 'd'; |
| 3393 | assign(res, mkexpr(arg_m)); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3394 | DIP("vmov %c%u, %c%u\n", reg_t, dreg, reg_t, mreg); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3395 | } |
| 3396 | break; |
| 3397 | case 3:{ |
| 3398 | /* VORN */ |
| 3399 | HChar reg_t = Q ? 'q' : 'd'; |
| 3400 | if (Q) { |
| 3401 | assign(res, binop(Iop_OrV128,mkexpr(arg_n), |
| 3402 | unop(Iop_NotV128, mkexpr(arg_m)))); |
| 3403 | } else { |
| 3404 | assign(res, binop(Iop_Or64, mkexpr(arg_n), |
| 3405 | unop(Iop_Not64, mkexpr(arg_m)))); |
| 3406 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3407 | DIP("vorn %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3408 | reg_t, dreg, reg_t, nreg, reg_t, mreg); |
| 3409 | break; |
| 3410 | } |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 3411 | default: |
| 3412 | vassert(0); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3413 | } |
| 3414 | } else { |
| 3415 | switch(C) { |
| 3416 | case 0: |
| 3417 | /* VEOR (XOR) */ |
| 3418 | if (Q) { |
| 3419 | assign(res, binop(Iop_XorV128, mkexpr(arg_n), |
| 3420 | mkexpr(arg_m))); |
| 3421 | } else { |
| 3422 | assign(res, binop(Iop_Xor64, mkexpr(arg_n), |
| 3423 | mkexpr(arg_m))); |
| 3424 | } |
| 3425 | DIP("veor %c%u, %c%u, %c%u\n", Q ? 'q' : 'd', dreg, |
| 3426 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 3427 | break; |
| 3428 | case 1: |
| 3429 | /* VBSL */ |
| 3430 | if (Q) { |
| 3431 | IRTemp reg_d = newTemp(Ity_V128); |
| 3432 | assign(reg_d, getQReg(dreg)); |
| 3433 | assign(res, |
| 3434 | binop(Iop_OrV128, |
| 3435 | binop(Iop_AndV128, mkexpr(arg_n), |
| 3436 | mkexpr(reg_d)), |
| 3437 | binop(Iop_AndV128, |
| 3438 | mkexpr(arg_m), |
| 3439 | unop(Iop_NotV128, |
| 3440 | mkexpr(reg_d)) ) ) ); |
| 3441 | } else { |
| 3442 | IRTemp reg_d = newTemp(Ity_I64); |
| 3443 | assign(reg_d, getDRegI64(dreg)); |
| 3444 | assign(res, |
| 3445 | binop(Iop_Or64, |
| 3446 | binop(Iop_And64, mkexpr(arg_n), |
| 3447 | mkexpr(reg_d)), |
| 3448 | binop(Iop_And64, |
| 3449 | mkexpr(arg_m), |
| 3450 | unop(Iop_Not64, mkexpr(reg_d))))); |
| 3451 | } |
| 3452 | DIP("vbsl %c%u, %c%u, %c%u\n", |
| 3453 | Q ? 'q' : 'd', dreg, |
| 3454 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 3455 | break; |
| 3456 | case 2: |
| 3457 | /* VBIT */ |
| 3458 | if (Q) { |
| 3459 | IRTemp reg_d = newTemp(Ity_V128); |
| 3460 | assign(reg_d, getQReg(dreg)); |
| 3461 | assign(res, |
| 3462 | binop(Iop_OrV128, |
| 3463 | binop(Iop_AndV128, mkexpr(arg_n), |
| 3464 | mkexpr(arg_m)), |
| 3465 | binop(Iop_AndV128, |
| 3466 | mkexpr(reg_d), |
| 3467 | unop(Iop_NotV128, mkexpr(arg_m))))); |
| 3468 | } else { |
| 3469 | IRTemp reg_d = newTemp(Ity_I64); |
| 3470 | assign(reg_d, getDRegI64(dreg)); |
| 3471 | assign(res, |
| 3472 | binop(Iop_Or64, |
| 3473 | binop(Iop_And64, mkexpr(arg_n), |
| 3474 | mkexpr(arg_m)), |
| 3475 | binop(Iop_And64, |
| 3476 | mkexpr(reg_d), |
| 3477 | unop(Iop_Not64, mkexpr(arg_m))))); |
| 3478 | } |
| 3479 | DIP("vbit %c%u, %c%u, %c%u\n", |
| 3480 | Q ? 'q' : 'd', dreg, |
| 3481 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 3482 | break; |
| 3483 | case 3: |
| 3484 | /* VBIF */ |
| 3485 | if (Q) { |
| 3486 | IRTemp reg_d = newTemp(Ity_V128); |
| 3487 | assign(reg_d, getQReg(dreg)); |
| 3488 | assign(res, |
| 3489 | binop(Iop_OrV128, |
| 3490 | binop(Iop_AndV128, mkexpr(reg_d), |
| 3491 | mkexpr(arg_m)), |
| 3492 | binop(Iop_AndV128, |
| 3493 | mkexpr(arg_n), |
| 3494 | unop(Iop_NotV128, mkexpr(arg_m))))); |
| 3495 | } else { |
| 3496 | IRTemp reg_d = newTemp(Ity_I64); |
| 3497 | assign(reg_d, getDRegI64(dreg)); |
| 3498 | assign(res, |
| 3499 | binop(Iop_Or64, |
| 3500 | binop(Iop_And64, mkexpr(reg_d), |
| 3501 | mkexpr(arg_m)), |
| 3502 | binop(Iop_And64, |
| 3503 | mkexpr(arg_n), |
| 3504 | unop(Iop_Not64, mkexpr(arg_m))))); |
| 3505 | } |
| 3506 | DIP("vbif %c%u, %c%u, %c%u\n", |
| 3507 | Q ? 'q' : 'd', dreg, |
| 3508 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 3509 | break; |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 3510 | default: |
| 3511 | vassert(0); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3512 | } |
| 3513 | } |
| 3514 | } |
| 3515 | break; |
| 3516 | case 2: |
| 3517 | if (B == 0) { |
| 3518 | /* VHSUB */ |
| 3519 | /* (A >> 1) - (B >> 1) - (NOT (A) & B & 1) */ |
| 3520 | ULong imm = 0; |
| 3521 | IRExpr *imm_val; |
| 3522 | IROp subOp; |
| 3523 | IROp notOp; |
| 3524 | IROp andOp; |
| 3525 | IROp shOp; |
| 3526 | if (size == 3) |
| 3527 | return False; |
| 3528 | switch(size) { |
| 3529 | case 0: imm = 0x101010101010101LL; break; |
| 3530 | case 1: imm = 0x1000100010001LL; break; |
| 3531 | case 2: imm = 0x100000001LL; break; |
| 3532 | default: vassert(0); |
| 3533 | } |
| 3534 | if (Q) { |
| 3535 | imm_val = binop(Iop_64HLtoV128, mkU64(imm), mkU64(imm)); |
| 3536 | andOp = Iop_AndV128; |
| 3537 | notOp = Iop_NotV128; |
| 3538 | } else { |
| 3539 | imm_val = mkU64(imm); |
| 3540 | andOp = Iop_And64; |
| 3541 | notOp = Iop_Not64; |
| 3542 | } |
| 3543 | if (U) { |
| 3544 | switch(size) { |
| 3545 | case 0: |
| 3546 | subOp = Q ? Iop_Sub8x16 : Iop_Sub8x8; |
| 3547 | shOp = Q ? Iop_ShrN8x16 : Iop_ShrN8x8; |
| 3548 | break; |
| 3549 | case 1: |
| 3550 | subOp = Q ? Iop_Sub16x8 : Iop_Sub16x4; |
| 3551 | shOp = Q ? Iop_ShrN16x8 : Iop_ShrN16x4; |
| 3552 | break; |
| 3553 | case 2: |
| 3554 | subOp = Q ? Iop_Sub32x4 : Iop_Sub32x2; |
| 3555 | shOp = Q ? Iop_ShrN32x4 : Iop_ShrN32x2; |
| 3556 | break; |
| 3557 | default: |
| 3558 | vassert(0); |
| 3559 | } |
| 3560 | } else { |
| 3561 | switch(size) { |
| 3562 | case 0: |
| 3563 | subOp = Q ? Iop_Sub8x16 : Iop_Sub8x8; |
| 3564 | shOp = Q ? Iop_SarN8x16 : Iop_SarN8x8; |
| 3565 | break; |
| 3566 | case 1: |
| 3567 | subOp = Q ? Iop_Sub16x8 : Iop_Sub16x4; |
| 3568 | shOp = Q ? Iop_SarN16x8 : Iop_SarN16x4; |
| 3569 | break; |
| 3570 | case 2: |
| 3571 | subOp = Q ? Iop_Sub32x4 : Iop_Sub32x2; |
| 3572 | shOp = Q ? Iop_SarN32x4 : Iop_SarN32x2; |
| 3573 | break; |
| 3574 | default: |
| 3575 | vassert(0); |
| 3576 | } |
| 3577 | } |
| 3578 | assign(res, |
| 3579 | binop(subOp, |
| 3580 | binop(subOp, |
| 3581 | binop(shOp, mkexpr(arg_n), mkU8(1)), |
| 3582 | binop(shOp, mkexpr(arg_m), mkU8(1))), |
| 3583 | binop(andOp, |
| 3584 | binop(andOp, |
| 3585 | unop(notOp, mkexpr(arg_n)), |
| 3586 | mkexpr(arg_m)), |
| 3587 | imm_val))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3588 | DIP("vhsub.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3589 | U ? 'u' : 's', 8 << size, |
| 3590 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', |
| 3591 | mreg); |
| 3592 | } else { |
| 3593 | /* VQSUB */ |
| 3594 | IROp op, op2; |
| 3595 | IRTemp tmp; |
| 3596 | if (Q) { |
| 3597 | switch (size) { |
| 3598 | case 0: |
| 3599 | op = U ? Iop_QSub8Ux16 : Iop_QSub8Sx16; |
| 3600 | op2 = Iop_Sub8x16; |
| 3601 | break; |
| 3602 | case 1: |
| 3603 | op = U ? Iop_QSub16Ux8 : Iop_QSub16Sx8; |
| 3604 | op2 = Iop_Sub16x8; |
| 3605 | break; |
| 3606 | case 2: |
| 3607 | op = U ? Iop_QSub32Ux4 : Iop_QSub32Sx4; |
| 3608 | op2 = Iop_Sub32x4; |
| 3609 | break; |
| 3610 | case 3: |
| 3611 | op = U ? Iop_QSub64Ux2 : Iop_QSub64Sx2; |
| 3612 | op2 = Iop_Sub64x2; |
| 3613 | break; |
| 3614 | default: |
| 3615 | vassert(0); |
| 3616 | } |
| 3617 | } else { |
| 3618 | switch (size) { |
| 3619 | case 0: |
| 3620 | op = U ? Iop_QSub8Ux8 : Iop_QSub8Sx8; |
| 3621 | op2 = Iop_Sub8x8; |
| 3622 | break; |
| 3623 | case 1: |
| 3624 | op = U ? Iop_QSub16Ux4 : Iop_QSub16Sx4; |
| 3625 | op2 = Iop_Sub16x4; |
| 3626 | break; |
| 3627 | case 2: |
| 3628 | op = U ? Iop_QSub32Ux2 : Iop_QSub32Sx2; |
| 3629 | op2 = Iop_Sub32x2; |
| 3630 | break; |
| 3631 | case 3: |
| 3632 | op = U ? Iop_QSub64Ux1 : Iop_QSub64Sx1; |
| 3633 | op2 = Iop_Sub64; |
| 3634 | break; |
| 3635 | default: |
| 3636 | vassert(0); |
| 3637 | } |
| 3638 | } |
| 3639 | if (Q) |
| 3640 | tmp = newTemp(Ity_V128); |
| 3641 | else |
| 3642 | tmp = newTemp(Ity_I64); |
| 3643 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3644 | assign(tmp, binop(op2, mkexpr(arg_n), mkexpr(arg_m))); |
| 3645 | setFlag_QC(mkexpr(res), mkexpr(tmp), Q, condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3646 | DIP("vqsub.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3647 | U ? 'u' : 's', 8 << size, |
| 3648 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', |
| 3649 | mreg); |
| 3650 | } |
| 3651 | break; |
| 3652 | case 3: { |
| 3653 | IROp op; |
| 3654 | if (Q) { |
| 3655 | switch (size) { |
| 3656 | case 0: op = U ? Iop_CmpGT8Ux16 : Iop_CmpGT8Sx16; break; |
| 3657 | case 1: op = U ? Iop_CmpGT16Ux8 : Iop_CmpGT16Sx8; break; |
| 3658 | case 2: op = U ? Iop_CmpGT32Ux4 : Iop_CmpGT32Sx4; break; |
| 3659 | case 3: return False; |
| 3660 | default: vassert(0); |
| 3661 | } |
| 3662 | } else { |
| 3663 | switch (size) { |
| 3664 | case 0: op = U ? Iop_CmpGT8Ux8 : Iop_CmpGT8Sx8; break; |
| 3665 | case 1: op = U ? Iop_CmpGT16Ux4 : Iop_CmpGT16Sx4; break; |
| 3666 | case 2: op = U ? Iop_CmpGT32Ux2: Iop_CmpGT32Sx2; break; |
| 3667 | case 3: return False; |
| 3668 | default: vassert(0); |
| 3669 | } |
| 3670 | } |
| 3671 | if (B == 0) { |
| 3672 | /* VCGT */ |
| 3673 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3674 | DIP("vcgt.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3675 | U ? 'u' : 's', 8 << size, |
| 3676 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', |
| 3677 | mreg); |
| 3678 | } else { |
| 3679 | /* VCGE */ |
| 3680 | /* VCGE res, argn, argm |
| 3681 | is equal to |
| 3682 | VCGT tmp, argm, argn |
| 3683 | VNOT res, tmp */ |
| 3684 | assign(res, |
| 3685 | unop(Q ? Iop_NotV128 : Iop_Not64, |
| 3686 | binop(op, mkexpr(arg_m), mkexpr(arg_n)))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3687 | DIP("vcge.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3688 | U ? 'u' : 's', 8 << size, |
| 3689 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', |
| 3690 | mreg); |
| 3691 | } |
| 3692 | } |
| 3693 | break; |
| 3694 | case 4: |
| 3695 | if (B == 0) { |
| 3696 | /* VSHL */ |
sewardj | 7e67bb6 | 2014-07-21 07:55:45 +0000 | [diff] [blame] | 3697 | IROp op = Iop_INVALID, sub_op = Iop_INVALID; |
| 3698 | IRTemp tmp = IRTemp_INVALID; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3699 | if (U) { |
| 3700 | switch (size) { |
| 3701 | case 0: op = Q ? Iop_Shl8x16 : Iop_Shl8x8; break; |
| 3702 | case 1: op = Q ? Iop_Shl16x8 : Iop_Shl16x4; break; |
| 3703 | case 2: op = Q ? Iop_Shl32x4 : Iop_Shl32x2; break; |
| 3704 | case 3: op = Q ? Iop_Shl64x2 : Iop_Shl64; break; |
| 3705 | default: vassert(0); |
| 3706 | } |
| 3707 | } else { |
| 3708 | tmp = newTemp(Q ? Ity_V128 : Ity_I64); |
| 3709 | switch (size) { |
| 3710 | case 0: |
| 3711 | op = Q ? Iop_Sar8x16 : Iop_Sar8x8; |
| 3712 | sub_op = Q ? Iop_Sub8x16 : Iop_Sub8x8; |
| 3713 | break; |
| 3714 | case 1: |
| 3715 | op = Q ? Iop_Sar16x8 : Iop_Sar16x4; |
| 3716 | sub_op = Q ? Iop_Sub16x8 : Iop_Sub16x4; |
| 3717 | break; |
| 3718 | case 2: |
| 3719 | op = Q ? Iop_Sar32x4 : Iop_Sar32x2; |
| 3720 | sub_op = Q ? Iop_Sub32x4 : Iop_Sub32x2; |
| 3721 | break; |
| 3722 | case 3: |
| 3723 | op = Q ? Iop_Sar64x2 : Iop_Sar64; |
| 3724 | sub_op = Q ? Iop_Sub64x2 : Iop_Sub64; |
| 3725 | break; |
| 3726 | default: |
| 3727 | vassert(0); |
| 3728 | } |
| 3729 | } |
| 3730 | if (U) { |
| 3731 | if (!Q && (size == 3)) |
| 3732 | assign(res, binop(op, mkexpr(arg_m), |
| 3733 | unop(Iop_64to8, mkexpr(arg_n)))); |
| 3734 | else |
| 3735 | assign(res, binop(op, mkexpr(arg_m), mkexpr(arg_n))); |
| 3736 | } else { |
| 3737 | if (Q) |
| 3738 | assign(tmp, binop(sub_op, |
| 3739 | binop(Iop_64HLtoV128, mkU64(0), mkU64(0)), |
| 3740 | mkexpr(arg_n))); |
| 3741 | else |
| 3742 | assign(tmp, binop(sub_op, mkU64(0), mkexpr(arg_n))); |
| 3743 | if (!Q && (size == 3)) |
| 3744 | assign(res, binop(op, mkexpr(arg_m), |
| 3745 | unop(Iop_64to8, mkexpr(tmp)))); |
| 3746 | else |
| 3747 | assign(res, binop(op, mkexpr(arg_m), mkexpr(tmp))); |
| 3748 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3749 | DIP("vshl.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3750 | U ? 'u' : 's', 8 << size, |
| 3751 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg, Q ? 'q' : 'd', |
| 3752 | nreg); |
| 3753 | } else { |
| 3754 | /* VQSHL */ |
| 3755 | IROp op, op_rev, op_shrn, op_shln, cmp_neq, cmp_gt; |
| 3756 | IRTemp tmp, shval, mask, old_shval; |
| 3757 | UInt i; |
| 3758 | ULong esize; |
| 3759 | cmp_neq = Q ? Iop_CmpNEZ8x16 : Iop_CmpNEZ8x8; |
| 3760 | cmp_gt = Q ? Iop_CmpGT8Sx16 : Iop_CmpGT8Sx8; |
| 3761 | if (U) { |
| 3762 | switch (size) { |
| 3763 | case 0: |
| 3764 | op = Q ? Iop_QShl8x16 : Iop_QShl8x8; |
| 3765 | op_rev = Q ? Iop_Shr8x16 : Iop_Shr8x8; |
| 3766 | op_shrn = Q ? Iop_ShrN8x16 : Iop_ShrN8x8; |
| 3767 | op_shln = Q ? Iop_ShlN8x16 : Iop_ShlN8x8; |
| 3768 | break; |
| 3769 | case 1: |
| 3770 | op = Q ? Iop_QShl16x8 : Iop_QShl16x4; |
| 3771 | op_rev = Q ? Iop_Shr16x8 : Iop_Shr16x4; |
| 3772 | op_shrn = Q ? Iop_ShrN16x8 : Iop_ShrN16x4; |
| 3773 | op_shln = Q ? Iop_ShlN16x8 : Iop_ShlN16x4; |
| 3774 | break; |
| 3775 | case 2: |
| 3776 | op = Q ? Iop_QShl32x4 : Iop_QShl32x2; |
| 3777 | op_rev = Q ? Iop_Shr32x4 : Iop_Shr32x2; |
| 3778 | op_shrn = Q ? Iop_ShrN32x4 : Iop_ShrN32x2; |
| 3779 | op_shln = Q ? Iop_ShlN32x4 : Iop_ShlN32x2; |
| 3780 | break; |
| 3781 | case 3: |
| 3782 | op = Q ? Iop_QShl64x2 : Iop_QShl64x1; |
| 3783 | op_rev = Q ? Iop_Shr64x2 : Iop_Shr64; |
| 3784 | op_shrn = Q ? Iop_ShrN64x2 : Iop_Shr64; |
| 3785 | op_shln = Q ? Iop_ShlN64x2 : Iop_Shl64; |
| 3786 | break; |
| 3787 | default: |
| 3788 | vassert(0); |
| 3789 | } |
| 3790 | } else { |
| 3791 | switch (size) { |
| 3792 | case 0: |
| 3793 | op = Q ? Iop_QSal8x16 : Iop_QSal8x8; |
| 3794 | op_rev = Q ? Iop_Sar8x16 : Iop_Sar8x8; |
| 3795 | op_shrn = Q ? Iop_ShrN8x16 : Iop_ShrN8x8; |
| 3796 | op_shln = Q ? Iop_ShlN8x16 : Iop_ShlN8x8; |
| 3797 | break; |
| 3798 | case 1: |
| 3799 | op = Q ? Iop_QSal16x8 : Iop_QSal16x4; |
| 3800 | op_rev = Q ? Iop_Sar16x8 : Iop_Sar16x4; |
| 3801 | op_shrn = Q ? Iop_ShrN16x8 : Iop_ShrN16x4; |
| 3802 | op_shln = Q ? Iop_ShlN16x8 : Iop_ShlN16x4; |
| 3803 | break; |
| 3804 | case 2: |
| 3805 | op = Q ? Iop_QSal32x4 : Iop_QSal32x2; |
| 3806 | op_rev = Q ? Iop_Sar32x4 : Iop_Sar32x2; |
| 3807 | op_shrn = Q ? Iop_ShrN32x4 : Iop_ShrN32x2; |
| 3808 | op_shln = Q ? Iop_ShlN32x4 : Iop_ShlN32x2; |
| 3809 | break; |
| 3810 | case 3: |
| 3811 | op = Q ? Iop_QSal64x2 : Iop_QSal64x1; |
| 3812 | op_rev = Q ? Iop_Sar64x2 : Iop_Sar64; |
| 3813 | op_shrn = Q ? Iop_ShrN64x2 : Iop_Shr64; |
| 3814 | op_shln = Q ? Iop_ShlN64x2 : Iop_Shl64; |
| 3815 | break; |
| 3816 | default: |
| 3817 | vassert(0); |
| 3818 | } |
| 3819 | } |
| 3820 | if (Q) { |
| 3821 | tmp = newTemp(Ity_V128); |
| 3822 | shval = newTemp(Ity_V128); |
| 3823 | mask = newTemp(Ity_V128); |
| 3824 | } else { |
| 3825 | tmp = newTemp(Ity_I64); |
| 3826 | shval = newTemp(Ity_I64); |
| 3827 | mask = newTemp(Ity_I64); |
| 3828 | } |
| 3829 | assign(res, binop(op, mkexpr(arg_m), mkexpr(arg_n))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3830 | /* Only least significant byte from second argument is used. |
| 3831 | Copy this byte to the whole vector element. */ |
| 3832 | assign(shval, binop(op_shrn, |
| 3833 | binop(op_shln, |
| 3834 | mkexpr(arg_n), |
| 3835 | mkU8((8 << size) - 8)), |
| 3836 | mkU8((8 << size) - 8))); |
| 3837 | for(i = 0; i < size; i++) { |
| 3838 | old_shval = shval; |
| 3839 | shval = newTemp(Q ? Ity_V128 : Ity_I64); |
| 3840 | assign(shval, binop(Q ? Iop_OrV128 : Iop_Or64, |
| 3841 | mkexpr(old_shval), |
| 3842 | binop(op_shln, |
| 3843 | mkexpr(old_shval), |
| 3844 | mkU8(8 << i)))); |
| 3845 | } |
| 3846 | /* If shift is greater or equal to the element size and |
| 3847 | element is non-zero, then QC flag should be set. */ |
| 3848 | esize = (8 << size) - 1; |
| 3849 | esize = (esize << 8) | esize; |
| 3850 | esize = (esize << 16) | esize; |
| 3851 | esize = (esize << 32) | esize; |
| 3852 | setFlag_QC(binop(Q ? Iop_AndV128 : Iop_And64, |
| 3853 | binop(cmp_gt, mkexpr(shval), |
| 3854 | Q ? mkU128(esize) : mkU64(esize)), |
| 3855 | unop(cmp_neq, mkexpr(arg_m))), |
| 3856 | Q ? mkU128(0) : mkU64(0), |
| 3857 | Q, condT); |
| 3858 | /* Othervise QC flag should be set if shift value is positive and |
| 3859 | result beign rightshifted the same value is not equal to left |
| 3860 | argument. */ |
| 3861 | assign(mask, binop(cmp_gt, mkexpr(shval), |
| 3862 | Q ? mkU128(0) : mkU64(0))); |
| 3863 | if (!Q && size == 3) |
| 3864 | assign(tmp, binop(op_rev, mkexpr(res), |
| 3865 | unop(Iop_64to8, mkexpr(arg_n)))); |
| 3866 | else |
| 3867 | assign(tmp, binop(op_rev, mkexpr(res), mkexpr(arg_n))); |
| 3868 | setFlag_QC(binop(Q ? Iop_AndV128 : Iop_And64, |
| 3869 | mkexpr(tmp), mkexpr(mask)), |
| 3870 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 3871 | mkexpr(arg_m), mkexpr(mask)), |
| 3872 | Q, condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 3873 | DIP("vqshl.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3874 | U ? 'u' : 's', 8 << size, |
| 3875 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg, Q ? 'q' : 'd', |
| 3876 | nreg); |
| 3877 | } |
| 3878 | break; |
| 3879 | case 5: |
| 3880 | if (B == 0) { |
| 3881 | /* VRSHL */ |
sewardj | 06122e7 | 2011-03-28 12:14:48 +0000 | [diff] [blame] | 3882 | IROp op, op_shrn, op_shln, cmp_gt, op_add; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3883 | IRTemp shval, old_shval, imm_val, round; |
| 3884 | UInt i; |
| 3885 | ULong imm; |
| 3886 | cmp_gt = Q ? Iop_CmpGT8Sx16 : Iop_CmpGT8Sx8; |
| 3887 | imm = 1L; |
| 3888 | switch (size) { |
| 3889 | case 0: imm = (imm << 8) | imm; /* fall through */ |
| 3890 | case 1: imm = (imm << 16) | imm; /* fall through */ |
| 3891 | case 2: imm = (imm << 32) | imm; /* fall through */ |
| 3892 | case 3: break; |
| 3893 | default: vassert(0); |
| 3894 | } |
| 3895 | imm_val = newTemp(Q ? Ity_V128 : Ity_I64); |
| 3896 | round = newTemp(Q ? Ity_V128 : Ity_I64); |
| 3897 | assign(imm_val, Q ? mkU128(imm) : mkU64(imm)); |
| 3898 | if (U) { |
| 3899 | switch (size) { |
| 3900 | case 0: |
| 3901 | op = Q ? Iop_Shl8x16 : Iop_Shl8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3902 | op_add = Q ? Iop_Add8x16 : Iop_Add8x8; |
| 3903 | op_shrn = Q ? Iop_ShrN8x16 : Iop_ShrN8x8; |
| 3904 | op_shln = Q ? Iop_ShlN8x16 : Iop_ShlN8x8; |
| 3905 | break; |
| 3906 | case 1: |
| 3907 | op = Q ? Iop_Shl16x8 : Iop_Shl16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3908 | op_add = Q ? Iop_Add16x8 : Iop_Add16x4; |
| 3909 | op_shrn = Q ? Iop_ShrN16x8 : Iop_ShrN16x4; |
| 3910 | op_shln = Q ? Iop_ShlN16x8 : Iop_ShlN16x4; |
| 3911 | break; |
| 3912 | case 2: |
| 3913 | op = Q ? Iop_Shl32x4 : Iop_Shl32x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3914 | op_add = Q ? Iop_Add32x4 : Iop_Add32x2; |
| 3915 | op_shrn = Q ? Iop_ShrN32x4 : Iop_ShrN32x2; |
| 3916 | op_shln = Q ? Iop_ShlN32x4 : Iop_ShlN32x2; |
| 3917 | break; |
| 3918 | case 3: |
| 3919 | op = Q ? Iop_Shl64x2 : Iop_Shl64; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3920 | op_add = Q ? Iop_Add64x2 : Iop_Add64; |
| 3921 | op_shrn = Q ? Iop_ShrN64x2 : Iop_Shr64; |
| 3922 | op_shln = Q ? Iop_ShlN64x2 : Iop_Shl64; |
| 3923 | break; |
| 3924 | default: |
| 3925 | vassert(0); |
| 3926 | } |
| 3927 | } else { |
| 3928 | switch (size) { |
| 3929 | case 0: |
| 3930 | op = Q ? Iop_Sal8x16 : Iop_Sal8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3931 | op_add = Q ? Iop_Add8x16 : Iop_Add8x8; |
| 3932 | op_shrn = Q ? Iop_ShrN8x16 : Iop_ShrN8x8; |
| 3933 | op_shln = Q ? Iop_ShlN8x16 : Iop_ShlN8x8; |
| 3934 | break; |
| 3935 | case 1: |
| 3936 | op = Q ? Iop_Sal16x8 : Iop_Sal16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3937 | op_add = Q ? Iop_Add16x8 : Iop_Add16x4; |
| 3938 | op_shrn = Q ? Iop_ShrN16x8 : Iop_ShrN16x4; |
| 3939 | op_shln = Q ? Iop_ShlN16x8 : Iop_ShlN16x4; |
| 3940 | break; |
| 3941 | case 2: |
| 3942 | op = Q ? Iop_Sal32x4 : Iop_Sal32x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3943 | op_add = Q ? Iop_Add32x4 : Iop_Add32x2; |
| 3944 | op_shrn = Q ? Iop_ShrN32x4 : Iop_ShrN32x2; |
| 3945 | op_shln = Q ? Iop_ShlN32x4 : Iop_ShlN32x2; |
| 3946 | break; |
| 3947 | case 3: |
| 3948 | op = Q ? Iop_Sal64x2 : Iop_Sal64x1; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3949 | op_add = Q ? Iop_Add64x2 : Iop_Add64; |
| 3950 | op_shrn = Q ? Iop_ShrN64x2 : Iop_Shr64; |
| 3951 | op_shln = Q ? Iop_ShlN64x2 : Iop_Shl64; |
| 3952 | break; |
| 3953 | default: |
| 3954 | vassert(0); |
| 3955 | } |
| 3956 | } |
| 3957 | if (Q) { |
| 3958 | shval = newTemp(Ity_V128); |
| 3959 | } else { |
| 3960 | shval = newTemp(Ity_I64); |
| 3961 | } |
| 3962 | /* Only least significant byte from second argument is used. |
| 3963 | Copy this byte to the whole vector element. */ |
| 3964 | assign(shval, binop(op_shrn, |
| 3965 | binop(op_shln, |
| 3966 | mkexpr(arg_n), |
| 3967 | mkU8((8 << size) - 8)), |
| 3968 | mkU8((8 << size) - 8))); |
| 3969 | for (i = 0; i < size; i++) { |
| 3970 | old_shval = shval; |
| 3971 | shval = newTemp(Q ? Ity_V128 : Ity_I64); |
| 3972 | assign(shval, binop(Q ? Iop_OrV128 : Iop_Or64, |
| 3973 | mkexpr(old_shval), |
| 3974 | binop(op_shln, |
| 3975 | mkexpr(old_shval), |
| 3976 | mkU8(8 << i)))); |
| 3977 | } |
| 3978 | /* Compute the result */ |
| 3979 | if (!Q && size == 3 && U) { |
| 3980 | assign(round, binop(Q ? Iop_AndV128 : Iop_And64, |
| 3981 | binop(op, |
| 3982 | mkexpr(arg_m), |
| 3983 | unop(Iop_64to8, |
sewardj | b0d80ea | 2010-10-06 20:47:22 +0000 | [diff] [blame] | 3984 | binop(op_add, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 3985 | mkexpr(arg_n), |
| 3986 | mkexpr(imm_val)))), |
| 3987 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 3988 | mkexpr(imm_val), |
| 3989 | binop(cmp_gt, |
| 3990 | Q ? mkU128(0) : mkU64(0), |
| 3991 | mkexpr(arg_n))))); |
| 3992 | assign(res, binop(op_add, |
| 3993 | binop(op, |
| 3994 | mkexpr(arg_m), |
| 3995 | unop(Iop_64to8, mkexpr(arg_n))), |
| 3996 | mkexpr(round))); |
| 3997 | } else { |
| 3998 | assign(round, binop(Q ? Iop_AndV128 : Iop_And64, |
| 3999 | binop(op, |
| 4000 | mkexpr(arg_m), |
| 4001 | binop(op_add, |
| 4002 | mkexpr(arg_n), |
| 4003 | mkexpr(imm_val))), |
| 4004 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 4005 | mkexpr(imm_val), |
| 4006 | binop(cmp_gt, |
| 4007 | Q ? mkU128(0) : mkU64(0), |
| 4008 | mkexpr(arg_n))))); |
| 4009 | assign(res, binop(op_add, |
| 4010 | binop(op, mkexpr(arg_m), mkexpr(arg_n)), |
| 4011 | mkexpr(round))); |
| 4012 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4013 | DIP("vrshl.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4014 | U ? 'u' : 's', 8 << size, |
| 4015 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg, Q ? 'q' : 'd', |
| 4016 | nreg); |
| 4017 | } else { |
| 4018 | /* VQRSHL */ |
sewardj | 06122e7 | 2011-03-28 12:14:48 +0000 | [diff] [blame] | 4019 | IROp op, op_rev, op_shrn, op_shln, cmp_neq, cmp_gt, op_add; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4020 | IRTemp tmp, shval, mask, old_shval, imm_val, round; |
| 4021 | UInt i; |
| 4022 | ULong esize, imm; |
| 4023 | cmp_neq = Q ? Iop_CmpNEZ8x16 : Iop_CmpNEZ8x8; |
| 4024 | cmp_gt = Q ? Iop_CmpGT8Sx16 : Iop_CmpGT8Sx8; |
| 4025 | imm = 1L; |
| 4026 | switch (size) { |
| 4027 | case 0: imm = (imm << 8) | imm; /* fall through */ |
| 4028 | case 1: imm = (imm << 16) | imm; /* fall through */ |
| 4029 | case 2: imm = (imm << 32) | imm; /* fall through */ |
| 4030 | case 3: break; |
| 4031 | default: vassert(0); |
| 4032 | } |
| 4033 | imm_val = newTemp(Q ? Ity_V128 : Ity_I64); |
| 4034 | round = newTemp(Q ? Ity_V128 : Ity_I64); |
| 4035 | assign(imm_val, Q ? mkU128(imm) : mkU64(imm)); |
| 4036 | if (U) { |
| 4037 | switch (size) { |
| 4038 | case 0: |
| 4039 | op = Q ? Iop_QShl8x16 : Iop_QShl8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4040 | op_add = Q ? Iop_Add8x16 : Iop_Add8x8; |
| 4041 | op_rev = Q ? Iop_Shr8x16 : Iop_Shr8x8; |
| 4042 | op_shrn = Q ? Iop_ShrN8x16 : Iop_ShrN8x8; |
| 4043 | op_shln = Q ? Iop_ShlN8x16 : Iop_ShlN8x8; |
| 4044 | break; |
| 4045 | case 1: |
| 4046 | op = Q ? Iop_QShl16x8 : Iop_QShl16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4047 | op_add = Q ? Iop_Add16x8 : Iop_Add16x4; |
| 4048 | op_rev = Q ? Iop_Shr16x8 : Iop_Shr16x4; |
| 4049 | op_shrn = Q ? Iop_ShrN16x8 : Iop_ShrN16x4; |
| 4050 | op_shln = Q ? Iop_ShlN16x8 : Iop_ShlN16x4; |
| 4051 | break; |
| 4052 | case 2: |
| 4053 | op = Q ? Iop_QShl32x4 : Iop_QShl32x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4054 | op_add = Q ? Iop_Add32x4 : Iop_Add32x2; |
| 4055 | op_rev = Q ? Iop_Shr32x4 : Iop_Shr32x2; |
| 4056 | op_shrn = Q ? Iop_ShrN32x4 : Iop_ShrN32x2; |
| 4057 | op_shln = Q ? Iop_ShlN32x4 : Iop_ShlN32x2; |
| 4058 | break; |
| 4059 | case 3: |
| 4060 | op = Q ? Iop_QShl64x2 : Iop_QShl64x1; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4061 | op_add = Q ? Iop_Add64x2 : Iop_Add64; |
| 4062 | op_rev = Q ? Iop_Shr64x2 : Iop_Shr64; |
| 4063 | op_shrn = Q ? Iop_ShrN64x2 : Iop_Shr64; |
| 4064 | op_shln = Q ? Iop_ShlN64x2 : Iop_Shl64; |
| 4065 | break; |
| 4066 | default: |
| 4067 | vassert(0); |
| 4068 | } |
| 4069 | } else { |
| 4070 | switch (size) { |
| 4071 | case 0: |
| 4072 | op = Q ? Iop_QSal8x16 : Iop_QSal8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4073 | op_add = Q ? Iop_Add8x16 : Iop_Add8x8; |
| 4074 | op_rev = Q ? Iop_Sar8x16 : Iop_Sar8x8; |
| 4075 | op_shrn = Q ? Iop_ShrN8x16 : Iop_ShrN8x8; |
| 4076 | op_shln = Q ? Iop_ShlN8x16 : Iop_ShlN8x8; |
| 4077 | break; |
| 4078 | case 1: |
| 4079 | op = Q ? Iop_QSal16x8 : Iop_QSal16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4080 | op_add = Q ? Iop_Add16x8 : Iop_Add16x4; |
| 4081 | op_rev = Q ? Iop_Sar16x8 : Iop_Sar16x4; |
| 4082 | op_shrn = Q ? Iop_ShrN16x8 : Iop_ShrN16x4; |
| 4083 | op_shln = Q ? Iop_ShlN16x8 : Iop_ShlN16x4; |
| 4084 | break; |
| 4085 | case 2: |
| 4086 | op = Q ? Iop_QSal32x4 : Iop_QSal32x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4087 | op_add = Q ? Iop_Add32x4 : Iop_Add32x2; |
| 4088 | op_rev = Q ? Iop_Sar32x4 : Iop_Sar32x2; |
| 4089 | op_shrn = Q ? Iop_ShrN32x4 : Iop_ShrN32x2; |
| 4090 | op_shln = Q ? Iop_ShlN32x4 : Iop_ShlN32x2; |
| 4091 | break; |
| 4092 | case 3: |
| 4093 | op = Q ? Iop_QSal64x2 : Iop_QSal64x1; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4094 | op_add = Q ? Iop_Add64x2 : Iop_Add64; |
| 4095 | op_rev = Q ? Iop_Sar64x2 : Iop_Sar64; |
| 4096 | op_shrn = Q ? Iop_ShrN64x2 : Iop_Shr64; |
| 4097 | op_shln = Q ? Iop_ShlN64x2 : Iop_Shl64; |
| 4098 | break; |
| 4099 | default: |
| 4100 | vassert(0); |
| 4101 | } |
| 4102 | } |
| 4103 | if (Q) { |
| 4104 | tmp = newTemp(Ity_V128); |
| 4105 | shval = newTemp(Ity_V128); |
| 4106 | mask = newTemp(Ity_V128); |
| 4107 | } else { |
| 4108 | tmp = newTemp(Ity_I64); |
| 4109 | shval = newTemp(Ity_I64); |
| 4110 | mask = newTemp(Ity_I64); |
| 4111 | } |
| 4112 | /* Only least significant byte from second argument is used. |
| 4113 | Copy this byte to the whole vector element. */ |
| 4114 | assign(shval, binop(op_shrn, |
| 4115 | binop(op_shln, |
| 4116 | mkexpr(arg_n), |
| 4117 | mkU8((8 << size) - 8)), |
| 4118 | mkU8((8 << size) - 8))); |
| 4119 | for (i = 0; i < size; i++) { |
| 4120 | old_shval = shval; |
| 4121 | shval = newTemp(Q ? Ity_V128 : Ity_I64); |
| 4122 | assign(shval, binop(Q ? Iop_OrV128 : Iop_Or64, |
| 4123 | mkexpr(old_shval), |
| 4124 | binop(op_shln, |
| 4125 | mkexpr(old_shval), |
| 4126 | mkU8(8 << i)))); |
| 4127 | } |
| 4128 | /* Compute the result */ |
| 4129 | assign(round, binop(Q ? Iop_AndV128 : Iop_And64, |
| 4130 | binop(op, |
| 4131 | mkexpr(arg_m), |
| 4132 | binop(op_add, |
| 4133 | mkexpr(arg_n), |
| 4134 | mkexpr(imm_val))), |
| 4135 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 4136 | mkexpr(imm_val), |
| 4137 | binop(cmp_gt, |
| 4138 | Q ? mkU128(0) : mkU64(0), |
| 4139 | mkexpr(arg_n))))); |
| 4140 | assign(res, binop(op_add, |
| 4141 | binop(op, mkexpr(arg_m), mkexpr(arg_n)), |
| 4142 | mkexpr(round))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4143 | /* If shift is greater or equal to the element size and element is |
| 4144 | non-zero, then QC flag should be set. */ |
| 4145 | esize = (8 << size) - 1; |
| 4146 | esize = (esize << 8) | esize; |
| 4147 | esize = (esize << 16) | esize; |
| 4148 | esize = (esize << 32) | esize; |
| 4149 | setFlag_QC(binop(Q ? Iop_AndV128 : Iop_And64, |
| 4150 | binop(cmp_gt, mkexpr(shval), |
| 4151 | Q ? mkU128(esize) : mkU64(esize)), |
| 4152 | unop(cmp_neq, mkexpr(arg_m))), |
| 4153 | Q ? mkU128(0) : mkU64(0), |
| 4154 | Q, condT); |
| 4155 | /* Othervise QC flag should be set if shift value is positive and |
| 4156 | result beign rightshifted the same value is not equal to left |
| 4157 | argument. */ |
| 4158 | assign(mask, binop(cmp_gt, mkexpr(shval), |
| 4159 | Q ? mkU128(0) : mkU64(0))); |
| 4160 | if (!Q && size == 3) |
| 4161 | assign(tmp, binop(op_rev, mkexpr(res), |
| 4162 | unop(Iop_64to8, mkexpr(arg_n)))); |
| 4163 | else |
| 4164 | assign(tmp, binop(op_rev, mkexpr(res), mkexpr(arg_n))); |
| 4165 | setFlag_QC(binop(Q ? Iop_AndV128 : Iop_And64, |
| 4166 | mkexpr(tmp), mkexpr(mask)), |
| 4167 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 4168 | mkexpr(arg_m), mkexpr(mask)), |
| 4169 | Q, condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4170 | DIP("vqrshl.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4171 | U ? 'u' : 's', 8 << size, |
| 4172 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg, Q ? 'q' : 'd', |
| 4173 | nreg); |
| 4174 | } |
| 4175 | break; |
| 4176 | case 6: |
| 4177 | /* VMAX, VMIN */ |
| 4178 | if (B == 0) { |
| 4179 | /* VMAX */ |
| 4180 | IROp op; |
| 4181 | if (U == 0) { |
| 4182 | switch (size) { |
| 4183 | case 0: op = Q ? Iop_Max8Sx16 : Iop_Max8Sx8; break; |
| 4184 | case 1: op = Q ? Iop_Max16Sx8 : Iop_Max16Sx4; break; |
| 4185 | case 2: op = Q ? Iop_Max32Sx4 : Iop_Max32Sx2; break; |
| 4186 | case 3: return False; |
| 4187 | default: vassert(0); |
| 4188 | } |
| 4189 | } else { |
| 4190 | switch (size) { |
| 4191 | case 0: op = Q ? Iop_Max8Ux16 : Iop_Max8Ux8; break; |
| 4192 | case 1: op = Q ? Iop_Max16Ux8 : Iop_Max16Ux4; break; |
| 4193 | case 2: op = Q ? Iop_Max32Ux4 : Iop_Max32Ux2; break; |
| 4194 | case 3: return False; |
| 4195 | default: vassert(0); |
| 4196 | } |
| 4197 | } |
| 4198 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4199 | DIP("vmax.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4200 | U ? 'u' : 's', 8 << size, |
| 4201 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', |
| 4202 | mreg); |
| 4203 | } else { |
| 4204 | /* VMIN */ |
| 4205 | IROp op; |
| 4206 | if (U == 0) { |
| 4207 | switch (size) { |
| 4208 | case 0: op = Q ? Iop_Min8Sx16 : Iop_Min8Sx8; break; |
| 4209 | case 1: op = Q ? Iop_Min16Sx8 : Iop_Min16Sx4; break; |
| 4210 | case 2: op = Q ? Iop_Min32Sx4 : Iop_Min32Sx2; break; |
| 4211 | case 3: return False; |
| 4212 | default: vassert(0); |
| 4213 | } |
| 4214 | } else { |
| 4215 | switch (size) { |
sewardj | b0d80ea | 2010-10-06 20:47:22 +0000 | [diff] [blame] | 4216 | case 0: op = Q ? Iop_Min8Ux16 : Iop_Min8Ux8; break; |
| 4217 | case 1: op = Q ? Iop_Min16Ux8 : Iop_Min16Ux4; break; |
| 4218 | case 2: op = Q ? Iop_Min32Ux4 : Iop_Min32Ux2; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4219 | case 3: return False; |
| 4220 | default: vassert(0); |
| 4221 | } |
| 4222 | } |
| 4223 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4224 | DIP("vmin.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4225 | U ? 'u' : 's', 8 << size, |
| 4226 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', |
| 4227 | mreg); |
| 4228 | } |
| 4229 | break; |
| 4230 | case 7: |
| 4231 | if (B == 0) { |
| 4232 | /* VABD */ |
| 4233 | IROp op_cmp, op_sub; |
| 4234 | IRTemp cond; |
| 4235 | if ((theInstr >> 23) & 1) { |
| 4236 | vpanic("VABDL should not be in dis_neon_data_3same\n"); |
| 4237 | } |
| 4238 | if (Q) { |
| 4239 | switch (size) { |
| 4240 | case 0: |
| 4241 | op_cmp = U ? Iop_CmpGT8Ux16 : Iop_CmpGT8Sx16; |
| 4242 | op_sub = Iop_Sub8x16; |
| 4243 | break; |
| 4244 | case 1: |
| 4245 | op_cmp = U ? Iop_CmpGT16Ux8 : Iop_CmpGT16Sx8; |
| 4246 | op_sub = Iop_Sub16x8; |
| 4247 | break; |
| 4248 | case 2: |
| 4249 | op_cmp = U ? Iop_CmpGT32Ux4 : Iop_CmpGT32Sx4; |
| 4250 | op_sub = Iop_Sub32x4; |
| 4251 | break; |
| 4252 | case 3: |
| 4253 | return False; |
| 4254 | default: |
| 4255 | vassert(0); |
| 4256 | } |
| 4257 | } else { |
| 4258 | switch (size) { |
| 4259 | case 0: |
| 4260 | op_cmp = U ? Iop_CmpGT8Ux8 : Iop_CmpGT8Sx8; |
| 4261 | op_sub = Iop_Sub8x8; |
| 4262 | break; |
| 4263 | case 1: |
| 4264 | op_cmp = U ? Iop_CmpGT16Ux4 : Iop_CmpGT16Sx4; |
| 4265 | op_sub = Iop_Sub16x4; |
| 4266 | break; |
| 4267 | case 2: |
| 4268 | op_cmp = U ? Iop_CmpGT32Ux2 : Iop_CmpGT32Sx2; |
| 4269 | op_sub = Iop_Sub32x2; |
| 4270 | break; |
| 4271 | case 3: |
| 4272 | return False; |
| 4273 | default: |
| 4274 | vassert(0); |
| 4275 | } |
| 4276 | } |
| 4277 | if (Q) { |
| 4278 | cond = newTemp(Ity_V128); |
| 4279 | } else { |
| 4280 | cond = newTemp(Ity_I64); |
| 4281 | } |
| 4282 | assign(cond, binop(op_cmp, mkexpr(arg_n), mkexpr(arg_m))); |
| 4283 | assign(res, binop(Q ? Iop_OrV128 : Iop_Or64, |
| 4284 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 4285 | binop(op_sub, mkexpr(arg_n), |
| 4286 | mkexpr(arg_m)), |
| 4287 | mkexpr(cond)), |
| 4288 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 4289 | binop(op_sub, mkexpr(arg_m), |
| 4290 | mkexpr(arg_n)), |
| 4291 | unop(Q ? Iop_NotV128 : Iop_Not64, |
| 4292 | mkexpr(cond))))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4293 | DIP("vabd.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4294 | U ? 'u' : 's', 8 << size, |
| 4295 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', |
| 4296 | mreg); |
| 4297 | } else { |
| 4298 | /* VABA */ |
| 4299 | IROp op_cmp, op_sub, op_add; |
| 4300 | IRTemp cond, acc, tmp; |
| 4301 | if ((theInstr >> 23) & 1) { |
| 4302 | vpanic("VABAL should not be in dis_neon_data_3same"); |
| 4303 | } |
| 4304 | if (Q) { |
| 4305 | switch (size) { |
| 4306 | case 0: |
| 4307 | op_cmp = U ? Iop_CmpGT8Ux16 : Iop_CmpGT8Sx16; |
| 4308 | op_sub = Iop_Sub8x16; |
| 4309 | op_add = Iop_Add8x16; |
| 4310 | break; |
| 4311 | case 1: |
| 4312 | op_cmp = U ? Iop_CmpGT16Ux8 : Iop_CmpGT16Sx8; |
| 4313 | op_sub = Iop_Sub16x8; |
| 4314 | op_add = Iop_Add16x8; |
| 4315 | break; |
| 4316 | case 2: |
| 4317 | op_cmp = U ? Iop_CmpGT32Ux4 : Iop_CmpGT32Sx4; |
| 4318 | op_sub = Iop_Sub32x4; |
| 4319 | op_add = Iop_Add32x4; |
| 4320 | break; |
| 4321 | case 3: |
| 4322 | return False; |
| 4323 | default: |
| 4324 | vassert(0); |
| 4325 | } |
| 4326 | } else { |
| 4327 | switch (size) { |
| 4328 | case 0: |
| 4329 | op_cmp = U ? Iop_CmpGT8Ux8 : Iop_CmpGT8Sx8; |
| 4330 | op_sub = Iop_Sub8x8; |
| 4331 | op_add = Iop_Add8x8; |
| 4332 | break; |
| 4333 | case 1: |
| 4334 | op_cmp = U ? Iop_CmpGT16Ux4 : Iop_CmpGT16Sx4; |
| 4335 | op_sub = Iop_Sub16x4; |
| 4336 | op_add = Iop_Add16x4; |
| 4337 | break; |
| 4338 | case 2: |
| 4339 | op_cmp = U ? Iop_CmpGT32Ux2 : Iop_CmpGT32Sx2; |
| 4340 | op_sub = Iop_Sub32x2; |
| 4341 | op_add = Iop_Add32x2; |
| 4342 | break; |
| 4343 | case 3: |
| 4344 | return False; |
| 4345 | default: |
| 4346 | vassert(0); |
| 4347 | } |
| 4348 | } |
| 4349 | if (Q) { |
| 4350 | cond = newTemp(Ity_V128); |
| 4351 | acc = newTemp(Ity_V128); |
| 4352 | tmp = newTemp(Ity_V128); |
| 4353 | assign(acc, getQReg(dreg)); |
| 4354 | } else { |
| 4355 | cond = newTemp(Ity_I64); |
| 4356 | acc = newTemp(Ity_I64); |
| 4357 | tmp = newTemp(Ity_I64); |
| 4358 | assign(acc, getDRegI64(dreg)); |
| 4359 | } |
| 4360 | assign(cond, binop(op_cmp, mkexpr(arg_n), mkexpr(arg_m))); |
| 4361 | assign(tmp, binop(Q ? Iop_OrV128 : Iop_Or64, |
| 4362 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 4363 | binop(op_sub, mkexpr(arg_n), |
| 4364 | mkexpr(arg_m)), |
| 4365 | mkexpr(cond)), |
| 4366 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 4367 | binop(op_sub, mkexpr(arg_m), |
| 4368 | mkexpr(arg_n)), |
| 4369 | unop(Q ? Iop_NotV128 : Iop_Not64, |
| 4370 | mkexpr(cond))))); |
| 4371 | assign(res, binop(op_add, mkexpr(acc), mkexpr(tmp))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4372 | DIP("vaba.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4373 | U ? 'u' : 's', 8 << size, |
| 4374 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', |
| 4375 | mreg); |
| 4376 | } |
| 4377 | break; |
| 4378 | case 8: |
| 4379 | if (B == 0) { |
| 4380 | IROp op; |
| 4381 | if (U == 0) { |
| 4382 | /* VADD */ |
| 4383 | switch (size) { |
| 4384 | case 0: op = Q ? Iop_Add8x16 : Iop_Add8x8; break; |
| 4385 | case 1: op = Q ? Iop_Add16x8 : Iop_Add16x4; break; |
| 4386 | case 2: op = Q ? Iop_Add32x4 : Iop_Add32x2; break; |
| 4387 | case 3: op = Q ? Iop_Add64x2 : Iop_Add64; break; |
| 4388 | default: vassert(0); |
| 4389 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4390 | DIP("vadd.i%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4391 | 8 << size, Q ? 'q' : 'd', |
| 4392 | dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4393 | } else { |
| 4394 | /* VSUB */ |
| 4395 | switch (size) { |
| 4396 | case 0: op = Q ? Iop_Sub8x16 : Iop_Sub8x8; break; |
| 4397 | case 1: op = Q ? Iop_Sub16x8 : Iop_Sub16x4; break; |
| 4398 | case 2: op = Q ? Iop_Sub32x4 : Iop_Sub32x2; break; |
| 4399 | case 3: op = Q ? Iop_Sub64x2 : Iop_Sub64; break; |
| 4400 | default: vassert(0); |
| 4401 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4402 | DIP("vsub.i%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4403 | 8 << size, Q ? 'q' : 'd', |
| 4404 | dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4405 | } |
| 4406 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
| 4407 | } else { |
| 4408 | IROp op; |
| 4409 | switch (size) { |
| 4410 | case 0: op = Q ? Iop_CmpNEZ8x16 : Iop_CmpNEZ8x8; break; |
| 4411 | case 1: op = Q ? Iop_CmpNEZ16x8 : Iop_CmpNEZ16x4; break; |
| 4412 | case 2: op = Q ? Iop_CmpNEZ32x4 : Iop_CmpNEZ32x2; break; |
| 4413 | case 3: op = Q ? Iop_CmpNEZ64x2 : Iop_CmpwNEZ64; break; |
| 4414 | default: vassert(0); |
| 4415 | } |
| 4416 | if (U == 0) { |
| 4417 | /* VTST */ |
| 4418 | assign(res, unop(op, binop(Q ? Iop_AndV128 : Iop_And64, |
| 4419 | mkexpr(arg_n), |
| 4420 | mkexpr(arg_m)))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4421 | DIP("vtst.%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4422 | 8 << size, Q ? 'q' : 'd', |
| 4423 | dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4424 | } else { |
| 4425 | /* VCEQ */ |
| 4426 | assign(res, unop(Q ? Iop_NotV128 : Iop_Not64, |
| 4427 | unop(op, |
| 4428 | binop(Q ? Iop_XorV128 : Iop_Xor64, |
| 4429 | mkexpr(arg_n), |
| 4430 | mkexpr(arg_m))))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4431 | DIP("vceq.i%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4432 | 8 << size, Q ? 'q' : 'd', |
| 4433 | dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4434 | } |
| 4435 | } |
| 4436 | break; |
| 4437 | case 9: |
| 4438 | if (B == 0) { |
| 4439 | /* VMLA, VMLS (integer) */ |
| 4440 | IROp op, op2; |
| 4441 | UInt P = (theInstr >> 24) & 1; |
| 4442 | if (P) { |
| 4443 | switch (size) { |
| 4444 | case 0: |
| 4445 | op = Q ? Iop_Mul8x16 : Iop_Mul8x8; |
| 4446 | op2 = Q ? Iop_Sub8x16 : Iop_Sub8x8; |
| 4447 | break; |
| 4448 | case 1: |
| 4449 | op = Q ? Iop_Mul16x8 : Iop_Mul16x4; |
| 4450 | op2 = Q ? Iop_Sub16x8 : Iop_Sub16x4; |
| 4451 | break; |
| 4452 | case 2: |
| 4453 | op = Q ? Iop_Mul32x4 : Iop_Mul32x2; |
| 4454 | op2 = Q ? Iop_Sub32x4 : Iop_Sub32x2; |
| 4455 | break; |
| 4456 | case 3: |
| 4457 | return False; |
| 4458 | default: |
| 4459 | vassert(0); |
| 4460 | } |
| 4461 | } else { |
| 4462 | switch (size) { |
| 4463 | case 0: |
| 4464 | op = Q ? Iop_Mul8x16 : Iop_Mul8x8; |
| 4465 | op2 = Q ? Iop_Add8x16 : Iop_Add8x8; |
| 4466 | break; |
| 4467 | case 1: |
| 4468 | op = Q ? Iop_Mul16x8 : Iop_Mul16x4; |
| 4469 | op2 = Q ? Iop_Add16x8 : Iop_Add16x4; |
| 4470 | break; |
| 4471 | case 2: |
| 4472 | op = Q ? Iop_Mul32x4 : Iop_Mul32x2; |
| 4473 | op2 = Q ? Iop_Add32x4 : Iop_Add32x2; |
| 4474 | break; |
| 4475 | case 3: |
| 4476 | return False; |
| 4477 | default: |
| 4478 | vassert(0); |
| 4479 | } |
| 4480 | } |
| 4481 | assign(res, binop(op2, |
| 4482 | Q ? getQReg(dreg) : getDRegI64(dreg), |
| 4483 | binop(op, mkexpr(arg_n), mkexpr(arg_m)))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4484 | DIP("vml%c.i%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4485 | P ? 's' : 'a', 8 << size, |
| 4486 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', |
| 4487 | mreg); |
| 4488 | } else { |
| 4489 | /* VMUL */ |
| 4490 | IROp op; |
| 4491 | UInt P = (theInstr >> 24) & 1; |
| 4492 | if (P) { |
| 4493 | switch (size) { |
| 4494 | case 0: |
| 4495 | op = Q ? Iop_PolynomialMul8x16 : Iop_PolynomialMul8x8; |
| 4496 | break; |
| 4497 | case 1: case 2: case 3: return False; |
| 4498 | default: vassert(0); |
| 4499 | } |
| 4500 | } else { |
| 4501 | switch (size) { |
| 4502 | case 0: op = Q ? Iop_Mul8x16 : Iop_Mul8x8; break; |
| 4503 | case 1: op = Q ? Iop_Mul16x8 : Iop_Mul16x4; break; |
| 4504 | case 2: op = Q ? Iop_Mul32x4 : Iop_Mul32x2; break; |
| 4505 | case 3: return False; |
| 4506 | default: vassert(0); |
| 4507 | } |
| 4508 | } |
| 4509 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4510 | DIP("vmul.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4511 | P ? 'p' : 'i', 8 << size, |
| 4512 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', |
| 4513 | mreg); |
| 4514 | } |
| 4515 | break; |
| 4516 | case 10: { |
| 4517 | /* VPMAX, VPMIN */ |
| 4518 | UInt P = (theInstr >> 4) & 1; |
| 4519 | IROp op; |
| 4520 | if (Q) |
| 4521 | return False; |
| 4522 | if (P) { |
| 4523 | switch (size) { |
| 4524 | case 0: op = U ? Iop_PwMin8Ux8 : Iop_PwMin8Sx8; break; |
| 4525 | case 1: op = U ? Iop_PwMin16Ux4 : Iop_PwMin16Sx4; break; |
| 4526 | case 2: op = U ? Iop_PwMin32Ux2 : Iop_PwMin32Sx2; break; |
| 4527 | case 3: return False; |
| 4528 | default: vassert(0); |
| 4529 | } |
| 4530 | } else { |
| 4531 | switch (size) { |
| 4532 | case 0: op = U ? Iop_PwMax8Ux8 : Iop_PwMax8Sx8; break; |
| 4533 | case 1: op = U ? Iop_PwMax16Ux4 : Iop_PwMax16Sx4; break; |
| 4534 | case 2: op = U ? Iop_PwMax32Ux2 : Iop_PwMax32Sx2; break; |
| 4535 | case 3: return False; |
| 4536 | default: vassert(0); |
| 4537 | } |
| 4538 | } |
| 4539 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4540 | DIP("vp%s.%c%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4541 | P ? "min" : "max", U ? 'u' : 's', |
| 4542 | 8 << size, Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', nreg, |
| 4543 | Q ? 'q' : 'd', mreg); |
| 4544 | break; |
| 4545 | } |
| 4546 | case 11: |
| 4547 | if (B == 0) { |
| 4548 | if (U == 0) { |
| 4549 | /* VQDMULH */ |
| 4550 | IROp op ,op2; |
| 4551 | ULong imm; |
| 4552 | switch (size) { |
| 4553 | case 0: case 3: |
| 4554 | return False; |
| 4555 | case 1: |
| 4556 | op = Q ? Iop_QDMulHi16Sx8 : Iop_QDMulHi16Sx4; |
| 4557 | op2 = Q ? Iop_CmpEQ16x8 : Iop_CmpEQ16x4; |
| 4558 | imm = 1LL << 15; |
| 4559 | imm = (imm << 16) | imm; |
| 4560 | imm = (imm << 32) | imm; |
| 4561 | break; |
| 4562 | case 2: |
| 4563 | op = Q ? Iop_QDMulHi32Sx4 : Iop_QDMulHi32Sx2; |
| 4564 | op2 = Q ? Iop_CmpEQ32x4 : Iop_CmpEQ32x2; |
| 4565 | imm = 1LL << 31; |
| 4566 | imm = (imm << 32) | imm; |
| 4567 | break; |
| 4568 | default: |
| 4569 | vassert(0); |
| 4570 | } |
| 4571 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4572 | setFlag_QC(binop(Q ? Iop_AndV128 : Iop_And64, |
| 4573 | binop(op2, mkexpr(arg_n), |
| 4574 | Q ? mkU128(imm) : mkU64(imm)), |
| 4575 | binop(op2, mkexpr(arg_m), |
| 4576 | Q ? mkU128(imm) : mkU64(imm))), |
| 4577 | Q ? mkU128(0) : mkU64(0), |
| 4578 | Q, condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4579 | DIP("vqdmulh.s%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4580 | 8 << size, Q ? 'q' : 'd', |
| 4581 | dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4582 | } else { |
| 4583 | /* VQRDMULH */ |
| 4584 | IROp op ,op2; |
| 4585 | ULong imm; |
| 4586 | switch(size) { |
| 4587 | case 0: case 3: |
| 4588 | return False; |
| 4589 | case 1: |
| 4590 | imm = 1LL << 15; |
| 4591 | imm = (imm << 16) | imm; |
| 4592 | imm = (imm << 32) | imm; |
| 4593 | op = Q ? Iop_QRDMulHi16Sx8 : Iop_QRDMulHi16Sx4; |
| 4594 | op2 = Q ? Iop_CmpEQ16x8 : Iop_CmpEQ16x4; |
| 4595 | break; |
| 4596 | case 2: |
| 4597 | imm = 1LL << 31; |
| 4598 | imm = (imm << 32) | imm; |
| 4599 | op = Q ? Iop_QRDMulHi32Sx4 : Iop_QRDMulHi32Sx2; |
| 4600 | op2 = Q ? Iop_CmpEQ32x4 : Iop_CmpEQ32x2; |
| 4601 | break; |
| 4602 | default: |
| 4603 | vassert(0); |
| 4604 | } |
| 4605 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4606 | setFlag_QC(binop(Q ? Iop_AndV128 : Iop_And64, |
| 4607 | binop(op2, mkexpr(arg_n), |
| 4608 | Q ? mkU128(imm) : mkU64(imm)), |
| 4609 | binop(op2, mkexpr(arg_m), |
| 4610 | Q ? mkU128(imm) : mkU64(imm))), |
| 4611 | Q ? mkU128(0) : mkU64(0), |
| 4612 | Q, condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4613 | DIP("vqrdmulh.s%d %c%u, %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4614 | 8 << size, Q ? 'q' : 'd', |
| 4615 | dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4616 | } |
| 4617 | } else { |
| 4618 | if (U == 0) { |
| 4619 | /* VPADD */ |
| 4620 | IROp op; |
| 4621 | if (Q) |
| 4622 | return False; |
| 4623 | switch (size) { |
| 4624 | case 0: op = Q ? Iop_PwAdd8x16 : Iop_PwAdd8x8; break; |
| 4625 | case 1: op = Q ? Iop_PwAdd16x8 : Iop_PwAdd16x4; break; |
| 4626 | case 2: op = Q ? Iop_PwAdd32x4 : Iop_PwAdd32x2; break; |
| 4627 | case 3: return False; |
| 4628 | default: vassert(0); |
| 4629 | } |
| 4630 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
| 4631 | DIP("vpadd.i%d %c%u, %c%u, %c%u\n", |
| 4632 | 8 << size, Q ? 'q' : 'd', |
| 4633 | dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 4634 | } else { |
| 4635 | return False; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4636 | } |
| 4637 | } |
| 4638 | break; |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 4639 | case 12: { |
| 4640 | return False; |
| 4641 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4642 | /* Starting from here these are FP SIMD cases */ |
| 4643 | case 13: |
| 4644 | if (B == 0) { |
| 4645 | IROp op; |
| 4646 | if (U == 0) { |
| 4647 | if ((C >> 1) == 0) { |
| 4648 | /* VADD */ |
| 4649 | op = Q ? Iop_Add32Fx4 : Iop_Add32Fx2 ; |
| 4650 | DIP("vadd.f32 %c%u, %c%u, %c%u\n", |
| 4651 | Q ? 'q' : 'd', dreg, |
| 4652 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4653 | } else { |
| 4654 | /* VSUB */ |
| 4655 | op = Q ? Iop_Sub32Fx4 : Iop_Sub32Fx2 ; |
| 4656 | DIP("vsub.f32 %c%u, %c%u, %c%u\n", |
| 4657 | Q ? 'q' : 'd', dreg, |
| 4658 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4659 | } |
| 4660 | } else { |
| 4661 | if ((C >> 1) == 0) { |
| 4662 | /* VPADD */ |
| 4663 | if (Q) |
| 4664 | return False; |
| 4665 | op = Iop_PwAdd32Fx2; |
| 4666 | DIP("vpadd.f32 d%u, d%u, d%u\n", dreg, nreg, mreg); |
| 4667 | } else { |
| 4668 | /* VABD */ |
| 4669 | if (Q) { |
| 4670 | assign(res, unop(Iop_Abs32Fx4, |
sewardj | e84eeb4 | 2014-03-15 11:41:39 +0000 | [diff] [blame] | 4671 | triop(Iop_Sub32Fx4, |
| 4672 | get_FAKE_roundingmode(), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4673 | mkexpr(arg_n), |
| 4674 | mkexpr(arg_m)))); |
| 4675 | } else { |
| 4676 | assign(res, unop(Iop_Abs32Fx2, |
| 4677 | binop(Iop_Sub32Fx2, |
| 4678 | mkexpr(arg_n), |
| 4679 | mkexpr(arg_m)))); |
| 4680 | } |
| 4681 | DIP("vabd.f32 %c%u, %c%u, %c%u\n", |
| 4682 | Q ? 'q' : 'd', dreg, |
| 4683 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4684 | break; |
| 4685 | } |
| 4686 | } |
sewardj | e84eeb4 | 2014-03-15 11:41:39 +0000 | [diff] [blame] | 4687 | assign(res, binop_w_fake_RM(op, mkexpr(arg_n), mkexpr(arg_m))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4688 | } else { |
| 4689 | if (U == 0) { |
| 4690 | /* VMLA, VMLS */ |
| 4691 | IROp op, op2; |
| 4692 | UInt P = (theInstr >> 21) & 1; |
| 4693 | if (P) { |
| 4694 | switch (size & 1) { |
| 4695 | case 0: |
| 4696 | op = Q ? Iop_Mul32Fx4 : Iop_Mul32Fx2; |
| 4697 | op2 = Q ? Iop_Sub32Fx4 : Iop_Sub32Fx2; |
| 4698 | break; |
| 4699 | case 1: return False; |
| 4700 | default: vassert(0); |
| 4701 | } |
| 4702 | } else { |
| 4703 | switch (size & 1) { |
| 4704 | case 0: |
| 4705 | op = Q ? Iop_Mul32Fx4 : Iop_Mul32Fx2; |
| 4706 | op2 = Q ? Iop_Add32Fx4 : Iop_Add32Fx2; |
| 4707 | break; |
| 4708 | case 1: return False; |
| 4709 | default: vassert(0); |
| 4710 | } |
| 4711 | } |
sewardj | e84eeb4 | 2014-03-15 11:41:39 +0000 | [diff] [blame] | 4712 | assign(res, binop_w_fake_RM( |
| 4713 | op2, |
| 4714 | Q ? getQReg(dreg) : getDRegI64(dreg), |
| 4715 | binop_w_fake_RM(op, mkexpr(arg_n), |
| 4716 | mkexpr(arg_m)))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4717 | |
| 4718 | DIP("vml%c.f32 %c%u, %c%u, %c%u\n", |
| 4719 | P ? 's' : 'a', Q ? 'q' : 'd', |
| 4720 | dreg, Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4721 | } else { |
| 4722 | /* VMUL */ |
| 4723 | IROp op; |
| 4724 | if ((C >> 1) != 0) |
| 4725 | return False; |
| 4726 | op = Q ? Iop_Mul32Fx4 : Iop_Mul32Fx2 ; |
sewardj | e84eeb4 | 2014-03-15 11:41:39 +0000 | [diff] [blame] | 4727 | assign(res, binop_w_fake_RM(op, mkexpr(arg_n), mkexpr(arg_m))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4728 | DIP("vmul.f32 %c%u, %c%u, %c%u\n", |
| 4729 | Q ? 'q' : 'd', dreg, |
| 4730 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4731 | } |
| 4732 | } |
| 4733 | break; |
| 4734 | case 14: |
| 4735 | if (B == 0) { |
| 4736 | if (U == 0) { |
| 4737 | if ((C >> 1) == 0) { |
| 4738 | /* VCEQ */ |
| 4739 | IROp op; |
| 4740 | if ((theInstr >> 20) & 1) |
| 4741 | return False; |
| 4742 | op = Q ? Iop_CmpEQ32Fx4 : Iop_CmpEQ32Fx2; |
| 4743 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
| 4744 | DIP("vceq.f32 %c%u, %c%u, %c%u\n", |
| 4745 | Q ? 'q' : 'd', dreg, |
| 4746 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4747 | } else { |
| 4748 | return False; |
| 4749 | } |
| 4750 | } else { |
| 4751 | if ((C >> 1) == 0) { |
| 4752 | /* VCGE */ |
| 4753 | IROp op; |
| 4754 | if ((theInstr >> 20) & 1) |
| 4755 | return False; |
| 4756 | op = Q ? Iop_CmpGE32Fx4 : Iop_CmpGE32Fx2; |
| 4757 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
| 4758 | DIP("vcge.f32 %c%u, %c%u, %c%u\n", |
| 4759 | Q ? 'q' : 'd', dreg, |
| 4760 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4761 | } else { |
| 4762 | /* VCGT */ |
| 4763 | IROp op; |
| 4764 | if ((theInstr >> 20) & 1) |
| 4765 | return False; |
| 4766 | op = Q ? Iop_CmpGT32Fx4 : Iop_CmpGT32Fx2; |
| 4767 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
| 4768 | DIP("vcgt.f32 %c%u, %c%u, %c%u\n", |
| 4769 | Q ? 'q' : 'd', dreg, |
| 4770 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4771 | } |
| 4772 | } |
| 4773 | } else { |
| 4774 | if (U == 1) { |
| 4775 | /* VACGE, VACGT */ |
| 4776 | UInt op_bit = (theInstr >> 21) & 1; |
| 4777 | IROp op, op2; |
| 4778 | op2 = Q ? Iop_Abs32Fx4 : Iop_Abs32Fx2; |
| 4779 | if (op_bit) { |
| 4780 | op = Q ? Iop_CmpGT32Fx4 : Iop_CmpGT32Fx2; |
| 4781 | assign(res, binop(op, |
| 4782 | unop(op2, mkexpr(arg_n)), |
| 4783 | unop(op2, mkexpr(arg_m)))); |
| 4784 | } else { |
| 4785 | op = Q ? Iop_CmpGE32Fx4 : Iop_CmpGE32Fx2; |
| 4786 | assign(res, binop(op, |
| 4787 | unop(op2, mkexpr(arg_n)), |
| 4788 | unop(op2, mkexpr(arg_m)))); |
| 4789 | } |
| 4790 | DIP("vacg%c.f32 %c%u, %c%u, %c%u\n", op_bit ? 't' : 'e', |
| 4791 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', nreg, |
| 4792 | Q ? 'q' : 'd', mreg); |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 4793 | } else { |
| 4794 | return False; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4795 | } |
| 4796 | } |
| 4797 | break; |
| 4798 | case 15: |
| 4799 | if (B == 0) { |
| 4800 | if (U == 0) { |
| 4801 | /* VMAX, VMIN */ |
| 4802 | IROp op; |
| 4803 | if ((theInstr >> 20) & 1) |
| 4804 | return False; |
| 4805 | if ((theInstr >> 21) & 1) { |
| 4806 | op = Q ? Iop_Min32Fx4 : Iop_Min32Fx2; |
| 4807 | DIP("vmin.f32 %c%u, %c%u, %c%u\n", Q ? 'q' : 'd', dreg, |
| 4808 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4809 | } else { |
| 4810 | op = Q ? Iop_Max32Fx4 : Iop_Max32Fx2; |
| 4811 | DIP("vmax.f32 %c%u, %c%u, %c%u\n", Q ? 'q' : 'd', dreg, |
| 4812 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4813 | } |
| 4814 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
| 4815 | } else { |
| 4816 | /* VPMAX, VPMIN */ |
| 4817 | IROp op; |
| 4818 | if (Q) |
| 4819 | return False; |
| 4820 | if ((theInstr >> 20) & 1) |
| 4821 | return False; |
| 4822 | if ((theInstr >> 21) & 1) { |
| 4823 | op = Iop_PwMin32Fx2; |
| 4824 | DIP("vpmin.f32 d%u, d%u, d%u\n", dreg, nreg, mreg); |
| 4825 | } else { |
| 4826 | op = Iop_PwMax32Fx2; |
| 4827 | DIP("vpmax.f32 d%u, d%u, d%u\n", dreg, nreg, mreg); |
| 4828 | } |
| 4829 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
| 4830 | } |
| 4831 | } else { |
| 4832 | if (U == 0) { |
| 4833 | if ((C >> 1) == 0) { |
| 4834 | /* VRECPS */ |
| 4835 | if ((theInstr >> 20) & 1) |
| 4836 | return False; |
sewardj | 1ddee21 | 2014-08-24 14:00:19 +0000 | [diff] [blame] | 4837 | assign(res, binop(Q ? Iop_RecipStep32Fx4 |
| 4838 | : Iop_RecipStep32Fx2, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4839 | mkexpr(arg_n), |
| 4840 | mkexpr(arg_m))); |
| 4841 | DIP("vrecps.f32 %c%u, %c%u, %c%u\n", Q ? 'q' : 'd', dreg, |
| 4842 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4843 | } else { |
| 4844 | /* VRSQRTS */ |
| 4845 | if ((theInstr >> 20) & 1) |
| 4846 | return False; |
sewardj | 1ddee21 | 2014-08-24 14:00:19 +0000 | [diff] [blame] | 4847 | assign(res, binop(Q ? Iop_RSqrtStep32Fx4 |
| 4848 | : Iop_RSqrtStep32Fx2, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4849 | mkexpr(arg_n), |
| 4850 | mkexpr(arg_m))); |
| 4851 | DIP("vrsqrts.f32 %c%u, %c%u, %c%u\n", Q ? 'q' : 'd', dreg, |
| 4852 | Q ? 'q' : 'd', nreg, Q ? 'q' : 'd', mreg); |
| 4853 | } |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 4854 | } else { |
| 4855 | return False; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4856 | } |
| 4857 | } |
| 4858 | break; |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 4859 | default: |
| 4860 | /*NOTREACHED*/ |
| 4861 | vassert(0); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4862 | } |
| 4863 | |
| 4864 | if (Q) { |
| 4865 | putQReg(dreg, mkexpr(res), condT); |
| 4866 | } else { |
| 4867 | putDRegI64(dreg, mkexpr(res), condT); |
| 4868 | } |
| 4869 | |
| 4870 | return True; |
| 4871 | } |
| 4872 | |
| 4873 | /* A7.4.2 Three registers of different length */ |
| 4874 | static |
| 4875 | Bool dis_neon_data_3diff ( UInt theInstr, IRTemp condT ) |
| 4876 | { |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 4877 | /* In paths where this returns False, indicating a non-decodable |
| 4878 | instruction, there may still be some IR assignments to temporaries |
| 4879 | generated. This is inconvenient but harmless, and the post-front-end |
| 4880 | IR optimisation pass will just remove them anyway. So there's no |
| 4881 | effort made here to tidy it up. |
| 4882 | */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4883 | UInt A = (theInstr >> 8) & 0xf; |
| 4884 | UInt B = (theInstr >> 20) & 3; |
| 4885 | UInt U = (theInstr >> 24) & 1; |
| 4886 | UInt P = (theInstr >> 9) & 1; |
| 4887 | UInt mreg = get_neon_m_regno(theInstr); |
| 4888 | UInt nreg = get_neon_n_regno(theInstr); |
| 4889 | UInt dreg = get_neon_d_regno(theInstr); |
| 4890 | UInt size = B; |
| 4891 | ULong imm; |
| 4892 | IRTemp res, arg_m, arg_n, cond, tmp; |
| 4893 | IROp cvt, cvt2, cmp, op, op2, sh, add; |
| 4894 | switch (A) { |
| 4895 | case 0: case 1: case 2: case 3: |
| 4896 | /* VADDL, VADDW, VSUBL, VSUBW */ |
| 4897 | if (dreg & 1) |
| 4898 | return False; |
| 4899 | dreg >>= 1; |
| 4900 | size = B; |
| 4901 | switch (size) { |
| 4902 | case 0: |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 4903 | cvt = U ? Iop_Widen8Uto16x8 : Iop_Widen8Sto16x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4904 | op = (A & 2) ? Iop_Sub16x8 : Iop_Add16x8; |
| 4905 | break; |
| 4906 | case 1: |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 4907 | cvt = U ? Iop_Widen16Uto32x4 : Iop_Widen16Sto32x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4908 | op = (A & 2) ? Iop_Sub32x4 : Iop_Add32x4; |
| 4909 | break; |
| 4910 | case 2: |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 4911 | cvt = U ? Iop_Widen32Uto64x2 : Iop_Widen32Sto64x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4912 | op = (A & 2) ? Iop_Sub64x2 : Iop_Add64x2; |
| 4913 | break; |
| 4914 | case 3: |
| 4915 | return False; |
| 4916 | default: |
| 4917 | vassert(0); |
| 4918 | } |
| 4919 | arg_n = newTemp(Ity_V128); |
| 4920 | arg_m = newTemp(Ity_V128); |
| 4921 | if (A & 1) { |
| 4922 | if (nreg & 1) |
| 4923 | return False; |
| 4924 | nreg >>= 1; |
| 4925 | assign(arg_n, getQReg(nreg)); |
| 4926 | } else { |
| 4927 | assign(arg_n, unop(cvt, getDRegI64(nreg))); |
| 4928 | } |
| 4929 | assign(arg_m, unop(cvt, getDRegI64(mreg))); |
| 4930 | putQReg(dreg, binop(op, mkexpr(arg_n), mkexpr(arg_m)), |
| 4931 | condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4932 | DIP("v%s%c.%c%d q%u, %c%u, d%u\n", (A & 2) ? "sub" : "add", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4933 | (A & 1) ? 'w' : 'l', U ? 'u' : 's', 8 << size, dreg, |
| 4934 | (A & 1) ? 'q' : 'd', nreg, mreg); |
| 4935 | return True; |
| 4936 | case 4: |
| 4937 | /* VADDHN, VRADDHN */ |
| 4938 | if (mreg & 1) |
| 4939 | return False; |
| 4940 | mreg >>= 1; |
| 4941 | if (nreg & 1) |
| 4942 | return False; |
| 4943 | nreg >>= 1; |
| 4944 | size = B; |
| 4945 | switch (size) { |
| 4946 | case 0: |
| 4947 | op = Iop_Add16x8; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 4948 | cvt = Iop_NarrowUn16to8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4949 | sh = Iop_ShrN16x8; |
| 4950 | imm = 1U << 7; |
| 4951 | imm = (imm << 16) | imm; |
| 4952 | imm = (imm << 32) | imm; |
| 4953 | break; |
| 4954 | case 1: |
| 4955 | op = Iop_Add32x4; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 4956 | cvt = Iop_NarrowUn32to16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4957 | sh = Iop_ShrN32x4; |
| 4958 | imm = 1U << 15; |
| 4959 | imm = (imm << 32) | imm; |
| 4960 | break; |
| 4961 | case 2: |
| 4962 | op = Iop_Add64x2; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 4963 | cvt = Iop_NarrowUn64to32x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4964 | sh = Iop_ShrN64x2; |
| 4965 | imm = 1U << 31; |
| 4966 | break; |
| 4967 | case 3: |
| 4968 | return False; |
| 4969 | default: |
| 4970 | vassert(0); |
| 4971 | } |
| 4972 | tmp = newTemp(Ity_V128); |
| 4973 | res = newTemp(Ity_V128); |
| 4974 | assign(tmp, binop(op, getQReg(nreg), getQReg(mreg))); |
| 4975 | if (U) { |
| 4976 | /* VRADDHN */ |
| 4977 | assign(res, binop(op, mkexpr(tmp), |
| 4978 | binop(Iop_64HLtoV128, mkU64(imm), mkU64(imm)))); |
| 4979 | } else { |
| 4980 | assign(res, mkexpr(tmp)); |
| 4981 | } |
| 4982 | putDRegI64(dreg, unop(cvt, binop(sh, mkexpr(res), mkU8(8 << size))), |
| 4983 | condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 4984 | DIP("v%saddhn.i%d d%u, q%u, q%u\n", U ? "r" : "", 16 << size, dreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 4985 | nreg, mreg); |
| 4986 | return True; |
| 4987 | case 5: |
| 4988 | /* VABAL */ |
| 4989 | if (!((theInstr >> 23) & 1)) { |
| 4990 | vpanic("VABA should not be in dis_neon_data_3diff\n"); |
| 4991 | } |
| 4992 | if (dreg & 1) |
| 4993 | return False; |
| 4994 | dreg >>= 1; |
| 4995 | switch (size) { |
| 4996 | case 0: |
| 4997 | cmp = U ? Iop_CmpGT8Ux8 : Iop_CmpGT8Sx8; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 4998 | cvt = U ? Iop_Widen8Uto16x8 : Iop_Widen8Sto16x8; |
| 4999 | cvt2 = Iop_Widen8Sto16x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5000 | op = Iop_Sub16x8; |
| 5001 | op2 = Iop_Add16x8; |
| 5002 | break; |
| 5003 | case 1: |
| 5004 | cmp = U ? Iop_CmpGT16Ux4 : Iop_CmpGT16Sx4; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 5005 | cvt = U ? Iop_Widen16Uto32x4 : Iop_Widen16Sto32x4; |
| 5006 | cvt2 = Iop_Widen16Sto32x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5007 | op = Iop_Sub32x4; |
| 5008 | op2 = Iop_Add32x4; |
| 5009 | break; |
| 5010 | case 2: |
| 5011 | cmp = U ? Iop_CmpGT32Ux2 : Iop_CmpGT32Sx2; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 5012 | cvt = U ? Iop_Widen32Uto64x2 : Iop_Widen32Sto64x2; |
| 5013 | cvt2 = Iop_Widen32Sto64x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5014 | op = Iop_Sub64x2; |
| 5015 | op2 = Iop_Add64x2; |
| 5016 | break; |
| 5017 | case 3: |
| 5018 | return False; |
| 5019 | default: |
| 5020 | vassert(0); |
| 5021 | } |
| 5022 | arg_n = newTemp(Ity_V128); |
| 5023 | arg_m = newTemp(Ity_V128); |
| 5024 | cond = newTemp(Ity_V128); |
| 5025 | res = newTemp(Ity_V128); |
| 5026 | assign(arg_n, unop(cvt, getDRegI64(nreg))); |
| 5027 | assign(arg_m, unop(cvt, getDRegI64(mreg))); |
| 5028 | assign(cond, unop(cvt2, binop(cmp, getDRegI64(nreg), |
| 5029 | getDRegI64(mreg)))); |
| 5030 | assign(res, binop(op2, |
| 5031 | binop(Iop_OrV128, |
| 5032 | binop(Iop_AndV128, |
| 5033 | binop(op, mkexpr(arg_n), mkexpr(arg_m)), |
| 5034 | mkexpr(cond)), |
| 5035 | binop(Iop_AndV128, |
| 5036 | binop(op, mkexpr(arg_m), mkexpr(arg_n)), |
| 5037 | unop(Iop_NotV128, mkexpr(cond)))), |
| 5038 | getQReg(dreg))); |
| 5039 | putQReg(dreg, mkexpr(res), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5040 | DIP("vabal.%c%d q%u, d%u, d%u\n", U ? 'u' : 's', 8 << size, dreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5041 | nreg, mreg); |
| 5042 | return True; |
| 5043 | case 6: |
| 5044 | /* VSUBHN, VRSUBHN */ |
| 5045 | if (mreg & 1) |
| 5046 | return False; |
| 5047 | mreg >>= 1; |
| 5048 | if (nreg & 1) |
| 5049 | return False; |
| 5050 | nreg >>= 1; |
| 5051 | size = B; |
| 5052 | switch (size) { |
| 5053 | case 0: |
| 5054 | op = Iop_Sub16x8; |
| 5055 | op2 = Iop_Add16x8; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 5056 | cvt = Iop_NarrowUn16to8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5057 | sh = Iop_ShrN16x8; |
| 5058 | imm = 1U << 7; |
| 5059 | imm = (imm << 16) | imm; |
| 5060 | imm = (imm << 32) | imm; |
| 5061 | break; |
| 5062 | case 1: |
| 5063 | op = Iop_Sub32x4; |
| 5064 | op2 = Iop_Add32x4; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 5065 | cvt = Iop_NarrowUn32to16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5066 | sh = Iop_ShrN32x4; |
| 5067 | imm = 1U << 15; |
| 5068 | imm = (imm << 32) | imm; |
| 5069 | break; |
| 5070 | case 2: |
| 5071 | op = Iop_Sub64x2; |
| 5072 | op2 = Iop_Add64x2; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 5073 | cvt = Iop_NarrowUn64to32x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5074 | sh = Iop_ShrN64x2; |
| 5075 | imm = 1U << 31; |
| 5076 | break; |
| 5077 | case 3: |
| 5078 | return False; |
| 5079 | default: |
| 5080 | vassert(0); |
| 5081 | } |
| 5082 | tmp = newTemp(Ity_V128); |
| 5083 | res = newTemp(Ity_V128); |
| 5084 | assign(tmp, binop(op, getQReg(nreg), getQReg(mreg))); |
| 5085 | if (U) { |
| 5086 | /* VRSUBHN */ |
| 5087 | assign(res, binop(op2, mkexpr(tmp), |
| 5088 | binop(Iop_64HLtoV128, mkU64(imm), mkU64(imm)))); |
| 5089 | } else { |
| 5090 | assign(res, mkexpr(tmp)); |
| 5091 | } |
| 5092 | putDRegI64(dreg, unop(cvt, binop(sh, mkexpr(res), mkU8(8 << size))), |
| 5093 | condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5094 | DIP("v%ssubhn.i%d d%u, q%u, q%u\n", U ? "r" : "", 16 << size, dreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5095 | nreg, mreg); |
| 5096 | return True; |
| 5097 | case 7: |
| 5098 | /* VABDL */ |
| 5099 | if (!((theInstr >> 23) & 1)) { |
| 5100 | vpanic("VABL should not be in dis_neon_data_3diff\n"); |
| 5101 | } |
| 5102 | if (dreg & 1) |
| 5103 | return False; |
| 5104 | dreg >>= 1; |
| 5105 | switch (size) { |
| 5106 | case 0: |
| 5107 | cmp = U ? Iop_CmpGT8Ux8 : Iop_CmpGT8Sx8; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 5108 | cvt = U ? Iop_Widen8Uto16x8 : Iop_Widen8Sto16x8; |
| 5109 | cvt2 = Iop_Widen8Sto16x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5110 | op = Iop_Sub16x8; |
| 5111 | break; |
| 5112 | case 1: |
| 5113 | cmp = U ? Iop_CmpGT16Ux4 : Iop_CmpGT16Sx4; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 5114 | cvt = U ? Iop_Widen16Uto32x4 : Iop_Widen16Sto32x4; |
| 5115 | cvt2 = Iop_Widen16Sto32x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5116 | op = Iop_Sub32x4; |
| 5117 | break; |
| 5118 | case 2: |
| 5119 | cmp = U ? Iop_CmpGT32Ux2 : Iop_CmpGT32Sx2; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 5120 | cvt = U ? Iop_Widen32Uto64x2 : Iop_Widen32Sto64x2; |
| 5121 | cvt2 = Iop_Widen32Sto64x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5122 | op = Iop_Sub64x2; |
| 5123 | break; |
| 5124 | case 3: |
| 5125 | return False; |
| 5126 | default: |
| 5127 | vassert(0); |
| 5128 | } |
| 5129 | arg_n = newTemp(Ity_V128); |
| 5130 | arg_m = newTemp(Ity_V128); |
| 5131 | cond = newTemp(Ity_V128); |
| 5132 | res = newTemp(Ity_V128); |
| 5133 | assign(arg_n, unop(cvt, getDRegI64(nreg))); |
| 5134 | assign(arg_m, unop(cvt, getDRegI64(mreg))); |
| 5135 | assign(cond, unop(cvt2, binop(cmp, getDRegI64(nreg), |
| 5136 | getDRegI64(mreg)))); |
| 5137 | assign(res, binop(Iop_OrV128, |
| 5138 | binop(Iop_AndV128, |
| 5139 | binop(op, mkexpr(arg_n), mkexpr(arg_m)), |
| 5140 | mkexpr(cond)), |
| 5141 | binop(Iop_AndV128, |
| 5142 | binop(op, mkexpr(arg_m), mkexpr(arg_n)), |
| 5143 | unop(Iop_NotV128, mkexpr(cond))))); |
| 5144 | putQReg(dreg, mkexpr(res), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5145 | DIP("vabdl.%c%d q%u, d%u, d%u\n", U ? 'u' : 's', 8 << size, dreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5146 | nreg, mreg); |
| 5147 | return True; |
| 5148 | case 8: |
| 5149 | case 10: |
| 5150 | /* VMLAL, VMLSL (integer) */ |
| 5151 | if (dreg & 1) |
| 5152 | return False; |
| 5153 | dreg >>= 1; |
| 5154 | size = B; |
| 5155 | switch (size) { |
| 5156 | case 0: |
| 5157 | op = U ? Iop_Mull8Ux8 : Iop_Mull8Sx8; |
| 5158 | op2 = P ? Iop_Sub16x8 : Iop_Add16x8; |
| 5159 | break; |
| 5160 | case 1: |
| 5161 | op = U ? Iop_Mull16Ux4 : Iop_Mull16Sx4; |
| 5162 | op2 = P ? Iop_Sub32x4 : Iop_Add32x4; |
| 5163 | break; |
| 5164 | case 2: |
| 5165 | op = U ? Iop_Mull32Ux2 : Iop_Mull32Sx2; |
| 5166 | op2 = P ? Iop_Sub64x2 : Iop_Add64x2; |
| 5167 | break; |
| 5168 | case 3: |
| 5169 | return False; |
| 5170 | default: |
| 5171 | vassert(0); |
| 5172 | } |
| 5173 | res = newTemp(Ity_V128); |
| 5174 | assign(res, binop(op, getDRegI64(nreg),getDRegI64(mreg))); |
| 5175 | putQReg(dreg, binop(op2, getQReg(dreg), mkexpr(res)), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5176 | DIP("vml%cl.%c%d q%u, d%u, d%u\n", P ? 's' : 'a', U ? 'u' : 's', |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5177 | 8 << size, dreg, nreg, mreg); |
| 5178 | return True; |
| 5179 | case 9: |
| 5180 | case 11: |
| 5181 | /* VQDMLAL, VQDMLSL */ |
| 5182 | if (U) |
| 5183 | return False; |
| 5184 | if (dreg & 1) |
| 5185 | return False; |
| 5186 | dreg >>= 1; |
| 5187 | size = B; |
| 5188 | switch (size) { |
| 5189 | case 0: case 3: |
| 5190 | return False; |
| 5191 | case 1: |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 5192 | op = Iop_QDMull16Sx4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5193 | cmp = Iop_CmpEQ16x4; |
| 5194 | add = P ? Iop_QSub32Sx4 : Iop_QAdd32Sx4; |
| 5195 | op2 = P ? Iop_Sub32x4 : Iop_Add32x4; |
| 5196 | imm = 1LL << 15; |
| 5197 | imm = (imm << 16) | imm; |
| 5198 | imm = (imm << 32) | imm; |
| 5199 | break; |
| 5200 | case 2: |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 5201 | op = Iop_QDMull32Sx2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5202 | cmp = Iop_CmpEQ32x2; |
| 5203 | add = P ? Iop_QSub64Sx2 : Iop_QAdd64Sx2; |
| 5204 | op2 = P ? Iop_Sub64x2 : Iop_Add64x2; |
| 5205 | imm = 1LL << 31; |
| 5206 | imm = (imm << 32) | imm; |
| 5207 | break; |
| 5208 | default: |
| 5209 | vassert(0); |
| 5210 | } |
| 5211 | res = newTemp(Ity_V128); |
| 5212 | tmp = newTemp(Ity_V128); |
| 5213 | assign(res, binop(op, getDRegI64(nreg), getDRegI64(mreg))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5214 | assign(tmp, binop(op2, getQReg(dreg), mkexpr(res))); |
| 5215 | setFlag_QC(mkexpr(tmp), binop(add, getQReg(dreg), mkexpr(res)), |
| 5216 | True, condT); |
| 5217 | setFlag_QC(binop(Iop_And64, |
| 5218 | binop(cmp, getDRegI64(nreg), mkU64(imm)), |
| 5219 | binop(cmp, getDRegI64(mreg), mkU64(imm))), |
| 5220 | mkU64(0), |
| 5221 | False, condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5222 | putQReg(dreg, binop(add, getQReg(dreg), mkexpr(res)), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5223 | DIP("vqdml%cl.s%d q%u, d%u, d%u\n", P ? 's' : 'a', 8 << size, dreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5224 | nreg, mreg); |
| 5225 | return True; |
| 5226 | case 12: |
| 5227 | case 14: |
| 5228 | /* VMULL (integer or polynomial) */ |
| 5229 | if (dreg & 1) |
| 5230 | return False; |
| 5231 | dreg >>= 1; |
| 5232 | size = B; |
| 5233 | switch (size) { |
| 5234 | case 0: |
| 5235 | op = (U) ? Iop_Mull8Ux8 : Iop_Mull8Sx8; |
| 5236 | if (P) |
| 5237 | op = Iop_PolynomialMull8x8; |
| 5238 | break; |
| 5239 | case 1: |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 5240 | if (P) return False; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5241 | op = (U) ? Iop_Mull16Ux4 : Iop_Mull16Sx4; |
| 5242 | break; |
| 5243 | case 2: |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 5244 | if (P) return False; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5245 | op = (U) ? Iop_Mull32Ux2 : Iop_Mull32Sx2; |
| 5246 | break; |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 5247 | case 3: |
| 5248 | return False; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5249 | default: |
| 5250 | vassert(0); |
| 5251 | } |
| 5252 | putQReg(dreg, binop(op, getDRegI64(nreg), |
| 5253 | getDRegI64(mreg)), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5254 | DIP("vmull.%c%d q%u, d%u, d%u\n", P ? 'p' : (U ? 'u' : 's'), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5255 | 8 << size, dreg, nreg, mreg); |
| 5256 | return True; |
| 5257 | case 13: |
| 5258 | /* VQDMULL */ |
| 5259 | if (U) |
| 5260 | return False; |
| 5261 | if (dreg & 1) |
| 5262 | return False; |
| 5263 | dreg >>= 1; |
| 5264 | size = B; |
| 5265 | switch (size) { |
| 5266 | case 0: |
| 5267 | case 3: |
| 5268 | return False; |
| 5269 | case 1: |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 5270 | op = Iop_QDMull16Sx4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5271 | op2 = Iop_CmpEQ16x4; |
| 5272 | imm = 1LL << 15; |
| 5273 | imm = (imm << 16) | imm; |
| 5274 | imm = (imm << 32) | imm; |
| 5275 | break; |
| 5276 | case 2: |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 5277 | op = Iop_QDMull32Sx2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5278 | op2 = Iop_CmpEQ32x2; |
| 5279 | imm = 1LL << 31; |
| 5280 | imm = (imm << 32) | imm; |
| 5281 | break; |
| 5282 | default: |
| 5283 | vassert(0); |
| 5284 | } |
| 5285 | putQReg(dreg, binop(op, getDRegI64(nreg), getDRegI64(mreg)), |
| 5286 | condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5287 | setFlag_QC(binop(Iop_And64, |
| 5288 | binop(op2, getDRegI64(nreg), mkU64(imm)), |
| 5289 | binop(op2, getDRegI64(mreg), mkU64(imm))), |
| 5290 | mkU64(0), |
| 5291 | False, condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5292 | DIP("vqdmull.s%d q%u, d%u, d%u\n", 8 << size, dreg, nreg, mreg); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5293 | return True; |
| 5294 | default: |
| 5295 | return False; |
| 5296 | } |
| 5297 | return False; |
| 5298 | } |
| 5299 | |
| 5300 | /* A7.4.3 Two registers and a scalar */ |
| 5301 | static |
| 5302 | Bool dis_neon_data_2reg_and_scalar ( UInt theInstr, IRTemp condT ) |
| 5303 | { |
| 5304 | # define INSN(_bMax,_bMin) SLICE_UInt(theInstr, (_bMax), (_bMin)) |
| 5305 | UInt U = INSN(24,24); |
| 5306 | UInt dreg = get_neon_d_regno(theInstr & ~(1 << 6)); |
| 5307 | UInt nreg = get_neon_n_regno(theInstr & ~(1 << 6)); |
| 5308 | UInt mreg = get_neon_m_regno(theInstr & ~(1 << 6)); |
| 5309 | UInt size = INSN(21,20); |
| 5310 | UInt index; |
| 5311 | UInt Q = INSN(24,24); |
| 5312 | |
| 5313 | if (INSN(27,25) != 1 || INSN(23,23) != 1 |
| 5314 | || INSN(6,6) != 1 || INSN(4,4) != 0) |
| 5315 | return False; |
| 5316 | |
| 5317 | /* VMLA, VMLS (scalar) */ |
| 5318 | if ((INSN(11,8) & BITS4(1,0,1,0)) == BITS4(0,0,0,0)) { |
| 5319 | IRTemp res, arg_m, arg_n; |
| 5320 | IROp dup, get, op, op2, add, sub; |
| 5321 | if (Q) { |
| 5322 | if ((dreg & 1) || (nreg & 1)) |
| 5323 | return False; |
| 5324 | dreg >>= 1; |
| 5325 | nreg >>= 1; |
| 5326 | res = newTemp(Ity_V128); |
| 5327 | arg_m = newTemp(Ity_V128); |
| 5328 | arg_n = newTemp(Ity_V128); |
| 5329 | assign(arg_n, getQReg(nreg)); |
| 5330 | switch(size) { |
| 5331 | case 1: |
| 5332 | dup = Iop_Dup16x8; |
| 5333 | get = Iop_GetElem16x4; |
| 5334 | index = mreg >> 3; |
| 5335 | mreg &= 7; |
| 5336 | break; |
| 5337 | case 2: |
| 5338 | dup = Iop_Dup32x4; |
| 5339 | get = Iop_GetElem32x2; |
| 5340 | index = mreg >> 4; |
| 5341 | mreg &= 0xf; |
| 5342 | break; |
| 5343 | case 0: |
| 5344 | case 3: |
| 5345 | return False; |
| 5346 | default: |
| 5347 | vassert(0); |
| 5348 | } |
| 5349 | assign(arg_m, unop(dup, binop(get, getDRegI64(mreg), mkU8(index)))); |
| 5350 | } else { |
| 5351 | res = newTemp(Ity_I64); |
| 5352 | arg_m = newTemp(Ity_I64); |
| 5353 | arg_n = newTemp(Ity_I64); |
| 5354 | assign(arg_n, getDRegI64(nreg)); |
| 5355 | switch(size) { |
| 5356 | case 1: |
| 5357 | dup = Iop_Dup16x4; |
| 5358 | get = Iop_GetElem16x4; |
| 5359 | index = mreg >> 3; |
| 5360 | mreg &= 7; |
| 5361 | break; |
| 5362 | case 2: |
| 5363 | dup = Iop_Dup32x2; |
| 5364 | get = Iop_GetElem32x2; |
| 5365 | index = mreg >> 4; |
| 5366 | mreg &= 0xf; |
| 5367 | break; |
| 5368 | case 0: |
| 5369 | case 3: |
| 5370 | return False; |
| 5371 | default: |
| 5372 | vassert(0); |
| 5373 | } |
| 5374 | assign(arg_m, unop(dup, binop(get, getDRegI64(mreg), mkU8(index)))); |
| 5375 | } |
| 5376 | if (INSN(8,8)) { |
| 5377 | switch (size) { |
| 5378 | case 2: |
| 5379 | op = Q ? Iop_Mul32Fx4 : Iop_Mul32Fx2; |
| 5380 | add = Q ? Iop_Add32Fx4 : Iop_Add32Fx2; |
| 5381 | sub = Q ? Iop_Sub32Fx4 : Iop_Sub32Fx2; |
| 5382 | break; |
| 5383 | case 0: |
| 5384 | case 1: |
| 5385 | case 3: |
| 5386 | return False; |
| 5387 | default: |
| 5388 | vassert(0); |
| 5389 | } |
| 5390 | } else { |
| 5391 | switch (size) { |
| 5392 | case 1: |
| 5393 | op = Q ? Iop_Mul16x8 : Iop_Mul16x4; |
| 5394 | add = Q ? Iop_Add16x8 : Iop_Add16x4; |
| 5395 | sub = Q ? Iop_Sub16x8 : Iop_Sub16x4; |
| 5396 | break; |
| 5397 | case 2: |
| 5398 | op = Q ? Iop_Mul32x4 : Iop_Mul32x2; |
| 5399 | add = Q ? Iop_Add32x4 : Iop_Add32x2; |
| 5400 | sub = Q ? Iop_Sub32x4 : Iop_Sub32x2; |
| 5401 | break; |
| 5402 | case 0: |
| 5403 | case 3: |
| 5404 | return False; |
| 5405 | default: |
| 5406 | vassert(0); |
| 5407 | } |
| 5408 | } |
| 5409 | op2 = INSN(10,10) ? sub : add; |
sewardj | e84eeb4 | 2014-03-15 11:41:39 +0000 | [diff] [blame] | 5410 | assign(res, binop_w_fake_RM(op, mkexpr(arg_n), mkexpr(arg_m))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5411 | if (Q) |
sewardj | e84eeb4 | 2014-03-15 11:41:39 +0000 | [diff] [blame] | 5412 | putQReg(dreg, binop_w_fake_RM(op2, getQReg(dreg), mkexpr(res)), |
| 5413 | condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5414 | else |
| 5415 | putDRegI64(dreg, binop(op2, getDRegI64(dreg), mkexpr(res)), |
| 5416 | condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5417 | DIP("vml%c.%c%d %c%u, %c%u, d%u[%u]\n", INSN(10,10) ? 's' : 'a', |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5418 | INSN(8,8) ? 'f' : 'i', 8 << size, |
| 5419 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', nreg, mreg, index); |
| 5420 | return True; |
| 5421 | } |
| 5422 | |
| 5423 | /* VMLAL, VMLSL (scalar) */ |
| 5424 | if ((INSN(11,8) & BITS4(1,0,1,1)) == BITS4(0,0,1,0)) { |
| 5425 | IRTemp res, arg_m, arg_n; |
| 5426 | IROp dup, get, op, op2, add, sub; |
| 5427 | if (dreg & 1) |
| 5428 | return False; |
| 5429 | dreg >>= 1; |
| 5430 | res = newTemp(Ity_V128); |
| 5431 | arg_m = newTemp(Ity_I64); |
| 5432 | arg_n = newTemp(Ity_I64); |
| 5433 | assign(arg_n, getDRegI64(nreg)); |
| 5434 | switch(size) { |
| 5435 | case 1: |
| 5436 | dup = Iop_Dup16x4; |
| 5437 | get = Iop_GetElem16x4; |
| 5438 | index = mreg >> 3; |
| 5439 | mreg &= 7; |
| 5440 | break; |
| 5441 | case 2: |
| 5442 | dup = Iop_Dup32x2; |
| 5443 | get = Iop_GetElem32x2; |
| 5444 | index = mreg >> 4; |
| 5445 | mreg &= 0xf; |
| 5446 | break; |
| 5447 | case 0: |
| 5448 | case 3: |
| 5449 | return False; |
| 5450 | default: |
| 5451 | vassert(0); |
| 5452 | } |
| 5453 | assign(arg_m, unop(dup, binop(get, getDRegI64(mreg), mkU8(index)))); |
| 5454 | switch (size) { |
| 5455 | case 1: |
| 5456 | op = U ? Iop_Mull16Ux4 : Iop_Mull16Sx4; |
| 5457 | add = Iop_Add32x4; |
| 5458 | sub = Iop_Sub32x4; |
| 5459 | break; |
| 5460 | case 2: |
| 5461 | op = U ? Iop_Mull32Ux2 : Iop_Mull32Sx2; |
| 5462 | add = Iop_Add64x2; |
| 5463 | sub = Iop_Sub64x2; |
| 5464 | break; |
| 5465 | case 0: |
| 5466 | case 3: |
| 5467 | return False; |
| 5468 | default: |
| 5469 | vassert(0); |
| 5470 | } |
| 5471 | op2 = INSN(10,10) ? sub : add; |
| 5472 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
| 5473 | putQReg(dreg, binop(op2, getQReg(dreg), mkexpr(res)), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5474 | DIP("vml%cl.%c%d q%u, d%u, d%u[%u]\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5475 | INSN(10,10) ? 's' : 'a', U ? 'u' : 's', |
| 5476 | 8 << size, dreg, nreg, mreg, index); |
| 5477 | return True; |
| 5478 | } |
| 5479 | |
| 5480 | /* VQDMLAL, VQDMLSL (scalar) */ |
| 5481 | if ((INSN(11,8) & BITS4(1,0,1,1)) == BITS4(0,0,1,1) && !U) { |
| 5482 | IRTemp res, arg_m, arg_n, tmp; |
| 5483 | IROp dup, get, op, op2, add, cmp; |
| 5484 | UInt P = INSN(10,10); |
| 5485 | ULong imm; |
| 5486 | if (dreg & 1) |
| 5487 | return False; |
| 5488 | dreg >>= 1; |
| 5489 | res = newTemp(Ity_V128); |
| 5490 | arg_m = newTemp(Ity_I64); |
| 5491 | arg_n = newTemp(Ity_I64); |
| 5492 | assign(arg_n, getDRegI64(nreg)); |
| 5493 | switch(size) { |
| 5494 | case 1: |
| 5495 | dup = Iop_Dup16x4; |
| 5496 | get = Iop_GetElem16x4; |
| 5497 | index = mreg >> 3; |
| 5498 | mreg &= 7; |
| 5499 | break; |
| 5500 | case 2: |
| 5501 | dup = Iop_Dup32x2; |
| 5502 | get = Iop_GetElem32x2; |
| 5503 | index = mreg >> 4; |
| 5504 | mreg &= 0xf; |
| 5505 | break; |
| 5506 | case 0: |
| 5507 | case 3: |
| 5508 | return False; |
| 5509 | default: |
| 5510 | vassert(0); |
| 5511 | } |
| 5512 | assign(arg_m, unop(dup, binop(get, getDRegI64(mreg), mkU8(index)))); |
| 5513 | switch (size) { |
| 5514 | case 0: |
| 5515 | case 3: |
| 5516 | return False; |
| 5517 | case 1: |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 5518 | op = Iop_QDMull16Sx4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5519 | cmp = Iop_CmpEQ16x4; |
| 5520 | add = P ? Iop_QSub32Sx4 : Iop_QAdd32Sx4; |
| 5521 | op2 = P ? Iop_Sub32x4 : Iop_Add32x4; |
| 5522 | imm = 1LL << 15; |
| 5523 | imm = (imm << 16) | imm; |
| 5524 | imm = (imm << 32) | imm; |
| 5525 | break; |
| 5526 | case 2: |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 5527 | op = Iop_QDMull32Sx2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5528 | cmp = Iop_CmpEQ32x2; |
| 5529 | add = P ? Iop_QSub64Sx2 : Iop_QAdd64Sx2; |
| 5530 | op2 = P ? Iop_Sub64x2 : Iop_Add64x2; |
| 5531 | imm = 1LL << 31; |
| 5532 | imm = (imm << 32) | imm; |
| 5533 | break; |
| 5534 | default: |
| 5535 | vassert(0); |
| 5536 | } |
| 5537 | res = newTemp(Ity_V128); |
| 5538 | tmp = newTemp(Ity_V128); |
| 5539 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5540 | assign(tmp, binop(op2, getQReg(dreg), mkexpr(res))); |
| 5541 | setFlag_QC(binop(Iop_And64, |
| 5542 | binop(cmp, mkexpr(arg_n), mkU64(imm)), |
| 5543 | binop(cmp, mkexpr(arg_m), mkU64(imm))), |
| 5544 | mkU64(0), |
| 5545 | False, condT); |
| 5546 | setFlag_QC(mkexpr(tmp), binop(add, getQReg(dreg), mkexpr(res)), |
| 5547 | True, condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5548 | putQReg(dreg, binop(add, getQReg(dreg), mkexpr(res)), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5549 | DIP("vqdml%cl.s%d q%u, d%u, d%u[%u]\n", P ? 's' : 'a', 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5550 | dreg, nreg, mreg, index); |
| 5551 | return True; |
| 5552 | } |
| 5553 | |
| 5554 | /* VMUL (by scalar) */ |
| 5555 | if ((INSN(11,8) & BITS4(1,1,1,0)) == BITS4(1,0,0,0)) { |
| 5556 | IRTemp res, arg_m, arg_n; |
| 5557 | IROp dup, get, op; |
| 5558 | if (Q) { |
| 5559 | if ((dreg & 1) || (nreg & 1)) |
| 5560 | return False; |
| 5561 | dreg >>= 1; |
| 5562 | nreg >>= 1; |
| 5563 | res = newTemp(Ity_V128); |
| 5564 | arg_m = newTemp(Ity_V128); |
| 5565 | arg_n = newTemp(Ity_V128); |
| 5566 | assign(arg_n, getQReg(nreg)); |
| 5567 | switch(size) { |
| 5568 | case 1: |
| 5569 | dup = Iop_Dup16x8; |
| 5570 | get = Iop_GetElem16x4; |
| 5571 | index = mreg >> 3; |
| 5572 | mreg &= 7; |
| 5573 | break; |
| 5574 | case 2: |
| 5575 | dup = Iop_Dup32x4; |
| 5576 | get = Iop_GetElem32x2; |
| 5577 | index = mreg >> 4; |
| 5578 | mreg &= 0xf; |
| 5579 | break; |
| 5580 | case 0: |
| 5581 | case 3: |
| 5582 | return False; |
| 5583 | default: |
| 5584 | vassert(0); |
| 5585 | } |
| 5586 | assign(arg_m, unop(dup, binop(get, getDRegI64(mreg), mkU8(index)))); |
| 5587 | } else { |
| 5588 | res = newTemp(Ity_I64); |
| 5589 | arg_m = newTemp(Ity_I64); |
| 5590 | arg_n = newTemp(Ity_I64); |
| 5591 | assign(arg_n, getDRegI64(nreg)); |
| 5592 | switch(size) { |
| 5593 | case 1: |
| 5594 | dup = Iop_Dup16x4; |
| 5595 | get = Iop_GetElem16x4; |
| 5596 | index = mreg >> 3; |
| 5597 | mreg &= 7; |
| 5598 | break; |
| 5599 | case 2: |
| 5600 | dup = Iop_Dup32x2; |
| 5601 | get = Iop_GetElem32x2; |
| 5602 | index = mreg >> 4; |
| 5603 | mreg &= 0xf; |
| 5604 | break; |
| 5605 | case 0: |
| 5606 | case 3: |
| 5607 | return False; |
| 5608 | default: |
| 5609 | vassert(0); |
| 5610 | } |
| 5611 | assign(arg_m, unop(dup, binop(get, getDRegI64(mreg), mkU8(index)))); |
| 5612 | } |
sewardj | a561f89 | 2011-07-19 07:37:03 +0000 | [diff] [blame] | 5613 | if (INSN(8,8)) { |
| 5614 | switch (size) { |
| 5615 | case 2: |
| 5616 | op = Q ? Iop_Mul32Fx4 : Iop_Mul32Fx2; |
| 5617 | break; |
| 5618 | case 0: |
| 5619 | case 1: |
| 5620 | case 3: |
| 5621 | return False; |
| 5622 | default: |
| 5623 | vassert(0); |
| 5624 | } |
| 5625 | } else { |
| 5626 | switch (size) { |
| 5627 | case 1: |
| 5628 | op = Q ? Iop_Mul16x8 : Iop_Mul16x4; |
| 5629 | break; |
| 5630 | case 2: |
| 5631 | op = Q ? Iop_Mul32x4 : Iop_Mul32x2; |
| 5632 | break; |
| 5633 | case 0: |
| 5634 | case 3: |
| 5635 | return False; |
| 5636 | default: |
| 5637 | vassert(0); |
| 5638 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5639 | } |
sewardj | e84eeb4 | 2014-03-15 11:41:39 +0000 | [diff] [blame] | 5640 | assign(res, binop_w_fake_RM(op, mkexpr(arg_n), mkexpr(arg_m))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5641 | if (Q) |
| 5642 | putQReg(dreg, mkexpr(res), condT); |
| 5643 | else |
| 5644 | putDRegI64(dreg, mkexpr(res), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5645 | DIP("vmul.%c%d %c%u, %c%u, d%u[%u]\n", INSN(8,8) ? 'f' : 'i', |
sewardj | a561f89 | 2011-07-19 07:37:03 +0000 | [diff] [blame] | 5646 | 8 << size, Q ? 'q' : 'd', dreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5647 | Q ? 'q' : 'd', nreg, mreg, index); |
| 5648 | return True; |
| 5649 | } |
| 5650 | |
| 5651 | /* VMULL (scalar) */ |
| 5652 | if (INSN(11,8) == BITS4(1,0,1,0)) { |
| 5653 | IRTemp res, arg_m, arg_n; |
| 5654 | IROp dup, get, op; |
| 5655 | if (dreg & 1) |
| 5656 | return False; |
| 5657 | dreg >>= 1; |
| 5658 | res = newTemp(Ity_V128); |
| 5659 | arg_m = newTemp(Ity_I64); |
| 5660 | arg_n = newTemp(Ity_I64); |
| 5661 | assign(arg_n, getDRegI64(nreg)); |
| 5662 | switch(size) { |
| 5663 | case 1: |
| 5664 | dup = Iop_Dup16x4; |
| 5665 | get = Iop_GetElem16x4; |
| 5666 | index = mreg >> 3; |
| 5667 | mreg &= 7; |
| 5668 | break; |
| 5669 | case 2: |
| 5670 | dup = Iop_Dup32x2; |
| 5671 | get = Iop_GetElem32x2; |
| 5672 | index = mreg >> 4; |
| 5673 | mreg &= 0xf; |
| 5674 | break; |
| 5675 | case 0: |
| 5676 | case 3: |
| 5677 | return False; |
| 5678 | default: |
| 5679 | vassert(0); |
| 5680 | } |
| 5681 | assign(arg_m, unop(dup, binop(get, getDRegI64(mreg), mkU8(index)))); |
| 5682 | switch (size) { |
| 5683 | case 1: op = U ? Iop_Mull16Ux4 : Iop_Mull16Sx4; break; |
| 5684 | case 2: op = U ? Iop_Mull32Ux2 : Iop_Mull32Sx2; break; |
| 5685 | case 0: case 3: return False; |
| 5686 | default: vassert(0); |
| 5687 | } |
| 5688 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
| 5689 | putQReg(dreg, mkexpr(res), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5690 | DIP("vmull.%c%d q%u, d%u, d%u[%u]\n", U ? 'u' : 's', 8 << size, dreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5691 | nreg, mreg, index); |
| 5692 | return True; |
| 5693 | } |
| 5694 | |
| 5695 | /* VQDMULL */ |
| 5696 | if (INSN(11,8) == BITS4(1,0,1,1) && !U) { |
| 5697 | IROp op ,op2, dup, get; |
| 5698 | ULong imm; |
sewardj | 06122e7 | 2011-03-28 12:14:48 +0000 | [diff] [blame] | 5699 | IRTemp arg_m, arg_n; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5700 | if (dreg & 1) |
| 5701 | return False; |
| 5702 | dreg >>= 1; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5703 | arg_m = newTemp(Ity_I64); |
| 5704 | arg_n = newTemp(Ity_I64); |
| 5705 | assign(arg_n, getDRegI64(nreg)); |
| 5706 | switch(size) { |
| 5707 | case 1: |
| 5708 | dup = Iop_Dup16x4; |
| 5709 | get = Iop_GetElem16x4; |
| 5710 | index = mreg >> 3; |
| 5711 | mreg &= 7; |
| 5712 | break; |
| 5713 | case 2: |
| 5714 | dup = Iop_Dup32x2; |
| 5715 | get = Iop_GetElem32x2; |
| 5716 | index = mreg >> 4; |
| 5717 | mreg &= 0xf; |
| 5718 | break; |
| 5719 | case 0: |
| 5720 | case 3: |
| 5721 | return False; |
| 5722 | default: |
| 5723 | vassert(0); |
| 5724 | } |
| 5725 | assign(arg_m, unop(dup, binop(get, getDRegI64(mreg), mkU8(index)))); |
| 5726 | switch (size) { |
| 5727 | case 0: |
| 5728 | case 3: |
| 5729 | return False; |
| 5730 | case 1: |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 5731 | op = Iop_QDMull16Sx4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5732 | op2 = Iop_CmpEQ16x4; |
| 5733 | imm = 1LL << 15; |
| 5734 | imm = (imm << 16) | imm; |
| 5735 | imm = (imm << 32) | imm; |
| 5736 | break; |
| 5737 | case 2: |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 5738 | op = Iop_QDMull32Sx2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5739 | op2 = Iop_CmpEQ32x2; |
| 5740 | imm = 1LL << 31; |
| 5741 | imm = (imm << 32) | imm; |
| 5742 | break; |
| 5743 | default: |
| 5744 | vassert(0); |
| 5745 | } |
| 5746 | putQReg(dreg, binop(op, mkexpr(arg_n), mkexpr(arg_m)), |
| 5747 | condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5748 | setFlag_QC(binop(Iop_And64, |
| 5749 | binop(op2, mkexpr(arg_n), mkU64(imm)), |
| 5750 | binop(op2, mkexpr(arg_m), mkU64(imm))), |
| 5751 | mkU64(0), |
| 5752 | False, condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5753 | DIP("vqdmull.s%d q%u, d%u, d%u[%u]\n", 8 << size, dreg, nreg, mreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5754 | index); |
| 5755 | return True; |
| 5756 | } |
| 5757 | |
| 5758 | /* VQDMULH */ |
| 5759 | if (INSN(11,8) == BITS4(1,1,0,0)) { |
| 5760 | IROp op ,op2, dup, get; |
| 5761 | ULong imm; |
| 5762 | IRTemp res, arg_m, arg_n; |
| 5763 | if (Q) { |
| 5764 | if ((dreg & 1) || (nreg & 1)) |
| 5765 | return False; |
| 5766 | dreg >>= 1; |
| 5767 | nreg >>= 1; |
| 5768 | res = newTemp(Ity_V128); |
| 5769 | arg_m = newTemp(Ity_V128); |
| 5770 | arg_n = newTemp(Ity_V128); |
| 5771 | assign(arg_n, getQReg(nreg)); |
| 5772 | switch(size) { |
| 5773 | case 1: |
| 5774 | dup = Iop_Dup16x8; |
| 5775 | get = Iop_GetElem16x4; |
| 5776 | index = mreg >> 3; |
| 5777 | mreg &= 7; |
| 5778 | break; |
| 5779 | case 2: |
| 5780 | dup = Iop_Dup32x4; |
| 5781 | get = Iop_GetElem32x2; |
| 5782 | index = mreg >> 4; |
| 5783 | mreg &= 0xf; |
| 5784 | break; |
| 5785 | case 0: |
| 5786 | case 3: |
| 5787 | return False; |
| 5788 | default: |
| 5789 | vassert(0); |
| 5790 | } |
| 5791 | assign(arg_m, unop(dup, binop(get, getDRegI64(mreg), mkU8(index)))); |
| 5792 | } else { |
| 5793 | res = newTemp(Ity_I64); |
| 5794 | arg_m = newTemp(Ity_I64); |
| 5795 | arg_n = newTemp(Ity_I64); |
| 5796 | assign(arg_n, getDRegI64(nreg)); |
| 5797 | switch(size) { |
| 5798 | case 1: |
| 5799 | dup = Iop_Dup16x4; |
| 5800 | get = Iop_GetElem16x4; |
| 5801 | index = mreg >> 3; |
| 5802 | mreg &= 7; |
| 5803 | break; |
| 5804 | case 2: |
| 5805 | dup = Iop_Dup32x2; |
| 5806 | get = Iop_GetElem32x2; |
| 5807 | index = mreg >> 4; |
| 5808 | mreg &= 0xf; |
| 5809 | break; |
| 5810 | case 0: |
| 5811 | case 3: |
| 5812 | return False; |
| 5813 | default: |
| 5814 | vassert(0); |
| 5815 | } |
| 5816 | assign(arg_m, unop(dup, binop(get, getDRegI64(mreg), mkU8(index)))); |
| 5817 | } |
| 5818 | switch (size) { |
| 5819 | case 0: |
| 5820 | case 3: |
| 5821 | return False; |
| 5822 | case 1: |
| 5823 | op = Q ? Iop_QDMulHi16Sx8 : Iop_QDMulHi16Sx4; |
| 5824 | op2 = Q ? Iop_CmpEQ16x8 : Iop_CmpEQ16x4; |
| 5825 | imm = 1LL << 15; |
| 5826 | imm = (imm << 16) | imm; |
| 5827 | imm = (imm << 32) | imm; |
| 5828 | break; |
| 5829 | case 2: |
| 5830 | op = Q ? Iop_QDMulHi32Sx4 : Iop_QDMulHi32Sx2; |
| 5831 | op2 = Q ? Iop_CmpEQ32x4 : Iop_CmpEQ32x2; |
| 5832 | imm = 1LL << 31; |
| 5833 | imm = (imm << 32) | imm; |
| 5834 | break; |
| 5835 | default: |
| 5836 | vassert(0); |
| 5837 | } |
| 5838 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5839 | setFlag_QC(binop(Q ? Iop_AndV128 : Iop_And64, |
| 5840 | binop(op2, mkexpr(arg_n), |
| 5841 | Q ? mkU128(imm) : mkU64(imm)), |
| 5842 | binop(op2, mkexpr(arg_m), |
| 5843 | Q ? mkU128(imm) : mkU64(imm))), |
| 5844 | Q ? mkU128(0) : mkU64(0), |
| 5845 | Q, condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5846 | if (Q) |
| 5847 | putQReg(dreg, mkexpr(res), condT); |
| 5848 | else |
| 5849 | putDRegI64(dreg, mkexpr(res), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5850 | DIP("vqdmulh.s%d %c%u, %c%u, d%u[%u]\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5851 | 8 << size, Q ? 'q' : 'd', dreg, |
| 5852 | Q ? 'q' : 'd', nreg, mreg, index); |
| 5853 | return True; |
| 5854 | } |
| 5855 | |
| 5856 | /* VQRDMULH (scalar) */ |
| 5857 | if (INSN(11,8) == BITS4(1,1,0,1)) { |
| 5858 | IROp op ,op2, dup, get; |
| 5859 | ULong imm; |
| 5860 | IRTemp res, arg_m, arg_n; |
| 5861 | if (Q) { |
| 5862 | if ((dreg & 1) || (nreg & 1)) |
| 5863 | return False; |
| 5864 | dreg >>= 1; |
| 5865 | nreg >>= 1; |
| 5866 | res = newTemp(Ity_V128); |
| 5867 | arg_m = newTemp(Ity_V128); |
| 5868 | arg_n = newTemp(Ity_V128); |
| 5869 | assign(arg_n, getQReg(nreg)); |
| 5870 | switch(size) { |
| 5871 | case 1: |
| 5872 | dup = Iop_Dup16x8; |
| 5873 | get = Iop_GetElem16x4; |
| 5874 | index = mreg >> 3; |
| 5875 | mreg &= 7; |
| 5876 | break; |
| 5877 | case 2: |
| 5878 | dup = Iop_Dup32x4; |
| 5879 | get = Iop_GetElem32x2; |
| 5880 | index = mreg >> 4; |
| 5881 | mreg &= 0xf; |
| 5882 | break; |
| 5883 | case 0: |
| 5884 | case 3: |
| 5885 | return False; |
| 5886 | default: |
| 5887 | vassert(0); |
| 5888 | } |
| 5889 | assign(arg_m, unop(dup, binop(get, getDRegI64(mreg), mkU8(index)))); |
| 5890 | } else { |
| 5891 | res = newTemp(Ity_I64); |
| 5892 | arg_m = newTemp(Ity_I64); |
| 5893 | arg_n = newTemp(Ity_I64); |
| 5894 | assign(arg_n, getDRegI64(nreg)); |
| 5895 | switch(size) { |
| 5896 | case 1: |
| 5897 | dup = Iop_Dup16x4; |
| 5898 | get = Iop_GetElem16x4; |
| 5899 | index = mreg >> 3; |
| 5900 | mreg &= 7; |
| 5901 | break; |
| 5902 | case 2: |
| 5903 | dup = Iop_Dup32x2; |
| 5904 | get = Iop_GetElem32x2; |
| 5905 | index = mreg >> 4; |
| 5906 | mreg &= 0xf; |
| 5907 | break; |
| 5908 | case 0: |
| 5909 | case 3: |
| 5910 | return False; |
| 5911 | default: |
| 5912 | vassert(0); |
| 5913 | } |
| 5914 | assign(arg_m, unop(dup, binop(get, getDRegI64(mreg), mkU8(index)))); |
| 5915 | } |
| 5916 | switch (size) { |
| 5917 | case 0: |
| 5918 | case 3: |
| 5919 | return False; |
| 5920 | case 1: |
| 5921 | op = Q ? Iop_QRDMulHi16Sx8 : Iop_QRDMulHi16Sx4; |
| 5922 | op2 = Q ? Iop_CmpEQ16x8 : Iop_CmpEQ16x4; |
| 5923 | imm = 1LL << 15; |
| 5924 | imm = (imm << 16) | imm; |
| 5925 | imm = (imm << 32) | imm; |
| 5926 | break; |
| 5927 | case 2: |
| 5928 | op = Q ? Iop_QRDMulHi32Sx4 : Iop_QRDMulHi32Sx2; |
| 5929 | op2 = Q ? Iop_CmpEQ32x4 : Iop_CmpEQ32x2; |
| 5930 | imm = 1LL << 31; |
| 5931 | imm = (imm << 32) | imm; |
| 5932 | break; |
| 5933 | default: |
| 5934 | vassert(0); |
| 5935 | } |
| 5936 | assign(res, binop(op, mkexpr(arg_n), mkexpr(arg_m))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5937 | setFlag_QC(binop(Q ? Iop_AndV128 : Iop_And64, |
| 5938 | binop(op2, mkexpr(arg_n), |
| 5939 | Q ? mkU128(imm) : mkU64(imm)), |
| 5940 | binop(op2, mkexpr(arg_m), |
| 5941 | Q ? mkU128(imm) : mkU64(imm))), |
| 5942 | Q ? mkU128(0) : mkU64(0), |
| 5943 | Q, condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5944 | if (Q) |
| 5945 | putQReg(dreg, mkexpr(res), condT); |
| 5946 | else |
| 5947 | putDRegI64(dreg, mkexpr(res), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 5948 | DIP("vqrdmulh.s%d %c%u, %c%u, d%u[%u]\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 5949 | 8 << size, Q ? 'q' : 'd', dreg, |
| 5950 | Q ? 'q' : 'd', nreg, mreg, index); |
| 5951 | return True; |
| 5952 | } |
| 5953 | |
| 5954 | return False; |
| 5955 | # undef INSN |
| 5956 | } |
| 5957 | |
| 5958 | /* A7.4.4 Two registers and a shift amount */ |
| 5959 | static |
| 5960 | Bool dis_neon_data_2reg_and_shift ( UInt theInstr, IRTemp condT ) |
| 5961 | { |
| 5962 | UInt A = (theInstr >> 8) & 0xf; |
| 5963 | UInt B = (theInstr >> 6) & 1; |
| 5964 | UInt L = (theInstr >> 7) & 1; |
| 5965 | UInt U = (theInstr >> 24) & 1; |
| 5966 | UInt Q = B; |
| 5967 | UInt imm6 = (theInstr >> 16) & 0x3f; |
| 5968 | UInt shift_imm; |
| 5969 | UInt size = 4; |
| 5970 | UInt tmp; |
| 5971 | UInt mreg = get_neon_m_regno(theInstr); |
| 5972 | UInt dreg = get_neon_d_regno(theInstr); |
| 5973 | ULong imm = 0; |
| 5974 | IROp op, cvt, add = Iop_INVALID, cvt2, op_rev; |
| 5975 | IRTemp reg_m, res, mask; |
| 5976 | |
| 5977 | if (L == 0 && ((theInstr >> 19) & 7) == 0) |
| 5978 | /* It is one reg and immediate */ |
| 5979 | return False; |
| 5980 | |
| 5981 | tmp = (L << 6) | imm6; |
| 5982 | if (tmp & 0x40) { |
| 5983 | size = 3; |
| 5984 | shift_imm = 64 - imm6; |
| 5985 | } else if (tmp & 0x20) { |
| 5986 | size = 2; |
| 5987 | shift_imm = 64 - imm6; |
| 5988 | } else if (tmp & 0x10) { |
| 5989 | size = 1; |
| 5990 | shift_imm = 32 - imm6; |
| 5991 | } else if (tmp & 0x8) { |
| 5992 | size = 0; |
| 5993 | shift_imm = 16 - imm6; |
| 5994 | } else { |
| 5995 | return False; |
| 5996 | } |
| 5997 | |
| 5998 | switch (A) { |
| 5999 | case 3: |
| 6000 | case 2: |
| 6001 | /* VRSHR, VRSRA */ |
| 6002 | if (shift_imm > 0) { |
| 6003 | IRExpr *imm_val; |
| 6004 | imm = 1L; |
| 6005 | switch (size) { |
| 6006 | case 0: |
| 6007 | imm = (imm << 8) | imm; |
| 6008 | /* fall through */ |
| 6009 | case 1: |
| 6010 | imm = (imm << 16) | imm; |
| 6011 | /* fall through */ |
| 6012 | case 2: |
| 6013 | imm = (imm << 32) | imm; |
| 6014 | /* fall through */ |
| 6015 | case 3: |
| 6016 | break; |
| 6017 | default: |
| 6018 | vassert(0); |
| 6019 | } |
| 6020 | if (Q) { |
| 6021 | reg_m = newTemp(Ity_V128); |
| 6022 | res = newTemp(Ity_V128); |
| 6023 | imm_val = binop(Iop_64HLtoV128, mkU64(imm), mkU64(imm)); |
| 6024 | assign(reg_m, getQReg(mreg)); |
| 6025 | switch (size) { |
| 6026 | case 0: |
| 6027 | add = Iop_Add8x16; |
| 6028 | op = U ? Iop_ShrN8x16 : Iop_SarN8x16; |
| 6029 | break; |
| 6030 | case 1: |
| 6031 | add = Iop_Add16x8; |
| 6032 | op = U ? Iop_ShrN16x8 : Iop_SarN16x8; |
| 6033 | break; |
| 6034 | case 2: |
| 6035 | add = Iop_Add32x4; |
| 6036 | op = U ? Iop_ShrN32x4 : Iop_SarN32x4; |
| 6037 | break; |
| 6038 | case 3: |
| 6039 | add = Iop_Add64x2; |
| 6040 | op = U ? Iop_ShrN64x2 : Iop_SarN64x2; |
| 6041 | break; |
| 6042 | default: |
| 6043 | vassert(0); |
| 6044 | } |
| 6045 | } else { |
| 6046 | reg_m = newTemp(Ity_I64); |
| 6047 | res = newTemp(Ity_I64); |
| 6048 | imm_val = mkU64(imm); |
| 6049 | assign(reg_m, getDRegI64(mreg)); |
| 6050 | switch (size) { |
| 6051 | case 0: |
| 6052 | add = Iop_Add8x8; |
| 6053 | op = U ? Iop_ShrN8x8 : Iop_SarN8x8; |
| 6054 | break; |
| 6055 | case 1: |
| 6056 | add = Iop_Add16x4; |
| 6057 | op = U ? Iop_ShrN16x4 : Iop_SarN16x4; |
| 6058 | break; |
| 6059 | case 2: |
| 6060 | add = Iop_Add32x2; |
| 6061 | op = U ? Iop_ShrN32x2 : Iop_SarN32x2; |
| 6062 | break; |
| 6063 | case 3: |
| 6064 | add = Iop_Add64; |
| 6065 | op = U ? Iop_Shr64 : Iop_Sar64; |
| 6066 | break; |
| 6067 | default: |
| 6068 | vassert(0); |
| 6069 | } |
| 6070 | } |
| 6071 | assign(res, |
| 6072 | binop(add, |
| 6073 | binop(op, |
| 6074 | mkexpr(reg_m), |
| 6075 | mkU8(shift_imm)), |
| 6076 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 6077 | binop(op, |
| 6078 | mkexpr(reg_m), |
| 6079 | mkU8(shift_imm - 1)), |
| 6080 | imm_val))); |
| 6081 | } else { |
| 6082 | if (Q) { |
| 6083 | res = newTemp(Ity_V128); |
| 6084 | assign(res, getQReg(mreg)); |
| 6085 | } else { |
| 6086 | res = newTemp(Ity_I64); |
| 6087 | assign(res, getDRegI64(mreg)); |
| 6088 | } |
| 6089 | } |
| 6090 | if (A == 3) { |
| 6091 | if (Q) { |
| 6092 | putQReg(dreg, binop(add, mkexpr(res), getQReg(dreg)), |
| 6093 | condT); |
| 6094 | } else { |
| 6095 | putDRegI64(dreg, binop(add, mkexpr(res), getDRegI64(dreg)), |
| 6096 | condT); |
| 6097 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6098 | DIP("vrsra.%c%d %c%u, %c%u, #%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6099 | U ? 'u' : 's', 8 << size, |
| 6100 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg, shift_imm); |
| 6101 | } else { |
| 6102 | if (Q) { |
| 6103 | putQReg(dreg, mkexpr(res), condT); |
| 6104 | } else { |
| 6105 | putDRegI64(dreg, mkexpr(res), condT); |
| 6106 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6107 | DIP("vrshr.%c%d %c%u, %c%u, #%u\n", U ? 'u' : 's', 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6108 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg, shift_imm); |
| 6109 | } |
| 6110 | return True; |
| 6111 | case 1: |
| 6112 | case 0: |
| 6113 | /* VSHR, VSRA */ |
| 6114 | if (Q) { |
| 6115 | reg_m = newTemp(Ity_V128); |
| 6116 | assign(reg_m, getQReg(mreg)); |
| 6117 | res = newTemp(Ity_V128); |
| 6118 | } else { |
| 6119 | reg_m = newTemp(Ity_I64); |
| 6120 | assign(reg_m, getDRegI64(mreg)); |
| 6121 | res = newTemp(Ity_I64); |
| 6122 | } |
| 6123 | if (Q) { |
| 6124 | switch (size) { |
| 6125 | case 0: |
| 6126 | op = U ? Iop_ShrN8x16 : Iop_SarN8x16; |
| 6127 | add = Iop_Add8x16; |
| 6128 | break; |
| 6129 | case 1: |
| 6130 | op = U ? Iop_ShrN16x8 : Iop_SarN16x8; |
| 6131 | add = Iop_Add16x8; |
| 6132 | break; |
| 6133 | case 2: |
| 6134 | op = U ? Iop_ShrN32x4 : Iop_SarN32x4; |
| 6135 | add = Iop_Add32x4; |
| 6136 | break; |
| 6137 | case 3: |
| 6138 | op = U ? Iop_ShrN64x2 : Iop_SarN64x2; |
| 6139 | add = Iop_Add64x2; |
| 6140 | break; |
| 6141 | default: |
| 6142 | vassert(0); |
| 6143 | } |
| 6144 | } else { |
| 6145 | switch (size) { |
| 6146 | case 0: |
| 6147 | op = U ? Iop_ShrN8x8 : Iop_SarN8x8; |
| 6148 | add = Iop_Add8x8; |
| 6149 | break; |
| 6150 | case 1: |
| 6151 | op = U ? Iop_ShrN16x4 : Iop_SarN16x4; |
| 6152 | add = Iop_Add16x4; |
| 6153 | break; |
| 6154 | case 2: |
| 6155 | op = U ? Iop_ShrN32x2 : Iop_SarN32x2; |
| 6156 | add = Iop_Add32x2; |
| 6157 | break; |
| 6158 | case 3: |
| 6159 | op = U ? Iop_Shr64 : Iop_Sar64; |
| 6160 | add = Iop_Add64; |
| 6161 | break; |
| 6162 | default: |
| 6163 | vassert(0); |
| 6164 | } |
| 6165 | } |
| 6166 | assign(res, binop(op, mkexpr(reg_m), mkU8(shift_imm))); |
| 6167 | if (A == 1) { |
| 6168 | if (Q) { |
| 6169 | putQReg(dreg, binop(add, mkexpr(res), getQReg(dreg)), |
| 6170 | condT); |
| 6171 | } else { |
| 6172 | putDRegI64(dreg, binop(add, mkexpr(res), getDRegI64(dreg)), |
| 6173 | condT); |
| 6174 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6175 | DIP("vsra.%c%d %c%u, %c%u, #%u\n", U ? 'u' : 's', 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6176 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg, shift_imm); |
| 6177 | } else { |
| 6178 | if (Q) { |
| 6179 | putQReg(dreg, mkexpr(res), condT); |
| 6180 | } else { |
| 6181 | putDRegI64(dreg, mkexpr(res), condT); |
| 6182 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6183 | DIP("vshr.%c%d %c%u, %c%u, #%u\n", U ? 'u' : 's', 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6184 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg, shift_imm); |
| 6185 | } |
| 6186 | return True; |
| 6187 | case 4: |
| 6188 | /* VSRI */ |
| 6189 | if (!U) |
| 6190 | return False; |
| 6191 | if (Q) { |
| 6192 | res = newTemp(Ity_V128); |
| 6193 | mask = newTemp(Ity_V128); |
| 6194 | } else { |
| 6195 | res = newTemp(Ity_I64); |
| 6196 | mask = newTemp(Ity_I64); |
| 6197 | } |
| 6198 | switch (size) { |
| 6199 | case 0: op = Q ? Iop_ShrN8x16 : Iop_ShrN8x8; break; |
| 6200 | case 1: op = Q ? Iop_ShrN16x8 : Iop_ShrN16x4; break; |
| 6201 | case 2: op = Q ? Iop_ShrN32x4 : Iop_ShrN32x2; break; |
| 6202 | case 3: op = Q ? Iop_ShrN64x2 : Iop_Shr64; break; |
| 6203 | default: vassert(0); |
| 6204 | } |
| 6205 | if (Q) { |
| 6206 | assign(mask, binop(op, binop(Iop_64HLtoV128, |
| 6207 | mkU64(0xFFFFFFFFFFFFFFFFLL), |
| 6208 | mkU64(0xFFFFFFFFFFFFFFFFLL)), |
| 6209 | mkU8(shift_imm))); |
| 6210 | assign(res, binop(Iop_OrV128, |
| 6211 | binop(Iop_AndV128, |
| 6212 | getQReg(dreg), |
| 6213 | unop(Iop_NotV128, |
| 6214 | mkexpr(mask))), |
| 6215 | binop(op, |
| 6216 | getQReg(mreg), |
| 6217 | mkU8(shift_imm)))); |
| 6218 | putQReg(dreg, mkexpr(res), condT); |
| 6219 | } else { |
| 6220 | assign(mask, binop(op, mkU64(0xFFFFFFFFFFFFFFFFLL), |
| 6221 | mkU8(shift_imm))); |
| 6222 | assign(res, binop(Iop_Or64, |
| 6223 | binop(Iop_And64, |
| 6224 | getDRegI64(dreg), |
| 6225 | unop(Iop_Not64, |
| 6226 | mkexpr(mask))), |
| 6227 | binop(op, |
| 6228 | getDRegI64(mreg), |
| 6229 | mkU8(shift_imm)))); |
| 6230 | putDRegI64(dreg, mkexpr(res), condT); |
| 6231 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6232 | DIP("vsri.%d %c%u, %c%u, #%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6233 | 8 << size, Q ? 'q' : 'd', dreg, |
| 6234 | Q ? 'q' : 'd', mreg, shift_imm); |
| 6235 | return True; |
| 6236 | case 5: |
| 6237 | if (U) { |
| 6238 | /* VSLI */ |
| 6239 | shift_imm = 8 * (1 << size) - shift_imm; |
| 6240 | if (Q) { |
| 6241 | res = newTemp(Ity_V128); |
| 6242 | mask = newTemp(Ity_V128); |
| 6243 | } else { |
| 6244 | res = newTemp(Ity_I64); |
| 6245 | mask = newTemp(Ity_I64); |
| 6246 | } |
| 6247 | switch (size) { |
| 6248 | case 0: op = Q ? Iop_ShlN8x16 : Iop_ShlN8x8; break; |
| 6249 | case 1: op = Q ? Iop_ShlN16x8 : Iop_ShlN16x4; break; |
| 6250 | case 2: op = Q ? Iop_ShlN32x4 : Iop_ShlN32x2; break; |
| 6251 | case 3: op = Q ? Iop_ShlN64x2 : Iop_Shl64; break; |
| 6252 | default: vassert(0); |
| 6253 | } |
| 6254 | if (Q) { |
| 6255 | assign(mask, binop(op, binop(Iop_64HLtoV128, |
| 6256 | mkU64(0xFFFFFFFFFFFFFFFFLL), |
| 6257 | mkU64(0xFFFFFFFFFFFFFFFFLL)), |
| 6258 | mkU8(shift_imm))); |
| 6259 | assign(res, binop(Iop_OrV128, |
| 6260 | binop(Iop_AndV128, |
| 6261 | getQReg(dreg), |
| 6262 | unop(Iop_NotV128, |
| 6263 | mkexpr(mask))), |
| 6264 | binop(op, |
| 6265 | getQReg(mreg), |
| 6266 | mkU8(shift_imm)))); |
| 6267 | putQReg(dreg, mkexpr(res), condT); |
| 6268 | } else { |
| 6269 | assign(mask, binop(op, mkU64(0xFFFFFFFFFFFFFFFFLL), |
| 6270 | mkU8(shift_imm))); |
| 6271 | assign(res, binop(Iop_Or64, |
| 6272 | binop(Iop_And64, |
| 6273 | getDRegI64(dreg), |
| 6274 | unop(Iop_Not64, |
| 6275 | mkexpr(mask))), |
| 6276 | binop(op, |
| 6277 | getDRegI64(mreg), |
| 6278 | mkU8(shift_imm)))); |
| 6279 | putDRegI64(dreg, mkexpr(res), condT); |
| 6280 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6281 | DIP("vsli.%d %c%u, %c%u, #%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6282 | 8 << size, Q ? 'q' : 'd', dreg, |
| 6283 | Q ? 'q' : 'd', mreg, shift_imm); |
| 6284 | return True; |
| 6285 | } else { |
| 6286 | /* VSHL #imm */ |
| 6287 | shift_imm = 8 * (1 << size) - shift_imm; |
| 6288 | if (Q) { |
| 6289 | res = newTemp(Ity_V128); |
| 6290 | } else { |
| 6291 | res = newTemp(Ity_I64); |
| 6292 | } |
| 6293 | switch (size) { |
| 6294 | case 0: op = Q ? Iop_ShlN8x16 : Iop_ShlN8x8; break; |
| 6295 | case 1: op = Q ? Iop_ShlN16x8 : Iop_ShlN16x4; break; |
| 6296 | case 2: op = Q ? Iop_ShlN32x4 : Iop_ShlN32x2; break; |
| 6297 | case 3: op = Q ? Iop_ShlN64x2 : Iop_Shl64; break; |
| 6298 | default: vassert(0); |
| 6299 | } |
| 6300 | assign(res, binop(op, Q ? getQReg(mreg) : getDRegI64(mreg), |
| 6301 | mkU8(shift_imm))); |
| 6302 | if (Q) { |
| 6303 | putQReg(dreg, mkexpr(res), condT); |
| 6304 | } else { |
| 6305 | putDRegI64(dreg, mkexpr(res), condT); |
| 6306 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6307 | DIP("vshl.i%d %c%u, %c%u, #%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6308 | 8 << size, Q ? 'q' : 'd', dreg, |
| 6309 | Q ? 'q' : 'd', mreg, shift_imm); |
| 6310 | return True; |
| 6311 | } |
| 6312 | break; |
| 6313 | case 6: |
| 6314 | case 7: |
| 6315 | /* VQSHL, VQSHLU */ |
| 6316 | shift_imm = 8 * (1 << size) - shift_imm; |
| 6317 | if (U) { |
| 6318 | if (A & 1) { |
| 6319 | switch (size) { |
| 6320 | case 0: |
sewardj | 1dd3ec1 | 2014-08-15 09:11:08 +0000 | [diff] [blame] | 6321 | op = Q ? Iop_QShlNsatUU8x16 : Iop_QShlNsatUU8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6322 | op_rev = Q ? Iop_ShrN8x16 : Iop_ShrN8x8; |
| 6323 | break; |
| 6324 | case 1: |
sewardj | 1dd3ec1 | 2014-08-15 09:11:08 +0000 | [diff] [blame] | 6325 | op = Q ? Iop_QShlNsatUU16x8 : Iop_QShlNsatUU16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6326 | op_rev = Q ? Iop_ShrN16x8 : Iop_ShrN16x4; |
| 6327 | break; |
| 6328 | case 2: |
sewardj | 1dd3ec1 | 2014-08-15 09:11:08 +0000 | [diff] [blame] | 6329 | op = Q ? Iop_QShlNsatUU32x4 : Iop_QShlNsatUU32x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6330 | op_rev = Q ? Iop_ShrN32x4 : Iop_ShrN32x2; |
| 6331 | break; |
| 6332 | case 3: |
sewardj | 1dd3ec1 | 2014-08-15 09:11:08 +0000 | [diff] [blame] | 6333 | op = Q ? Iop_QShlNsatUU64x2 : Iop_QShlNsatUU64x1; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6334 | op_rev = Q ? Iop_ShrN64x2 : Iop_Shr64; |
| 6335 | break; |
| 6336 | default: |
| 6337 | vassert(0); |
| 6338 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6339 | DIP("vqshl.u%d %c%u, %c%u, #%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6340 | 8 << size, |
| 6341 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg, shift_imm); |
| 6342 | } else { |
| 6343 | switch (size) { |
| 6344 | case 0: |
sewardj | 1dd3ec1 | 2014-08-15 09:11:08 +0000 | [diff] [blame] | 6345 | op = Q ? Iop_QShlNsatSU8x16 : Iop_QShlNsatSU8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6346 | op_rev = Q ? Iop_ShrN8x16 : Iop_ShrN8x8; |
| 6347 | break; |
| 6348 | case 1: |
sewardj | 1dd3ec1 | 2014-08-15 09:11:08 +0000 | [diff] [blame] | 6349 | op = Q ? Iop_QShlNsatSU16x8 : Iop_QShlNsatSU16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6350 | op_rev = Q ? Iop_ShrN16x8 : Iop_ShrN16x4; |
| 6351 | break; |
| 6352 | case 2: |
sewardj | 1dd3ec1 | 2014-08-15 09:11:08 +0000 | [diff] [blame] | 6353 | op = Q ? Iop_QShlNsatSU32x4 : Iop_QShlNsatSU32x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6354 | op_rev = Q ? Iop_ShrN32x4 : Iop_ShrN32x2; |
| 6355 | break; |
| 6356 | case 3: |
sewardj | 1dd3ec1 | 2014-08-15 09:11:08 +0000 | [diff] [blame] | 6357 | op = Q ? Iop_QShlNsatSU64x2 : Iop_QShlNsatSU64x1; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6358 | op_rev = Q ? Iop_ShrN64x2 : Iop_Shr64; |
| 6359 | break; |
| 6360 | default: |
| 6361 | vassert(0); |
| 6362 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6363 | DIP("vqshlu.s%d %c%u, %c%u, #%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6364 | 8 << size, |
| 6365 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg, shift_imm); |
| 6366 | } |
| 6367 | } else { |
| 6368 | if (!(A & 1)) |
| 6369 | return False; |
| 6370 | switch (size) { |
| 6371 | case 0: |
sewardj | 1dd3ec1 | 2014-08-15 09:11:08 +0000 | [diff] [blame] | 6372 | op = Q ? Iop_QShlNsatSS8x16 : Iop_QShlNsatSS8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6373 | op_rev = Q ? Iop_SarN8x16 : Iop_SarN8x8; |
| 6374 | break; |
| 6375 | case 1: |
sewardj | 1dd3ec1 | 2014-08-15 09:11:08 +0000 | [diff] [blame] | 6376 | op = Q ? Iop_QShlNsatSS16x8 : Iop_QShlNsatSS16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6377 | op_rev = Q ? Iop_SarN16x8 : Iop_SarN16x4; |
| 6378 | break; |
| 6379 | case 2: |
sewardj | 1dd3ec1 | 2014-08-15 09:11:08 +0000 | [diff] [blame] | 6380 | op = Q ? Iop_QShlNsatSS32x4 : Iop_QShlNsatSS32x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6381 | op_rev = Q ? Iop_SarN32x4 : Iop_SarN32x2; |
| 6382 | break; |
| 6383 | case 3: |
sewardj | 1dd3ec1 | 2014-08-15 09:11:08 +0000 | [diff] [blame] | 6384 | op = Q ? Iop_QShlNsatSS64x2 : Iop_QShlNsatSS64x1; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6385 | op_rev = Q ? Iop_SarN64x2 : Iop_Sar64; |
| 6386 | break; |
| 6387 | default: |
| 6388 | vassert(0); |
| 6389 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6390 | DIP("vqshl.s%d %c%u, %c%u, #%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6391 | 8 << size, |
| 6392 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg, shift_imm); |
| 6393 | } |
| 6394 | if (Q) { |
| 6395 | tmp = newTemp(Ity_V128); |
| 6396 | res = newTemp(Ity_V128); |
| 6397 | reg_m = newTemp(Ity_V128); |
| 6398 | assign(reg_m, getQReg(mreg)); |
| 6399 | } else { |
| 6400 | tmp = newTemp(Ity_I64); |
| 6401 | res = newTemp(Ity_I64); |
| 6402 | reg_m = newTemp(Ity_I64); |
| 6403 | assign(reg_m, getDRegI64(mreg)); |
| 6404 | } |
| 6405 | assign(res, binop(op, mkexpr(reg_m), mkU8(shift_imm))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6406 | assign(tmp, binop(op_rev, mkexpr(res), mkU8(shift_imm))); |
| 6407 | setFlag_QC(mkexpr(tmp), mkexpr(reg_m), Q, condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6408 | if (Q) |
| 6409 | putQReg(dreg, mkexpr(res), condT); |
| 6410 | else |
| 6411 | putDRegI64(dreg, mkexpr(res), condT); |
| 6412 | return True; |
| 6413 | case 8: |
| 6414 | if (!U) { |
| 6415 | if (L == 1) |
| 6416 | return False; |
| 6417 | size++; |
| 6418 | dreg = ((theInstr >> 18) & 0x10) | ((theInstr >> 12) & 0xF); |
| 6419 | mreg = ((theInstr >> 1) & 0x10) | (theInstr & 0xF); |
| 6420 | if (mreg & 1) |
| 6421 | return False; |
| 6422 | mreg >>= 1; |
| 6423 | if (!B) { |
| 6424 | /* VSHRN*/ |
| 6425 | IROp narOp; |
| 6426 | reg_m = newTemp(Ity_V128); |
| 6427 | assign(reg_m, getQReg(mreg)); |
| 6428 | res = newTemp(Ity_I64); |
| 6429 | switch (size) { |
| 6430 | case 1: |
| 6431 | op = Iop_ShrN16x8; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6432 | narOp = Iop_NarrowUn16to8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6433 | break; |
| 6434 | case 2: |
| 6435 | op = Iop_ShrN32x4; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6436 | narOp = Iop_NarrowUn32to16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6437 | break; |
| 6438 | case 3: |
| 6439 | op = Iop_ShrN64x2; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6440 | narOp = Iop_NarrowUn64to32x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6441 | break; |
| 6442 | default: |
| 6443 | vassert(0); |
| 6444 | } |
| 6445 | assign(res, unop(narOp, |
| 6446 | binop(op, |
| 6447 | mkexpr(reg_m), |
| 6448 | mkU8(shift_imm)))); |
| 6449 | putDRegI64(dreg, mkexpr(res), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6450 | DIP("vshrn.i%d d%u, q%u, #%u\n", 8 << size, dreg, mreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6451 | shift_imm); |
| 6452 | return True; |
| 6453 | } else { |
| 6454 | /* VRSHRN */ |
| 6455 | IROp addOp, shOp, narOp; |
| 6456 | IRExpr *imm_val; |
| 6457 | reg_m = newTemp(Ity_V128); |
| 6458 | assign(reg_m, getQReg(mreg)); |
| 6459 | res = newTemp(Ity_I64); |
| 6460 | imm = 1L; |
| 6461 | switch (size) { |
| 6462 | case 0: imm = (imm << 8) | imm; /* fall through */ |
| 6463 | case 1: imm = (imm << 16) | imm; /* fall through */ |
| 6464 | case 2: imm = (imm << 32) | imm; /* fall through */ |
| 6465 | case 3: break; |
| 6466 | default: vassert(0); |
| 6467 | } |
| 6468 | imm_val = binop(Iop_64HLtoV128, mkU64(imm), mkU64(imm)); |
| 6469 | switch (size) { |
| 6470 | case 1: |
| 6471 | addOp = Iop_Add16x8; |
| 6472 | shOp = Iop_ShrN16x8; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6473 | narOp = Iop_NarrowUn16to8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6474 | break; |
| 6475 | case 2: |
| 6476 | addOp = Iop_Add32x4; |
| 6477 | shOp = Iop_ShrN32x4; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6478 | narOp = Iop_NarrowUn32to16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6479 | break; |
| 6480 | case 3: |
| 6481 | addOp = Iop_Add64x2; |
| 6482 | shOp = Iop_ShrN64x2; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6483 | narOp = Iop_NarrowUn64to32x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6484 | break; |
| 6485 | default: |
| 6486 | vassert(0); |
| 6487 | } |
| 6488 | assign(res, unop(narOp, |
| 6489 | binop(addOp, |
| 6490 | binop(shOp, |
| 6491 | mkexpr(reg_m), |
| 6492 | mkU8(shift_imm)), |
| 6493 | binop(Iop_AndV128, |
| 6494 | binop(shOp, |
| 6495 | mkexpr(reg_m), |
| 6496 | mkU8(shift_imm - 1)), |
| 6497 | imm_val)))); |
| 6498 | putDRegI64(dreg, mkexpr(res), condT); |
| 6499 | if (shift_imm == 0) { |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6500 | DIP("vmov%d d%u, q%u, #%u\n", 8 << size, dreg, mreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6501 | shift_imm); |
| 6502 | } else { |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6503 | DIP("vrshrn.i%d d%u, q%u, #%u\n", 8 << size, dreg, mreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6504 | shift_imm); |
| 6505 | } |
| 6506 | return True; |
| 6507 | } |
| 6508 | } else { |
| 6509 | /* fall through */ |
| 6510 | } |
| 6511 | case 9: |
| 6512 | dreg = ((theInstr >> 18) & 0x10) | ((theInstr >> 12) & 0xF); |
| 6513 | mreg = ((theInstr >> 1) & 0x10) | (theInstr & 0xF); |
| 6514 | if (mreg & 1) |
| 6515 | return False; |
| 6516 | mreg >>= 1; |
| 6517 | size++; |
| 6518 | if ((theInstr >> 8) & 1) { |
| 6519 | switch (size) { |
| 6520 | case 1: |
| 6521 | op = U ? Iop_ShrN16x8 : Iop_SarN16x8; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6522 | cvt = U ? Iop_QNarrowUn16Uto8Ux8 : Iop_QNarrowUn16Sto8Sx8; |
| 6523 | cvt2 = U ? Iop_Widen8Uto16x8 : Iop_Widen8Sto16x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6524 | break; |
| 6525 | case 2: |
| 6526 | op = U ? Iop_ShrN32x4 : Iop_SarN32x4; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6527 | cvt = U ? Iop_QNarrowUn32Uto16Ux4 : Iop_QNarrowUn32Sto16Sx4; |
| 6528 | cvt2 = U ? Iop_Widen16Uto32x4 : Iop_Widen16Sto32x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6529 | break; |
| 6530 | case 3: |
| 6531 | op = U ? Iop_ShrN64x2 : Iop_SarN64x2; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6532 | cvt = U ? Iop_QNarrowUn64Uto32Ux2 : Iop_QNarrowUn64Sto32Sx2; |
| 6533 | cvt2 = U ? Iop_Widen32Uto64x2 : Iop_Widen32Sto64x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6534 | break; |
| 6535 | default: |
| 6536 | vassert(0); |
| 6537 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6538 | DIP("vq%sshrn.%c%d d%u, q%u, #%u\n", B ? "r" : "", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6539 | U ? 'u' : 's', 8 << size, dreg, mreg, shift_imm); |
| 6540 | } else { |
| 6541 | vassert(U); |
| 6542 | switch (size) { |
| 6543 | case 1: |
| 6544 | op = Iop_SarN16x8; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6545 | cvt = Iop_QNarrowUn16Sto8Ux8; |
| 6546 | cvt2 = Iop_Widen8Uto16x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6547 | break; |
| 6548 | case 2: |
| 6549 | op = Iop_SarN32x4; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6550 | cvt = Iop_QNarrowUn32Sto16Ux4; |
| 6551 | cvt2 = Iop_Widen16Uto32x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6552 | break; |
| 6553 | case 3: |
| 6554 | op = Iop_SarN64x2; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6555 | cvt = Iop_QNarrowUn64Sto32Ux2; |
| 6556 | cvt2 = Iop_Widen32Uto64x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6557 | break; |
| 6558 | default: |
| 6559 | vassert(0); |
| 6560 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6561 | DIP("vq%sshrun.s%d d%u, q%u, #%u\n", B ? "r" : "", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6562 | 8 << size, dreg, mreg, shift_imm); |
| 6563 | } |
| 6564 | if (B) { |
| 6565 | if (shift_imm > 0) { |
| 6566 | imm = 1; |
| 6567 | switch (size) { |
| 6568 | case 1: imm = (imm << 16) | imm; /* fall through */ |
| 6569 | case 2: imm = (imm << 32) | imm; /* fall through */ |
| 6570 | case 3: break; |
| 6571 | case 0: default: vassert(0); |
| 6572 | } |
| 6573 | switch (size) { |
| 6574 | case 1: add = Iop_Add16x8; break; |
| 6575 | case 2: add = Iop_Add32x4; break; |
| 6576 | case 3: add = Iop_Add64x2; break; |
| 6577 | case 0: default: vassert(0); |
| 6578 | } |
| 6579 | } |
| 6580 | } |
| 6581 | reg_m = newTemp(Ity_V128); |
| 6582 | res = newTemp(Ity_V128); |
| 6583 | assign(reg_m, getQReg(mreg)); |
| 6584 | if (B) { |
| 6585 | /* VQRSHRN, VQRSHRUN */ |
| 6586 | assign(res, binop(add, |
| 6587 | binop(op, mkexpr(reg_m), mkU8(shift_imm)), |
| 6588 | binop(Iop_AndV128, |
| 6589 | binop(op, |
| 6590 | mkexpr(reg_m), |
| 6591 | mkU8(shift_imm - 1)), |
| 6592 | mkU128(imm)))); |
| 6593 | } else { |
| 6594 | /* VQSHRN, VQSHRUN */ |
| 6595 | assign(res, binop(op, mkexpr(reg_m), mkU8(shift_imm))); |
| 6596 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6597 | setFlag_QC(unop(cvt2, unop(cvt, mkexpr(res))), mkexpr(res), |
| 6598 | True, condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6599 | putDRegI64(dreg, unop(cvt, mkexpr(res)), condT); |
| 6600 | return True; |
| 6601 | case 10: |
| 6602 | /* VSHLL |
| 6603 | VMOVL ::= VSHLL #0 */ |
| 6604 | if (B) |
| 6605 | return False; |
| 6606 | if (dreg & 1) |
| 6607 | return False; |
| 6608 | dreg >>= 1; |
| 6609 | shift_imm = (8 << size) - shift_imm; |
| 6610 | res = newTemp(Ity_V128); |
| 6611 | switch (size) { |
| 6612 | case 0: |
| 6613 | op = Iop_ShlN16x8; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6614 | cvt = U ? Iop_Widen8Uto16x8 : Iop_Widen8Sto16x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6615 | break; |
| 6616 | case 1: |
| 6617 | op = Iop_ShlN32x4; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6618 | cvt = U ? Iop_Widen16Uto32x4 : Iop_Widen16Sto32x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6619 | break; |
| 6620 | case 2: |
| 6621 | op = Iop_ShlN64x2; |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 6622 | cvt = U ? Iop_Widen32Uto64x2 : Iop_Widen32Sto64x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6623 | break; |
| 6624 | case 3: |
| 6625 | return False; |
| 6626 | default: |
| 6627 | vassert(0); |
| 6628 | } |
| 6629 | assign(res, binop(op, unop(cvt, getDRegI64(mreg)), mkU8(shift_imm))); |
| 6630 | putQReg(dreg, mkexpr(res), condT); |
| 6631 | if (shift_imm == 0) { |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6632 | DIP("vmovl.%c%d q%u, d%u\n", U ? 'u' : 's', 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6633 | dreg, mreg); |
| 6634 | } else { |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6635 | DIP("vshll.%c%d q%u, d%u, #%u\n", U ? 'u' : 's', 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6636 | dreg, mreg, shift_imm); |
| 6637 | } |
| 6638 | return True; |
| 6639 | case 14: |
| 6640 | case 15: |
| 6641 | /* VCVT floating-point <-> fixed-point */ |
| 6642 | if ((theInstr >> 8) & 1) { |
| 6643 | if (U) { |
| 6644 | op = Q ? Iop_F32ToFixed32Ux4_RZ : Iop_F32ToFixed32Ux2_RZ; |
| 6645 | } else { |
| 6646 | op = Q ? Iop_F32ToFixed32Sx4_RZ : Iop_F32ToFixed32Sx2_RZ; |
| 6647 | } |
| 6648 | DIP("vcvt.%c32.f32 %c%u, %c%u, #%u\n", U ? 'u' : 's', |
| 6649 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg, |
| 6650 | 64 - ((theInstr >> 16) & 0x3f)); |
| 6651 | } else { |
| 6652 | if (U) { |
| 6653 | op = Q ? Iop_Fixed32UToF32x4_RN : Iop_Fixed32UToF32x2_RN; |
| 6654 | } else { |
| 6655 | op = Q ? Iop_Fixed32SToF32x4_RN : Iop_Fixed32SToF32x2_RN; |
| 6656 | } |
| 6657 | DIP("vcvt.f32.%c32 %c%u, %c%u, #%u\n", U ? 'u' : 's', |
| 6658 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg, |
| 6659 | 64 - ((theInstr >> 16) & 0x3f)); |
| 6660 | } |
| 6661 | if (((theInstr >> 21) & 1) == 0) |
| 6662 | return False; |
| 6663 | if (Q) { |
| 6664 | putQReg(dreg, binop(op, getQReg(mreg), |
| 6665 | mkU8(64 - ((theInstr >> 16) & 0x3f))), condT); |
| 6666 | } else { |
| 6667 | putDRegI64(dreg, binop(op, getDRegI64(mreg), |
| 6668 | mkU8(64 - ((theInstr >> 16) & 0x3f))), condT); |
| 6669 | } |
| 6670 | return True; |
| 6671 | default: |
| 6672 | return False; |
| 6673 | |
| 6674 | } |
| 6675 | return False; |
| 6676 | } |
| 6677 | |
| 6678 | /* A7.4.5 Two registers, miscellaneous */ |
| 6679 | static |
| 6680 | Bool dis_neon_data_2reg_misc ( UInt theInstr, IRTemp condT ) |
| 6681 | { |
| 6682 | UInt A = (theInstr >> 16) & 3; |
| 6683 | UInt B = (theInstr >> 6) & 0x1f; |
| 6684 | UInt Q = (theInstr >> 6) & 1; |
| 6685 | UInt U = (theInstr >> 24) & 1; |
| 6686 | UInt size = (theInstr >> 18) & 3; |
| 6687 | UInt dreg = get_neon_d_regno(theInstr); |
| 6688 | UInt mreg = get_neon_m_regno(theInstr); |
| 6689 | UInt F = (theInstr >> 10) & 1; |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 6690 | IRTemp arg_d = IRTemp_INVALID; |
| 6691 | IRTemp arg_m = IRTemp_INVALID; |
| 6692 | IRTemp res = IRTemp_INVALID; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6693 | switch (A) { |
| 6694 | case 0: |
| 6695 | if (Q) { |
| 6696 | arg_m = newTemp(Ity_V128); |
| 6697 | res = newTemp(Ity_V128); |
| 6698 | assign(arg_m, getQReg(mreg)); |
| 6699 | } else { |
| 6700 | arg_m = newTemp(Ity_I64); |
| 6701 | res = newTemp(Ity_I64); |
| 6702 | assign(arg_m, getDRegI64(mreg)); |
| 6703 | } |
| 6704 | switch (B >> 1) { |
| 6705 | case 0: { |
| 6706 | /* VREV64 */ |
| 6707 | IROp op; |
| 6708 | switch (size) { |
| 6709 | case 0: |
sewardj | 3368035 | 2014-06-26 10:49:33 +0000 | [diff] [blame] | 6710 | op = Q ? Iop_Reverse8sIn64_x2 : Iop_Reverse8sIn64_x1; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6711 | break; |
| 6712 | case 1: |
sewardj | 3368035 | 2014-06-26 10:49:33 +0000 | [diff] [blame] | 6713 | op = Q ? Iop_Reverse16sIn64_x2 : Iop_Reverse16sIn64_x1; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6714 | break; |
| 6715 | case 2: |
sewardj | 3368035 | 2014-06-26 10:49:33 +0000 | [diff] [blame] | 6716 | op = Q ? Iop_Reverse32sIn64_x2 : Iop_Reverse32sIn64_x1; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6717 | break; |
| 6718 | case 3: |
| 6719 | return False; |
| 6720 | default: |
| 6721 | vassert(0); |
| 6722 | } |
| 6723 | assign(res, unop(op, mkexpr(arg_m))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6724 | DIP("vrev64.%d %c%u, %c%u\n", 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6725 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 6726 | break; |
| 6727 | } |
| 6728 | case 1: { |
| 6729 | /* VREV32 */ |
| 6730 | IROp op; |
| 6731 | switch (size) { |
| 6732 | case 0: |
sewardj | 3368035 | 2014-06-26 10:49:33 +0000 | [diff] [blame] | 6733 | op = Q ? Iop_Reverse8sIn32_x4 : Iop_Reverse8sIn32_x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6734 | break; |
| 6735 | case 1: |
sewardj | 3368035 | 2014-06-26 10:49:33 +0000 | [diff] [blame] | 6736 | op = Q ? Iop_Reverse16sIn32_x4 : Iop_Reverse16sIn32_x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6737 | break; |
| 6738 | case 2: |
| 6739 | case 3: |
| 6740 | return False; |
| 6741 | default: |
| 6742 | vassert(0); |
| 6743 | } |
| 6744 | assign(res, unop(op, mkexpr(arg_m))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6745 | DIP("vrev32.%d %c%u, %c%u\n", 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6746 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 6747 | break; |
| 6748 | } |
| 6749 | case 2: { |
| 6750 | /* VREV16 */ |
| 6751 | IROp op; |
| 6752 | switch (size) { |
| 6753 | case 0: |
sewardj | 3368035 | 2014-06-26 10:49:33 +0000 | [diff] [blame] | 6754 | op = Q ? Iop_Reverse8sIn16_x8 : Iop_Reverse8sIn16_x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6755 | break; |
| 6756 | case 1: |
| 6757 | case 2: |
| 6758 | case 3: |
| 6759 | return False; |
| 6760 | default: |
| 6761 | vassert(0); |
| 6762 | } |
| 6763 | assign(res, unop(op, mkexpr(arg_m))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6764 | DIP("vrev16.%d %c%u, %c%u\n", 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6765 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 6766 | break; |
| 6767 | } |
| 6768 | case 3: |
| 6769 | return False; |
| 6770 | case 4: |
| 6771 | case 5: { |
| 6772 | /* VPADDL */ |
| 6773 | IROp op; |
| 6774 | U = (theInstr >> 7) & 1; |
| 6775 | if (Q) { |
| 6776 | switch (size) { |
| 6777 | case 0: op = U ? Iop_PwAddL8Ux16 : Iop_PwAddL8Sx16; break; |
| 6778 | case 1: op = U ? Iop_PwAddL16Ux8 : Iop_PwAddL16Sx8; break; |
| 6779 | case 2: op = U ? Iop_PwAddL32Ux4 : Iop_PwAddL32Sx4; break; |
| 6780 | case 3: return False; |
| 6781 | default: vassert(0); |
| 6782 | } |
| 6783 | } else { |
| 6784 | switch (size) { |
| 6785 | case 0: op = U ? Iop_PwAddL8Ux8 : Iop_PwAddL8Sx8; break; |
| 6786 | case 1: op = U ? Iop_PwAddL16Ux4 : Iop_PwAddL16Sx4; break; |
| 6787 | case 2: op = U ? Iop_PwAddL32Ux2 : Iop_PwAddL32Sx2; break; |
| 6788 | case 3: return False; |
| 6789 | default: vassert(0); |
| 6790 | } |
| 6791 | } |
| 6792 | assign(res, unop(op, mkexpr(arg_m))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6793 | DIP("vpaddl.%c%d %c%u, %c%u\n", U ? 'u' : 's', 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6794 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 6795 | break; |
| 6796 | } |
| 6797 | case 6: |
| 6798 | case 7: |
| 6799 | return False; |
| 6800 | case 8: { |
| 6801 | /* VCLS */ |
| 6802 | IROp op; |
| 6803 | switch (size) { |
sewardj | a8c7b0f | 2014-06-26 08:18:08 +0000 | [diff] [blame] | 6804 | case 0: op = Q ? Iop_Cls8x16 : Iop_Cls8x8; break; |
| 6805 | case 1: op = Q ? Iop_Cls16x8 : Iop_Cls16x4; break; |
| 6806 | case 2: op = Q ? Iop_Cls32x4 : Iop_Cls32x2; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6807 | case 3: return False; |
| 6808 | default: vassert(0); |
| 6809 | } |
| 6810 | assign(res, unop(op, mkexpr(arg_m))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6811 | DIP("vcls.s%d %c%u, %c%u\n", 8 << size, Q ? 'q' : 'd', dreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6812 | Q ? 'q' : 'd', mreg); |
| 6813 | break; |
| 6814 | } |
| 6815 | case 9: { |
| 6816 | /* VCLZ */ |
| 6817 | IROp op; |
| 6818 | switch (size) { |
sewardj | a8c7b0f | 2014-06-26 08:18:08 +0000 | [diff] [blame] | 6819 | case 0: op = Q ? Iop_Clz8x16 : Iop_Clz8x8; break; |
| 6820 | case 1: op = Q ? Iop_Clz16x8 : Iop_Clz16x4; break; |
| 6821 | case 2: op = Q ? Iop_Clz32x4 : Iop_Clz32x2; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6822 | case 3: return False; |
| 6823 | default: vassert(0); |
| 6824 | } |
| 6825 | assign(res, unop(op, mkexpr(arg_m))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6826 | DIP("vclz.i%d %c%u, %c%u\n", 8 << size, Q ? 'q' : 'd', dreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6827 | Q ? 'q' : 'd', mreg); |
| 6828 | break; |
| 6829 | } |
| 6830 | case 10: |
| 6831 | /* VCNT */ |
| 6832 | assign(res, unop(Q ? Iop_Cnt8x16 : Iop_Cnt8x8, mkexpr(arg_m))); |
| 6833 | DIP("vcnt.8 %c%u, %c%u\n", Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', |
| 6834 | mreg); |
| 6835 | break; |
| 6836 | case 11: |
| 6837 | /* VMVN */ |
| 6838 | if (Q) |
| 6839 | assign(res, unop(Iop_NotV128, mkexpr(arg_m))); |
| 6840 | else |
| 6841 | assign(res, unop(Iop_Not64, mkexpr(arg_m))); |
| 6842 | DIP("vmvn %c%u, %c%u\n", Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', |
| 6843 | mreg); |
| 6844 | break; |
| 6845 | case 12: |
| 6846 | case 13: { |
| 6847 | /* VPADAL */ |
| 6848 | IROp op, add_op; |
| 6849 | U = (theInstr >> 7) & 1; |
| 6850 | if (Q) { |
| 6851 | switch (size) { |
| 6852 | case 0: |
| 6853 | op = U ? Iop_PwAddL8Ux16 : Iop_PwAddL8Sx16; |
| 6854 | add_op = Iop_Add16x8; |
| 6855 | break; |
| 6856 | case 1: |
| 6857 | op = U ? Iop_PwAddL16Ux8 : Iop_PwAddL16Sx8; |
| 6858 | add_op = Iop_Add32x4; |
| 6859 | break; |
| 6860 | case 2: |
| 6861 | op = U ? Iop_PwAddL32Ux4 : Iop_PwAddL32Sx4; |
| 6862 | add_op = Iop_Add64x2; |
| 6863 | break; |
| 6864 | case 3: |
| 6865 | return False; |
| 6866 | default: |
| 6867 | vassert(0); |
| 6868 | } |
| 6869 | } else { |
| 6870 | switch (size) { |
| 6871 | case 0: |
| 6872 | op = U ? Iop_PwAddL8Ux8 : Iop_PwAddL8Sx8; |
| 6873 | add_op = Iop_Add16x4; |
| 6874 | break; |
| 6875 | case 1: |
| 6876 | op = U ? Iop_PwAddL16Ux4 : Iop_PwAddL16Sx4; |
| 6877 | add_op = Iop_Add32x2; |
| 6878 | break; |
| 6879 | case 2: |
| 6880 | op = U ? Iop_PwAddL32Ux2 : Iop_PwAddL32Sx2; |
| 6881 | add_op = Iop_Add64; |
| 6882 | break; |
| 6883 | case 3: |
| 6884 | return False; |
| 6885 | default: |
| 6886 | vassert(0); |
| 6887 | } |
| 6888 | } |
| 6889 | if (Q) { |
| 6890 | arg_d = newTemp(Ity_V128); |
| 6891 | assign(arg_d, getQReg(dreg)); |
| 6892 | } else { |
| 6893 | arg_d = newTemp(Ity_I64); |
| 6894 | assign(arg_d, getDRegI64(dreg)); |
| 6895 | } |
| 6896 | assign(res, binop(add_op, unop(op, mkexpr(arg_m)), |
| 6897 | mkexpr(arg_d))); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6898 | DIP("vpadal.%c%d %c%u, %c%u\n", U ? 'u' : 's', 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6899 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 6900 | break; |
| 6901 | } |
| 6902 | case 14: { |
| 6903 | /* VQABS */ |
| 6904 | IROp op_sub, op_qsub, op_cmp; |
| 6905 | IRTemp mask, tmp; |
| 6906 | IRExpr *zero1, *zero2; |
| 6907 | IRExpr *neg, *neg2; |
| 6908 | if (Q) { |
| 6909 | zero1 = binop(Iop_64HLtoV128, mkU64(0), mkU64(0)); |
| 6910 | zero2 = binop(Iop_64HLtoV128, mkU64(0), mkU64(0)); |
| 6911 | mask = newTemp(Ity_V128); |
| 6912 | tmp = newTemp(Ity_V128); |
| 6913 | } else { |
| 6914 | zero1 = mkU64(0); |
| 6915 | zero2 = mkU64(0); |
| 6916 | mask = newTemp(Ity_I64); |
| 6917 | tmp = newTemp(Ity_I64); |
| 6918 | } |
| 6919 | switch (size) { |
| 6920 | case 0: |
| 6921 | op_sub = Q ? Iop_Sub8x16 : Iop_Sub8x8; |
| 6922 | op_qsub = Q ? Iop_QSub8Sx16 : Iop_QSub8Sx8; |
| 6923 | op_cmp = Q ? Iop_CmpGT8Sx16 : Iop_CmpGT8Sx8; |
| 6924 | break; |
| 6925 | case 1: |
| 6926 | op_sub = Q ? Iop_Sub16x8 : Iop_Sub16x4; |
| 6927 | op_qsub = Q ? Iop_QSub16Sx8 : Iop_QSub16Sx4; |
| 6928 | op_cmp = Q ? Iop_CmpGT16Sx8 : Iop_CmpGT16Sx4; |
| 6929 | break; |
| 6930 | case 2: |
| 6931 | op_sub = Q ? Iop_Sub32x4 : Iop_Sub32x2; |
| 6932 | op_qsub = Q ? Iop_QSub32Sx4 : Iop_QSub32Sx2; |
| 6933 | op_cmp = Q ? Iop_CmpGT32Sx4 : Iop_CmpGT32Sx2; |
| 6934 | break; |
| 6935 | case 3: |
| 6936 | return False; |
| 6937 | default: |
| 6938 | vassert(0); |
| 6939 | } |
| 6940 | assign(mask, binop(op_cmp, mkexpr(arg_m), zero1)); |
| 6941 | neg = binop(op_qsub, zero2, mkexpr(arg_m)); |
| 6942 | neg2 = binop(op_sub, zero2, mkexpr(arg_m)); |
| 6943 | assign(res, binop(Q ? Iop_OrV128 : Iop_Or64, |
| 6944 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 6945 | mkexpr(mask), |
| 6946 | mkexpr(arg_m)), |
| 6947 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 6948 | unop(Q ? Iop_NotV128 : Iop_Not64, |
| 6949 | mkexpr(mask)), |
| 6950 | neg))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6951 | assign(tmp, binop(Q ? Iop_OrV128 : Iop_Or64, |
| 6952 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 6953 | mkexpr(mask), |
| 6954 | mkexpr(arg_m)), |
| 6955 | binop(Q ? Iop_AndV128 : Iop_And64, |
| 6956 | unop(Q ? Iop_NotV128 : Iop_Not64, |
| 6957 | mkexpr(mask)), |
| 6958 | neg2))); |
| 6959 | setFlag_QC(mkexpr(res), mkexpr(tmp), Q, condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6960 | DIP("vqabs.s%d %c%u, %c%u\n", 8 << size, Q ? 'q' : 'd', dreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6961 | Q ? 'q' : 'd', mreg); |
| 6962 | break; |
| 6963 | } |
| 6964 | case 15: { |
| 6965 | /* VQNEG */ |
| 6966 | IROp op, op2; |
| 6967 | IRExpr *zero; |
| 6968 | if (Q) { |
| 6969 | zero = binop(Iop_64HLtoV128, mkU64(0), mkU64(0)); |
| 6970 | } else { |
| 6971 | zero = mkU64(0); |
| 6972 | } |
| 6973 | switch (size) { |
| 6974 | case 0: |
| 6975 | op = Q ? Iop_QSub8Sx16 : Iop_QSub8Sx8; |
| 6976 | op2 = Q ? Iop_Sub8x16 : Iop_Sub8x8; |
| 6977 | break; |
| 6978 | case 1: |
| 6979 | op = Q ? Iop_QSub16Sx8 : Iop_QSub16Sx4; |
| 6980 | op2 = Q ? Iop_Sub16x8 : Iop_Sub16x4; |
| 6981 | break; |
| 6982 | case 2: |
| 6983 | op = Q ? Iop_QSub32Sx4 : Iop_QSub32Sx2; |
| 6984 | op2 = Q ? Iop_Sub32x4 : Iop_Sub32x2; |
| 6985 | break; |
| 6986 | case 3: |
| 6987 | return False; |
| 6988 | default: |
| 6989 | vassert(0); |
| 6990 | } |
| 6991 | assign(res, binop(op, zero, mkexpr(arg_m))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6992 | setFlag_QC(mkexpr(res), binop(op2, zero, mkexpr(arg_m)), |
| 6993 | Q, condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 6994 | DIP("vqneg.s%d %c%u, %c%u\n", 8 << size, Q ? 'q' : 'd', dreg, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 6995 | Q ? 'q' : 'd', mreg); |
| 6996 | break; |
| 6997 | } |
| 6998 | default: |
| 6999 | vassert(0); |
| 7000 | } |
| 7001 | if (Q) { |
| 7002 | putQReg(dreg, mkexpr(res), condT); |
| 7003 | } else { |
| 7004 | putDRegI64(dreg, mkexpr(res), condT); |
| 7005 | } |
| 7006 | return True; |
| 7007 | case 1: |
| 7008 | if (Q) { |
| 7009 | arg_m = newTemp(Ity_V128); |
| 7010 | res = newTemp(Ity_V128); |
| 7011 | assign(arg_m, getQReg(mreg)); |
| 7012 | } else { |
| 7013 | arg_m = newTemp(Ity_I64); |
| 7014 | res = newTemp(Ity_I64); |
| 7015 | assign(arg_m, getDRegI64(mreg)); |
| 7016 | } |
| 7017 | switch ((B >> 1) & 0x7) { |
| 7018 | case 0: { |
| 7019 | /* VCGT #0 */ |
| 7020 | IRExpr *zero; |
| 7021 | IROp op; |
| 7022 | if (Q) { |
| 7023 | zero = binop(Iop_64HLtoV128, mkU64(0), mkU64(0)); |
| 7024 | } else { |
| 7025 | zero = mkU64(0); |
| 7026 | } |
| 7027 | if (F) { |
| 7028 | switch (size) { |
| 7029 | case 0: case 1: case 3: return False; |
| 7030 | case 2: op = Q ? Iop_CmpGT32Fx4 : Iop_CmpGT32Fx2; break; |
| 7031 | default: vassert(0); |
| 7032 | } |
| 7033 | } else { |
| 7034 | switch (size) { |
| 7035 | case 0: op = Q ? Iop_CmpGT8Sx16 : Iop_CmpGT8Sx8; break; |
| 7036 | case 1: op = Q ? Iop_CmpGT16Sx8 : Iop_CmpGT16Sx4; break; |
| 7037 | case 2: op = Q ? Iop_CmpGT32Sx4 : Iop_CmpGT32Sx2; break; |
| 7038 | case 3: return False; |
| 7039 | default: vassert(0); |
| 7040 | } |
| 7041 | } |
| 7042 | assign(res, binop(op, mkexpr(arg_m), zero)); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7043 | DIP("vcgt.%c%d %c%u, %c%u, #0\n", F ? 'f' : 's', 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7044 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 7045 | break; |
| 7046 | } |
| 7047 | case 1: { |
| 7048 | /* VCGE #0 */ |
| 7049 | IROp op; |
| 7050 | IRExpr *zero; |
| 7051 | if (Q) { |
| 7052 | zero = binop(Iop_64HLtoV128, mkU64(0), mkU64(0)); |
| 7053 | } else { |
| 7054 | zero = mkU64(0); |
| 7055 | } |
| 7056 | if (F) { |
| 7057 | switch (size) { |
| 7058 | case 0: case 1: case 3: return False; |
| 7059 | case 2: op = Q ? Iop_CmpGE32Fx4 : Iop_CmpGE32Fx2; break; |
| 7060 | default: vassert(0); |
| 7061 | } |
| 7062 | assign(res, binop(op, mkexpr(arg_m), zero)); |
| 7063 | } else { |
| 7064 | switch (size) { |
| 7065 | case 0: op = Q ? Iop_CmpGT8Sx16 : Iop_CmpGT8Sx8; break; |
| 7066 | case 1: op = Q ? Iop_CmpGT16Sx8 : Iop_CmpGT16Sx4; break; |
| 7067 | case 2: op = Q ? Iop_CmpGT32Sx4 : Iop_CmpGT32Sx2; break; |
| 7068 | case 3: return False; |
| 7069 | default: vassert(0); |
| 7070 | } |
| 7071 | assign(res, unop(Q ? Iop_NotV128 : Iop_Not64, |
| 7072 | binop(op, zero, mkexpr(arg_m)))); |
| 7073 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7074 | DIP("vcge.%c%d %c%u, %c%u, #0\n", F ? 'f' : 's', 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7075 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 7076 | break; |
| 7077 | } |
| 7078 | case 2: { |
| 7079 | /* VCEQ #0 */ |
| 7080 | IROp op; |
| 7081 | IRExpr *zero; |
| 7082 | if (F) { |
| 7083 | if (Q) { |
| 7084 | zero = binop(Iop_64HLtoV128, mkU64(0), mkU64(0)); |
| 7085 | } else { |
| 7086 | zero = mkU64(0); |
| 7087 | } |
| 7088 | switch (size) { |
| 7089 | case 0: case 1: case 3: return False; |
| 7090 | case 2: op = Q ? Iop_CmpEQ32Fx4 : Iop_CmpEQ32Fx2; break; |
| 7091 | default: vassert(0); |
| 7092 | } |
| 7093 | assign(res, binop(op, zero, mkexpr(arg_m))); |
| 7094 | } else { |
| 7095 | switch (size) { |
| 7096 | case 0: op = Q ? Iop_CmpNEZ8x16 : Iop_CmpNEZ8x8; break; |
| 7097 | case 1: op = Q ? Iop_CmpNEZ16x8 : Iop_CmpNEZ16x4; break; |
| 7098 | case 2: op = Q ? Iop_CmpNEZ32x4 : Iop_CmpNEZ32x2; break; |
| 7099 | case 3: return False; |
| 7100 | default: vassert(0); |
| 7101 | } |
| 7102 | assign(res, unop(Q ? Iop_NotV128 : Iop_Not64, |
| 7103 | unop(op, mkexpr(arg_m)))); |
| 7104 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7105 | DIP("vceq.%c%d %c%u, %c%u, #0\n", F ? 'f' : 'i', 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7106 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 7107 | break; |
| 7108 | } |
| 7109 | case 3: { |
| 7110 | /* VCLE #0 */ |
| 7111 | IRExpr *zero; |
| 7112 | IROp op; |
| 7113 | if (Q) { |
| 7114 | zero = binop(Iop_64HLtoV128, mkU64(0), mkU64(0)); |
| 7115 | } else { |
| 7116 | zero = mkU64(0); |
| 7117 | } |
| 7118 | if (F) { |
| 7119 | switch (size) { |
| 7120 | case 0: case 1: case 3: return False; |
| 7121 | case 2: op = Q ? Iop_CmpGE32Fx4 : Iop_CmpGE32Fx2; break; |
| 7122 | default: vassert(0); |
| 7123 | } |
| 7124 | assign(res, binop(op, zero, mkexpr(arg_m))); |
| 7125 | } else { |
| 7126 | switch (size) { |
| 7127 | case 0: op = Q ? Iop_CmpGT8Sx16 : Iop_CmpGT8Sx8; break; |
| 7128 | case 1: op = Q ? Iop_CmpGT16Sx8 : Iop_CmpGT16Sx4; break; |
| 7129 | case 2: op = Q ? Iop_CmpGT32Sx4 : Iop_CmpGT32Sx2; break; |
| 7130 | case 3: return False; |
| 7131 | default: vassert(0); |
| 7132 | } |
| 7133 | assign(res, unop(Q ? Iop_NotV128 : Iop_Not64, |
| 7134 | binop(op, mkexpr(arg_m), zero))); |
| 7135 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7136 | DIP("vcle.%c%d %c%u, %c%u, #0\n", F ? 'f' : 's', 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7137 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 7138 | break; |
| 7139 | } |
| 7140 | case 4: { |
| 7141 | /* VCLT #0 */ |
| 7142 | IROp op; |
| 7143 | IRExpr *zero; |
| 7144 | if (Q) { |
| 7145 | zero = binop(Iop_64HLtoV128, mkU64(0), mkU64(0)); |
| 7146 | } else { |
| 7147 | zero = mkU64(0); |
| 7148 | } |
| 7149 | if (F) { |
| 7150 | switch (size) { |
| 7151 | case 0: case 1: case 3: return False; |
| 7152 | case 2: op = Q ? Iop_CmpGT32Fx4 : Iop_CmpGT32Fx2; break; |
| 7153 | default: vassert(0); |
| 7154 | } |
| 7155 | assign(res, binop(op, zero, mkexpr(arg_m))); |
| 7156 | } else { |
| 7157 | switch (size) { |
| 7158 | case 0: op = Q ? Iop_CmpGT8Sx16 : Iop_CmpGT8Sx8; break; |
| 7159 | case 1: op = Q ? Iop_CmpGT16Sx8 : Iop_CmpGT16Sx4; break; |
| 7160 | case 2: op = Q ? Iop_CmpGT32Sx4 : Iop_CmpGT32Sx2; break; |
| 7161 | case 3: return False; |
| 7162 | default: vassert(0); |
| 7163 | } |
| 7164 | assign(res, binop(op, zero, mkexpr(arg_m))); |
| 7165 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7166 | DIP("vclt.%c%d %c%u, %c%u, #0\n", F ? 'f' : 's', 8 << size, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7167 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 7168 | break; |
| 7169 | } |
| 7170 | case 5: |
| 7171 | return False; |
| 7172 | case 6: { |
| 7173 | /* VABS */ |
| 7174 | if (!F) { |
| 7175 | IROp op; |
| 7176 | switch(size) { |
| 7177 | case 0: op = Q ? Iop_Abs8x16 : Iop_Abs8x8; break; |
| 7178 | case 1: op = Q ? Iop_Abs16x8 : Iop_Abs16x4; break; |
| 7179 | case 2: op = Q ? Iop_Abs32x4 : Iop_Abs32x2; break; |
| 7180 | case 3: return False; |
| 7181 | default: vassert(0); |
| 7182 | } |
| 7183 | assign(res, unop(op, mkexpr(arg_m))); |
| 7184 | } else { |
| 7185 | assign(res, unop(Q ? Iop_Abs32Fx4 : Iop_Abs32Fx2, |
| 7186 | mkexpr(arg_m))); |
| 7187 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7188 | DIP("vabs.%c%d %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7189 | F ? 'f' : 's', 8 << size, Q ? 'q' : 'd', dreg, |
| 7190 | Q ? 'q' : 'd', mreg); |
| 7191 | break; |
| 7192 | } |
| 7193 | case 7: { |
| 7194 | /* VNEG */ |
| 7195 | IROp op; |
| 7196 | IRExpr *zero; |
| 7197 | if (F) { |
| 7198 | switch (size) { |
| 7199 | case 0: case 1: case 3: return False; |
| 7200 | case 2: op = Q ? Iop_Neg32Fx4 : Iop_Neg32Fx2; break; |
| 7201 | default: vassert(0); |
| 7202 | } |
| 7203 | assign(res, unop(op, mkexpr(arg_m))); |
| 7204 | } else { |
| 7205 | if (Q) { |
| 7206 | zero = binop(Iop_64HLtoV128, mkU64(0), mkU64(0)); |
| 7207 | } else { |
| 7208 | zero = mkU64(0); |
| 7209 | } |
| 7210 | switch (size) { |
| 7211 | case 0: op = Q ? Iop_Sub8x16 : Iop_Sub8x8; break; |
| 7212 | case 1: op = Q ? Iop_Sub16x8 : Iop_Sub16x4; break; |
| 7213 | case 2: op = Q ? Iop_Sub32x4 : Iop_Sub32x2; break; |
| 7214 | case 3: return False; |
| 7215 | default: vassert(0); |
| 7216 | } |
| 7217 | assign(res, binop(op, zero, mkexpr(arg_m))); |
| 7218 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7219 | DIP("vneg.%c%d %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7220 | F ? 'f' : 's', 8 << size, Q ? 'q' : 'd', dreg, |
| 7221 | Q ? 'q' : 'd', mreg); |
| 7222 | break; |
| 7223 | } |
| 7224 | default: |
| 7225 | vassert(0); |
| 7226 | } |
| 7227 | if (Q) { |
| 7228 | putQReg(dreg, mkexpr(res), condT); |
| 7229 | } else { |
| 7230 | putDRegI64(dreg, mkexpr(res), condT); |
| 7231 | } |
| 7232 | return True; |
| 7233 | case 2: |
| 7234 | if ((B >> 1) == 0) { |
| 7235 | /* VSWP */ |
| 7236 | if (Q) { |
| 7237 | arg_m = newTemp(Ity_V128); |
| 7238 | assign(arg_m, getQReg(mreg)); |
| 7239 | putQReg(mreg, getQReg(dreg), condT); |
| 7240 | putQReg(dreg, mkexpr(arg_m), condT); |
| 7241 | } else { |
| 7242 | arg_m = newTemp(Ity_I64); |
| 7243 | assign(arg_m, getDRegI64(mreg)); |
| 7244 | putDRegI64(mreg, getDRegI64(dreg), condT); |
| 7245 | putDRegI64(dreg, mkexpr(arg_m), condT); |
| 7246 | } |
| 7247 | DIP("vswp %c%u, %c%u\n", |
| 7248 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 7249 | return True; |
| 7250 | } else if ((B >> 1) == 1) { |
| 7251 | /* VTRN */ |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7252 | IROp op_odd = Iop_INVALID, op_even = Iop_INVALID; |
| 7253 | IRTemp old_m, old_d, new_d, new_m; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7254 | if (Q) { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7255 | old_m = newTemp(Ity_V128); |
| 7256 | old_d = newTemp(Ity_V128); |
| 7257 | new_m = newTemp(Ity_V128); |
| 7258 | new_d = newTemp(Ity_V128); |
| 7259 | assign(old_m, getQReg(mreg)); |
| 7260 | assign(old_d, getQReg(dreg)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7261 | } else { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7262 | old_m = newTemp(Ity_I64); |
| 7263 | old_d = newTemp(Ity_I64); |
| 7264 | new_m = newTemp(Ity_I64); |
| 7265 | new_d = newTemp(Ity_I64); |
| 7266 | assign(old_m, getDRegI64(mreg)); |
| 7267 | assign(old_d, getDRegI64(dreg)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7268 | } |
| 7269 | if (Q) { |
| 7270 | switch (size) { |
| 7271 | case 0: |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7272 | op_odd = Iop_InterleaveOddLanes8x16; |
| 7273 | op_even = Iop_InterleaveEvenLanes8x16; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7274 | break; |
| 7275 | case 1: |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7276 | op_odd = Iop_InterleaveOddLanes16x8; |
| 7277 | op_even = Iop_InterleaveEvenLanes16x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7278 | break; |
| 7279 | case 2: |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7280 | op_odd = Iop_InterleaveOddLanes32x4; |
| 7281 | op_even = Iop_InterleaveEvenLanes32x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7282 | break; |
| 7283 | case 3: |
| 7284 | return False; |
| 7285 | default: |
| 7286 | vassert(0); |
| 7287 | } |
| 7288 | } else { |
| 7289 | switch (size) { |
| 7290 | case 0: |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7291 | op_odd = Iop_InterleaveOddLanes8x8; |
| 7292 | op_even = Iop_InterleaveEvenLanes8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7293 | break; |
| 7294 | case 1: |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7295 | op_odd = Iop_InterleaveOddLanes16x4; |
| 7296 | op_even = Iop_InterleaveEvenLanes16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7297 | break; |
| 7298 | case 2: |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7299 | op_odd = Iop_InterleaveHI32x2; |
| 7300 | op_even = Iop_InterleaveLO32x2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7301 | break; |
| 7302 | case 3: |
| 7303 | return False; |
| 7304 | default: |
| 7305 | vassert(0); |
| 7306 | } |
| 7307 | } |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7308 | assign(new_d, binop(op_even, mkexpr(old_m), mkexpr(old_d))); |
| 7309 | assign(new_m, binop(op_odd, mkexpr(old_m), mkexpr(old_d))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7310 | if (Q) { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7311 | putQReg(dreg, mkexpr(new_d), condT); |
| 7312 | putQReg(mreg, mkexpr(new_m), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7313 | } else { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7314 | putDRegI64(dreg, mkexpr(new_d), condT); |
| 7315 | putDRegI64(mreg, mkexpr(new_m), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7316 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7317 | DIP("vtrn.%d %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7318 | 8 << size, Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 7319 | return True; |
| 7320 | } else if ((B >> 1) == 2) { |
| 7321 | /* VUZP */ |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7322 | IROp op_even, op_odd; |
| 7323 | IRTemp old_m, old_d, new_m, new_d; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7324 | if (!Q && size == 2) |
| 7325 | return False; |
| 7326 | if (Q) { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7327 | old_m = newTemp(Ity_V128); |
| 7328 | old_d = newTemp(Ity_V128); |
| 7329 | new_m = newTemp(Ity_V128); |
| 7330 | new_d = newTemp(Ity_V128); |
| 7331 | assign(old_m, getQReg(mreg)); |
| 7332 | assign(old_d, getQReg(dreg)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7333 | } else { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7334 | old_m = newTemp(Ity_I64); |
| 7335 | old_d = newTemp(Ity_I64); |
| 7336 | new_m = newTemp(Ity_I64); |
| 7337 | new_d = newTemp(Ity_I64); |
| 7338 | assign(old_m, getDRegI64(mreg)); |
| 7339 | assign(old_d, getDRegI64(dreg)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7340 | } |
| 7341 | switch (size) { |
| 7342 | case 0: |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7343 | op_odd = Q ? Iop_CatOddLanes8x16 : Iop_CatOddLanes8x8; |
| 7344 | op_even = Q ? Iop_CatEvenLanes8x16 : Iop_CatEvenLanes8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7345 | break; |
| 7346 | case 1: |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7347 | op_odd = Q ? Iop_CatOddLanes16x8 : Iop_CatOddLanes16x4; |
| 7348 | op_even = Q ? Iop_CatEvenLanes16x8 : Iop_CatEvenLanes16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7349 | break; |
| 7350 | case 2: |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7351 | op_odd = Iop_CatOddLanes32x4; |
| 7352 | op_even = Iop_CatEvenLanes32x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7353 | break; |
| 7354 | case 3: |
| 7355 | return False; |
| 7356 | default: |
| 7357 | vassert(0); |
| 7358 | } |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7359 | assign(new_d, binop(op_even, mkexpr(old_m), mkexpr(old_d))); |
| 7360 | assign(new_m, binop(op_odd, mkexpr(old_m), mkexpr(old_d))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7361 | if (Q) { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7362 | putQReg(dreg, mkexpr(new_d), condT); |
| 7363 | putQReg(mreg, mkexpr(new_m), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7364 | } else { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7365 | putDRegI64(dreg, mkexpr(new_d), condT); |
| 7366 | putDRegI64(mreg, mkexpr(new_m), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7367 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7368 | DIP("vuzp.%d %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7369 | 8 << size, Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 7370 | return True; |
| 7371 | } else if ((B >> 1) == 3) { |
| 7372 | /* VZIP */ |
| 7373 | IROp op_lo, op_hi; |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7374 | IRTemp old_m, old_d, new_m, new_d; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7375 | if (!Q && size == 2) |
| 7376 | return False; |
| 7377 | if (Q) { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7378 | old_m = newTemp(Ity_V128); |
| 7379 | old_d = newTemp(Ity_V128); |
| 7380 | new_m = newTemp(Ity_V128); |
| 7381 | new_d = newTemp(Ity_V128); |
| 7382 | assign(old_m, getQReg(mreg)); |
| 7383 | assign(old_d, getQReg(dreg)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7384 | } else { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7385 | old_m = newTemp(Ity_I64); |
| 7386 | old_d = newTemp(Ity_I64); |
| 7387 | new_m = newTemp(Ity_I64); |
| 7388 | new_d = newTemp(Ity_I64); |
| 7389 | assign(old_m, getDRegI64(mreg)); |
| 7390 | assign(old_d, getDRegI64(dreg)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7391 | } |
| 7392 | switch (size) { |
| 7393 | case 0: |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7394 | op_hi = Q ? Iop_InterleaveHI8x16 : Iop_InterleaveHI8x8; |
| 7395 | op_lo = Q ? Iop_InterleaveLO8x16 : Iop_InterleaveLO8x8; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7396 | break; |
| 7397 | case 1: |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7398 | op_hi = Q ? Iop_InterleaveHI16x8 : Iop_InterleaveHI16x4; |
| 7399 | op_lo = Q ? Iop_InterleaveLO16x8 : Iop_InterleaveLO16x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7400 | break; |
| 7401 | case 2: |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7402 | op_hi = Iop_InterleaveHI32x4; |
| 7403 | op_lo = Iop_InterleaveLO32x4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7404 | break; |
| 7405 | case 3: |
| 7406 | return False; |
| 7407 | default: |
| 7408 | vassert(0); |
| 7409 | } |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7410 | assign(new_d, binop(op_lo, mkexpr(old_m), mkexpr(old_d))); |
| 7411 | assign(new_m, binop(op_hi, mkexpr(old_m), mkexpr(old_d))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7412 | if (Q) { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7413 | putQReg(dreg, mkexpr(new_d), condT); |
| 7414 | putQReg(mreg, mkexpr(new_m), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7415 | } else { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7416 | putDRegI64(dreg, mkexpr(new_d), condT); |
| 7417 | putDRegI64(mreg, mkexpr(new_m), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7418 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7419 | DIP("vzip.%d %c%u, %c%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7420 | 8 << size, Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 7421 | return True; |
| 7422 | } else if (B == 8) { |
| 7423 | /* VMOVN */ |
| 7424 | IROp op; |
| 7425 | mreg >>= 1; |
| 7426 | switch (size) { |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 7427 | case 0: op = Iop_NarrowUn16to8x8; break; |
| 7428 | case 1: op = Iop_NarrowUn32to16x4; break; |
| 7429 | case 2: op = Iop_NarrowUn64to32x2; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7430 | case 3: return False; |
| 7431 | default: vassert(0); |
| 7432 | } |
| 7433 | putDRegI64(dreg, unop(op, getQReg(mreg)), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7434 | DIP("vmovn.i%d d%u, q%u\n", 16 << size, dreg, mreg); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7435 | return True; |
| 7436 | } else if (B == 9 || (B >> 1) == 5) { |
| 7437 | /* VQMOVN, VQMOVUN */ |
| 7438 | IROp op, op2; |
| 7439 | IRTemp tmp; |
| 7440 | dreg = ((theInstr >> 18) & 0x10) | ((theInstr >> 12) & 0xF); |
| 7441 | mreg = ((theInstr >> 1) & 0x10) | (theInstr & 0xF); |
| 7442 | if (mreg & 1) |
| 7443 | return False; |
| 7444 | mreg >>= 1; |
| 7445 | switch (size) { |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 7446 | case 0: op2 = Iop_NarrowUn16to8x8; break; |
| 7447 | case 1: op2 = Iop_NarrowUn32to16x4; break; |
| 7448 | case 2: op2 = Iop_NarrowUn64to32x2; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7449 | case 3: return False; |
| 7450 | default: vassert(0); |
| 7451 | } |
| 7452 | switch (B & 3) { |
| 7453 | case 0: |
| 7454 | vassert(0); |
| 7455 | case 1: |
| 7456 | switch (size) { |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 7457 | case 0: op = Iop_QNarrowUn16Sto8Ux8; break; |
| 7458 | case 1: op = Iop_QNarrowUn32Sto16Ux4; break; |
| 7459 | case 2: op = Iop_QNarrowUn64Sto32Ux2; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7460 | case 3: return False; |
| 7461 | default: vassert(0); |
| 7462 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7463 | DIP("vqmovun.s%d d%u, q%u\n", 16 << size, dreg, mreg); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7464 | break; |
| 7465 | case 2: |
| 7466 | switch (size) { |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 7467 | case 0: op = Iop_QNarrowUn16Sto8Sx8; break; |
| 7468 | case 1: op = Iop_QNarrowUn32Sto16Sx4; break; |
| 7469 | case 2: op = Iop_QNarrowUn64Sto32Sx2; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7470 | case 3: return False; |
| 7471 | default: vassert(0); |
| 7472 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7473 | DIP("vqmovn.s%d d%u, q%u\n", 16 << size, dreg, mreg); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7474 | break; |
| 7475 | case 3: |
| 7476 | switch (size) { |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 7477 | case 0: op = Iop_QNarrowUn16Uto8Ux8; break; |
| 7478 | case 1: op = Iop_QNarrowUn32Uto16Ux4; break; |
| 7479 | case 2: op = Iop_QNarrowUn64Uto32Ux2; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7480 | case 3: return False; |
| 7481 | default: vassert(0); |
| 7482 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7483 | DIP("vqmovn.u%d d%u, q%u\n", 16 << size, dreg, mreg); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7484 | break; |
| 7485 | default: |
| 7486 | vassert(0); |
| 7487 | } |
| 7488 | res = newTemp(Ity_I64); |
| 7489 | tmp = newTemp(Ity_I64); |
| 7490 | assign(res, unop(op, getQReg(mreg))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7491 | assign(tmp, unop(op2, getQReg(mreg))); |
| 7492 | setFlag_QC(mkexpr(res), mkexpr(tmp), False, condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7493 | putDRegI64(dreg, mkexpr(res), condT); |
| 7494 | return True; |
| 7495 | } else if (B == 12) { |
| 7496 | /* VSHLL (maximum shift) */ |
| 7497 | IROp op, cvt; |
| 7498 | UInt shift_imm; |
| 7499 | if (Q) |
| 7500 | return False; |
| 7501 | if (dreg & 1) |
| 7502 | return False; |
| 7503 | dreg >>= 1; |
| 7504 | shift_imm = 8 << size; |
| 7505 | res = newTemp(Ity_V128); |
| 7506 | switch (size) { |
sewardj | 5f438dd | 2011-06-16 11:36:23 +0000 | [diff] [blame] | 7507 | case 0: op = Iop_ShlN16x8; cvt = Iop_Widen8Uto16x8; break; |
| 7508 | case 1: op = Iop_ShlN32x4; cvt = Iop_Widen16Uto32x4; break; |
| 7509 | case 2: op = Iop_ShlN64x2; cvt = Iop_Widen32Uto64x2; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7510 | case 3: return False; |
| 7511 | default: vassert(0); |
| 7512 | } |
| 7513 | assign(res, binop(op, unop(cvt, getDRegI64(mreg)), |
| 7514 | mkU8(shift_imm))); |
| 7515 | putQReg(dreg, mkexpr(res), condT); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 7516 | DIP("vshll.i%d q%u, d%u, #%d\n", 8 << size, dreg, mreg, 8 << size); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7517 | return True; |
| 7518 | } else if ((B >> 3) == 3 && (B & 3) == 0) { |
| 7519 | /* VCVT (half<->single) */ |
| 7520 | /* Half-precision extensions are needed to run this */ |
| 7521 | vassert(0); // ATC |
| 7522 | if (((theInstr >> 18) & 3) != 1) |
| 7523 | return False; |
| 7524 | if ((theInstr >> 8) & 1) { |
| 7525 | if (dreg & 1) |
| 7526 | return False; |
| 7527 | dreg >>= 1; |
| 7528 | putQReg(dreg, unop(Iop_F16toF32x4, getDRegI64(mreg)), |
| 7529 | condT); |
| 7530 | DIP("vcvt.f32.f16 q%u, d%u\n", dreg, mreg); |
| 7531 | } else { |
| 7532 | if (mreg & 1) |
| 7533 | return False; |
| 7534 | mreg >>= 1; |
| 7535 | putDRegI64(dreg, unop(Iop_F32toF16x4, getQReg(mreg)), |
| 7536 | condT); |
| 7537 | DIP("vcvt.f16.f32 d%u, q%u\n", dreg, mreg); |
| 7538 | } |
| 7539 | return True; |
| 7540 | } else { |
| 7541 | return False; |
| 7542 | } |
| 7543 | vassert(0); |
| 7544 | return True; |
| 7545 | case 3: |
| 7546 | if (((B >> 1) & BITS4(1,1,0,1)) == BITS4(1,0,0,0)) { |
| 7547 | /* VRECPE */ |
| 7548 | IROp op; |
| 7549 | F = (theInstr >> 8) & 1; |
| 7550 | if (size != 2) |
| 7551 | return False; |
| 7552 | if (Q) { |
sewardj | 1ddee21 | 2014-08-24 14:00:19 +0000 | [diff] [blame] | 7553 | op = F ? Iop_RecipEst32Fx4 : Iop_RecipEst32Ux4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7554 | putQReg(dreg, unop(op, getQReg(mreg)), condT); |
| 7555 | DIP("vrecpe.%c32 q%u, q%u\n", F ? 'f' : 'u', dreg, mreg); |
| 7556 | } else { |
sewardj | 1ddee21 | 2014-08-24 14:00:19 +0000 | [diff] [blame] | 7557 | op = F ? Iop_RecipEst32Fx2 : Iop_RecipEst32Ux2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7558 | putDRegI64(dreg, unop(op, getDRegI64(mreg)), condT); |
| 7559 | DIP("vrecpe.%c32 d%u, d%u\n", F ? 'f' : 'u', dreg, mreg); |
| 7560 | } |
| 7561 | return True; |
| 7562 | } else if (((B >> 1) & BITS4(1,1,0,1)) == BITS4(1,0,0,1)) { |
| 7563 | /* VRSQRTE */ |
| 7564 | IROp op; |
| 7565 | F = (B >> 2) & 1; |
| 7566 | if (size != 2) |
| 7567 | return False; |
| 7568 | if (F) { |
| 7569 | /* fp */ |
sewardj | 1ddee21 | 2014-08-24 14:00:19 +0000 | [diff] [blame] | 7570 | op = Q ? Iop_RSqrtEst32Fx4 : Iop_RSqrtEst32Fx2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7571 | } else { |
| 7572 | /* unsigned int */ |
sewardj | 1ddee21 | 2014-08-24 14:00:19 +0000 | [diff] [blame] | 7573 | op = Q ? Iop_RSqrtEst32Ux4 : Iop_RSqrtEst32Ux2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7574 | } |
| 7575 | if (Q) { |
| 7576 | putQReg(dreg, unop(op, getQReg(mreg)), condT); |
| 7577 | DIP("vrsqrte.%c32 q%u, q%u\n", F ? 'f' : 'u', dreg, mreg); |
| 7578 | } else { |
| 7579 | putDRegI64(dreg, unop(op, getDRegI64(mreg)), condT); |
| 7580 | DIP("vrsqrte.%c32 d%u, d%u\n", F ? 'f' : 'u', dreg, mreg); |
| 7581 | } |
| 7582 | return True; |
| 7583 | } else if ((B >> 3) == 3) { |
| 7584 | /* VCVT (fp<->integer) */ |
| 7585 | IROp op; |
| 7586 | if (size != 2) |
| 7587 | return False; |
| 7588 | switch ((B >> 1) & 3) { |
| 7589 | case 0: |
| 7590 | op = Q ? Iop_I32StoFx4 : Iop_I32StoFx2; |
| 7591 | DIP("vcvt.f32.s32 %c%u, %c%u\n", |
| 7592 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 7593 | break; |
| 7594 | case 1: |
| 7595 | op = Q ? Iop_I32UtoFx4 : Iop_I32UtoFx2; |
| 7596 | DIP("vcvt.f32.u32 %c%u, %c%u\n", |
| 7597 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 7598 | break; |
| 7599 | case 2: |
| 7600 | op = Q ? Iop_FtoI32Sx4_RZ : Iop_FtoI32Sx2_RZ; |
| 7601 | DIP("vcvt.s32.f32 %c%u, %c%u\n", |
| 7602 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 7603 | break; |
| 7604 | case 3: |
| 7605 | op = Q ? Iop_FtoI32Ux4_RZ : Iop_FtoI32Ux2_RZ; |
| 7606 | DIP("vcvt.u32.f32 %c%u, %c%u\n", |
| 7607 | Q ? 'q' : 'd', dreg, Q ? 'q' : 'd', mreg); |
| 7608 | break; |
| 7609 | default: |
| 7610 | vassert(0); |
| 7611 | } |
| 7612 | if (Q) { |
| 7613 | putQReg(dreg, unop(op, getQReg(mreg)), condT); |
| 7614 | } else { |
| 7615 | putDRegI64(dreg, unop(op, getDRegI64(mreg)), condT); |
| 7616 | } |
| 7617 | return True; |
| 7618 | } else { |
| 7619 | return False; |
| 7620 | } |
| 7621 | vassert(0); |
| 7622 | return True; |
| 7623 | default: |
| 7624 | vassert(0); |
| 7625 | } |
| 7626 | return False; |
| 7627 | } |
| 7628 | |
| 7629 | /* A7.4.6 One register and a modified immediate value */ |
| 7630 | static |
| 7631 | void ppNeonImm(UInt imm, UInt cmode, UInt op) |
| 7632 | { |
| 7633 | int i; |
| 7634 | switch (cmode) { |
| 7635 | case 0: case 1: case 8: case 9: |
| 7636 | vex_printf("0x%x", imm); |
| 7637 | break; |
| 7638 | case 2: case 3: case 10: case 11: |
| 7639 | vex_printf("0x%x00", imm); |
| 7640 | break; |
| 7641 | case 4: case 5: |
| 7642 | vex_printf("0x%x0000", imm); |
| 7643 | break; |
| 7644 | case 6: case 7: |
| 7645 | vex_printf("0x%x000000", imm); |
| 7646 | break; |
| 7647 | case 12: |
| 7648 | vex_printf("0x%xff", imm); |
| 7649 | break; |
| 7650 | case 13: |
| 7651 | vex_printf("0x%xffff", imm); |
| 7652 | break; |
| 7653 | case 14: |
| 7654 | if (op) { |
| 7655 | vex_printf("0x"); |
| 7656 | for (i = 7; i >= 0; i--) |
| 7657 | vex_printf("%s", (imm & (1 << i)) ? "ff" : "00"); |
| 7658 | } else { |
| 7659 | vex_printf("0x%x", imm); |
| 7660 | } |
| 7661 | break; |
| 7662 | case 15: |
| 7663 | vex_printf("0x%x", imm); |
| 7664 | break; |
| 7665 | } |
| 7666 | } |
| 7667 | |
| 7668 | static |
| 7669 | const char *ppNeonImmType(UInt cmode, UInt op) |
| 7670 | { |
| 7671 | switch (cmode) { |
| 7672 | case 0 ... 7: |
| 7673 | case 12: case 13: |
| 7674 | return "i32"; |
| 7675 | case 8 ... 11: |
| 7676 | return "i16"; |
| 7677 | case 14: |
| 7678 | if (op) |
| 7679 | return "i64"; |
| 7680 | else |
| 7681 | return "i8"; |
| 7682 | case 15: |
| 7683 | if (op) |
| 7684 | vassert(0); |
| 7685 | else |
| 7686 | return "f32"; |
| 7687 | default: |
| 7688 | vassert(0); |
| 7689 | } |
| 7690 | } |
| 7691 | |
| 7692 | static |
| 7693 | void DIPimm(UInt imm, UInt cmode, UInt op, |
| 7694 | const char *instr, UInt Q, UInt dreg) |
| 7695 | { |
| 7696 | if (vex_traceflags & VEX_TRACE_FE) { |
| 7697 | vex_printf("%s.%s %c%u, #", instr, |
| 7698 | ppNeonImmType(cmode, op), Q ? 'q' : 'd', dreg); |
| 7699 | ppNeonImm(imm, cmode, op); |
| 7700 | vex_printf("\n"); |
| 7701 | } |
| 7702 | } |
| 7703 | |
| 7704 | static |
| 7705 | Bool dis_neon_data_1reg_and_imm ( UInt theInstr, IRTemp condT ) |
| 7706 | { |
| 7707 | UInt dreg = get_neon_d_regno(theInstr); |
| 7708 | ULong imm_raw = ((theInstr >> 17) & 0x80) | ((theInstr >> 12) & 0x70) | |
| 7709 | (theInstr & 0xf); |
| 7710 | ULong imm_raw_pp = imm_raw; |
| 7711 | UInt cmode = (theInstr >> 8) & 0xf; |
| 7712 | UInt op_bit = (theInstr >> 5) & 1; |
| 7713 | ULong imm = 0; |
| 7714 | UInt Q = (theInstr >> 6) & 1; |
| 7715 | int i, j; |
| 7716 | UInt tmp; |
| 7717 | IRExpr *imm_val; |
| 7718 | IRExpr *expr; |
| 7719 | IRTemp tmp_var; |
| 7720 | switch(cmode) { |
| 7721 | case 7: case 6: |
| 7722 | imm_raw = imm_raw << 8; |
| 7723 | /* fallthrough */ |
| 7724 | case 5: case 4: |
| 7725 | imm_raw = imm_raw << 8; |
| 7726 | /* fallthrough */ |
| 7727 | case 3: case 2: |
| 7728 | imm_raw = imm_raw << 8; |
| 7729 | /* fallthrough */ |
| 7730 | case 0: case 1: |
| 7731 | imm = (imm_raw << 32) | imm_raw; |
| 7732 | break; |
| 7733 | case 11: case 10: |
| 7734 | imm_raw = imm_raw << 8; |
| 7735 | /* fallthrough */ |
| 7736 | case 9: case 8: |
| 7737 | imm_raw = (imm_raw << 16) | imm_raw; |
| 7738 | imm = (imm_raw << 32) | imm_raw; |
| 7739 | break; |
| 7740 | case 13: |
| 7741 | imm_raw = (imm_raw << 8) | 0xff; |
| 7742 | /* fallthrough */ |
| 7743 | case 12: |
| 7744 | imm_raw = (imm_raw << 8) | 0xff; |
| 7745 | imm = (imm_raw << 32) | imm_raw; |
| 7746 | break; |
| 7747 | case 14: |
| 7748 | if (! op_bit) { |
| 7749 | for(i = 0; i < 8; i++) { |
| 7750 | imm = (imm << 8) | imm_raw; |
| 7751 | } |
| 7752 | } else { |
| 7753 | for(i = 7; i >= 0; i--) { |
| 7754 | tmp = 0; |
| 7755 | for(j = 0; j < 8; j++) { |
| 7756 | tmp = (tmp << 1) | ((imm_raw >> i) & 1); |
| 7757 | } |
| 7758 | imm = (imm << 8) | tmp; |
| 7759 | } |
| 7760 | } |
| 7761 | break; |
| 7762 | case 15: |
| 7763 | imm = (imm_raw & 0x80) << 5; |
sewardj | f755839 | 2011-07-19 15:48:29 +0000 | [diff] [blame] | 7764 | imm |= ((~imm_raw & 0x40) << 5); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 7765 | for(i = 1; i <= 4; i++) |
| 7766 | imm |= (imm_raw & 0x40) << i; |
| 7767 | imm |= (imm_raw & 0x7f); |
| 7768 | imm = imm << 19; |
| 7769 | imm = (imm << 32) | imm; |
| 7770 | break; |
| 7771 | default: |
| 7772 | return False; |
| 7773 | } |
| 7774 | if (Q) { |
| 7775 | imm_val = binop(Iop_64HLtoV128, mkU64(imm), mkU64(imm)); |
| 7776 | } else { |
| 7777 | imm_val = mkU64(imm); |
| 7778 | } |
| 7779 | if (((op_bit == 0) && |
| 7780 | (((cmode & 9) == 0) || ((cmode & 13) == 8) || ((cmode & 12) == 12))) || |
| 7781 | ((op_bit == 1) && (cmode == 14))) { |
| 7782 | /* VMOV (immediate) */ |
| 7783 | if (Q) { |
| 7784 | putQReg(dreg, imm_val, condT); |
| 7785 | } else { |
| 7786 | putDRegI64(dreg, imm_val, condT); |
| 7787 | } |
| 7788 | DIPimm(imm_raw_pp, cmode, op_bit, "vmov", Q, dreg); |
| 7789 | return True; |
| 7790 | } |
| 7791 | if ((op_bit == 1) && |
| 7792 | (((cmode & 9) == 0) || ((cmode & 13) == 8) || ((cmode & 14) == 12))) { |
| 7793 | /* VMVN (immediate) */ |
| 7794 | if (Q) { |
| 7795 | putQReg(dreg, unop(Iop_NotV128, imm_val), condT); |
| 7796 | } else { |
| 7797 | putDRegI64(dreg, unop(Iop_Not64, imm_val), condT); |
| 7798 | } |
| 7799 | DIPimm(imm_raw_pp, cmode, op_bit, "vmvn", Q, dreg); |
| 7800 | return True; |
| 7801 | } |
| 7802 | if (Q) { |
| 7803 | tmp_var = newTemp(Ity_V128); |
| 7804 | assign(tmp_var, getQReg(dreg)); |
| 7805 | } else { |
| 7806 | tmp_var = newTemp(Ity_I64); |
| 7807 | assign(tmp_var, getDRegI64(dreg)); |
| 7808 | } |
| 7809 | if ((op_bit == 0) && (((cmode & 9) == 1) || ((cmode & 13) == 9))) { |
| 7810 | /* VORR (immediate) */ |
| 7811 | if (Q) |
| 7812 | expr = binop(Iop_OrV128, mkexpr(tmp_var), imm_val); |
| 7813 | else |
| 7814 | expr = binop(Iop_Or64, mkexpr(tmp_var), imm_val); |
| 7815 | DIPimm(imm_raw_pp, cmode, op_bit, "vorr", Q, dreg); |
| 7816 | } else if ((op_bit == 1) && (((cmode & 9) == 1) || ((cmode & 13) == 9))) { |
| 7817 | /* VBIC (immediate) */ |
| 7818 | if (Q) |
| 7819 | expr = binop(Iop_AndV128, mkexpr(tmp_var), |
| 7820 | unop(Iop_NotV128, imm_val)); |
| 7821 | else |
| 7822 | expr = binop(Iop_And64, mkexpr(tmp_var), unop(Iop_Not64, imm_val)); |
| 7823 | DIPimm(imm_raw_pp, cmode, op_bit, "vbic", Q, dreg); |
| 7824 | } else { |
| 7825 | return False; |
| 7826 | } |
| 7827 | if (Q) |
| 7828 | putQReg(dreg, expr, condT); |
| 7829 | else |
| 7830 | putDRegI64(dreg, expr, condT); |
| 7831 | return True; |
| 7832 | } |
| 7833 | |
| 7834 | /* A7.4 Advanced SIMD data-processing instructions */ |
| 7835 | static |
| 7836 | Bool dis_neon_data_processing ( UInt theInstr, IRTemp condT ) |
| 7837 | { |
| 7838 | UInt A = (theInstr >> 19) & 0x1F; |
| 7839 | UInt B = (theInstr >> 8) & 0xF; |
| 7840 | UInt C = (theInstr >> 4) & 0xF; |
| 7841 | UInt U = (theInstr >> 24) & 0x1; |
| 7842 | |
| 7843 | if (! (A & 0x10)) { |
| 7844 | return dis_neon_data_3same(theInstr, condT); |
| 7845 | } |
| 7846 | if (((A & 0x17) == 0x10) && ((C & 0x9) == 0x1)) { |
| 7847 | return dis_neon_data_1reg_and_imm(theInstr, condT); |
| 7848 | } |
| 7849 | if ((C & 1) == 1) { |
| 7850 | return dis_neon_data_2reg_and_shift(theInstr, condT); |
| 7851 | } |
| 7852 | if (((C & 5) == 0) && (((A & 0x14) == 0x10) || ((A & 0x16) == 0x14))) { |
| 7853 | return dis_neon_data_3diff(theInstr, condT); |
| 7854 | } |
| 7855 | if (((C & 5) == 4) && (((A & 0x14) == 0x10) || ((A & 0x16) == 0x14))) { |
| 7856 | return dis_neon_data_2reg_and_scalar(theInstr, condT); |
| 7857 | } |
| 7858 | if ((A & 0x16) == 0x16) { |
| 7859 | if ((U == 0) && ((C & 1) == 0)) { |
| 7860 | return dis_neon_vext(theInstr, condT); |
| 7861 | } |
| 7862 | if ((U != 1) || ((C & 1) == 1)) |
| 7863 | return False; |
| 7864 | if ((B & 8) == 0) { |
| 7865 | return dis_neon_data_2reg_misc(theInstr, condT); |
| 7866 | } |
| 7867 | if ((B & 12) == 8) { |
| 7868 | return dis_neon_vtb(theInstr, condT); |
| 7869 | } |
| 7870 | if ((B == 12) && ((C & 9) == 0)) { |
| 7871 | return dis_neon_vdup(theInstr, condT); |
| 7872 | } |
| 7873 | } |
| 7874 | return False; |
| 7875 | } |
| 7876 | |
| 7877 | |
| 7878 | /*------------------------------------------------------------*/ |
| 7879 | /*--- NEON loads and stores ---*/ |
| 7880 | /*------------------------------------------------------------*/ |
| 7881 | |
| 7882 | /* For NEON memory operations, we use the standard scheme to handle |
| 7883 | conditionalisation: generate a jump around the instruction if the |
| 7884 | condition is false. That's only necessary in Thumb mode, however, |
| 7885 | since in ARM mode NEON instructions are unconditional. */ |
| 7886 | |
| 7887 | /* A helper function for what follows. It assumes we already went |
| 7888 | uncond as per comments at the top of this section. */ |
| 7889 | static |
| 7890 | void mk_neon_elem_load_to_one_lane( UInt rD, UInt inc, UInt index, |
| 7891 | UInt N, UInt size, IRTemp addr ) |
| 7892 | { |
| 7893 | UInt i; |
| 7894 | switch (size) { |
| 7895 | case 0: |
| 7896 | putDRegI64(rD, triop(Iop_SetElem8x8, getDRegI64(rD), mkU8(index), |
| 7897 | loadLE(Ity_I8, mkexpr(addr))), IRTemp_INVALID); |
| 7898 | break; |
| 7899 | case 1: |
| 7900 | putDRegI64(rD, triop(Iop_SetElem16x4, getDRegI64(rD), mkU8(index), |
| 7901 | loadLE(Ity_I16, mkexpr(addr))), IRTemp_INVALID); |
| 7902 | break; |
| 7903 | case 2: |
| 7904 | putDRegI64(rD, triop(Iop_SetElem32x2, getDRegI64(rD), mkU8(index), |
| 7905 | loadLE(Ity_I32, mkexpr(addr))), IRTemp_INVALID); |
| 7906 | break; |
| 7907 | default: |
| 7908 | vassert(0); |
| 7909 | } |
| 7910 | for (i = 1; i <= N; i++) { |
| 7911 | switch (size) { |
| 7912 | case 0: |
| 7913 | putDRegI64(rD + i * inc, |
| 7914 | triop(Iop_SetElem8x8, |
| 7915 | getDRegI64(rD + i * inc), |
| 7916 | mkU8(index), |
| 7917 | loadLE(Ity_I8, binop(Iop_Add32, |
| 7918 | mkexpr(addr), |
| 7919 | mkU32(i * 1)))), |
| 7920 | IRTemp_INVALID); |
| 7921 | break; |
| 7922 | case 1: |
| 7923 | putDRegI64(rD + i * inc, |
| 7924 | triop(Iop_SetElem16x4, |
| 7925 | getDRegI64(rD + i * inc), |
| 7926 | mkU8(index), |
| 7927 | loadLE(Ity_I16, binop(Iop_Add32, |
| 7928 | mkexpr(addr), |
| 7929 | mkU32(i * 2)))), |
| 7930 | IRTemp_INVALID); |
| 7931 | break; |
| 7932 | case 2: |
| 7933 | putDRegI64(rD + i * inc, |
| 7934 | triop(Iop_SetElem32x2, |
| 7935 | getDRegI64(rD + i * inc), |
| 7936 | mkU8(index), |
| 7937 | loadLE(Ity_I32, binop(Iop_Add32, |
| 7938 | mkexpr(addr), |
| 7939 | mkU32(i * 4)))), |
| 7940 | IRTemp_INVALID); |
| 7941 | break; |
| 7942 | default: |
| 7943 | vassert(0); |
| 7944 | } |
| 7945 | } |
| 7946 | } |
| 7947 | |
| 7948 | /* A(nother) helper function for what follows. It assumes we already |
| 7949 | went uncond as per comments at the top of this section. */ |
| 7950 | static |
| 7951 | void mk_neon_elem_store_from_one_lane( UInt rD, UInt inc, UInt index, |
| 7952 | UInt N, UInt size, IRTemp addr ) |
| 7953 | { |
| 7954 | UInt i; |
| 7955 | switch (size) { |
| 7956 | case 0: |
| 7957 | storeLE(mkexpr(addr), |
| 7958 | binop(Iop_GetElem8x8, getDRegI64(rD), mkU8(index))); |
| 7959 | break; |
| 7960 | case 1: |
| 7961 | storeLE(mkexpr(addr), |
| 7962 | binop(Iop_GetElem16x4, getDRegI64(rD), mkU8(index))); |
| 7963 | break; |
| 7964 | case 2: |
| 7965 | storeLE(mkexpr(addr), |
| 7966 | binop(Iop_GetElem32x2, getDRegI64(rD), mkU8(index))); |
| 7967 | break; |
| 7968 | default: |
| 7969 | vassert(0); |
| 7970 | } |
| 7971 | for (i = 1; i <= N; i++) { |
| 7972 | switch (size) { |
| 7973 | case 0: |
| 7974 | storeLE(binop(Iop_Add32, mkexpr(addr), mkU32(i * 1)), |
| 7975 | binop(Iop_GetElem8x8, getDRegI64(rD + i * inc), |
| 7976 | mkU8(index))); |
| 7977 | break; |
| 7978 | case 1: |
| 7979 | storeLE(binop(Iop_Add32, mkexpr(addr), mkU32(i * 2)), |
| 7980 | binop(Iop_GetElem16x4, getDRegI64(rD + i * inc), |
| 7981 | mkU8(index))); |
| 7982 | break; |
| 7983 | case 2: |
| 7984 | storeLE(binop(Iop_Add32, mkexpr(addr), mkU32(i * 4)), |
| 7985 | binop(Iop_GetElem32x2, getDRegI64(rD + i * inc), |
| 7986 | mkU8(index))); |
| 7987 | break; |
| 7988 | default: |
| 7989 | vassert(0); |
| 7990 | } |
| 7991 | } |
| 7992 | } |
| 7993 | |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 7994 | /* Generate 2x64 -> 2x64 deinterleave code, for VLD2. Caller must |
| 7995 | make *u0 and *u1 be valid IRTemps before the call. */ |
| 7996 | static void math_DEINTERLEAVE_2 (/*OUT*/IRTemp* u0, /*OUT*/IRTemp* u1, |
| 7997 | IRTemp i0, IRTemp i1, Int laneszB) |
| 7998 | { |
| 7999 | /* The following assumes that the guest is little endian, and hence |
| 8000 | that the memory-side (interleaved) data is stored |
| 8001 | little-endianly. */ |
| 8002 | vassert(u0 && u1); |
| 8003 | /* This is pretty easy, since we have primitives directly to |
| 8004 | hand. */ |
| 8005 | if (laneszB == 4) { |
| 8006 | // memLE(128 bits) == A0 B0 A1 B1 |
| 8007 | // i0 == B0 A0, i1 == B1 A1 |
| 8008 | // u0 == A1 A0, u1 == B1 B0 |
| 8009 | assign(*u0, binop(Iop_InterleaveLO32x2, mkexpr(i1), mkexpr(i0))); |
| 8010 | assign(*u1, binop(Iop_InterleaveHI32x2, mkexpr(i1), mkexpr(i0))); |
| 8011 | } else if (laneszB == 2) { |
| 8012 | // memLE(128 bits) == A0 B0 A1 B1 A2 B2 A3 B3 |
| 8013 | // i0 == B1 A1 B0 A0, i1 == B3 A3 B2 A2 |
| 8014 | // u0 == A3 A2 A1 A0, u1 == B3 B2 B1 B0 |
| 8015 | assign(*u0, binop(Iop_CatEvenLanes16x4, mkexpr(i1), mkexpr(i0))); |
| 8016 | assign(*u1, binop(Iop_CatOddLanes16x4, mkexpr(i1), mkexpr(i0))); |
| 8017 | } else if (laneszB == 1) { |
| 8018 | // memLE(128 bits) == A0 B0 A1 B1 A2 B2 A3 B3 A4 B4 A5 B5 A6 B6 A7 B7 |
| 8019 | // i0 == B3 A3 B2 A2 B1 A1 B0 A0, i1 == B7 A7 B6 A6 B5 A5 B4 A4 |
| 8020 | // u0 == A7 A6 A5 A4 A3 A2 A1 A0, u1 == B7 B6 B5 B4 B3 B2 B1 B0 |
| 8021 | assign(*u0, binop(Iop_CatEvenLanes8x8, mkexpr(i1), mkexpr(i0))); |
| 8022 | assign(*u1, binop(Iop_CatOddLanes8x8, mkexpr(i1), mkexpr(i0))); |
| 8023 | } else { |
| 8024 | // Can never happen, since VLD2 only has valid lane widths of 32, |
| 8025 | // 16 or 8 bits. |
| 8026 | vpanic("math_DEINTERLEAVE_2"); |
| 8027 | } |
| 8028 | } |
| 8029 | |
| 8030 | /* Generate 2x64 -> 2x64 interleave code, for VST2. Caller must make |
| 8031 | *u0 and *u1 be valid IRTemps before the call. */ |
| 8032 | static void math_INTERLEAVE_2 (/*OUT*/IRTemp* i0, /*OUT*/IRTemp* i1, |
| 8033 | IRTemp u0, IRTemp u1, Int laneszB) |
| 8034 | { |
| 8035 | /* The following assumes that the guest is little endian, and hence |
| 8036 | that the memory-side (interleaved) data is stored |
| 8037 | little-endianly. */ |
sewardj | a73ab63 | 2014-06-25 11:59:24 +0000 | [diff] [blame] | 8038 | vassert(i0 && i1); |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 8039 | /* This is pretty easy, since we have primitives directly to |
| 8040 | hand. */ |
| 8041 | if (laneszB == 4) { |
| 8042 | // memLE(128 bits) == A0 B0 A1 B1 |
| 8043 | // i0 == B0 A0, i1 == B1 A1 |
| 8044 | // u0 == A1 A0, u1 == B1 B0 |
| 8045 | assign(*i0, binop(Iop_InterleaveLO32x2, mkexpr(u1), mkexpr(u0))); |
| 8046 | assign(*i1, binop(Iop_InterleaveHI32x2, mkexpr(u1), mkexpr(u0))); |
| 8047 | } else if (laneszB == 2) { |
| 8048 | // memLE(128 bits) == A0 B0 A1 B1 A2 B2 A3 B3 |
| 8049 | // i0 == B1 A1 B0 A0, i1 == B3 A3 B2 A2 |
| 8050 | // u0 == A3 A2 A1 A0, u1 == B3 B2 B1 B0 |
| 8051 | assign(*i0, binop(Iop_InterleaveLO16x4, mkexpr(u1), mkexpr(u0))); |
| 8052 | assign(*i1, binop(Iop_InterleaveHI16x4, mkexpr(u1), mkexpr(u0))); |
| 8053 | } else if (laneszB == 1) { |
| 8054 | // memLE(128 bits) == A0 B0 A1 B1 A2 B2 A3 B3 A4 B4 A5 B5 A6 B6 A7 B7 |
| 8055 | // i0 == B3 A3 B2 A2 B1 A1 B0 A0, i1 == B7 A7 B6 A6 B5 A5 B4 A4 |
| 8056 | // u0 == A7 A6 A5 A4 A3 A2 A1 A0, u1 == B7 B6 B5 B4 B3 B2 B1 B0 |
| 8057 | assign(*i0, binop(Iop_InterleaveLO8x8, mkexpr(u1), mkexpr(u0))); |
| 8058 | assign(*i1, binop(Iop_InterleaveHI8x8, mkexpr(u1), mkexpr(u0))); |
| 8059 | } else { |
| 8060 | // Can never happen, since VST2 only has valid lane widths of 32, |
| 8061 | // 16 or 8 bits. |
| 8062 | vpanic("math_INTERLEAVE_2"); |
| 8063 | } |
| 8064 | } |
| 8065 | |
sewardj | c1e23c4 | 2013-04-20 21:19:44 +0000 | [diff] [blame] | 8066 | // Helper function for generating arbitrary slicing 'n' dicing of |
| 8067 | // 3 8x8 vectors, as needed for VLD3.8 and VST3.8. |
| 8068 | static IRExpr* math_PERM_8x8x3(const UChar* desc, |
| 8069 | IRTemp s0, IRTemp s1, IRTemp s2) |
| 8070 | { |
| 8071 | // desc is an array of 8 pairs, encoded as 16 bytes, |
| 8072 | // that describe how to assemble the result lanes, starting with |
| 8073 | // lane 7. Each pair is: first component (0..2) says which of |
| 8074 | // s0/s1/s2 to use. Second component (0..7) is the lane number |
| 8075 | // in the source to use. |
| 8076 | UInt si; |
| 8077 | for (si = 0; si < 7; si++) { |
| 8078 | vassert(desc[2 * si + 0] <= 2); |
| 8079 | vassert(desc[2 * si + 1] <= 7); |
| 8080 | } |
| 8081 | IRTemp h3 = newTemp(Ity_I64); |
| 8082 | IRTemp h2 = newTemp(Ity_I64); |
| 8083 | IRTemp h1 = newTemp(Ity_I64); |
| 8084 | IRTemp h0 = newTemp(Ity_I64); |
| 8085 | IRTemp srcs[3] = {s0, s1, s2}; |
| 8086 | # define SRC_VEC(_lane) mkexpr(srcs[desc[2 * (7-(_lane)) + 0]]) |
| 8087 | # define SRC_SHIFT(_lane) mkU8(56-8*(desc[2 * (7-(_lane)) + 1])) |
| 8088 | assign(h3, binop(Iop_InterleaveHI8x8, |
| 8089 | binop(Iop_Shl64, SRC_VEC(7), SRC_SHIFT(7)), |
| 8090 | binop(Iop_Shl64, SRC_VEC(6), SRC_SHIFT(6)))); |
| 8091 | assign(h2, binop(Iop_InterleaveHI8x8, |
| 8092 | binop(Iop_Shl64, SRC_VEC(5), SRC_SHIFT(5)), |
| 8093 | binop(Iop_Shl64, SRC_VEC(4), SRC_SHIFT(4)))); |
| 8094 | assign(h1, binop(Iop_InterleaveHI8x8, |
| 8095 | binop(Iop_Shl64, SRC_VEC(3), SRC_SHIFT(3)), |
| 8096 | binop(Iop_Shl64, SRC_VEC(2), SRC_SHIFT(2)))); |
| 8097 | assign(h0, binop(Iop_InterleaveHI8x8, |
| 8098 | binop(Iop_Shl64, SRC_VEC(1), SRC_SHIFT(1)), |
| 8099 | binop(Iop_Shl64, SRC_VEC(0), SRC_SHIFT(0)))); |
| 8100 | # undef SRC_VEC |
| 8101 | # undef SRC_SHIFT |
| 8102 | // Now h3..h0 are 64 bit vectors with useful information only |
| 8103 | // in the top 16 bits. We now concatentate those four 16-bit |
| 8104 | // groups so as to produce the final result. |
| 8105 | IRTemp w1 = newTemp(Ity_I64); |
| 8106 | IRTemp w0 = newTemp(Ity_I64); |
| 8107 | assign(w1, binop(Iop_InterleaveHI16x4, mkexpr(h3), mkexpr(h2))); |
| 8108 | assign(w0, binop(Iop_InterleaveHI16x4, mkexpr(h1), mkexpr(h0))); |
| 8109 | return binop(Iop_InterleaveHI32x2, mkexpr(w1), mkexpr(w0)); |
| 8110 | } |
| 8111 | |
| 8112 | /* Generate 3x64 -> 3x64 deinterleave code, for VLD3. Caller must |
| 8113 | make *u0, *u1 and *u2 be valid IRTemps before the call. */ |
| 8114 | static void math_DEINTERLEAVE_3 ( |
| 8115 | /*OUT*/IRTemp* u0, /*OUT*/IRTemp* u1, /*OUT*/IRTemp* u2, |
| 8116 | IRTemp i0, IRTemp i1, IRTemp i2, Int laneszB |
| 8117 | ) |
| 8118 | { |
| 8119 | # define IHI32x2(_e1, _e2) binop(Iop_InterleaveHI32x2, (_e1), (_e2)) |
| 8120 | # define IHI16x4(_e1, _e2) binop(Iop_InterleaveHI16x4, (_e1), (_e2)) |
| 8121 | # define SHL64(_tmp, _amt) binop(Iop_Shl64, mkexpr(_tmp), mkU8(_amt)) |
| 8122 | /* The following assumes that the guest is little endian, and hence |
| 8123 | that the memory-side (interleaved) data is stored |
| 8124 | little-endianly. */ |
| 8125 | vassert(u0 && u1 && u2); |
| 8126 | if (laneszB == 4) { |
| 8127 | // memLE(192 bits) == A0 B0 C0 A1 B1 C1 |
| 8128 | // i0 == B0 A0, i1 == A1 C0, i2 == C1 B1 |
| 8129 | // u0 == A1 A0, u1 == B1 B0, u2 == C1 C0 |
| 8130 | assign(*u0, IHI32x2(SHL64(i1, 0), SHL64(i0, 32))); |
| 8131 | assign(*u1, IHI32x2(SHL64(i2, 32), SHL64(i0, 0))); |
| 8132 | assign(*u2, IHI32x2(SHL64(i2, 0), SHL64(i1, 32))); |
| 8133 | } else if (laneszB == 2) { |
| 8134 | // memLE(192 bits) == A0 B0 C0 A1, B1 C1 A2 B2, C2 A3 B3 C3 |
| 8135 | // i0 == A1 C0 B0 A0, i1 == B2 A2 C1 B1, i2 == C3 B3 A3 C2 |
| 8136 | // u0 == A3 A2 A1 A0, u1 == B3 B2 B1 B0, u2 == C3 C2 C1 C0 |
| 8137 | # define XXX(_tmp3,_la3,_tmp2,_la2,_tmp1,_la1,_tmp0,_la0) \ |
| 8138 | IHI32x2( \ |
| 8139 | IHI16x4(SHL64((_tmp3),48-16*(_la3)), \ |
| 8140 | SHL64((_tmp2),48-16*(_la2))), \ |
| 8141 | IHI16x4(SHL64((_tmp1),48-16*(_la1)), \ |
| 8142 | SHL64((_tmp0),48-16*(_la0)))) |
| 8143 | assign(*u0, XXX(i2,1, i1,2, i0,3, i0,0)); |
| 8144 | assign(*u1, XXX(i2,2, i1,3, i1,0, i0,1)); |
| 8145 | assign(*u2, XXX(i2,3, i2,0, i1,1, i0,2)); |
| 8146 | # undef XXX |
| 8147 | } else if (laneszB == 1) { |
| 8148 | // These describe how the result vectors [7..0] are |
| 8149 | // assembled from the source vectors. Each pair is |
| 8150 | // (source vector number, lane number). |
| 8151 | static const UChar de0[16] = {2,5, 2,2, 1,7, 1,4, 1,1, 0,6, 0,3, 0,0}; |
| 8152 | static const UChar de1[16] = {2,6, 2,3, 2,0, 1,5, 1,2, 0,7, 0,4, 0,1}; |
| 8153 | static const UChar de2[16] = {2,7, 2,4, 2,1, 1,6, 1,3, 1,0, 0,5, 0,2}; |
| 8154 | assign(*u0, math_PERM_8x8x3(de0, i0, i1, i2)); |
| 8155 | assign(*u1, math_PERM_8x8x3(de1, i0, i1, i2)); |
| 8156 | assign(*u2, math_PERM_8x8x3(de2, i0, i1, i2)); |
| 8157 | } else { |
| 8158 | // Can never happen, since VLD3 only has valid lane widths of 32, |
| 8159 | // 16 or 8 bits. |
| 8160 | vpanic("math_DEINTERLEAVE_3"); |
| 8161 | } |
| 8162 | # undef SHL64 |
| 8163 | # undef IHI16x4 |
| 8164 | # undef IHI32x2 |
| 8165 | } |
| 8166 | |
| 8167 | /* Generate 3x64 -> 3x64 interleave code, for VST3. Caller must |
| 8168 | make *i0, *i1 and *i2 be valid IRTemps before the call. */ |
| 8169 | static void math_INTERLEAVE_3 ( |
| 8170 | /*OUT*/IRTemp* i0, /*OUT*/IRTemp* i1, /*OUT*/IRTemp* i2, |
| 8171 | IRTemp u0, IRTemp u1, IRTemp u2, Int laneszB |
| 8172 | ) |
| 8173 | { |
| 8174 | # define IHI32x2(_e1, _e2) binop(Iop_InterleaveHI32x2, (_e1), (_e2)) |
| 8175 | # define IHI16x4(_e1, _e2) binop(Iop_InterleaveHI16x4, (_e1), (_e2)) |
| 8176 | # define SHL64(_tmp, _amt) binop(Iop_Shl64, mkexpr(_tmp), mkU8(_amt)) |
| 8177 | /* The following assumes that the guest is little endian, and hence |
| 8178 | that the memory-side (interleaved) data is stored |
| 8179 | little-endianly. */ |
| 8180 | vassert(i0 && i1 && i2); |
| 8181 | if (laneszB == 4) { |
| 8182 | // memLE(192 bits) == A0 B0 C0 A1 B1 C1 |
| 8183 | // i0 == B0 A0, i1 == A1 C0, i2 == C1 B1 |
| 8184 | // u0 == A1 A0, u1 == B1 B0, u2 == C1 C0 |
| 8185 | assign(*i0, IHI32x2(SHL64(u1, 32), SHL64(u0, 32))); |
| 8186 | assign(*i1, IHI32x2(SHL64(u0, 0), SHL64(u2, 32))); |
| 8187 | assign(*i2, IHI32x2(SHL64(u2, 0), SHL64(u1, 0))); |
| 8188 | } else if (laneszB == 2) { |
| 8189 | // memLE(192 bits) == A0 B0 C0 A1, B1 C1 A2 B2, C2 A3 B3 C3 |
| 8190 | // i0 == A1 C0 B0 A0, i1 == B2 A2 C1 B1, i2 == C3 B3 A3 C2 |
| 8191 | // u0 == A3 A2 A1 A0, u1 == B3 B2 B1 B0, u2 == C3 C2 C1 C0 |
| 8192 | # define XXX(_tmp3,_la3,_tmp2,_la2,_tmp1,_la1,_tmp0,_la0) \ |
| 8193 | IHI32x2( \ |
| 8194 | IHI16x4(SHL64((_tmp3),48-16*(_la3)), \ |
| 8195 | SHL64((_tmp2),48-16*(_la2))), \ |
| 8196 | IHI16x4(SHL64((_tmp1),48-16*(_la1)), \ |
| 8197 | SHL64((_tmp0),48-16*(_la0)))) |
| 8198 | assign(*i0, XXX(u0,1, u2,0, u1,0, u0,0)); |
| 8199 | assign(*i1, XXX(u1,2, u0,2, u2,1, u1,1)); |
| 8200 | assign(*i2, XXX(u2,3, u1,3, u0,3, u2,2)); |
| 8201 | # undef XXX |
| 8202 | } else if (laneszB == 1) { |
| 8203 | // These describe how the result vectors [7..0] are |
| 8204 | // assembled from the source vectors. Each pair is |
| 8205 | // (source vector number, lane number). |
| 8206 | static const UChar in0[16] = {1,2, 0,2, 2,1, 1,1, 0,1, 2,0, 1,0, 0,0}; |
| 8207 | static const UChar in1[16] = {0,5, 2,4, 1,4, 0,4, 2,3, 1,3, 0,3, 2,2}; |
| 8208 | static const UChar in2[16] = {2,7, 1,7, 0,7, 2,6, 1,6, 0,6, 2,5, 1,5}; |
| 8209 | assign(*i0, math_PERM_8x8x3(in0, u0, u1, u2)); |
| 8210 | assign(*i1, math_PERM_8x8x3(in1, u0, u1, u2)); |
| 8211 | assign(*i2, math_PERM_8x8x3(in2, u0, u1, u2)); |
| 8212 | } else { |
| 8213 | // Can never happen, since VST3 only has valid lane widths of 32, |
| 8214 | // 16 or 8 bits. |
| 8215 | vpanic("math_INTERLEAVE_3"); |
| 8216 | } |
| 8217 | # undef SHL64 |
| 8218 | # undef IHI16x4 |
| 8219 | # undef IHI32x2 |
| 8220 | } |
| 8221 | |
sewardj | 0adc2f2 | 2013-04-20 23:27:36 +0000 | [diff] [blame] | 8222 | /* Generate 4x64 -> 4x64 deinterleave code, for VLD4. Caller must |
| 8223 | make *u0, *u1, *u2 and *u3 be valid IRTemps before the call. */ |
| 8224 | static void math_DEINTERLEAVE_4 ( |
| 8225 | /*OUT*/IRTemp* u0, /*OUT*/IRTemp* u1, |
| 8226 | /*OUT*/IRTemp* u2, /*OUT*/IRTemp* u3, |
| 8227 | IRTemp i0, IRTemp i1, IRTemp i2, IRTemp i3, Int laneszB |
| 8228 | ) |
| 8229 | { |
| 8230 | # define IHI32x2(_t1, _t2) \ |
| 8231 | binop(Iop_InterleaveHI32x2, mkexpr(_t1), mkexpr(_t2)) |
| 8232 | # define ILO32x2(_t1, _t2) \ |
| 8233 | binop(Iop_InterleaveLO32x2, mkexpr(_t1), mkexpr(_t2)) |
| 8234 | # define IHI16x4(_t1, _t2) \ |
| 8235 | binop(Iop_InterleaveHI16x4, mkexpr(_t1), mkexpr(_t2)) |
| 8236 | # define ILO16x4(_t1, _t2) \ |
| 8237 | binop(Iop_InterleaveLO16x4, mkexpr(_t1), mkexpr(_t2)) |
| 8238 | # define IHI8x8(_t1, _e2) \ |
| 8239 | binop(Iop_InterleaveHI8x8, mkexpr(_t1), _e2) |
| 8240 | # define SHL64(_tmp, _amt) \ |
| 8241 | binop(Iop_Shl64, mkexpr(_tmp), mkU8(_amt)) |
| 8242 | /* The following assumes that the guest is little endian, and hence |
| 8243 | that the memory-side (interleaved) data is stored |
| 8244 | little-endianly. */ |
| 8245 | vassert(u0 && u1 && u2 && u3); |
| 8246 | if (laneszB == 4) { |
| 8247 | assign(*u0, ILO32x2(i2, i0)); |
| 8248 | assign(*u1, IHI32x2(i2, i0)); |
| 8249 | assign(*u2, ILO32x2(i3, i1)); |
| 8250 | assign(*u3, IHI32x2(i3, i1)); |
| 8251 | } else if (laneszB == 2) { |
| 8252 | IRTemp b1b0a1a0 = newTemp(Ity_I64); |
| 8253 | IRTemp b3b2a3a2 = newTemp(Ity_I64); |
| 8254 | IRTemp d1d0c1c0 = newTemp(Ity_I64); |
| 8255 | IRTemp d3d2c3c2 = newTemp(Ity_I64); |
| 8256 | assign(b1b0a1a0, ILO16x4(i1, i0)); |
| 8257 | assign(b3b2a3a2, ILO16x4(i3, i2)); |
| 8258 | assign(d1d0c1c0, IHI16x4(i1, i0)); |
| 8259 | assign(d3d2c3c2, IHI16x4(i3, i2)); |
| 8260 | // And now do what we did for the 32-bit case. |
| 8261 | assign(*u0, ILO32x2(b3b2a3a2, b1b0a1a0)); |
| 8262 | assign(*u1, IHI32x2(b3b2a3a2, b1b0a1a0)); |
| 8263 | assign(*u2, ILO32x2(d3d2c3c2, d1d0c1c0)); |
| 8264 | assign(*u3, IHI32x2(d3d2c3c2, d1d0c1c0)); |
| 8265 | } else if (laneszB == 1) { |
| 8266 | // Deinterleave into 16-bit chunks, then do as the 16-bit case. |
| 8267 | IRTemp i0x = newTemp(Ity_I64); |
| 8268 | IRTemp i1x = newTemp(Ity_I64); |
| 8269 | IRTemp i2x = newTemp(Ity_I64); |
| 8270 | IRTemp i3x = newTemp(Ity_I64); |
| 8271 | assign(i0x, IHI8x8(i0, SHL64(i0, 32))); |
| 8272 | assign(i1x, IHI8x8(i1, SHL64(i1, 32))); |
| 8273 | assign(i2x, IHI8x8(i2, SHL64(i2, 32))); |
| 8274 | assign(i3x, IHI8x8(i3, SHL64(i3, 32))); |
| 8275 | // From here on is like the 16 bit case. |
| 8276 | IRTemp b1b0a1a0 = newTemp(Ity_I64); |
| 8277 | IRTemp b3b2a3a2 = newTemp(Ity_I64); |
| 8278 | IRTemp d1d0c1c0 = newTemp(Ity_I64); |
| 8279 | IRTemp d3d2c3c2 = newTemp(Ity_I64); |
| 8280 | assign(b1b0a1a0, ILO16x4(i1x, i0x)); |
| 8281 | assign(b3b2a3a2, ILO16x4(i3x, i2x)); |
| 8282 | assign(d1d0c1c0, IHI16x4(i1x, i0x)); |
| 8283 | assign(d3d2c3c2, IHI16x4(i3x, i2x)); |
| 8284 | // And now do what we did for the 32-bit case. |
| 8285 | assign(*u0, ILO32x2(b3b2a3a2, b1b0a1a0)); |
| 8286 | assign(*u1, IHI32x2(b3b2a3a2, b1b0a1a0)); |
| 8287 | assign(*u2, ILO32x2(d3d2c3c2, d1d0c1c0)); |
| 8288 | assign(*u3, IHI32x2(d3d2c3c2, d1d0c1c0)); |
| 8289 | } else { |
| 8290 | // Can never happen, since VLD4 only has valid lane widths of 32, |
| 8291 | // 16 or 8 bits. |
| 8292 | vpanic("math_DEINTERLEAVE_4"); |
| 8293 | } |
| 8294 | # undef SHL64 |
| 8295 | # undef IHI8x8 |
| 8296 | # undef ILO16x4 |
| 8297 | # undef IHI16x4 |
| 8298 | # undef ILO32x2 |
| 8299 | # undef IHI32x2 |
| 8300 | } |
| 8301 | |
| 8302 | /* Generate 4x64 -> 4x64 interleave code, for VST4. Caller must |
| 8303 | make *i0, *i1, *i2 and *i3 be valid IRTemps before the call. */ |
| 8304 | static void math_INTERLEAVE_4 ( |
| 8305 | /*OUT*/IRTemp* i0, /*OUT*/IRTemp* i1, |
| 8306 | /*OUT*/IRTemp* i2, /*OUT*/IRTemp* i3, |
| 8307 | IRTemp u0, IRTemp u1, IRTemp u2, IRTemp u3, Int laneszB |
| 8308 | ) |
| 8309 | { |
| 8310 | # define IHI32x2(_t1, _t2) \ |
| 8311 | binop(Iop_InterleaveHI32x2, mkexpr(_t1), mkexpr(_t2)) |
| 8312 | # define ILO32x2(_t1, _t2) \ |
| 8313 | binop(Iop_InterleaveLO32x2, mkexpr(_t1), mkexpr(_t2)) |
| 8314 | # define CEV16x4(_t1, _t2) \ |
| 8315 | binop(Iop_CatEvenLanes16x4, mkexpr(_t1), mkexpr(_t2)) |
| 8316 | # define COD16x4(_t1, _t2) \ |
| 8317 | binop(Iop_CatOddLanes16x4, mkexpr(_t1), mkexpr(_t2)) |
| 8318 | # define COD8x8(_t1, _e2) \ |
| 8319 | binop(Iop_CatOddLanes8x8, mkexpr(_t1), _e2) |
| 8320 | # define SHL64(_tmp, _amt) \ |
| 8321 | binop(Iop_Shl64, mkexpr(_tmp), mkU8(_amt)) |
| 8322 | /* The following assumes that the guest is little endian, and hence |
| 8323 | that the memory-side (interleaved) data is stored |
| 8324 | little-endianly. */ |
| 8325 | vassert(u0 && u1 && u2 && u3); |
| 8326 | if (laneszB == 4) { |
| 8327 | assign(*i0, ILO32x2(u1, u0)); |
| 8328 | assign(*i1, ILO32x2(u3, u2)); |
| 8329 | assign(*i2, IHI32x2(u1, u0)); |
| 8330 | assign(*i3, IHI32x2(u3, u2)); |
| 8331 | } else if (laneszB == 2) { |
| 8332 | // First, interleave at the 32-bit lane size. |
| 8333 | IRTemp b1b0a1a0 = newTemp(Ity_I64); |
| 8334 | IRTemp b3b2a3a2 = newTemp(Ity_I64); |
| 8335 | IRTemp d1d0c1c0 = newTemp(Ity_I64); |
| 8336 | IRTemp d3d2c3c2 = newTemp(Ity_I64); |
| 8337 | assign(b1b0a1a0, ILO32x2(u1, u0)); |
| 8338 | assign(b3b2a3a2, IHI32x2(u1, u0)); |
| 8339 | assign(d1d0c1c0, ILO32x2(u3, u2)); |
| 8340 | assign(d3d2c3c2, IHI32x2(u3, u2)); |
| 8341 | // And interleave (cat) at the 16 bit size. |
| 8342 | assign(*i0, CEV16x4(d1d0c1c0, b1b0a1a0)); |
| 8343 | assign(*i1, COD16x4(d1d0c1c0, b1b0a1a0)); |
| 8344 | assign(*i2, CEV16x4(d3d2c3c2, b3b2a3a2)); |
| 8345 | assign(*i3, COD16x4(d3d2c3c2, b3b2a3a2)); |
| 8346 | } else if (laneszB == 1) { |
| 8347 | // First, interleave at the 32-bit lane size. |
| 8348 | IRTemp b1b0a1a0 = newTemp(Ity_I64); |
| 8349 | IRTemp b3b2a3a2 = newTemp(Ity_I64); |
| 8350 | IRTemp d1d0c1c0 = newTemp(Ity_I64); |
| 8351 | IRTemp d3d2c3c2 = newTemp(Ity_I64); |
| 8352 | assign(b1b0a1a0, ILO32x2(u1, u0)); |
| 8353 | assign(b3b2a3a2, IHI32x2(u1, u0)); |
| 8354 | assign(d1d0c1c0, ILO32x2(u3, u2)); |
| 8355 | assign(d3d2c3c2, IHI32x2(u3, u2)); |
| 8356 | // And interleave (cat) at the 16 bit size. |
| 8357 | IRTemp i0x = newTemp(Ity_I64); |
| 8358 | IRTemp i1x = newTemp(Ity_I64); |
| 8359 | IRTemp i2x = newTemp(Ity_I64); |
| 8360 | IRTemp i3x = newTemp(Ity_I64); |
| 8361 | assign(i0x, CEV16x4(d1d0c1c0, b1b0a1a0)); |
| 8362 | assign(i1x, COD16x4(d1d0c1c0, b1b0a1a0)); |
| 8363 | assign(i2x, CEV16x4(d3d2c3c2, b3b2a3a2)); |
| 8364 | assign(i3x, COD16x4(d3d2c3c2, b3b2a3a2)); |
| 8365 | // And rearrange within each word, to get the right 8 bit lanes. |
| 8366 | assign(*i0, COD8x8(i0x, SHL64(i0x, 8))); |
| 8367 | assign(*i1, COD8x8(i1x, SHL64(i1x, 8))); |
| 8368 | assign(*i2, COD8x8(i2x, SHL64(i2x, 8))); |
| 8369 | assign(*i3, COD8x8(i3x, SHL64(i3x, 8))); |
| 8370 | } else { |
| 8371 | // Can never happen, since VLD4 only has valid lane widths of 32, |
| 8372 | // 16 or 8 bits. |
| 8373 | vpanic("math_DEINTERLEAVE_4"); |
| 8374 | } |
| 8375 | # undef SHL64 |
| 8376 | # undef COD8x8 |
| 8377 | # undef COD16x4 |
| 8378 | # undef CEV16x4 |
| 8379 | # undef ILO32x2 |
| 8380 | # undef IHI32x2 |
| 8381 | } |
sewardj | c1e23c4 | 2013-04-20 21:19:44 +0000 | [diff] [blame] | 8382 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8383 | /* A7.7 Advanced SIMD element or structure load/store instructions */ |
| 8384 | static |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8385 | Bool dis_neon_load_or_store ( UInt theInstr, |
| 8386 | Bool isT, IRTemp condT ) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8387 | { |
| 8388 | # define INSN(_bMax,_bMin) SLICE_UInt(theInstr, (_bMax), (_bMin)) |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8389 | UInt bA = INSN(23,23); |
| 8390 | UInt fB = INSN(11,8); |
| 8391 | UInt bL = INSN(21,21); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8392 | UInt rD = (INSN(22,22) << 4) | INSN(15,12); |
| 8393 | UInt rN = INSN(19,16); |
| 8394 | UInt rM = INSN(3,0); |
| 8395 | UInt N, size, i, j; |
| 8396 | UInt inc; |
| 8397 | UInt regs = 1; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8398 | |
| 8399 | if (isT) { |
| 8400 | vassert(condT != IRTemp_INVALID); |
| 8401 | } else { |
| 8402 | vassert(condT == IRTemp_INVALID); |
| 8403 | } |
| 8404 | /* So now, if condT is not IRTemp_INVALID, we know we're |
| 8405 | dealing with Thumb code. */ |
| 8406 | |
| 8407 | if (INSN(20,20) != 0) |
| 8408 | return False; |
| 8409 | |
sewardj | 21eaf24 | 2010-08-31 09:18:22 +0000 | [diff] [blame] | 8410 | IRTemp initialRn = newTemp(Ity_I32); |
| 8411 | assign(initialRn, isT ? getIRegT(rN) : getIRegA(rN)); |
| 8412 | |
| 8413 | IRTemp initialRm = newTemp(Ity_I32); |
| 8414 | assign(initialRm, isT ? getIRegT(rM) : getIRegA(rM)); |
| 8415 | |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8416 | /* There are 3 cases: |
| 8417 | (1) VSTn / VLDn (n-element structure from/to one lane) |
| 8418 | (2) VLDn (single element to all lanes) |
| 8419 | (3) VSTn / VLDn (multiple n-element structures) |
| 8420 | */ |
| 8421 | if (bA) { |
| 8422 | N = fB & 3; |
| 8423 | if ((fB >> 2) < 3) { |
| 8424 | /* ------------ Case (1) ------------ |
| 8425 | VSTn / VLDn (n-element structure from/to one lane) */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8426 | |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8427 | size = fB >> 2; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8428 | |
| 8429 | switch (size) { |
| 8430 | case 0: i = INSN(7,5); inc = 1; break; |
| 8431 | case 1: i = INSN(7,6); inc = INSN(5,5) ? 2 : 1; break; |
| 8432 | case 2: i = INSN(7,7); inc = INSN(6,6) ? 2 : 1; break; |
| 8433 | case 3: return False; |
| 8434 | default: vassert(0); |
| 8435 | } |
| 8436 | |
sewardj | 21eaf24 | 2010-08-31 09:18:22 +0000 | [diff] [blame] | 8437 | IRTemp addr = newTemp(Ity_I32); |
| 8438 | assign(addr, mkexpr(initialRn)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8439 | |
| 8440 | // go uncond |
| 8441 | if (condT != IRTemp_INVALID) |
| 8442 | mk_skip_over_T32_if_cond_is_false(condT); |
| 8443 | // now uncond |
| 8444 | |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8445 | if (bL) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8446 | mk_neon_elem_load_to_one_lane(rD, inc, i, N, size, addr); |
| 8447 | else |
| 8448 | mk_neon_elem_store_from_one_lane(rD, inc, i, N, size, addr); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 8449 | DIP("v%s%u.%d {", bL ? "ld" : "st", N + 1, 8 << size); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8450 | for (j = 0; j <= N; j++) { |
| 8451 | if (j) |
| 8452 | DIP(", "); |
| 8453 | DIP("d%u[%u]", rD + j * inc, i); |
| 8454 | } |
sewardj | 17bf0aa | 2010-08-31 09:29:51 +0000 | [diff] [blame] | 8455 | DIP("}, [r%u]", rN); |
| 8456 | if (rM != 13 && rM != 15) { |
| 8457 | DIP(", r%u\n", rM); |
| 8458 | } else { |
| 8459 | DIP("%s\n", (rM != 15) ? "!" : ""); |
| 8460 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8461 | } else { |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8462 | /* ------------ Case (2) ------------ |
| 8463 | VLDn (single element to all lanes) */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8464 | UInt r; |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8465 | if (bL == 0) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8466 | return False; |
| 8467 | |
| 8468 | inc = INSN(5,5) + 1; |
| 8469 | size = INSN(7,6); |
| 8470 | |
| 8471 | /* size == 3 and size == 2 cases differ in alignment constraints */ |
| 8472 | if (size == 3 && N == 3 && INSN(4,4) == 1) |
| 8473 | size = 2; |
| 8474 | |
| 8475 | if (size == 0 && N == 0 && INSN(4,4) == 1) |
| 8476 | return False; |
| 8477 | if (N == 2 && INSN(4,4) == 1) |
| 8478 | return False; |
| 8479 | if (size == 3) |
| 8480 | return False; |
| 8481 | |
| 8482 | // go uncond |
| 8483 | if (condT != IRTemp_INVALID) |
| 8484 | mk_skip_over_T32_if_cond_is_false(condT); |
| 8485 | // now uncond |
| 8486 | |
sewardj | 21eaf24 | 2010-08-31 09:18:22 +0000 | [diff] [blame] | 8487 | IRTemp addr = newTemp(Ity_I32); |
| 8488 | assign(addr, mkexpr(initialRn)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8489 | |
| 8490 | if (N == 0 && INSN(5,5)) |
| 8491 | regs = 2; |
| 8492 | |
| 8493 | for (r = 0; r < regs; r++) { |
| 8494 | switch (size) { |
| 8495 | case 0: |
| 8496 | putDRegI64(rD + r, unop(Iop_Dup8x8, |
| 8497 | loadLE(Ity_I8, mkexpr(addr))), |
| 8498 | IRTemp_INVALID); |
| 8499 | break; |
| 8500 | case 1: |
| 8501 | putDRegI64(rD + r, unop(Iop_Dup16x4, |
| 8502 | loadLE(Ity_I16, mkexpr(addr))), |
| 8503 | IRTemp_INVALID); |
| 8504 | break; |
| 8505 | case 2: |
| 8506 | putDRegI64(rD + r, unop(Iop_Dup32x2, |
| 8507 | loadLE(Ity_I32, mkexpr(addr))), |
| 8508 | IRTemp_INVALID); |
| 8509 | break; |
| 8510 | default: |
| 8511 | vassert(0); |
| 8512 | } |
| 8513 | for (i = 1; i <= N; i++) { |
| 8514 | switch (size) { |
| 8515 | case 0: |
| 8516 | putDRegI64(rD + r + i * inc, |
| 8517 | unop(Iop_Dup8x8, |
| 8518 | loadLE(Ity_I8, binop(Iop_Add32, |
| 8519 | mkexpr(addr), |
| 8520 | mkU32(i * 1)))), |
| 8521 | IRTemp_INVALID); |
| 8522 | break; |
| 8523 | case 1: |
| 8524 | putDRegI64(rD + r + i * inc, |
| 8525 | unop(Iop_Dup16x4, |
| 8526 | loadLE(Ity_I16, binop(Iop_Add32, |
| 8527 | mkexpr(addr), |
| 8528 | mkU32(i * 2)))), |
| 8529 | IRTemp_INVALID); |
| 8530 | break; |
| 8531 | case 2: |
| 8532 | putDRegI64(rD + r + i * inc, |
| 8533 | unop(Iop_Dup32x2, |
| 8534 | loadLE(Ity_I32, binop(Iop_Add32, |
| 8535 | mkexpr(addr), |
| 8536 | mkU32(i * 4)))), |
| 8537 | IRTemp_INVALID); |
| 8538 | break; |
| 8539 | default: |
| 8540 | vassert(0); |
| 8541 | } |
| 8542 | } |
| 8543 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 8544 | DIP("vld%u.%d {", N + 1, 8 << size); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8545 | for (r = 0; r < regs; r++) { |
| 8546 | for (i = 0; i <= N; i++) { |
| 8547 | if (i || r) |
| 8548 | DIP(", "); |
| 8549 | DIP("d%u[]", rD + r + i * inc); |
| 8550 | } |
| 8551 | } |
sewardj | 17bf0aa | 2010-08-31 09:29:51 +0000 | [diff] [blame] | 8552 | DIP("}, [r%u]", rN); |
| 8553 | if (rM != 13 && rM != 15) { |
| 8554 | DIP(", r%u\n", rM); |
| 8555 | } else { |
| 8556 | DIP("%s\n", (rM != 15) ? "!" : ""); |
| 8557 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8558 | } |
| 8559 | /* Writeback. We're uncond here, so no condT-ing. */ |
| 8560 | if (rM != 15) { |
| 8561 | if (rM == 13) { |
| 8562 | IRExpr* e = binop(Iop_Add32, |
sewardj | 21eaf24 | 2010-08-31 09:18:22 +0000 | [diff] [blame] | 8563 | mkexpr(initialRn), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8564 | mkU32((1 << size) * (N + 1))); |
| 8565 | if (isT) |
| 8566 | putIRegT(rN, e, IRTemp_INVALID); |
| 8567 | else |
| 8568 | putIRegA(rN, e, IRTemp_INVALID, Ijk_Boring); |
| 8569 | } else { |
sewardj | 21eaf24 | 2010-08-31 09:18:22 +0000 | [diff] [blame] | 8570 | IRExpr* e = binop(Iop_Add32, |
| 8571 | mkexpr(initialRn), |
| 8572 | mkexpr(initialRm)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8573 | if (isT) |
| 8574 | putIRegT(rN, e, IRTemp_INVALID); |
| 8575 | else |
| 8576 | putIRegA(rN, e, IRTemp_INVALID, Ijk_Boring); |
| 8577 | } |
| 8578 | } |
| 8579 | return True; |
| 8580 | } else { |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8581 | /* ------------ Case (3) ------------ |
| 8582 | VSTn / VLDn (multiple n-element structures) */ |
sewardj | 0adc2f2 | 2013-04-20 23:27:36 +0000 | [diff] [blame] | 8583 | inc = (fB & 1) + 1; |
| 8584 | |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 8585 | if (fB == BITS4(0,0,1,0) // Dd, Dd+1, Dd+2, Dd+3 inc = 1 regs = 4 |
| 8586 | || fB == BITS4(0,1,1,0) // Dd, Dd+1, Dd+2 inc = 1 regs = 3 |
| 8587 | || fB == BITS4(0,1,1,1) // Dd inc = 2 regs = 1 |
| 8588 | || fB == BITS4(1,0,1,0)) { // Dd, Dd+1 inc = 1 regs = 2 |
| 8589 | N = 0; // VLD1/VST1. 'inc' does not appear to have any |
| 8590 | // meaning for the VLD1/VST1 cases. 'regs' is the number of |
| 8591 | // registers involved. |
sewardj | 0adc2f2 | 2013-04-20 23:27:36 +0000 | [diff] [blame] | 8592 | if (rD + regs > 32) return False; |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 8593 | } |
| 8594 | else |
| 8595 | if (fB == BITS4(0,0,1,1) // Dd, Dd+1, Dd+2, Dd+3 inc=2 regs = 2 |
| 8596 | || fB == BITS4(1,0,0,0) // Dd, Dd+1 inc=1 regs = 1 |
| 8597 | || fB == BITS4(1,0,0,1)) { // Dd, Dd+2 inc=2 regs = 1 |
| 8598 | N = 1; // VLD2/VST2. 'regs' is the number of register-pairs involved |
sewardj | 0adc2f2 | 2013-04-20 23:27:36 +0000 | [diff] [blame] | 8599 | if (regs == 1 && inc == 1 && rD + 1 >= 32) return False; |
| 8600 | if (regs == 1 && inc == 2 && rD + 2 >= 32) return False; |
| 8601 | if (regs == 2 && inc == 2 && rD + 3 >= 32) return False; |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8602 | } else if (fB == BITS4(0,1,0,0) || fB == BITS4(0,1,0,1)) { |
sewardj | 0adc2f2 | 2013-04-20 23:27:36 +0000 | [diff] [blame] | 8603 | N = 2; // VLD3/VST3 |
| 8604 | if (inc == 1 && rD + 2 >= 32) return False; |
| 8605 | if (inc == 2 && rD + 4 >= 32) return False; |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8606 | } else if (fB == BITS4(0,0,0,0) || fB == BITS4(0,0,0,1)) { |
sewardj | 0adc2f2 | 2013-04-20 23:27:36 +0000 | [diff] [blame] | 8607 | N = 3; // VLD4/VST4 |
| 8608 | if (inc == 1 && rD + 3 >= 32) return False; |
| 8609 | if (inc == 2 && rD + 6 >= 32) return False; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8610 | } else { |
| 8611 | return False; |
| 8612 | } |
sewardj | 0adc2f2 | 2013-04-20 23:27:36 +0000 | [diff] [blame] | 8613 | |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8614 | if (N == 1 && fB == BITS4(0,0,1,1)) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8615 | regs = 2; |
| 8616 | } else if (N == 0) { |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8617 | if (fB == BITS4(1,0,1,0)) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8618 | regs = 2; |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8619 | } else if (fB == BITS4(0,1,1,0)) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8620 | regs = 3; |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8621 | } else if (fB == BITS4(0,0,1,0)) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8622 | regs = 4; |
| 8623 | } |
| 8624 | } |
| 8625 | |
| 8626 | size = INSN(7,6); |
| 8627 | if (N == 0 && size == 3) |
| 8628 | size = 2; |
| 8629 | if (size == 3) |
| 8630 | return False; |
| 8631 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8632 | // go uncond |
| 8633 | if (condT != IRTemp_INVALID) |
| 8634 | mk_skip_over_T32_if_cond_is_false(condT); |
| 8635 | // now uncond |
| 8636 | |
sewardj | 21eaf24 | 2010-08-31 09:18:22 +0000 | [diff] [blame] | 8637 | IRTemp addr = newTemp(Ity_I32); |
| 8638 | assign(addr, mkexpr(initialRn)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8639 | |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 8640 | if (N == 0 /* No interleaving -- VLD1/VST1 */) { |
sewardj | 0adc2f2 | 2013-04-20 23:27:36 +0000 | [diff] [blame] | 8641 | UInt r; |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 8642 | vassert(regs == 1 || regs == 2 || regs == 3 || regs == 4); |
| 8643 | /* inc has no relevance here */ |
| 8644 | for (r = 0; r < regs; r++) { |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8645 | if (bL) |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 8646 | putDRegI64(rD+r, loadLE(Ity_I64, mkexpr(addr)), IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8647 | else |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 8648 | storeLE(mkexpr(addr), getDRegI64(rD+r)); |
| 8649 | IRTemp tmp = newTemp(Ity_I32); |
| 8650 | assign(tmp, binop(Iop_Add32, mkexpr(addr), mkU32(8))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8651 | addr = tmp; |
| 8652 | } |
| 8653 | } |
sewardj | c1e23c4 | 2013-04-20 21:19:44 +0000 | [diff] [blame] | 8654 | else |
| 8655 | if (N == 1 /* 2-interleaving -- VLD2/VST2 */) { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 8656 | vassert( (regs == 1 && (inc == 1 || inc == 2)) |
| 8657 | || (regs == 2 && inc == 2) ); |
| 8658 | // Make 'nregs' be the number of registers and 'regstep' |
| 8659 | // equal the actual register-step. The ARM encoding, using 'regs' |
| 8660 | // and 'inc', is bizarre. After this, we have: |
| 8661 | // Dd, Dd+1 regs = 1, inc = 1, nregs = 2, regstep = 1 |
| 8662 | // Dd, Dd+2 regs = 1, inc = 2, nregs = 2, regstep = 2 |
| 8663 | // Dd, Dd+1, Dd+2, Dd+3 regs = 2, inc = 2, nregs = 4, regstep = 1 |
| 8664 | UInt nregs = 2; |
| 8665 | UInt regstep = 1; |
| 8666 | if (regs == 1 && inc == 1) { |
| 8667 | /* nothing */ |
| 8668 | } else if (regs == 1 && inc == 2) { |
| 8669 | regstep = 2; |
| 8670 | } else if (regs == 2 && inc == 2) { |
| 8671 | nregs = 4; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8672 | } else { |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 8673 | vassert(0); |
| 8674 | } |
| 8675 | // 'a' is address, |
| 8676 | // 'di' is interleaved data, 'du' is uninterleaved data |
| 8677 | if (nregs == 2) { |
| 8678 | IRExpr* a0 = binop(Iop_Add32, mkexpr(addr), mkU32(0)); |
| 8679 | IRExpr* a1 = binop(Iop_Add32, mkexpr(addr), mkU32(8)); |
| 8680 | IRTemp di0 = newTemp(Ity_I64); |
| 8681 | IRTemp di1 = newTemp(Ity_I64); |
| 8682 | IRTemp du0 = newTemp(Ity_I64); |
| 8683 | IRTemp du1 = newTemp(Ity_I64); |
| 8684 | if (bL) { |
| 8685 | assign(di0, loadLE(Ity_I64, a0)); |
| 8686 | assign(di1, loadLE(Ity_I64, a1)); |
| 8687 | math_DEINTERLEAVE_2(&du0, &du1, di0, di1, 1 << size); |
| 8688 | putDRegI64(rD + 0 * regstep, mkexpr(du0), IRTemp_INVALID); |
| 8689 | putDRegI64(rD + 1 * regstep, mkexpr(du1), IRTemp_INVALID); |
| 8690 | } else { |
| 8691 | assign(du0, getDRegI64(rD + 0 * regstep)); |
| 8692 | assign(du1, getDRegI64(rD + 1 * regstep)); |
| 8693 | math_INTERLEAVE_2(&di0, &di1, du0, du1, 1 << size); |
| 8694 | storeLE(a0, mkexpr(di0)); |
| 8695 | storeLE(a1, mkexpr(di1)); |
| 8696 | } |
| 8697 | IRTemp tmp = newTemp(Ity_I32); |
| 8698 | assign(tmp, binop(Iop_Add32, mkexpr(addr), mkU32(16))); |
| 8699 | addr = tmp; |
| 8700 | } else { |
| 8701 | vassert(nregs == 4); |
| 8702 | vassert(regstep == 1); |
| 8703 | IRExpr* a0 = binop(Iop_Add32, mkexpr(addr), mkU32(0)); |
| 8704 | IRExpr* a1 = binop(Iop_Add32, mkexpr(addr), mkU32(8)); |
| 8705 | IRExpr* a2 = binop(Iop_Add32, mkexpr(addr), mkU32(16)); |
| 8706 | IRExpr* a3 = binop(Iop_Add32, mkexpr(addr), mkU32(24)); |
| 8707 | IRTemp di0 = newTemp(Ity_I64); |
| 8708 | IRTemp di1 = newTemp(Ity_I64); |
| 8709 | IRTemp di2 = newTemp(Ity_I64); |
| 8710 | IRTemp di3 = newTemp(Ity_I64); |
| 8711 | IRTemp du0 = newTemp(Ity_I64); |
| 8712 | IRTemp du1 = newTemp(Ity_I64); |
| 8713 | IRTemp du2 = newTemp(Ity_I64); |
| 8714 | IRTemp du3 = newTemp(Ity_I64); |
| 8715 | if (bL) { |
| 8716 | assign(di0, loadLE(Ity_I64, a0)); |
| 8717 | assign(di1, loadLE(Ity_I64, a1)); |
| 8718 | assign(di2, loadLE(Ity_I64, a2)); |
| 8719 | assign(di3, loadLE(Ity_I64, a3)); |
sewardj | c1e23c4 | 2013-04-20 21:19:44 +0000 | [diff] [blame] | 8720 | // Note spooky interleaving: du0, du2, di0, di1 etc |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 8721 | math_DEINTERLEAVE_2(&du0, &du2, di0, di1, 1 << size); |
| 8722 | math_DEINTERLEAVE_2(&du1, &du3, di2, di3, 1 << size); |
| 8723 | putDRegI64(rD + 0 * regstep, mkexpr(du0), IRTemp_INVALID); |
| 8724 | putDRegI64(rD + 1 * regstep, mkexpr(du1), IRTemp_INVALID); |
| 8725 | putDRegI64(rD + 2 * regstep, mkexpr(du2), IRTemp_INVALID); |
| 8726 | putDRegI64(rD + 3 * regstep, mkexpr(du3), IRTemp_INVALID); |
| 8727 | } else { |
| 8728 | assign(du0, getDRegI64(rD + 0 * regstep)); |
| 8729 | assign(du1, getDRegI64(rD + 1 * regstep)); |
| 8730 | assign(du2, getDRegI64(rD + 2 * regstep)); |
| 8731 | assign(du3, getDRegI64(rD + 3 * regstep)); |
sewardj | c1e23c4 | 2013-04-20 21:19:44 +0000 | [diff] [blame] | 8732 | // Note spooky interleaving: du0, du2, di0, di1 etc |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 8733 | math_INTERLEAVE_2(&di0, &di1, du0, du2, 1 << size); |
| 8734 | math_INTERLEAVE_2(&di2, &di3, du1, du3, 1 << size); |
| 8735 | storeLE(a0, mkexpr(di0)); |
| 8736 | storeLE(a1, mkexpr(di1)); |
| 8737 | storeLE(a2, mkexpr(di2)); |
| 8738 | storeLE(a3, mkexpr(di3)); |
| 8739 | } |
| 8740 | |
| 8741 | IRTemp tmp = newTemp(Ity_I32); |
| 8742 | assign(tmp, binop(Iop_Add32, mkexpr(addr), mkU32(32))); |
| 8743 | addr = tmp; |
| 8744 | } |
sewardj | c1e23c4 | 2013-04-20 21:19:44 +0000 | [diff] [blame] | 8745 | } |
sewardj | 0adc2f2 | 2013-04-20 23:27:36 +0000 | [diff] [blame] | 8746 | else |
| 8747 | if (N == 2 /* 3-interleaving -- VLD3/VST3 */) { |
sewardj | c1e23c4 | 2013-04-20 21:19:44 +0000 | [diff] [blame] | 8748 | // Dd, Dd+1, Dd+2 regs = 1, inc = 1 |
| 8749 | // Dd, Dd+2, Dd+4 regs = 1, inc = 2 |
| 8750 | vassert(regs == 1 && (inc == 1 || inc == 2)); |
| 8751 | IRExpr* a0 = binop(Iop_Add32, mkexpr(addr), mkU32(0)); |
| 8752 | IRExpr* a1 = binop(Iop_Add32, mkexpr(addr), mkU32(8)); |
| 8753 | IRExpr* a2 = binop(Iop_Add32, mkexpr(addr), mkU32(16)); |
| 8754 | IRTemp di0 = newTemp(Ity_I64); |
| 8755 | IRTemp di1 = newTemp(Ity_I64); |
| 8756 | IRTemp di2 = newTemp(Ity_I64); |
| 8757 | IRTemp du0 = newTemp(Ity_I64); |
| 8758 | IRTemp du1 = newTemp(Ity_I64); |
| 8759 | IRTemp du2 = newTemp(Ity_I64); |
| 8760 | if (bL) { |
| 8761 | assign(di0, loadLE(Ity_I64, a0)); |
| 8762 | assign(di1, loadLE(Ity_I64, a1)); |
| 8763 | assign(di2, loadLE(Ity_I64, a2)); |
| 8764 | math_DEINTERLEAVE_3(&du0, &du1, &du2, di0, di1, di2, 1 << size); |
| 8765 | putDRegI64(rD + 0 * inc, mkexpr(du0), IRTemp_INVALID); |
| 8766 | putDRegI64(rD + 1 * inc, mkexpr(du1), IRTemp_INVALID); |
| 8767 | putDRegI64(rD + 2 * inc, mkexpr(du2), IRTemp_INVALID); |
| 8768 | } else { |
| 8769 | assign(du0, getDRegI64(rD + 0 * inc)); |
| 8770 | assign(du1, getDRegI64(rD + 1 * inc)); |
| 8771 | assign(du2, getDRegI64(rD + 2 * inc)); |
| 8772 | math_INTERLEAVE_3(&di0, &di1, &di2, du0, du1, du2, 1 << size); |
| 8773 | storeLE(a0, mkexpr(di0)); |
| 8774 | storeLE(a1, mkexpr(di1)); |
| 8775 | storeLE(a2, mkexpr(di2)); |
| 8776 | } |
| 8777 | IRTemp tmp = newTemp(Ity_I32); |
| 8778 | assign(tmp, binop(Iop_Add32, mkexpr(addr), mkU32(24))); |
| 8779 | addr = tmp; |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 8780 | } |
sewardj | 0adc2f2 | 2013-04-20 23:27:36 +0000 | [diff] [blame] | 8781 | else |
| 8782 | if (N == 3 /* 4-interleaving -- VLD4/VST4 */) { |
| 8783 | // Dd, Dd+1, Dd+2, Dd+3 regs = 1, inc = 1 |
| 8784 | // Dd, Dd+2, Dd+4, Dd+6 regs = 1, inc = 2 |
| 8785 | vassert(regs == 1 && (inc == 1 || inc == 2)); |
| 8786 | IRExpr* a0 = binop(Iop_Add32, mkexpr(addr), mkU32(0)); |
| 8787 | IRExpr* a1 = binop(Iop_Add32, mkexpr(addr), mkU32(8)); |
| 8788 | IRExpr* a2 = binop(Iop_Add32, mkexpr(addr), mkU32(16)); |
| 8789 | IRExpr* a3 = binop(Iop_Add32, mkexpr(addr), mkU32(24)); |
| 8790 | IRTemp di0 = newTemp(Ity_I64); |
| 8791 | IRTemp di1 = newTemp(Ity_I64); |
| 8792 | IRTemp di2 = newTemp(Ity_I64); |
| 8793 | IRTemp di3 = newTemp(Ity_I64); |
| 8794 | IRTemp du0 = newTemp(Ity_I64); |
| 8795 | IRTemp du1 = newTemp(Ity_I64); |
| 8796 | IRTemp du2 = newTemp(Ity_I64); |
| 8797 | IRTemp du3 = newTemp(Ity_I64); |
| 8798 | if (bL) { |
| 8799 | assign(di0, loadLE(Ity_I64, a0)); |
| 8800 | assign(di1, loadLE(Ity_I64, a1)); |
| 8801 | assign(di2, loadLE(Ity_I64, a2)); |
| 8802 | assign(di3, loadLE(Ity_I64, a3)); |
| 8803 | math_DEINTERLEAVE_4(&du0, &du1, &du2, &du3, |
| 8804 | di0, di1, di2, di3, 1 << size); |
| 8805 | putDRegI64(rD + 0 * inc, mkexpr(du0), IRTemp_INVALID); |
| 8806 | putDRegI64(rD + 1 * inc, mkexpr(du1), IRTemp_INVALID); |
| 8807 | putDRegI64(rD + 2 * inc, mkexpr(du2), IRTemp_INVALID); |
| 8808 | putDRegI64(rD + 3 * inc, mkexpr(du3), IRTemp_INVALID); |
| 8809 | } else { |
| 8810 | assign(du0, getDRegI64(rD + 0 * inc)); |
| 8811 | assign(du1, getDRegI64(rD + 1 * inc)); |
| 8812 | assign(du2, getDRegI64(rD + 2 * inc)); |
| 8813 | assign(du3, getDRegI64(rD + 3 * inc)); |
| 8814 | math_INTERLEAVE_4(&di0, &di1, &di2, &di3, |
| 8815 | du0, du1, du2, du3, 1 << size); |
| 8816 | storeLE(a0, mkexpr(di0)); |
| 8817 | storeLE(a1, mkexpr(di1)); |
| 8818 | storeLE(a2, mkexpr(di2)); |
| 8819 | storeLE(a3, mkexpr(di3)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8820 | } |
sewardj | 0adc2f2 | 2013-04-20 23:27:36 +0000 | [diff] [blame] | 8821 | IRTemp tmp = newTemp(Ity_I32); |
| 8822 | assign(tmp, binop(Iop_Add32, mkexpr(addr), mkU32(32))); |
| 8823 | addr = tmp; |
| 8824 | } |
| 8825 | else { |
| 8826 | vassert(0); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8827 | } |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 8828 | |
| 8829 | /* Writeback */ |
| 8830 | if (rM != 15) { |
| 8831 | IRExpr* e; |
| 8832 | if (rM == 13) { |
| 8833 | e = binop(Iop_Add32, mkexpr(initialRn), |
| 8834 | mkU32(8 * (N + 1) * regs)); |
| 8835 | } else { |
| 8836 | e = binop(Iop_Add32, mkexpr(initialRn), |
| 8837 | mkexpr(initialRm)); |
| 8838 | } |
| 8839 | if (isT) |
| 8840 | putIRegT(rN, e, IRTemp_INVALID); |
| 8841 | else |
| 8842 | putIRegA(rN, e, IRTemp_INVALID, Ijk_Boring); |
| 8843 | } |
| 8844 | |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 8845 | DIP("v%s%u.%d {", bL ? "ld" : "st", N + 1, 8 << INSN(7,6)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8846 | if ((inc == 1 && regs * (N + 1) > 1) |
| 8847 | || (inc == 2 && regs > 1 && N > 0)) { |
| 8848 | DIP("d%u-d%u", rD, rD + regs * (N + 1) - 1); |
| 8849 | } else { |
sewardj | 0adc2f2 | 2013-04-20 23:27:36 +0000 | [diff] [blame] | 8850 | UInt r; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8851 | for (r = 0; r < regs; r++) { |
| 8852 | for (i = 0; i <= N; i++) { |
| 8853 | if (i || r) |
| 8854 | DIP(", "); |
| 8855 | DIP("d%u", rD + r + i * inc); |
| 8856 | } |
| 8857 | } |
| 8858 | } |
sewardj | 17bf0aa | 2010-08-31 09:29:51 +0000 | [diff] [blame] | 8859 | DIP("}, [r%u]", rN); |
| 8860 | if (rM != 13 && rM != 15) { |
| 8861 | DIP(", r%u\n", rM); |
| 8862 | } else { |
| 8863 | DIP("%s\n", (rM != 15) ? "!" : ""); |
| 8864 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8865 | return True; |
| 8866 | } |
| 8867 | # undef INSN |
| 8868 | } |
| 8869 | |
| 8870 | |
| 8871 | /*------------------------------------------------------------*/ |
| 8872 | /*--- NEON, top level control ---*/ |
| 8873 | /*------------------------------------------------------------*/ |
| 8874 | |
| 8875 | /* Both ARM and Thumb */ |
| 8876 | |
| 8877 | /* Translate a NEON instruction. If successful, returns |
| 8878 | True and *dres may or may not be updated. If failure, returns |
| 8879 | False and doesn't change *dres nor create any IR. |
| 8880 | |
| 8881 | The Thumb and ARM encodings are similar for the 24 bottom bits, but |
| 8882 | the top 8 bits are slightly different. In both cases, the caller |
| 8883 | must pass the entire 32 bits. Callers may pass any instruction; |
| 8884 | this ignores non-NEON ones. |
| 8885 | |
| 8886 | Caller must supply an IRTemp 'condT' holding the gating condition, |
| 8887 | or IRTemp_INVALID indicating the insn is always executed. In ARM |
| 8888 | code, this must always be IRTemp_INVALID because NEON insns are |
| 8889 | unconditional for ARM. |
| 8890 | |
| 8891 | Finally, the caller must indicate whether this occurs in ARM or in |
| 8892 | Thumb code. |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 8893 | |
| 8894 | This only handles NEON for ARMv7 and below. The NEON extensions |
| 8895 | for v8 are handled by decode_V8_instruction. |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8896 | */ |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 8897 | static Bool decode_NEON_instruction_ARMv7_and_below ( |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8898 | /*MOD*/DisResult* dres, |
| 8899 | UInt insn32, |
| 8900 | IRTemp condT, |
| 8901 | Bool isT |
| 8902 | ) |
| 8903 | { |
| 8904 | # define INSN(_bMax,_bMin) SLICE_UInt(insn32, (_bMax), (_bMin)) |
| 8905 | |
| 8906 | /* There are two kinds of instruction to deal with: load/store and |
| 8907 | data processing. In each case, in ARM mode we merely identify |
| 8908 | the kind, and pass it on to the relevant sub-handler. In Thumb |
| 8909 | mode we identify the kind, swizzle the bits around to make it |
| 8910 | have the same encoding as in ARM, and hand it on to the |
| 8911 | sub-handler. |
| 8912 | */ |
| 8913 | |
| 8914 | /* In ARM mode, NEON instructions can't be conditional. */ |
| 8915 | if (!isT) |
| 8916 | vassert(condT == IRTemp_INVALID); |
| 8917 | |
| 8918 | /* Data processing: |
| 8919 | Thumb: 111U 1111 AAAA Axxx xxxx BBBB CCCC xxxx |
| 8920 | ARM: 1111 001U AAAA Axxx xxxx BBBB CCCC xxxx |
| 8921 | */ |
| 8922 | if (!isT && INSN(31,25) == BITS7(1,1,1,1,0,0,1)) { |
| 8923 | // ARM, DP |
| 8924 | return dis_neon_data_processing(INSN(31,0), condT); |
| 8925 | } |
| 8926 | if (isT && INSN(31,29) == BITS3(1,1,1) |
| 8927 | && INSN(27,24) == BITS4(1,1,1,1)) { |
| 8928 | // Thumb, DP |
| 8929 | UInt reformatted = INSN(23,0); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 8930 | reformatted |= (((UInt)INSN(28,28)) << 24); // U bit |
| 8931 | reformatted |= (((UInt)BITS7(1,1,1,1,0,0,1)) << 25); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8932 | return dis_neon_data_processing(reformatted, condT); |
| 8933 | } |
| 8934 | |
| 8935 | /* Load/store: |
| 8936 | Thumb: 1111 1001 AxL0 xxxx xxxx BBBB xxxx xxxx |
| 8937 | ARM: 1111 0100 AxL0 xxxx xxxx BBBB xxxx xxxx |
| 8938 | */ |
| 8939 | if (!isT && INSN(31,24) == BITS8(1,1,1,1,0,1,0,0)) { |
| 8940 | // ARM, memory |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8941 | return dis_neon_load_or_store(INSN(31,0), isT, condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8942 | } |
| 8943 | if (isT && INSN(31,24) == BITS8(1,1,1,1,1,0,0,1)) { |
| 8944 | UInt reformatted = INSN(23,0); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 8945 | reformatted |= (((UInt)BITS8(1,1,1,1,0,1,0,0)) << 24); |
sewardj | 2895832 | 2011-07-21 22:45:42 +0000 | [diff] [blame] | 8946 | return dis_neon_load_or_store(reformatted, isT, condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 8947 | } |
| 8948 | |
| 8949 | /* Doesn't match. */ |
| 8950 | return False; |
| 8951 | |
| 8952 | # undef INSN |
| 8953 | } |
| 8954 | |
| 8955 | |
| 8956 | /*------------------------------------------------------------*/ |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 8957 | /*--- V6 MEDIA instructions ---*/ |
| 8958 | /*------------------------------------------------------------*/ |
| 8959 | |
| 8960 | /* Both ARM and Thumb */ |
| 8961 | |
| 8962 | /* Translate a V6 media instruction. If successful, returns |
| 8963 | True and *dres may or may not be updated. If failure, returns |
| 8964 | False and doesn't change *dres nor create any IR. |
| 8965 | |
| 8966 | The Thumb and ARM encodings are completely different. In Thumb |
| 8967 | mode, the caller must pass the entire 32 bits. In ARM mode it must |
| 8968 | pass the lower 28 bits. Apart from that, callers may pass any |
| 8969 | instruction; this function ignores anything it doesn't recognise. |
| 8970 | |
| 8971 | Caller must supply an IRTemp 'condT' holding the gating condition, |
| 8972 | or IRTemp_INVALID indicating the insn is always executed. |
| 8973 | |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 8974 | Caller must also supply an ARMCondcode 'conq'. This is only used |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 8975 | for debug printing, no other purpose. For ARM, this is simply the |
| 8976 | top 4 bits of the original instruction. For Thumb, the condition |
| 8977 | is not (really) known until run time, and so ARMCondAL should be |
| 8978 | passed, only so that printing of these instructions does not show |
| 8979 | any condition. |
| 8980 | |
| 8981 | Finally, the caller must indicate whether this occurs in ARM or in |
| 8982 | Thumb code. |
| 8983 | */ |
| 8984 | static Bool decode_V6MEDIA_instruction ( |
| 8985 | /*MOD*/DisResult* dres, |
| 8986 | UInt insnv6m, |
| 8987 | IRTemp condT, |
| 8988 | ARMCondcode conq, |
| 8989 | Bool isT |
| 8990 | ) |
| 8991 | { |
| 8992 | # define INSNA(_bMax,_bMin) SLICE_UInt(insnv6m, (_bMax), (_bMin)) |
| 8993 | # define INSNT0(_bMax,_bMin) SLICE_UInt( ((insnv6m >> 16) & 0xFFFF), \ |
| 8994 | (_bMax), (_bMin) ) |
| 8995 | # define INSNT1(_bMax,_bMin) SLICE_UInt( ((insnv6m >> 0) & 0xFFFF), \ |
| 8996 | (_bMax), (_bMin) ) |
| 8997 | HChar dis_buf[128]; |
| 8998 | dis_buf[0] = 0; |
| 8999 | |
| 9000 | if (isT) { |
| 9001 | vassert(conq == ARMCondAL); |
| 9002 | } else { |
| 9003 | vassert(INSNA(31,28) == BITS4(0,0,0,0)); // caller's obligation |
| 9004 | vassert(conq >= ARMCondEQ && conq <= ARMCondAL); |
| 9005 | } |
| 9006 | |
| 9007 | /* ----------- smulbb, smulbt, smultb, smultt ----------- */ |
| 9008 | { |
| 9009 | UInt regD = 99, regM = 99, regN = 99, bitM = 0, bitN = 0; |
| 9010 | Bool gate = False; |
| 9011 | |
| 9012 | if (isT) { |
| 9013 | if (INSNT0(15,4) == 0xFB1 && INSNT1(15,12) == BITS4(1,1,1,1) |
| 9014 | && INSNT1(7,6) == BITS2(0,0)) { |
| 9015 | regD = INSNT1(11,8); |
| 9016 | regM = INSNT1(3,0); |
| 9017 | regN = INSNT0(3,0); |
| 9018 | bitM = INSNT1(4,4); |
| 9019 | bitN = INSNT1(5,5); |
| 9020 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9021 | gate = True; |
| 9022 | } |
| 9023 | } else { |
| 9024 | if (BITS8(0,0,0,1,0,1,1,0) == INSNA(27,20) && |
| 9025 | BITS4(0,0,0,0) == INSNA(15,12) && |
| 9026 | BITS4(1,0,0,0) == (INSNA(7,4) & BITS4(1,0,0,1)) ) { |
| 9027 | regD = INSNA(19,16); |
| 9028 | regM = INSNA(11,8); |
| 9029 | regN = INSNA(3,0); |
| 9030 | bitM = INSNA(6,6); |
| 9031 | bitN = INSNA(5,5); |
| 9032 | if (regD != 15 && regN != 15 && regM != 15) |
| 9033 | gate = True; |
| 9034 | } |
| 9035 | } |
| 9036 | |
| 9037 | if (gate) { |
| 9038 | IRTemp srcN = newTemp(Ity_I32); |
| 9039 | IRTemp srcM = newTemp(Ity_I32); |
| 9040 | IRTemp res = newTemp(Ity_I32); |
| 9041 | |
| 9042 | assign( srcN, binop(Iop_Sar32, |
| 9043 | binop(Iop_Shl32, |
| 9044 | isT ? getIRegT(regN) : getIRegA(regN), |
| 9045 | mkU8(bitN ? 0 : 16)), mkU8(16)) ); |
| 9046 | assign( srcM, binop(Iop_Sar32, |
| 9047 | binop(Iop_Shl32, |
| 9048 | isT ? getIRegT(regM) : getIRegA(regM), |
| 9049 | mkU8(bitM ? 0 : 16)), mkU8(16)) ); |
| 9050 | assign( res, binop(Iop_Mul32, mkexpr(srcN), mkexpr(srcM)) ); |
| 9051 | |
| 9052 | if (isT) |
| 9053 | putIRegT( regD, mkexpr(res), condT ); |
| 9054 | else |
| 9055 | putIRegA( regD, mkexpr(res), condT, Ijk_Boring ); |
| 9056 | |
| 9057 | DIP( "smul%c%c%s r%u, r%u, r%u\n", bitN ? 't' : 'b', bitM ? 't' : 'b', |
| 9058 | nCC(conq), regD, regN, regM ); |
| 9059 | return True; |
| 9060 | } |
| 9061 | /* fall through */ |
| 9062 | } |
| 9063 | |
| 9064 | /* ------------ smulwb<y><c> <Rd>,<Rn>,<Rm> ------------- */ |
| 9065 | /* ------------ smulwt<y><c> <Rd>,<Rn>,<Rm> ------------- */ |
| 9066 | { |
| 9067 | UInt regD = 99, regN = 99, regM = 99, bitM = 0; |
| 9068 | Bool gate = False; |
| 9069 | |
| 9070 | if (isT) { |
| 9071 | if (INSNT0(15,4) == 0xFB3 && INSNT1(15,12) == BITS4(1,1,1,1) |
| 9072 | && INSNT1(7,5) == BITS3(0,0,0)) { |
| 9073 | regN = INSNT0(3,0); |
| 9074 | regD = INSNT1(11,8); |
| 9075 | regM = INSNT1(3,0); |
| 9076 | bitM = INSNT1(4,4); |
| 9077 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9078 | gate = True; |
| 9079 | } |
| 9080 | } else { |
| 9081 | if (INSNA(27,20) == BITS8(0,0,0,1,0,0,1,0) && |
| 9082 | INSNA(15,12) == BITS4(0,0,0,0) && |
| 9083 | (INSNA(7,4) & BITS4(1,0,1,1)) == BITS4(1,0,1,0)) { |
| 9084 | regD = INSNA(19,16); |
| 9085 | regN = INSNA(3,0); |
| 9086 | regM = INSNA(11,8); |
| 9087 | bitM = INSNA(6,6); |
| 9088 | if (regD != 15 && regN != 15 && regM != 15) |
| 9089 | gate = True; |
| 9090 | } |
| 9091 | } |
| 9092 | |
| 9093 | if (gate) { |
| 9094 | IRTemp irt_prod = newTemp(Ity_I64); |
| 9095 | |
| 9096 | assign( irt_prod, |
| 9097 | binop(Iop_MullS32, |
| 9098 | isT ? getIRegT(regN) : getIRegA(regN), |
| 9099 | binop(Iop_Sar32, |
| 9100 | binop(Iop_Shl32, |
| 9101 | isT ? getIRegT(regM) : getIRegA(regM), |
| 9102 | mkU8(bitM ? 0 : 16)), |
| 9103 | mkU8(16))) ); |
| 9104 | |
| 9105 | IRExpr* ire_result = binop(Iop_Or32, |
| 9106 | binop( Iop_Shl32, |
| 9107 | unop(Iop_64HIto32, mkexpr(irt_prod)), |
| 9108 | mkU8(16) ), |
| 9109 | binop( Iop_Shr32, |
| 9110 | unop(Iop_64to32, mkexpr(irt_prod)), |
| 9111 | mkU8(16) ) ); |
| 9112 | |
| 9113 | if (isT) |
| 9114 | putIRegT( regD, ire_result, condT ); |
| 9115 | else |
| 9116 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 9117 | |
| 9118 | DIP("smulw%c%s r%u, r%u, r%u\n", |
| 9119 | bitM ? 't' : 'b', nCC(conq),regD,regN,regM); |
| 9120 | return True; |
| 9121 | } |
| 9122 | /* fall through */ |
| 9123 | } |
| 9124 | |
| 9125 | /* ------------ pkhbt<c> Rd, Rn, Rm {,LSL #imm} ------------- */ |
| 9126 | /* ------------ pkhtb<c> Rd, Rn, Rm {,ASR #imm} ------------- */ |
| 9127 | { |
| 9128 | UInt regD = 99, regN = 99, regM = 99, imm5 = 99, shift_type = 99; |
| 9129 | Bool tbform = False; |
| 9130 | Bool gate = False; |
| 9131 | |
| 9132 | if (isT) { |
| 9133 | if (INSNT0(15,4) == 0xEAC |
| 9134 | && INSNT1(15,15) == 0 && INSNT1(4,4) == 0) { |
| 9135 | regN = INSNT0(3,0); |
| 9136 | regD = INSNT1(11,8); |
| 9137 | regM = INSNT1(3,0); |
| 9138 | imm5 = (INSNT1(14,12) << 2) | INSNT1(7,6); |
| 9139 | shift_type = (INSNT1(5,5) << 1) | 0; |
| 9140 | tbform = (INSNT1(5,5) == 0) ? False : True; |
| 9141 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9142 | gate = True; |
| 9143 | } |
| 9144 | } else { |
| 9145 | if (INSNA(27,20) == BITS8(0,1,1,0,1,0,0,0) && |
| 9146 | INSNA(5,4) == BITS2(0,1) && |
| 9147 | (INSNA(6,6) == 0 || INSNA(6,6) == 1) ) { |
| 9148 | regD = INSNA(15,12); |
| 9149 | regN = INSNA(19,16); |
| 9150 | regM = INSNA(3,0); |
| 9151 | imm5 = INSNA(11,7); |
| 9152 | shift_type = (INSNA(6,6) << 1) | 0; |
| 9153 | tbform = (INSNA(6,6) == 0) ? False : True; |
| 9154 | if (regD != 15 && regN != 15 && regM != 15) |
| 9155 | gate = True; |
| 9156 | } |
| 9157 | } |
| 9158 | |
| 9159 | if (gate) { |
| 9160 | IRTemp irt_regM = newTemp(Ity_I32); |
| 9161 | IRTemp irt_regM_shift = newTemp(Ity_I32); |
| 9162 | assign( irt_regM, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 9163 | compute_result_and_C_after_shift_by_imm5( |
| 9164 | dis_buf, &irt_regM_shift, NULL, irt_regM, shift_type, imm5, regM ); |
| 9165 | |
| 9166 | UInt mask = (tbform == True) ? 0x0000FFFF : 0xFFFF0000; |
| 9167 | IRExpr* ire_result |
| 9168 | = binop( Iop_Or32, |
| 9169 | binop(Iop_And32, mkexpr(irt_regM_shift), mkU32(mask)), |
| 9170 | binop(Iop_And32, isT ? getIRegT(regN) : getIRegA(regN), |
| 9171 | unop(Iop_Not32, mkU32(mask))) ); |
| 9172 | |
| 9173 | if (isT) |
| 9174 | putIRegT( regD, ire_result, condT ); |
| 9175 | else |
| 9176 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 9177 | |
| 9178 | DIP( "pkh%s%s r%u, r%u, r%u %s\n", tbform ? "tb" : "bt", |
| 9179 | nCC(conq), regD, regN, regM, dis_buf ); |
| 9180 | |
| 9181 | return True; |
| 9182 | } |
| 9183 | /* fall through */ |
| 9184 | } |
| 9185 | |
| 9186 | /* ---------- usat<c> <Rd>,#<imm5>,<Rn>{,<shift>} ----------- */ |
| 9187 | { |
| 9188 | UInt regD = 99, regN = 99, shift_type = 99, imm5 = 99, sat_imm = 99; |
| 9189 | Bool gate = False; |
| 9190 | |
| 9191 | if (isT) { |
| 9192 | if (INSNT0(15,6) == BITS10(1,1,1,1,0,0,1,1,1,0) |
| 9193 | && INSNT0(4,4) == 0 |
| 9194 | && INSNT1(15,15) == 0 && INSNT1(5,5) == 0) { |
| 9195 | regD = INSNT1(11,8); |
| 9196 | regN = INSNT0(3,0); |
| 9197 | shift_type = (INSNT0(5,5) << 1) | 0; |
| 9198 | imm5 = (INSNT1(14,12) << 2) | INSNT1(7,6); |
| 9199 | sat_imm = INSNT1(4,0); |
| 9200 | if (!isBadRegT(regD) && !isBadRegT(regN)) |
| 9201 | gate = True; |
| 9202 | if (shift_type == BITS2(1,0) && imm5 == 0) |
| 9203 | gate = False; |
| 9204 | } |
| 9205 | } else { |
| 9206 | if (INSNA(27,21) == BITS7(0,1,1,0,1,1,1) && |
| 9207 | INSNA(5,4) == BITS2(0,1)) { |
| 9208 | regD = INSNA(15,12); |
| 9209 | regN = INSNA(3,0); |
| 9210 | shift_type = (INSNA(6,6) << 1) | 0; |
| 9211 | imm5 = INSNA(11,7); |
| 9212 | sat_imm = INSNA(20,16); |
| 9213 | if (regD != 15 && regN != 15) |
| 9214 | gate = True; |
| 9215 | } |
| 9216 | } |
| 9217 | |
| 9218 | if (gate) { |
| 9219 | IRTemp irt_regN = newTemp(Ity_I32); |
| 9220 | IRTemp irt_regN_shift = newTemp(Ity_I32); |
| 9221 | IRTemp irt_sat_Q = newTemp(Ity_I32); |
| 9222 | IRTemp irt_result = newTemp(Ity_I32); |
| 9223 | |
| 9224 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9225 | compute_result_and_C_after_shift_by_imm5( |
| 9226 | dis_buf, &irt_regN_shift, NULL, |
| 9227 | irt_regN, shift_type, imm5, regN ); |
| 9228 | |
| 9229 | armUnsignedSatQ( &irt_result, &irt_sat_Q, irt_regN_shift, sat_imm ); |
| 9230 | or_into_QFLAG32( mkexpr(irt_sat_Q), condT ); |
| 9231 | |
| 9232 | if (isT) |
| 9233 | putIRegT( regD, mkexpr(irt_result), condT ); |
| 9234 | else |
| 9235 | putIRegA( regD, mkexpr(irt_result), condT, Ijk_Boring ); |
| 9236 | |
| 9237 | DIP("usat%s r%u, #0x%04x, %s\n", |
| 9238 | nCC(conq), regD, imm5, dis_buf); |
| 9239 | return True; |
| 9240 | } |
| 9241 | /* fall through */ |
| 9242 | } |
| 9243 | |
| 9244 | /* ----------- ssat<c> <Rd>,#<imm5>,<Rn>{,<shift>} ----------- */ |
| 9245 | { |
| 9246 | UInt regD = 99, regN = 99, shift_type = 99, imm5 = 99, sat_imm = 99; |
| 9247 | Bool gate = False; |
| 9248 | |
| 9249 | if (isT) { |
| 9250 | if (INSNT0(15,6) == BITS10(1,1,1,1,0,0,1,1,0,0) |
| 9251 | && INSNT0(4,4) == 0 |
| 9252 | && INSNT1(15,15) == 0 && INSNT1(5,5) == 0) { |
| 9253 | regD = INSNT1(11,8); |
| 9254 | regN = INSNT0(3,0); |
| 9255 | shift_type = (INSNT0(5,5) << 1) | 0; |
| 9256 | imm5 = (INSNT1(14,12) << 2) | INSNT1(7,6); |
| 9257 | sat_imm = INSNT1(4,0) + 1; |
| 9258 | if (!isBadRegT(regD) && !isBadRegT(regN)) |
| 9259 | gate = True; |
| 9260 | if (shift_type == BITS2(1,0) && imm5 == 0) |
| 9261 | gate = False; |
| 9262 | } |
| 9263 | } else { |
| 9264 | if (INSNA(27,21) == BITS7(0,1,1,0,1,0,1) && |
| 9265 | INSNA(5,4) == BITS2(0,1)) { |
| 9266 | regD = INSNA(15,12); |
| 9267 | regN = INSNA(3,0); |
| 9268 | shift_type = (INSNA(6,6) << 1) | 0; |
| 9269 | imm5 = INSNA(11,7); |
| 9270 | sat_imm = INSNA(20,16) + 1; |
| 9271 | if (regD != 15 && regN != 15) |
| 9272 | gate = True; |
| 9273 | } |
| 9274 | } |
| 9275 | |
| 9276 | if (gate) { |
| 9277 | IRTemp irt_regN = newTemp(Ity_I32); |
| 9278 | IRTemp irt_regN_shift = newTemp(Ity_I32); |
| 9279 | IRTemp irt_sat_Q = newTemp(Ity_I32); |
| 9280 | IRTemp irt_result = newTemp(Ity_I32); |
| 9281 | |
| 9282 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9283 | compute_result_and_C_after_shift_by_imm5( |
| 9284 | dis_buf, &irt_regN_shift, NULL, |
| 9285 | irt_regN, shift_type, imm5, regN ); |
| 9286 | |
| 9287 | armSignedSatQ( irt_regN_shift, sat_imm, &irt_result, &irt_sat_Q ); |
| 9288 | or_into_QFLAG32( mkexpr(irt_sat_Q), condT ); |
| 9289 | |
| 9290 | if (isT) |
| 9291 | putIRegT( regD, mkexpr(irt_result), condT ); |
| 9292 | else |
| 9293 | putIRegA( regD, mkexpr(irt_result), condT, Ijk_Boring ); |
| 9294 | |
| 9295 | DIP( "ssat%s r%u, #0x%04x, %s\n", |
| 9296 | nCC(conq), regD, imm5, dis_buf); |
| 9297 | return True; |
| 9298 | } |
| 9299 | /* fall through */ |
| 9300 | } |
| 9301 | |
sewardj | e660158 | 2013-05-13 10:02:33 +0000 | [diff] [blame] | 9302 | /* ----------- ssat16<c> <Rd>,#<imm>,<Rn> ----------- */ |
| 9303 | { |
| 9304 | UInt regD = 99, regN = 99, sat_imm = 99; |
| 9305 | Bool gate = False; |
| 9306 | |
| 9307 | if (isT) { |
| 9308 | if (INSNT0(15,6) == BITS10(1,1,1,1,0,0,1,1,0,0) |
| 9309 | && INSNT0(5,4) == BITS2(1,0) |
| 9310 | && INSNT1(15,12) == BITS4(0,0,0,0) |
| 9311 | && INSNT1(7,4) == BITS4(0,0,0,0)) { |
| 9312 | regD = INSNT1(11,8); |
| 9313 | regN = INSNT0(3,0); |
| 9314 | sat_imm = INSNT1(3,0) + 1; |
| 9315 | if (!isBadRegT(regD) && !isBadRegT(regN)) |
| 9316 | gate = True; |
| 9317 | } |
| 9318 | } else { |
| 9319 | if (INSNA(27,20) == BITS8(0,1,1,0,1,0,1,0) && |
| 9320 | INSNA(11,4) == BITS8(1,1,1,1,0,0,1,1)) { |
| 9321 | regD = INSNA(15,12); |
| 9322 | regN = INSNA(3,0); |
| 9323 | sat_imm = INSNA(19,16) + 1; |
| 9324 | if (regD != 15 && regN != 15) |
| 9325 | gate = True; |
| 9326 | } |
| 9327 | } |
| 9328 | |
| 9329 | if (gate) { |
| 9330 | IRTemp irt_regN = newTemp(Ity_I32); |
| 9331 | IRTemp irt_regN_lo = newTemp(Ity_I32); |
| 9332 | IRTemp irt_regN_hi = newTemp(Ity_I32); |
| 9333 | IRTemp irt_Q_lo = newTemp(Ity_I32); |
| 9334 | IRTemp irt_Q_hi = newTemp(Ity_I32); |
| 9335 | IRTemp irt_res_lo = newTemp(Ity_I32); |
| 9336 | IRTemp irt_res_hi = newTemp(Ity_I32); |
| 9337 | |
| 9338 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9339 | assign( irt_regN_lo, |
| 9340 | binop( Iop_Sar32, |
| 9341 | binop(Iop_Shl32, mkexpr(irt_regN), mkU8(16)), |
| 9342 | mkU8(16)) ); |
| 9343 | assign( irt_regN_hi, binop(Iop_Sar32, mkexpr(irt_regN), mkU8(16)) ); |
| 9344 | |
| 9345 | armSignedSatQ( irt_regN_lo, sat_imm, &irt_res_lo, &irt_Q_lo ); |
| 9346 | or_into_QFLAG32( mkexpr(irt_Q_lo), condT ); |
| 9347 | |
| 9348 | armSignedSatQ( irt_regN_hi, sat_imm, &irt_res_hi, &irt_Q_hi ); |
| 9349 | or_into_QFLAG32( mkexpr(irt_Q_hi), condT ); |
| 9350 | |
| 9351 | IRExpr* ire_result |
| 9352 | = binop(Iop_Or32, |
| 9353 | binop(Iop_And32, mkexpr(irt_res_lo), mkU32(0xFFFF)), |
| 9354 | binop(Iop_Shl32, mkexpr(irt_res_hi), mkU8(16))); |
| 9355 | if (isT) |
| 9356 | putIRegT( regD, ire_result, condT ); |
| 9357 | else |
| 9358 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 9359 | |
| 9360 | DIP( "ssat16%s r%u, #0x%04x, r%u\n", nCC(conq), regD, sat_imm, regN ); |
| 9361 | return True; |
| 9362 | } |
| 9363 | /* fall through */ |
| 9364 | } |
| 9365 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9366 | /* -------------- usat16<c> <Rd>,#<imm4>,<Rn> --------------- */ |
| 9367 | { |
| 9368 | UInt regD = 99, regN = 99, sat_imm = 99; |
| 9369 | Bool gate = False; |
| 9370 | |
| 9371 | if (isT) { |
| 9372 | if (INSNT0(15,4) == 0xF3A && (INSNT1(15,0) & 0xF0F0) == 0x0000) { |
| 9373 | regN = INSNT0(3,0); |
| 9374 | regD = INSNT1(11,8); |
| 9375 | sat_imm = INSNT1(3,0); |
| 9376 | if (!isBadRegT(regD) && !isBadRegT(regN)) |
| 9377 | gate = True; |
| 9378 | } |
| 9379 | } else { |
| 9380 | if (INSNA(27,20) == BITS8(0,1,1,0,1,1,1,0) && |
| 9381 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 9382 | INSNA(7,4) == BITS4(0,0,1,1)) { |
| 9383 | regD = INSNA(15,12); |
| 9384 | regN = INSNA(3,0); |
| 9385 | sat_imm = INSNA(19,16); |
| 9386 | if (regD != 15 && regN != 15) |
| 9387 | gate = True; |
| 9388 | } |
| 9389 | } |
| 9390 | |
| 9391 | if (gate) { |
| 9392 | IRTemp irt_regN = newTemp(Ity_I32); |
| 9393 | IRTemp irt_regN_lo = newTemp(Ity_I32); |
| 9394 | IRTemp irt_regN_hi = newTemp(Ity_I32); |
| 9395 | IRTemp irt_Q_lo = newTemp(Ity_I32); |
| 9396 | IRTemp irt_Q_hi = newTemp(Ity_I32); |
| 9397 | IRTemp irt_res_lo = newTemp(Ity_I32); |
| 9398 | IRTemp irt_res_hi = newTemp(Ity_I32); |
| 9399 | |
| 9400 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9401 | assign( irt_regN_lo, binop( Iop_Sar32, |
| 9402 | binop(Iop_Shl32, mkexpr(irt_regN), mkU8(16)), |
| 9403 | mkU8(16)) ); |
| 9404 | assign( irt_regN_hi, binop(Iop_Sar32, mkexpr(irt_regN), mkU8(16)) ); |
| 9405 | |
| 9406 | armUnsignedSatQ( &irt_res_lo, &irt_Q_lo, irt_regN_lo, sat_imm ); |
| 9407 | or_into_QFLAG32( mkexpr(irt_Q_lo), condT ); |
| 9408 | |
| 9409 | armUnsignedSatQ( &irt_res_hi, &irt_Q_hi, irt_regN_hi, sat_imm ); |
| 9410 | or_into_QFLAG32( mkexpr(irt_Q_hi), condT ); |
| 9411 | |
| 9412 | IRExpr* ire_result = binop( Iop_Or32, |
| 9413 | binop(Iop_Shl32, mkexpr(irt_res_hi), mkU8(16)), |
| 9414 | mkexpr(irt_res_lo) ); |
| 9415 | |
| 9416 | if (isT) |
| 9417 | putIRegT( regD, ire_result, condT ); |
| 9418 | else |
| 9419 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 9420 | |
| 9421 | DIP( "usat16%s r%u, #0x%04x, r%u\n", nCC(conq), regD, sat_imm, regN ); |
| 9422 | return True; |
| 9423 | } |
| 9424 | /* fall through */ |
| 9425 | } |
| 9426 | |
| 9427 | /* -------------- uadd16<c> <Rd>,<Rn>,<Rm> -------------- */ |
| 9428 | { |
| 9429 | UInt regD = 99, regN = 99, regM = 99; |
| 9430 | Bool gate = False; |
| 9431 | |
| 9432 | if (isT) { |
| 9433 | if (INSNT0(15,4) == 0xFA9 && (INSNT1(15,0) & 0xF0F0) == 0xF040) { |
| 9434 | regN = INSNT0(3,0); |
| 9435 | regD = INSNT1(11,8); |
| 9436 | regM = INSNT1(3,0); |
| 9437 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9438 | gate = True; |
| 9439 | } |
| 9440 | } else { |
| 9441 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,0,1) && |
| 9442 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 9443 | INSNA(7,4) == BITS4(0,0,0,1)) { |
| 9444 | regD = INSNA(15,12); |
| 9445 | regN = INSNA(19,16); |
| 9446 | regM = INSNA(3,0); |
| 9447 | if (regD != 15 && regN != 15 && regM != 15) |
| 9448 | gate = True; |
| 9449 | } |
| 9450 | } |
| 9451 | |
| 9452 | if (gate) { |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9453 | IRTemp rNt = newTemp(Ity_I32); |
| 9454 | IRTemp rMt = newTemp(Ity_I32); |
| 9455 | IRTemp res = newTemp(Ity_I32); |
| 9456 | IRTemp reso = newTemp(Ity_I32); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9457 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9458 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9459 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9460 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9461 | assign(res, binop(Iop_Add16x2, mkexpr(rNt), mkexpr(rMt))); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9462 | if (isT) |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9463 | putIRegT( regD, mkexpr(res), condT ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9464 | else |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9465 | putIRegA( regD, mkexpr(res), condT, Ijk_Boring ); |
| 9466 | |
| 9467 | assign(reso, binop(Iop_HAdd16Ux2, mkexpr(rNt), mkexpr(rMt))); |
| 9468 | set_GE_32_10_from_bits_31_15(reso, condT); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9469 | |
| 9470 | DIP("uadd16%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 9471 | return True; |
| 9472 | } |
| 9473 | /* fall through */ |
| 9474 | } |
| 9475 | |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 9476 | /* -------------- sadd16<c> <Rd>,<Rn>,<Rm> -------------- */ |
| 9477 | { |
| 9478 | UInt regD = 99, regN = 99, regM = 99; |
| 9479 | Bool gate = False; |
| 9480 | |
| 9481 | if (isT) { |
| 9482 | if (INSNT0(15,4) == 0xFA9 && (INSNT1(15,0) & 0xF0F0) == 0xF000) { |
| 9483 | regN = INSNT0(3,0); |
| 9484 | regD = INSNT1(11,8); |
| 9485 | regM = INSNT1(3,0); |
| 9486 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9487 | gate = True; |
| 9488 | } |
| 9489 | } else { |
| 9490 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,0,1) && |
| 9491 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 9492 | INSNA(7,4) == BITS4(0,0,0,1)) { |
| 9493 | regD = INSNA(15,12); |
| 9494 | regN = INSNA(19,16); |
| 9495 | regM = INSNA(3,0); |
| 9496 | if (regD != 15 && regN != 15 && regM != 15) |
| 9497 | gate = True; |
| 9498 | } |
| 9499 | } |
| 9500 | |
| 9501 | if (gate) { |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9502 | IRTemp rNt = newTemp(Ity_I32); |
| 9503 | IRTemp rMt = newTemp(Ity_I32); |
| 9504 | IRTemp res = newTemp(Ity_I32); |
| 9505 | IRTemp reso = newTemp(Ity_I32); |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 9506 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9507 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9508 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 9509 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9510 | assign(res, binop(Iop_Add16x2, mkexpr(rNt), mkexpr(rMt))); |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 9511 | if (isT) |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9512 | putIRegT( regD, mkexpr(res), condT ); |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 9513 | else |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9514 | putIRegA( regD, mkexpr(res), condT, Ijk_Boring ); |
| 9515 | |
| 9516 | assign(reso, unop(Iop_Not32, |
| 9517 | binop(Iop_HAdd16Sx2, mkexpr(rNt), mkexpr(rMt)))); |
| 9518 | set_GE_32_10_from_bits_31_15(reso, condT); |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 9519 | |
| 9520 | DIP("sadd16%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 9521 | return True; |
| 9522 | } |
| 9523 | /* fall through */ |
| 9524 | } |
| 9525 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9526 | /* ---------------- usub16<c> <Rd>,<Rn>,<Rm> ---------------- */ |
| 9527 | { |
| 9528 | UInt regD = 99, regN = 99, regM = 99; |
| 9529 | Bool gate = False; |
| 9530 | |
| 9531 | if (isT) { |
| 9532 | if (INSNT0(15,4) == 0xFAD && (INSNT1(15,0) & 0xF0F0) == 0xF040) { |
| 9533 | regN = INSNT0(3,0); |
| 9534 | regD = INSNT1(11,8); |
| 9535 | regM = INSNT1(3,0); |
| 9536 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9537 | gate = True; |
| 9538 | } |
| 9539 | } else { |
| 9540 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,0,1) && |
| 9541 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 9542 | INSNA(7,4) == BITS4(0,1,1,1)) { |
| 9543 | regD = INSNA(15,12); |
| 9544 | regN = INSNA(19,16); |
| 9545 | regM = INSNA(3,0); |
| 9546 | if (regD != 15 && regN != 15 && regM != 15) |
| 9547 | gate = True; |
| 9548 | } |
| 9549 | } |
| 9550 | |
| 9551 | if (gate) { |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9552 | IRTemp rNt = newTemp(Ity_I32); |
| 9553 | IRTemp rMt = newTemp(Ity_I32); |
| 9554 | IRTemp res = newTemp(Ity_I32); |
| 9555 | IRTemp reso = newTemp(Ity_I32); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9556 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9557 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9558 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9559 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9560 | assign(res, binop(Iop_Sub16x2, mkexpr(rNt), mkexpr(rMt))); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9561 | if (isT) |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9562 | putIRegT( regD, mkexpr(res), condT ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9563 | else |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9564 | putIRegA( regD, mkexpr(res), condT, Ijk_Boring ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9565 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9566 | assign(reso, unop(Iop_Not32, |
| 9567 | binop(Iop_HSub16Ux2, mkexpr(rNt), mkexpr(rMt)))); |
| 9568 | set_GE_32_10_from_bits_31_15(reso, condT); |
| 9569 | |
| 9570 | DIP("usub16%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9571 | return True; |
| 9572 | } |
| 9573 | /* fall through */ |
| 9574 | } |
| 9575 | |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 9576 | /* -------------- ssub16<c> <Rd>,<Rn>,<Rm> -------------- */ |
| 9577 | { |
| 9578 | UInt regD = 99, regN = 99, regM = 99; |
| 9579 | Bool gate = False; |
| 9580 | |
| 9581 | if (isT) { |
| 9582 | if (INSNT0(15,4) == 0xFAD && (INSNT1(15,0) & 0xF0F0) == 0xF000) { |
| 9583 | regN = INSNT0(3,0); |
| 9584 | regD = INSNT1(11,8); |
| 9585 | regM = INSNT1(3,0); |
| 9586 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9587 | gate = True; |
| 9588 | } |
| 9589 | } else { |
| 9590 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,0,1) && |
| 9591 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 9592 | INSNA(7,4) == BITS4(0,1,1,1)) { |
| 9593 | regD = INSNA(15,12); |
| 9594 | regN = INSNA(19,16); |
| 9595 | regM = INSNA(3,0); |
| 9596 | if (regD != 15 && regN != 15 && regM != 15) |
| 9597 | gate = True; |
| 9598 | } |
| 9599 | } |
| 9600 | |
| 9601 | if (gate) { |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9602 | IRTemp rNt = newTemp(Ity_I32); |
| 9603 | IRTemp rMt = newTemp(Ity_I32); |
| 9604 | IRTemp res = newTemp(Ity_I32); |
| 9605 | IRTemp reso = newTemp(Ity_I32); |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 9606 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9607 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9608 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 9609 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9610 | assign(res, binop(Iop_Sub16x2, mkexpr(rNt), mkexpr(rMt))); |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 9611 | if (isT) |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9612 | putIRegT( regD, mkexpr(res), condT ); |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 9613 | else |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9614 | putIRegA( regD, mkexpr(res), condT, Ijk_Boring ); |
| 9615 | |
| 9616 | assign(reso, unop(Iop_Not32, |
| 9617 | binop(Iop_HSub16Sx2, mkexpr(rNt), mkexpr(rMt)))); |
| 9618 | set_GE_32_10_from_bits_31_15(reso, condT); |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 9619 | |
| 9620 | DIP("ssub16%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 9621 | return True; |
| 9622 | } |
| 9623 | /* fall through */ |
| 9624 | } |
| 9625 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9626 | /* ----------------- uadd8<c> <Rd>,<Rn>,<Rm> ---------------- */ |
| 9627 | { |
| 9628 | UInt regD = 99, regN = 99, regM = 99; |
| 9629 | Bool gate = False; |
| 9630 | |
| 9631 | if (isT) { |
| 9632 | if (INSNT0(15,4) == 0xFA8 && (INSNT1(15,0) & 0xF0F0) == 0xF040) { |
| 9633 | regN = INSNT0(3,0); |
| 9634 | regD = INSNT1(11,8); |
| 9635 | regM = INSNT1(3,0); |
| 9636 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9637 | gate = True; |
| 9638 | } |
| 9639 | } else { |
| 9640 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,0,1) && |
| 9641 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 9642 | (INSNA(7,4) == BITS4(1,0,0,1))) { |
| 9643 | regD = INSNA(15,12); |
| 9644 | regN = INSNA(19,16); |
| 9645 | regM = INSNA(3,0); |
| 9646 | if (regD != 15 && regN != 15 && regM != 15) |
| 9647 | gate = True; |
| 9648 | } |
| 9649 | } |
| 9650 | |
| 9651 | if (gate) { |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9652 | IRTemp rNt = newTemp(Ity_I32); |
| 9653 | IRTemp rMt = newTemp(Ity_I32); |
| 9654 | IRTemp res = newTemp(Ity_I32); |
| 9655 | IRTemp reso = newTemp(Ity_I32); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9656 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9657 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9658 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9659 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9660 | assign(res, binop(Iop_Add8x4, mkexpr(rNt), mkexpr(rMt))); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9661 | if (isT) |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9662 | putIRegT( regD, mkexpr(res), condT ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9663 | else |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9664 | putIRegA( regD, mkexpr(res), condT, Ijk_Boring ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9665 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9666 | assign(reso, binop(Iop_HAdd8Ux4, mkexpr(rNt), mkexpr(rMt))); |
| 9667 | set_GE_3_2_1_0_from_bits_31_23_15_7(reso, condT); |
| 9668 | |
| 9669 | DIP("uadd8%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 9670 | return True; |
| 9671 | } |
| 9672 | /* fall through */ |
| 9673 | } |
| 9674 | |
| 9675 | /* ------------------- sadd8<c> <Rd>,<Rn>,<Rm> ------------------ */ |
| 9676 | { |
| 9677 | UInt regD = 99, regN = 99, regM = 99; |
| 9678 | Bool gate = False; |
| 9679 | |
| 9680 | if (isT) { |
| 9681 | if (INSNT0(15,4) == 0xFA8 && (INSNT1(15,0) & 0xF0F0) == 0xF000) { |
| 9682 | regN = INSNT0(3,0); |
| 9683 | regD = INSNT1(11,8); |
| 9684 | regM = INSNT1(3,0); |
| 9685 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9686 | gate = True; |
| 9687 | } |
| 9688 | } else { |
| 9689 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,0,1) && |
| 9690 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 9691 | (INSNA(7,4) == BITS4(1,0,0,1))) { |
| 9692 | regD = INSNA(15,12); |
| 9693 | regN = INSNA(19,16); |
| 9694 | regM = INSNA(3,0); |
| 9695 | if (regD != 15 && regN != 15 && regM != 15) |
| 9696 | gate = True; |
| 9697 | } |
| 9698 | } |
| 9699 | |
| 9700 | if (gate) { |
| 9701 | IRTemp rNt = newTemp(Ity_I32); |
| 9702 | IRTemp rMt = newTemp(Ity_I32); |
| 9703 | IRTemp res = newTemp(Ity_I32); |
| 9704 | IRTemp reso = newTemp(Ity_I32); |
| 9705 | |
| 9706 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9707 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 9708 | |
| 9709 | assign(res, binop(Iop_Add8x4, mkexpr(rNt), mkexpr(rMt))); |
| 9710 | if (isT) |
| 9711 | putIRegT( regD, mkexpr(res), condT ); |
| 9712 | else |
| 9713 | putIRegA( regD, mkexpr(res), condT, Ijk_Boring ); |
| 9714 | |
| 9715 | assign(reso, unop(Iop_Not32, |
| 9716 | binop(Iop_HAdd8Sx4, mkexpr(rNt), mkexpr(rMt)))); |
| 9717 | set_GE_3_2_1_0_from_bits_31_23_15_7(reso, condT); |
| 9718 | |
| 9719 | DIP("sadd8%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9720 | return True; |
| 9721 | } |
| 9722 | /* fall through */ |
| 9723 | } |
| 9724 | |
| 9725 | /* ------------------- usub8<c> <Rd>,<Rn>,<Rm> ------------------ */ |
| 9726 | { |
| 9727 | UInt regD = 99, regN = 99, regM = 99; |
| 9728 | Bool gate = False; |
| 9729 | |
| 9730 | if (isT) { |
| 9731 | if (INSNT0(15,4) == 0xFAC && (INSNT1(15,0) & 0xF0F0) == 0xF040) { |
| 9732 | regN = INSNT0(3,0); |
| 9733 | regD = INSNT1(11,8); |
| 9734 | regM = INSNT1(3,0); |
| 9735 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9736 | gate = True; |
| 9737 | } |
| 9738 | } else { |
| 9739 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,0,1) && |
| 9740 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 9741 | (INSNA(7,4) == BITS4(1,1,1,1))) { |
| 9742 | regD = INSNA(15,12); |
| 9743 | regN = INSNA(19,16); |
| 9744 | regM = INSNA(3,0); |
| 9745 | if (regD != 15 && regN != 15 && regM != 15) |
| 9746 | gate = True; |
| 9747 | } |
| 9748 | } |
| 9749 | |
| 9750 | if (gate) { |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9751 | IRTemp rNt = newTemp(Ity_I32); |
| 9752 | IRTemp rMt = newTemp(Ity_I32); |
| 9753 | IRTemp res = newTemp(Ity_I32); |
| 9754 | IRTemp reso = newTemp(Ity_I32); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9755 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9756 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9757 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9758 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9759 | assign(res, binop(Iop_Sub8x4, mkexpr(rNt), mkexpr(rMt))); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9760 | if (isT) |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9761 | putIRegT( regD, mkexpr(res), condT ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9762 | else |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9763 | putIRegA( regD, mkexpr(res), condT, Ijk_Boring ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 9764 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 9765 | assign(reso, unop(Iop_Not32, |
| 9766 | binop(Iop_HSub8Ux4, mkexpr(rNt), mkexpr(rMt)))); |
| 9767 | set_GE_3_2_1_0_from_bits_31_23_15_7(reso, condT); |
| 9768 | |
| 9769 | DIP("usub8%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 9770 | return True; |
| 9771 | } |
| 9772 | /* fall through */ |
| 9773 | } |
| 9774 | |
| 9775 | /* ------------------- ssub8<c> <Rd>,<Rn>,<Rm> ------------------ */ |
| 9776 | { |
| 9777 | UInt regD = 99, regN = 99, regM = 99; |
| 9778 | Bool gate = False; |
| 9779 | |
| 9780 | if (isT) { |
| 9781 | if (INSNT0(15,4) == 0xFAC && (INSNT1(15,0) & 0xF0F0) == 0xF000) { |
| 9782 | regN = INSNT0(3,0); |
| 9783 | regD = INSNT1(11,8); |
| 9784 | regM = INSNT1(3,0); |
| 9785 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9786 | gate = True; |
| 9787 | } |
| 9788 | } else { |
| 9789 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,0,1) && |
| 9790 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 9791 | INSNA(7,4) == BITS4(1,1,1,1)) { |
| 9792 | regD = INSNA(15,12); |
| 9793 | regN = INSNA(19,16); |
| 9794 | regM = INSNA(3,0); |
| 9795 | if (regD != 15 && regN != 15 && regM != 15) |
| 9796 | gate = True; |
| 9797 | } |
| 9798 | } |
| 9799 | |
| 9800 | if (gate) { |
| 9801 | IRTemp rNt = newTemp(Ity_I32); |
| 9802 | IRTemp rMt = newTemp(Ity_I32); |
| 9803 | IRTemp res = newTemp(Ity_I32); |
| 9804 | IRTemp reso = newTemp(Ity_I32); |
| 9805 | |
| 9806 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9807 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 9808 | |
| 9809 | assign(res, binop(Iop_Sub8x4, mkexpr(rNt), mkexpr(rMt))); |
| 9810 | if (isT) |
| 9811 | putIRegT( regD, mkexpr(res), condT ); |
| 9812 | else |
| 9813 | putIRegA( regD, mkexpr(res), condT, Ijk_Boring ); |
| 9814 | |
| 9815 | assign(reso, unop(Iop_Not32, |
| 9816 | binop(Iop_HSub8Sx4, mkexpr(rNt), mkexpr(rMt)))); |
| 9817 | set_GE_3_2_1_0_from_bits_31_23_15_7(reso, condT); |
| 9818 | |
| 9819 | DIP("ssub8%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 9820 | return True; |
| 9821 | } |
| 9822 | /* fall through */ |
| 9823 | } |
| 9824 | |
| 9825 | /* ------------------ qadd8<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 9826 | { |
| 9827 | UInt regD = 99, regN = 99, regM = 99; |
| 9828 | Bool gate = False; |
| 9829 | |
| 9830 | if (isT) { |
| 9831 | if (INSNT0(15,4) == 0xFA8 && (INSNT1(15,0) & 0xF0F0) == 0xF010) { |
| 9832 | regN = INSNT0(3,0); |
| 9833 | regD = INSNT1(11,8); |
| 9834 | regM = INSNT1(3,0); |
| 9835 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9836 | gate = True; |
| 9837 | } |
| 9838 | } else { |
| 9839 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,1,0) && |
| 9840 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 9841 | INSNA(7,4) == BITS4(1,0,0,1)) { |
| 9842 | regD = INSNA(15,12); |
| 9843 | regN = INSNA(19,16); |
| 9844 | regM = INSNA(3,0); |
| 9845 | if (regD != 15 && regN != 15 && regM != 15) |
| 9846 | gate = True; |
| 9847 | } |
| 9848 | } |
| 9849 | |
| 9850 | if (gate) { |
| 9851 | IRTemp rNt = newTemp(Ity_I32); |
| 9852 | IRTemp rMt = newTemp(Ity_I32); |
| 9853 | IRTemp res_q = newTemp(Ity_I32); |
| 9854 | |
| 9855 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9856 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 9857 | |
| 9858 | assign(res_q, binop(Iop_QAdd8Sx4, mkexpr(rNt), mkexpr(rMt))); |
| 9859 | if (isT) |
| 9860 | putIRegT( regD, mkexpr(res_q), condT ); |
| 9861 | else |
| 9862 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 9863 | |
| 9864 | DIP("qadd8%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 9865 | return True; |
| 9866 | } |
| 9867 | /* fall through */ |
| 9868 | } |
| 9869 | |
| 9870 | /* ------------------ qsub8<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 9871 | { |
| 9872 | UInt regD = 99, regN = 99, regM = 99; |
| 9873 | Bool gate = False; |
| 9874 | |
| 9875 | if (isT) { |
| 9876 | if (INSNT0(15,4) == 0xFAC && (INSNT1(15,0) & 0xF0F0) == 0xF010) { |
| 9877 | regN = INSNT0(3,0); |
| 9878 | regD = INSNT1(11,8); |
| 9879 | regM = INSNT1(3,0); |
| 9880 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9881 | gate = True; |
| 9882 | } |
| 9883 | } else { |
| 9884 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,1,0) && |
| 9885 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 9886 | INSNA(7,4) == BITS4(1,1,1,1)) { |
| 9887 | regD = INSNA(15,12); |
| 9888 | regN = INSNA(19,16); |
| 9889 | regM = INSNA(3,0); |
| 9890 | if (regD != 15 && regN != 15 && regM != 15) |
| 9891 | gate = True; |
| 9892 | } |
| 9893 | } |
| 9894 | |
| 9895 | if (gate) { |
| 9896 | IRTemp rNt = newTemp(Ity_I32); |
| 9897 | IRTemp rMt = newTemp(Ity_I32); |
| 9898 | IRTemp res_q = newTemp(Ity_I32); |
| 9899 | |
| 9900 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9901 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 9902 | |
| 9903 | assign(res_q, binop(Iop_QSub8Sx4, mkexpr(rNt), mkexpr(rMt))); |
| 9904 | if (isT) |
| 9905 | putIRegT( regD, mkexpr(res_q), condT ); |
| 9906 | else |
| 9907 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 9908 | |
| 9909 | DIP("qsub8%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 9910 | return True; |
| 9911 | } |
| 9912 | /* fall through */ |
| 9913 | } |
| 9914 | |
| 9915 | /* ------------------ uqadd8<c> <Rd>,<Rn>,<Rm> ------------------ */ |
| 9916 | { |
| 9917 | UInt regD = 99, regN = 99, regM = 99; |
| 9918 | Bool gate = False; |
| 9919 | |
| 9920 | if (isT) { |
| 9921 | if (INSNT0(15,4) == 0xFA8 && (INSNT1(15,0) & 0xF0F0) == 0xF050) { |
| 9922 | regN = INSNT0(3,0); |
| 9923 | regD = INSNT1(11,8); |
| 9924 | regM = INSNT1(3,0); |
| 9925 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9926 | gate = True; |
| 9927 | } |
| 9928 | } else { |
| 9929 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,1,0) && |
| 9930 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 9931 | (INSNA(7,4) == BITS4(1,0,0,1))) { |
| 9932 | regD = INSNA(15,12); |
| 9933 | regN = INSNA(19,16); |
| 9934 | regM = INSNA(3,0); |
| 9935 | if (regD != 15 && regN != 15 && regM != 15) |
| 9936 | gate = True; |
| 9937 | } |
| 9938 | } |
| 9939 | |
| 9940 | if (gate) { |
| 9941 | IRTemp rNt = newTemp(Ity_I32); |
| 9942 | IRTemp rMt = newTemp(Ity_I32); |
| 9943 | IRTemp res_q = newTemp(Ity_I32); |
| 9944 | |
| 9945 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9946 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 9947 | |
| 9948 | assign(res_q, binop(Iop_QAdd8Ux4, mkexpr(rNt), mkexpr(rMt))); |
| 9949 | if (isT) |
| 9950 | putIRegT( regD, mkexpr(res_q), condT ); |
| 9951 | else |
| 9952 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 9953 | |
| 9954 | DIP("uqadd8%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 9955 | return True; |
| 9956 | } |
| 9957 | /* fall through */ |
| 9958 | } |
| 9959 | |
| 9960 | /* ------------------ uqsub8<c> <Rd>,<Rn>,<Rm> ------------------ */ |
| 9961 | { |
| 9962 | UInt regD = 99, regN = 99, regM = 99; |
| 9963 | Bool gate = False; |
| 9964 | |
| 9965 | if (isT) { |
| 9966 | if (INSNT0(15,4) == 0xFAC && (INSNT1(15,0) & 0xF0F0) == 0xF050) { |
| 9967 | regN = INSNT0(3,0); |
| 9968 | regD = INSNT1(11,8); |
| 9969 | regM = INSNT1(3,0); |
| 9970 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 9971 | gate = True; |
| 9972 | } |
| 9973 | } else { |
| 9974 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,1,0) && |
| 9975 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 9976 | (INSNA(7,4) == BITS4(1,1,1,1))) { |
| 9977 | regD = INSNA(15,12); |
| 9978 | regN = INSNA(19,16); |
| 9979 | regM = INSNA(3,0); |
| 9980 | if (regD != 15 && regN != 15 && regM != 15) |
| 9981 | gate = True; |
| 9982 | } |
| 9983 | } |
| 9984 | |
| 9985 | if (gate) { |
| 9986 | IRTemp rNt = newTemp(Ity_I32); |
| 9987 | IRTemp rMt = newTemp(Ity_I32); |
| 9988 | IRTemp res_q = newTemp(Ity_I32); |
| 9989 | |
| 9990 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 9991 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 9992 | |
| 9993 | assign(res_q, binop(Iop_QSub8Ux4, mkexpr(rNt), mkexpr(rMt))); |
| 9994 | if (isT) |
| 9995 | putIRegT( regD, mkexpr(res_q), condT ); |
| 9996 | else |
| 9997 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 9998 | |
| 9999 | DIP("uqsub8%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 10000 | return True; |
| 10001 | } |
| 10002 | /* fall through */ |
| 10003 | } |
| 10004 | |
| 10005 | /* ----------------- uhadd8<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 10006 | { |
| 10007 | UInt regD = 99, regN = 99, regM = 99; |
| 10008 | Bool gate = False; |
| 10009 | |
| 10010 | if (isT) { |
| 10011 | if (INSNT0(15,4) == 0xFA8 && (INSNT1(15,0) & 0xF0F0) == 0xF060) { |
| 10012 | regN = INSNT0(3,0); |
| 10013 | regD = INSNT1(11,8); |
| 10014 | regM = INSNT1(3,0); |
| 10015 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 10016 | gate = True; |
| 10017 | } |
| 10018 | } else { |
| 10019 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,1,1) && |
| 10020 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 10021 | INSNA(7,4) == BITS4(1,0,0,1)) { |
| 10022 | regD = INSNA(15,12); |
| 10023 | regN = INSNA(19,16); |
| 10024 | regM = INSNA(3,0); |
| 10025 | if (regD != 15 && regN != 15 && regM != 15) |
| 10026 | gate = True; |
| 10027 | } |
| 10028 | } |
| 10029 | |
| 10030 | if (gate) { |
| 10031 | IRTemp rNt = newTemp(Ity_I32); |
| 10032 | IRTemp rMt = newTemp(Ity_I32); |
| 10033 | IRTemp res_q = newTemp(Ity_I32); |
| 10034 | |
| 10035 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 10036 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 10037 | |
| 10038 | assign(res_q, binop(Iop_HAdd8Ux4, mkexpr(rNt), mkexpr(rMt))); |
| 10039 | if (isT) |
| 10040 | putIRegT( regD, mkexpr(res_q), condT ); |
| 10041 | else |
| 10042 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 10043 | |
| 10044 | DIP("uhadd8%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 10045 | return True; |
| 10046 | } |
| 10047 | /* fall through */ |
| 10048 | } |
| 10049 | |
sewardj | 9de2809 | 2012-06-22 09:27:54 +0000 | [diff] [blame] | 10050 | /* ----------------- uhadd16<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 10051 | { |
| 10052 | UInt regD = 99, regN = 99, regM = 99; |
| 10053 | Bool gate = False; |
| 10054 | |
| 10055 | if (isT) { |
| 10056 | if (INSNT0(15,4) == 0xFA9 && (INSNT1(15,0) & 0xF0F0) == 0xF060) { |
| 10057 | regN = INSNT0(3,0); |
| 10058 | regD = INSNT1(11,8); |
| 10059 | regM = INSNT1(3,0); |
| 10060 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 10061 | gate = True; |
| 10062 | } |
| 10063 | } else { |
| 10064 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,1,1) && |
| 10065 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 10066 | INSNA(7,4) == BITS4(0,0,0,1)) { |
| 10067 | regD = INSNA(15,12); |
| 10068 | regN = INSNA(19,16); |
| 10069 | regM = INSNA(3,0); |
| 10070 | if (regD != 15 && regN != 15 && regM != 15) |
| 10071 | gate = True; |
| 10072 | } |
| 10073 | } |
| 10074 | |
| 10075 | if (gate) { |
| 10076 | IRTemp rNt = newTemp(Ity_I32); |
| 10077 | IRTemp rMt = newTemp(Ity_I32); |
| 10078 | IRTemp res_q = newTemp(Ity_I32); |
| 10079 | |
| 10080 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 10081 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 10082 | |
| 10083 | assign(res_q, binop(Iop_HAdd16Ux2, mkexpr(rNt), mkexpr(rMt))); |
| 10084 | if (isT) |
| 10085 | putIRegT( regD, mkexpr(res_q), condT ); |
| 10086 | else |
| 10087 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 10088 | |
| 10089 | DIP("uhadd16%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 10090 | return True; |
| 10091 | } |
| 10092 | /* fall through */ |
| 10093 | } |
| 10094 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10095 | /* ----------------- shadd8<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 10096 | { |
| 10097 | UInt regD = 99, regN = 99, regM = 99; |
| 10098 | Bool gate = False; |
| 10099 | |
| 10100 | if (isT) { |
| 10101 | if (INSNT0(15,4) == 0xFA8 && (INSNT1(15,0) & 0xF0F0) == 0xF020) { |
| 10102 | regN = INSNT0(3,0); |
| 10103 | regD = INSNT1(11,8); |
| 10104 | regM = INSNT1(3,0); |
| 10105 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 10106 | gate = True; |
| 10107 | } |
| 10108 | } else { |
| 10109 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,1,1) && |
| 10110 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 10111 | INSNA(7,4) == BITS4(1,0,0,1)) { |
| 10112 | regD = INSNA(15,12); |
| 10113 | regN = INSNA(19,16); |
| 10114 | regM = INSNA(3,0); |
| 10115 | if (regD != 15 && regN != 15 && regM != 15) |
| 10116 | gate = True; |
| 10117 | } |
| 10118 | } |
| 10119 | |
| 10120 | if (gate) { |
| 10121 | IRTemp rNt = newTemp(Ity_I32); |
| 10122 | IRTemp rMt = newTemp(Ity_I32); |
| 10123 | IRTemp res_q = newTemp(Ity_I32); |
| 10124 | |
| 10125 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 10126 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 10127 | |
| 10128 | assign(res_q, binop(Iop_HAdd8Sx4, mkexpr(rNt), mkexpr(rMt))); |
| 10129 | if (isT) |
| 10130 | putIRegT( regD, mkexpr(res_q), condT ); |
| 10131 | else |
| 10132 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 10133 | |
| 10134 | DIP("shadd8%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10135 | return True; |
| 10136 | } |
| 10137 | /* fall through */ |
| 10138 | } |
| 10139 | |
| 10140 | /* ------------------ qadd16<c> <Rd>,<Rn>,<Rm> ------------------ */ |
| 10141 | { |
| 10142 | UInt regD = 99, regN = 99, regM = 99; |
| 10143 | Bool gate = False; |
| 10144 | |
| 10145 | if (isT) { |
| 10146 | if (INSNT0(15,4) == 0xFA9 && (INSNT1(15,0) & 0xF0F0) == 0xF010) { |
| 10147 | regN = INSNT0(3,0); |
| 10148 | regD = INSNT1(11,8); |
| 10149 | regM = INSNT1(3,0); |
| 10150 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 10151 | gate = True; |
| 10152 | } |
| 10153 | } else { |
| 10154 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,1,0) && |
| 10155 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 10156 | INSNA(7,4) == BITS4(0,0,0,1)) { |
| 10157 | regD = INSNA(15,12); |
| 10158 | regN = INSNA(19,16); |
| 10159 | regM = INSNA(3,0); |
| 10160 | if (regD != 15 && regN != 15 && regM != 15) |
| 10161 | gate = True; |
| 10162 | } |
| 10163 | } |
| 10164 | |
| 10165 | if (gate) { |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10166 | IRTemp rNt = newTemp(Ity_I32); |
| 10167 | IRTemp rMt = newTemp(Ity_I32); |
| 10168 | IRTemp res_q = newTemp(Ity_I32); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10169 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10170 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 10171 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10172 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10173 | assign(res_q, binop(Iop_QAdd16Sx2, mkexpr(rNt), mkexpr(rMt))); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10174 | if (isT) |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10175 | putIRegT( regD, mkexpr(res_q), condT ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10176 | else |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10177 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10178 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10179 | DIP("qadd16%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10180 | return True; |
| 10181 | } |
| 10182 | /* fall through */ |
| 10183 | } |
| 10184 | |
| 10185 | /* ------------------ qsub16<c> <Rd>,<Rn>,<Rm> ------------------ */ |
| 10186 | { |
| 10187 | UInt regD = 99, regN = 99, regM = 99; |
| 10188 | Bool gate = False; |
| 10189 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10190 | if (isT) { |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10191 | if (INSNT0(15,4) == 0xFAD && (INSNT1(15,0) & 0xF0F0) == 0xF010) { |
| 10192 | regN = INSNT0(3,0); |
| 10193 | regD = INSNT1(11,8); |
| 10194 | regM = INSNT1(3,0); |
| 10195 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 10196 | gate = True; |
| 10197 | } |
| 10198 | } else { |
| 10199 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,1,0) && |
| 10200 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 10201 | INSNA(7,4) == BITS4(0,1,1,1)) { |
| 10202 | regD = INSNA(15,12); |
| 10203 | regN = INSNA(19,16); |
| 10204 | regM = INSNA(3,0); |
| 10205 | if (regD != 15 && regN != 15 && regM != 15) |
| 10206 | gate = True; |
| 10207 | } |
| 10208 | } |
| 10209 | |
| 10210 | if (gate) { |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10211 | IRTemp rNt = newTemp(Ity_I32); |
| 10212 | IRTemp rMt = newTemp(Ity_I32); |
| 10213 | IRTemp res_q = newTemp(Ity_I32); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10214 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10215 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 10216 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10217 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10218 | assign(res_q, binop(Iop_QSub16Sx2, mkexpr(rNt), mkexpr(rMt))); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10219 | if (isT) |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10220 | putIRegT( regD, mkexpr(res_q), condT ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10221 | else |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10222 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10223 | |
sewardj | e2ea176 | 2010-09-22 00:56:37 +0000 | [diff] [blame] | 10224 | DIP("qsub16%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10225 | return True; |
| 10226 | } |
| 10227 | /* fall through */ |
| 10228 | } |
| 10229 | |
| 10230 | /* ------------------- qsax<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 10231 | /* note: the hardware seems to construct the result differently |
| 10232 | from wot the manual says. */ |
| 10233 | { |
| 10234 | UInt regD = 99, regN = 99, regM = 99; |
| 10235 | Bool gate = False; |
| 10236 | |
| 10237 | if (isT) { |
| 10238 | if (INSNT0(15,4) == 0xFAE && (INSNT1(15,0) & 0xF0F0) == 0xF010) { |
| 10239 | regN = INSNT0(3,0); |
| 10240 | regD = INSNT1(11,8); |
| 10241 | regM = INSNT1(3,0); |
| 10242 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 10243 | gate = True; |
| 10244 | } |
| 10245 | } else { |
| 10246 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,1,0) && |
| 10247 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 10248 | INSNA(7,4) == BITS4(0,1,0,1)) { |
| 10249 | regD = INSNA(15,12); |
| 10250 | regN = INSNA(19,16); |
| 10251 | regM = INSNA(3,0); |
| 10252 | if (regD != 15 && regN != 15 && regM != 15) |
| 10253 | gate = True; |
| 10254 | } |
| 10255 | } |
| 10256 | |
| 10257 | if (gate) { |
| 10258 | IRTemp irt_regN = newTemp(Ity_I32); |
| 10259 | IRTemp irt_regM = newTemp(Ity_I32); |
| 10260 | IRTemp irt_sum = newTemp(Ity_I32); |
| 10261 | IRTemp irt_diff = newTemp(Ity_I32); |
| 10262 | IRTemp irt_sum_res = newTemp(Ity_I32); |
| 10263 | IRTemp irt_diff_res = newTemp(Ity_I32); |
| 10264 | |
| 10265 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 10266 | assign( irt_regM, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 10267 | |
| 10268 | assign( irt_diff, |
| 10269 | binop( Iop_Sub32, |
| 10270 | binop( Iop_Sar32, mkexpr(irt_regN), mkU8(16) ), |
| 10271 | binop( Iop_Sar32, |
| 10272 | binop(Iop_Shl32, mkexpr(irt_regM), mkU8(16)), |
| 10273 | mkU8(16) ) ) ); |
| 10274 | armSignedSatQ( irt_diff, 0x10, &irt_diff_res, NULL); |
| 10275 | |
| 10276 | assign( irt_sum, |
| 10277 | binop( Iop_Add32, |
| 10278 | binop( Iop_Sar32, |
| 10279 | binop( Iop_Shl32, mkexpr(irt_regN), mkU8(16) ), |
| 10280 | mkU8(16) ), |
| 10281 | binop( Iop_Sar32, mkexpr(irt_regM), mkU8(16) )) ); |
| 10282 | armSignedSatQ( irt_sum, 0x10, &irt_sum_res, NULL ); |
| 10283 | |
| 10284 | IRExpr* ire_result = binop( Iop_Or32, |
| 10285 | binop( Iop_Shl32, mkexpr(irt_diff_res), |
| 10286 | mkU8(16) ), |
| 10287 | binop( Iop_And32, mkexpr(irt_sum_res), |
| 10288 | mkU32(0xFFFF)) ); |
| 10289 | |
| 10290 | if (isT) |
| 10291 | putIRegT( regD, ire_result, condT ); |
| 10292 | else |
| 10293 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 10294 | |
| 10295 | DIP( "qsax%s r%u, r%u, r%u\n", nCC(conq), regD, regN, regM ); |
| 10296 | return True; |
| 10297 | } |
| 10298 | /* fall through */ |
| 10299 | } |
| 10300 | |
| 10301 | /* ------------------- qasx<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 10302 | { |
| 10303 | UInt regD = 99, regN = 99, regM = 99; |
| 10304 | Bool gate = False; |
| 10305 | |
| 10306 | if (isT) { |
| 10307 | if (INSNT0(15,4) == 0xFAA && (INSNT1(15,0) & 0xF0F0) == 0xF010) { |
| 10308 | regN = INSNT0(3,0); |
| 10309 | regD = INSNT1(11,8); |
| 10310 | regM = INSNT1(3,0); |
| 10311 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 10312 | gate = True; |
| 10313 | } |
| 10314 | } else { |
| 10315 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,1,0) && |
| 10316 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 10317 | INSNA(7,4) == BITS4(0,0,1,1)) { |
| 10318 | regD = INSNA(15,12); |
| 10319 | regN = INSNA(19,16); |
| 10320 | regM = INSNA(3,0); |
| 10321 | if (regD != 15 && regN != 15 && regM != 15) |
| 10322 | gate = True; |
| 10323 | } |
| 10324 | } |
| 10325 | |
| 10326 | if (gate) { |
| 10327 | IRTemp irt_regN = newTemp(Ity_I32); |
| 10328 | IRTemp irt_regM = newTemp(Ity_I32); |
| 10329 | IRTemp irt_sum = newTemp(Ity_I32); |
| 10330 | IRTemp irt_diff = newTemp(Ity_I32); |
| 10331 | IRTemp irt_res_sum = newTemp(Ity_I32); |
| 10332 | IRTemp irt_res_diff = newTemp(Ity_I32); |
| 10333 | |
| 10334 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 10335 | assign( irt_regM, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 10336 | |
| 10337 | assign( irt_diff, |
| 10338 | binop( Iop_Sub32, |
| 10339 | binop( Iop_Sar32, |
| 10340 | binop( Iop_Shl32, mkexpr(irt_regN), mkU8(16) ), |
| 10341 | mkU8(16) ), |
| 10342 | binop( Iop_Sar32, mkexpr(irt_regM), mkU8(16) ) ) ); |
| 10343 | armSignedSatQ( irt_diff, 0x10, &irt_res_diff, NULL ); |
| 10344 | |
| 10345 | assign( irt_sum, |
| 10346 | binop( Iop_Add32, |
| 10347 | binop( Iop_Sar32, mkexpr(irt_regN), mkU8(16) ), |
| 10348 | binop( Iop_Sar32, |
| 10349 | binop( Iop_Shl32, mkexpr(irt_regM), mkU8(16) ), |
| 10350 | mkU8(16) ) ) ); |
| 10351 | armSignedSatQ( irt_sum, 0x10, &irt_res_sum, NULL ); |
| 10352 | |
| 10353 | IRExpr* ire_result |
| 10354 | = binop( Iop_Or32, |
| 10355 | binop( Iop_Shl32, mkexpr(irt_res_sum), mkU8(16) ), |
| 10356 | binop( Iop_And32, mkexpr(irt_res_diff), mkU32(0xFFFF) ) ); |
| 10357 | |
| 10358 | if (isT) |
| 10359 | putIRegT( regD, ire_result, condT ); |
| 10360 | else |
| 10361 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 10362 | |
| 10363 | DIP( "qasx%s r%u, r%u, r%u\n", nCC(conq), regD, regN, regM ); |
| 10364 | return True; |
| 10365 | } |
| 10366 | /* fall through */ |
| 10367 | } |
| 10368 | |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 10369 | /* ------------------- sasx<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 10370 | { |
| 10371 | UInt regD = 99, regN = 99, regM = 99; |
| 10372 | Bool gate = False; |
| 10373 | |
| 10374 | if (isT) { |
| 10375 | if (INSNT0(15,4) == 0xFAA && (INSNT1(15,0) & 0xF0F0) == 0xF000) { |
| 10376 | regN = INSNT0(3,0); |
| 10377 | regD = INSNT1(11,8); |
| 10378 | regM = INSNT1(3,0); |
| 10379 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 10380 | gate = True; |
| 10381 | } |
| 10382 | } else { |
| 10383 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,0,1) && |
| 10384 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 10385 | INSNA(7,4) == BITS4(0,0,1,1)) { |
| 10386 | regD = INSNA(15,12); |
| 10387 | regN = INSNA(19,16); |
| 10388 | regM = INSNA(3,0); |
| 10389 | if (regD != 15 && regN != 15 && regM != 15) |
| 10390 | gate = True; |
| 10391 | } |
| 10392 | } |
| 10393 | |
| 10394 | if (gate) { |
| 10395 | IRTemp irt_regN = newTemp(Ity_I32); |
| 10396 | IRTemp irt_regM = newTemp(Ity_I32); |
| 10397 | IRTemp irt_sum = newTemp(Ity_I32); |
| 10398 | IRTemp irt_diff = newTemp(Ity_I32); |
| 10399 | |
| 10400 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 10401 | assign( irt_regM, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 10402 | |
| 10403 | assign( irt_diff, |
| 10404 | binop( Iop_Sub32, |
| 10405 | binop( Iop_Sar32, |
| 10406 | binop( Iop_Shl32, mkexpr(irt_regN), mkU8(16) ), |
| 10407 | mkU8(16) ), |
| 10408 | binop( Iop_Sar32, mkexpr(irt_regM), mkU8(16) ) ) ); |
| 10409 | |
| 10410 | assign( irt_sum, |
| 10411 | binop( Iop_Add32, |
| 10412 | binop( Iop_Sar32, mkexpr(irt_regN), mkU8(16) ), |
| 10413 | binop( Iop_Sar32, |
| 10414 | binop( Iop_Shl32, mkexpr(irt_regM), mkU8(16) ), |
| 10415 | mkU8(16) ) ) ); |
| 10416 | |
| 10417 | IRExpr* ire_result |
| 10418 | = binop( Iop_Or32, |
| 10419 | binop( Iop_Shl32, mkexpr(irt_sum), mkU8(16) ), |
| 10420 | binop( Iop_And32, mkexpr(irt_diff), mkU32(0xFFFF) ) ); |
| 10421 | |
| 10422 | IRTemp ge10 = newTemp(Ity_I32); |
| 10423 | assign(ge10, unop(Iop_Not32, mkexpr(irt_diff))); |
| 10424 | put_GEFLAG32( 0, 31, mkexpr(ge10), condT ); |
| 10425 | put_GEFLAG32( 1, 31, mkexpr(ge10), condT ); |
| 10426 | |
| 10427 | IRTemp ge32 = newTemp(Ity_I32); |
| 10428 | assign(ge32, unop(Iop_Not32, mkexpr(irt_sum))); |
| 10429 | put_GEFLAG32( 2, 31, mkexpr(ge32), condT ); |
| 10430 | put_GEFLAG32( 3, 31, mkexpr(ge32), condT ); |
| 10431 | |
| 10432 | if (isT) |
| 10433 | putIRegT( regD, ire_result, condT ); |
| 10434 | else |
| 10435 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 10436 | |
| 10437 | DIP( "sasx%s r%u, r%u, r%u\n", nCC(conq), regD, regN, regM ); |
| 10438 | return True; |
| 10439 | } |
| 10440 | /* fall through */ |
| 10441 | } |
| 10442 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10443 | /* --------------- smuad, smuadx<c><Rd>,<Rn>,<Rm> --------------- */ |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10444 | /* --------------- smsad, smsadx<c><Rd>,<Rn>,<Rm> --------------- */ |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10445 | { |
| 10446 | UInt regD = 99, regN = 99, regM = 99, bitM = 99; |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10447 | Bool gate = False, isAD = False; |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10448 | |
| 10449 | if (isT) { |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10450 | if ((INSNT0(15,4) == 0xFB2 || INSNT0(15,4) == 0xFB4) |
| 10451 | && (INSNT1(15,0) & 0xF0E0) == 0xF000) { |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10452 | regN = INSNT0(3,0); |
| 10453 | regD = INSNT1(11,8); |
| 10454 | regM = INSNT1(3,0); |
| 10455 | bitM = INSNT1(4,4); |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10456 | isAD = INSNT0(15,4) == 0xFB2; |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10457 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 10458 | gate = True; |
| 10459 | } |
| 10460 | } else { |
| 10461 | if (INSNA(27,20) == BITS8(0,1,1,1,0,0,0,0) && |
| 10462 | INSNA(15,12) == BITS4(1,1,1,1) && |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10463 | (INSNA(7,4) & BITS4(1,0,0,1)) == BITS4(0,0,0,1) ) { |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10464 | regD = INSNA(19,16); |
| 10465 | regN = INSNA(3,0); |
| 10466 | regM = INSNA(11,8); |
| 10467 | bitM = INSNA(5,5); |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10468 | isAD = INSNA(6,6) == 0; |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10469 | if (regD != 15 && regN != 15 && regM != 15) |
| 10470 | gate = True; |
| 10471 | } |
| 10472 | } |
| 10473 | |
| 10474 | if (gate) { |
| 10475 | IRTemp irt_regN = newTemp(Ity_I32); |
| 10476 | IRTemp irt_regM = newTemp(Ity_I32); |
| 10477 | IRTemp irt_prod_lo = newTemp(Ity_I32); |
| 10478 | IRTemp irt_prod_hi = newTemp(Ity_I32); |
| 10479 | IRTemp tmpM = newTemp(Ity_I32); |
| 10480 | |
| 10481 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 10482 | |
| 10483 | assign( tmpM, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 10484 | assign( irt_regM, genROR32(tmpM, (bitM & 1) ? 16 : 0) ); |
| 10485 | |
| 10486 | assign( irt_prod_lo, |
| 10487 | binop( Iop_Mul32, |
| 10488 | binop( Iop_Sar32, |
| 10489 | binop(Iop_Shl32, mkexpr(irt_regN), mkU8(16)), |
| 10490 | mkU8(16) ), |
| 10491 | binop( Iop_Sar32, |
| 10492 | binop(Iop_Shl32, mkexpr(irt_regM), mkU8(16)), |
| 10493 | mkU8(16) ) ) ); |
| 10494 | assign( irt_prod_hi, binop(Iop_Mul32, |
| 10495 | binop(Iop_Sar32, mkexpr(irt_regN), mkU8(16)), |
| 10496 | binop(Iop_Sar32, mkexpr(irt_regM), mkU8(16))) ); |
| 10497 | IRExpr* ire_result |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10498 | = binop( isAD ? Iop_Add32 : Iop_Sub32, |
| 10499 | mkexpr(irt_prod_lo), mkexpr(irt_prod_hi) ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10500 | |
| 10501 | if (isT) |
| 10502 | putIRegT( regD, ire_result, condT ); |
| 10503 | else |
| 10504 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 10505 | |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10506 | if (isAD) { |
| 10507 | or_into_QFLAG32( |
| 10508 | signed_overflow_after_Add32( ire_result, |
| 10509 | irt_prod_lo, irt_prod_hi ), |
| 10510 | condT |
| 10511 | ); |
| 10512 | } |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10513 | |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10514 | DIP("smu%cd%s%s r%u, r%u, r%u\n", |
| 10515 | isAD ? 'a' : 's', |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10516 | bitM ? "x" : "", nCC(conq), regD, regN, regM); |
| 10517 | return True; |
| 10518 | } |
| 10519 | /* fall through */ |
| 10520 | } |
| 10521 | |
| 10522 | /* --------------- smlad{X}<c> <Rd>,<Rn>,<Rm>,<Ra> -------------- */ |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10523 | /* --------------- smlsd{X}<c> <Rd>,<Rn>,<Rm>,<Ra> -------------- */ |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10524 | { |
| 10525 | UInt regD = 99, regN = 99, regM = 99, regA = 99, bitM = 99; |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10526 | Bool gate = False, isAD = False; |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10527 | |
| 10528 | if (isT) { |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10529 | if ((INSNT0(15,4) == 0xFB2 || INSNT0(15,4) == 0xFB4) |
| 10530 | && INSNT1(7,5) == BITS3(0,0,0)) { |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10531 | regN = INSNT0(3,0); |
| 10532 | regD = INSNT1(11,8); |
| 10533 | regM = INSNT1(3,0); |
| 10534 | regA = INSNT1(15,12); |
| 10535 | bitM = INSNT1(4,4); |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10536 | isAD = INSNT0(15,4) == 0xFB2; |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10537 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM) |
| 10538 | && !isBadRegT(regA)) |
| 10539 | gate = True; |
| 10540 | } |
| 10541 | } else { |
| 10542 | if (INSNA(27,20) == BITS8(0,1,1,1,0,0,0,0) && |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10543 | (INSNA(7,4) & BITS4(1,0,0,1)) == BITS4(0,0,0,1)) { |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10544 | regD = INSNA(19,16); |
| 10545 | regA = INSNA(15,12); |
| 10546 | regN = INSNA(3,0); |
| 10547 | regM = INSNA(11,8); |
| 10548 | bitM = INSNA(5,5); |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10549 | isAD = INSNA(6,6) == 0; |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10550 | if (regD != 15 && regN != 15 && regM != 15 && regA != 15) |
| 10551 | gate = True; |
| 10552 | } |
| 10553 | } |
| 10554 | |
| 10555 | if (gate) { |
| 10556 | IRTemp irt_regN = newTemp(Ity_I32); |
| 10557 | IRTemp irt_regM = newTemp(Ity_I32); |
| 10558 | IRTemp irt_regA = newTemp(Ity_I32); |
| 10559 | IRTemp irt_prod_lo = newTemp(Ity_I32); |
| 10560 | IRTemp irt_prod_hi = newTemp(Ity_I32); |
| 10561 | IRTemp irt_sum = newTemp(Ity_I32); |
| 10562 | IRTemp tmpM = newTemp(Ity_I32); |
| 10563 | |
| 10564 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 10565 | assign( irt_regA, isT ? getIRegT(regA) : getIRegA(regA) ); |
| 10566 | |
| 10567 | assign( tmpM, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 10568 | assign( irt_regM, genROR32(tmpM, (bitM & 1) ? 16 : 0) ); |
| 10569 | |
| 10570 | assign( irt_prod_lo, |
| 10571 | binop(Iop_Mul32, |
| 10572 | binop(Iop_Sar32, |
| 10573 | binop( Iop_Shl32, mkexpr(irt_regN), mkU8(16) ), |
| 10574 | mkU8(16)), |
| 10575 | binop(Iop_Sar32, |
| 10576 | binop( Iop_Shl32, mkexpr(irt_regM), mkU8(16) ), |
| 10577 | mkU8(16))) ); |
| 10578 | assign( irt_prod_hi, |
| 10579 | binop( Iop_Mul32, |
| 10580 | binop( Iop_Sar32, mkexpr(irt_regN), mkU8(16) ), |
| 10581 | binop( Iop_Sar32, mkexpr(irt_regM), mkU8(16) ) ) ); |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10582 | assign( irt_sum, binop( isAD ? Iop_Add32 : Iop_Sub32, |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10583 | mkexpr(irt_prod_lo), mkexpr(irt_prod_hi) ) ); |
| 10584 | |
| 10585 | IRExpr* ire_result = binop(Iop_Add32, mkexpr(irt_sum), mkexpr(irt_regA)); |
| 10586 | |
| 10587 | if (isT) |
| 10588 | putIRegT( regD, ire_result, condT ); |
| 10589 | else |
| 10590 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 10591 | |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10592 | if (isAD) { |
| 10593 | or_into_QFLAG32( |
| 10594 | signed_overflow_after_Add32( mkexpr(irt_sum), |
| 10595 | irt_prod_lo, irt_prod_hi ), |
| 10596 | condT |
| 10597 | ); |
| 10598 | } |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10599 | |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10600 | or_into_QFLAG32( |
| 10601 | signed_overflow_after_Add32( ire_result, irt_sum, irt_regA ), |
| 10602 | condT |
| 10603 | ); |
| 10604 | |
| 10605 | DIP("sml%cd%s%s r%u, r%u, r%u, r%u\n", |
| 10606 | isAD ? 'a' : 's', |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10607 | bitM ? "x" : "", nCC(conq), regD, regN, regM, regA); |
| 10608 | return True; |
| 10609 | } |
| 10610 | /* fall through */ |
| 10611 | } |
| 10612 | |
| 10613 | /* ----- smlabb, smlabt, smlatb, smlatt <Rd>,<Rn>,<Rm>,<Ra> ----- */ |
| 10614 | { |
| 10615 | UInt regD = 99, regN = 99, regM = 99, regA = 99, bitM = 99, bitN = 99; |
| 10616 | Bool gate = False; |
| 10617 | |
| 10618 | if (isT) { |
| 10619 | if (INSNT0(15,4) == 0xFB1 && INSNT1(7,6) == BITS2(0,0)) { |
| 10620 | regN = INSNT0(3,0); |
| 10621 | regD = INSNT1(11,8); |
| 10622 | regM = INSNT1(3,0); |
| 10623 | regA = INSNT1(15,12); |
| 10624 | bitM = INSNT1(4,4); |
| 10625 | bitN = INSNT1(5,5); |
| 10626 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM) |
| 10627 | && !isBadRegT(regA)) |
| 10628 | gate = True; |
| 10629 | } |
| 10630 | } else { |
| 10631 | if (INSNA(27,20) == BITS8(0,0,0,1,0,0,0,0) && |
| 10632 | (INSNA(7,4) & BITS4(1,0,0,1)) == BITS4(1,0,0,0)) { |
| 10633 | regD = INSNA(19,16); |
| 10634 | regN = INSNA(3,0); |
| 10635 | regM = INSNA(11,8); |
| 10636 | regA = INSNA(15,12); |
| 10637 | bitM = INSNA(6,6); |
| 10638 | bitN = INSNA(5,5); |
| 10639 | if (regD != 15 && regN != 15 && regM != 15 && regA != 15) |
| 10640 | gate = True; |
| 10641 | } |
| 10642 | } |
| 10643 | |
| 10644 | if (gate) { |
| 10645 | IRTemp irt_regA = newTemp(Ity_I32); |
| 10646 | IRTemp irt_prod = newTemp(Ity_I32); |
| 10647 | |
| 10648 | assign( irt_prod, |
| 10649 | binop(Iop_Mul32, |
| 10650 | binop(Iop_Sar32, |
| 10651 | binop(Iop_Shl32, |
| 10652 | isT ? getIRegT(regN) : getIRegA(regN), |
| 10653 | mkU8(bitN ? 0 : 16)), |
| 10654 | mkU8(16)), |
| 10655 | binop(Iop_Sar32, |
| 10656 | binop(Iop_Shl32, |
| 10657 | isT ? getIRegT(regM) : getIRegA(regM), |
| 10658 | mkU8(bitM ? 0 : 16)), |
| 10659 | mkU8(16))) ); |
| 10660 | |
| 10661 | assign( irt_regA, isT ? getIRegT(regA) : getIRegA(regA) ); |
| 10662 | |
| 10663 | IRExpr* ire_result = binop(Iop_Add32, mkexpr(irt_prod), mkexpr(irt_regA)); |
| 10664 | |
| 10665 | if (isT) |
| 10666 | putIRegT( regD, ire_result, condT ); |
| 10667 | else |
| 10668 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 10669 | |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10670 | or_into_QFLAG32( |
| 10671 | signed_overflow_after_Add32( ire_result, irt_prod, irt_regA ), |
| 10672 | condT |
| 10673 | ); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10674 | |
| 10675 | DIP( "smla%c%c%s r%u, r%u, r%u, r%u\n", |
| 10676 | bitN ? 't' : 'b', bitM ? 't' : 'b', |
| 10677 | nCC(conq), regD, regN, regM, regA ); |
| 10678 | return True; |
| 10679 | } |
| 10680 | /* fall through */ |
| 10681 | } |
| 10682 | |
sewardj | 44db1e7 | 2013-03-04 18:33:56 +0000 | [diff] [blame] | 10683 | /* ----- smlalbb, smlalbt, smlaltb, smlaltt <Rd>,<Rn>,<Rm>,<Ra> ----- */ |
| 10684 | { |
| 10685 | UInt regDHi = 99, regN = 99, regM = 99, regDLo = 99, bitM = 99, bitN = 99; |
| 10686 | Bool gate = False; |
| 10687 | |
| 10688 | if (isT) { |
| 10689 | if (INSNT0(15,4) == 0xFBC && INSNT1(7,6) == BITS2(1,0)) { |
| 10690 | regN = INSNT0(3,0); |
| 10691 | regDHi = INSNT1(11,8); |
| 10692 | regM = INSNT1(3,0); |
| 10693 | regDLo = INSNT1(15,12); |
| 10694 | bitM = INSNT1(4,4); |
| 10695 | bitN = INSNT1(5,5); |
| 10696 | if (!isBadRegT(regDHi) && !isBadRegT(regN) && !isBadRegT(regM) |
| 10697 | && !isBadRegT(regDLo) && regDHi != regDLo) |
| 10698 | gate = True; |
| 10699 | } |
| 10700 | } else { |
| 10701 | if (INSNA(27,20) == BITS8(0,0,0,1,0,1,0,0) && |
| 10702 | (INSNA(7,4) & BITS4(1,0,0,1)) == BITS4(1,0,0,0)) { |
| 10703 | regDHi = INSNA(19,16); |
| 10704 | regN = INSNA(3,0); |
| 10705 | regM = INSNA(11,8); |
| 10706 | regDLo = INSNA(15,12); |
| 10707 | bitM = INSNA(6,6); |
| 10708 | bitN = INSNA(5,5); |
| 10709 | if (regDHi != 15 && regN != 15 && regM != 15 && regDLo != 15 && |
| 10710 | regDHi != regDLo) |
| 10711 | gate = True; |
| 10712 | } |
| 10713 | } |
| 10714 | |
| 10715 | if (gate) { |
| 10716 | IRTemp irt_regD = newTemp(Ity_I64); |
| 10717 | IRTemp irt_prod = newTemp(Ity_I64); |
| 10718 | IRTemp irt_res = newTemp(Ity_I64); |
| 10719 | IRTemp irt_resHi = newTemp(Ity_I32); |
| 10720 | IRTemp irt_resLo = newTemp(Ity_I32); |
| 10721 | |
| 10722 | assign( irt_prod, |
| 10723 | binop(Iop_MullS32, |
| 10724 | binop(Iop_Sar32, |
| 10725 | binop(Iop_Shl32, |
| 10726 | isT ? getIRegT(regN) : getIRegA(regN), |
| 10727 | mkU8(bitN ? 0 : 16)), |
| 10728 | mkU8(16)), |
| 10729 | binop(Iop_Sar32, |
| 10730 | binop(Iop_Shl32, |
| 10731 | isT ? getIRegT(regM) : getIRegA(regM), |
| 10732 | mkU8(bitM ? 0 : 16)), |
| 10733 | mkU8(16))) ); |
| 10734 | |
| 10735 | assign( irt_regD, binop(Iop_32HLto64, |
| 10736 | isT ? getIRegT(regDHi) : getIRegA(regDHi), |
| 10737 | isT ? getIRegT(regDLo) : getIRegA(regDLo)) ); |
| 10738 | assign( irt_res, binop(Iop_Add64, mkexpr(irt_regD), mkexpr(irt_prod)) ); |
| 10739 | assign( irt_resHi, unop(Iop_64HIto32, mkexpr(irt_res)) ); |
| 10740 | assign( irt_resLo, unop(Iop_64to32, mkexpr(irt_res)) ); |
| 10741 | |
| 10742 | if (isT) { |
| 10743 | putIRegT( regDHi, mkexpr(irt_resHi), condT ); |
| 10744 | putIRegT( regDLo, mkexpr(irt_resLo), condT ); |
| 10745 | } else { |
| 10746 | putIRegA( regDHi, mkexpr(irt_resHi), condT, Ijk_Boring ); |
| 10747 | putIRegA( regDLo, mkexpr(irt_resLo), condT, Ijk_Boring ); |
| 10748 | } |
| 10749 | |
| 10750 | DIP( "smlal%c%c%s r%u, r%u, r%u, r%u\n", |
| 10751 | bitN ? 't' : 'b', bitM ? 't' : 'b', |
| 10752 | nCC(conq), regDHi, regN, regM, regDLo ); |
| 10753 | return True; |
| 10754 | } |
| 10755 | /* fall through */ |
| 10756 | } |
| 10757 | |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 10758 | /* ----- smlawb, smlawt <Rd>,<Rn>,<Rm>,<Ra> ----- */ |
| 10759 | { |
| 10760 | UInt regD = 99, regN = 99, regM = 99, regA = 99, bitM = 99; |
| 10761 | Bool gate = False; |
| 10762 | |
| 10763 | if (isT) { |
| 10764 | if (INSNT0(15,4) == 0xFB3 && INSNT1(7,5) == BITS3(0,0,0)) { |
| 10765 | regN = INSNT0(3,0); |
| 10766 | regD = INSNT1(11,8); |
| 10767 | regM = INSNT1(3,0); |
| 10768 | regA = INSNT1(15,12); |
| 10769 | bitM = INSNT1(4,4); |
| 10770 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM) |
| 10771 | && !isBadRegT(regA)) |
| 10772 | gate = True; |
| 10773 | } |
| 10774 | } else { |
| 10775 | if (INSNA(27,20) == BITS8(0,0,0,1,0,0,1,0) && |
| 10776 | (INSNA(7,4) & BITS4(1,0,1,1)) == BITS4(1,0,0,0)) { |
| 10777 | regD = INSNA(19,16); |
| 10778 | regN = INSNA(3,0); |
| 10779 | regM = INSNA(11,8); |
| 10780 | regA = INSNA(15,12); |
| 10781 | bitM = INSNA(6,6); |
| 10782 | if (regD != 15 && regN != 15 && regM != 15 && regA != 15) |
| 10783 | gate = True; |
| 10784 | } |
| 10785 | } |
| 10786 | |
| 10787 | if (gate) { |
| 10788 | IRTemp irt_regA = newTemp(Ity_I32); |
| 10789 | IRTemp irt_prod = newTemp(Ity_I64); |
| 10790 | |
| 10791 | assign( irt_prod, |
| 10792 | binop(Iop_MullS32, |
| 10793 | isT ? getIRegT(regN) : getIRegA(regN), |
| 10794 | binop(Iop_Sar32, |
| 10795 | binop(Iop_Shl32, |
| 10796 | isT ? getIRegT(regM) : getIRegA(regM), |
| 10797 | mkU8(bitM ? 0 : 16)), |
| 10798 | mkU8(16))) ); |
| 10799 | |
| 10800 | assign( irt_regA, isT ? getIRegT(regA) : getIRegA(regA) ); |
| 10801 | |
| 10802 | IRTemp prod32 = newTemp(Ity_I32); |
| 10803 | assign(prod32, |
| 10804 | binop(Iop_Or32, |
| 10805 | binop(Iop_Shl32, unop(Iop_64HIto32, mkexpr(irt_prod)), mkU8(16)), |
| 10806 | binop(Iop_Shr32, unop(Iop_64to32, mkexpr(irt_prod)), mkU8(16)) |
| 10807 | )); |
| 10808 | |
| 10809 | IRExpr* ire_result = binop(Iop_Add32, mkexpr(prod32), mkexpr(irt_regA)); |
| 10810 | |
| 10811 | if (isT) |
| 10812 | putIRegT( regD, ire_result, condT ); |
| 10813 | else |
| 10814 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 10815 | |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10816 | or_into_QFLAG32( |
| 10817 | signed_overflow_after_Add32( ire_result, prod32, irt_regA ), |
| 10818 | condT |
| 10819 | ); |
sewardj | 22ac596 | 2010-09-20 22:35:35 +0000 | [diff] [blame] | 10820 | |
| 10821 | DIP( "smlaw%c%s r%u, r%u, r%u, r%u\n", |
| 10822 | bitM ? 't' : 'b', |
| 10823 | nCC(conq), regD, regN, regM, regA ); |
| 10824 | return True; |
| 10825 | } |
| 10826 | /* fall through */ |
| 10827 | } |
| 10828 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10829 | /* ------------------- sel<c> <Rd>,<Rn>,<Rm> -------------------- */ |
| 10830 | /* fixme: fix up the test in v6media.c so that we can pass the ge |
| 10831 | flags as part of the test. */ |
| 10832 | { |
| 10833 | UInt regD = 99, regN = 99, regM = 99; |
| 10834 | Bool gate = False; |
| 10835 | |
| 10836 | if (isT) { |
| 10837 | if (INSNT0(15,4) == 0xFAA && (INSNT1(15,0) & 0xF0F0) == 0xF080) { |
| 10838 | regN = INSNT0(3,0); |
| 10839 | regD = INSNT1(11,8); |
| 10840 | regM = INSNT1(3,0); |
| 10841 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 10842 | gate = True; |
| 10843 | } |
| 10844 | } else { |
| 10845 | if (INSNA(27,20) == BITS8(0,1,1,0,1,0,0,0) && |
| 10846 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 10847 | INSNA(7,4) == BITS4(1,0,1,1)) { |
| 10848 | regD = INSNA(15,12); |
| 10849 | regN = INSNA(19,16); |
| 10850 | regM = INSNA(3,0); |
| 10851 | if (regD != 15 && regN != 15 && regM != 15) |
| 10852 | gate = True; |
| 10853 | } |
| 10854 | } |
| 10855 | |
| 10856 | if (gate) { |
| 10857 | IRTemp irt_ge_flag0 = newTemp(Ity_I32); |
| 10858 | IRTemp irt_ge_flag1 = newTemp(Ity_I32); |
| 10859 | IRTemp irt_ge_flag2 = newTemp(Ity_I32); |
| 10860 | IRTemp irt_ge_flag3 = newTemp(Ity_I32); |
| 10861 | |
| 10862 | assign( irt_ge_flag0, get_GEFLAG32(0) ); |
| 10863 | assign( irt_ge_flag1, get_GEFLAG32(1) ); |
| 10864 | assign( irt_ge_flag2, get_GEFLAG32(2) ); |
| 10865 | assign( irt_ge_flag3, get_GEFLAG32(3) ); |
| 10866 | |
| 10867 | IRExpr* ire_ge_flag0_or |
| 10868 | = binop(Iop_Or32, mkexpr(irt_ge_flag0), |
| 10869 | binop(Iop_Sub32, mkU32(0), mkexpr(irt_ge_flag0))); |
| 10870 | IRExpr* ire_ge_flag1_or |
| 10871 | = binop(Iop_Or32, mkexpr(irt_ge_flag1), |
| 10872 | binop(Iop_Sub32, mkU32(0), mkexpr(irt_ge_flag1))); |
| 10873 | IRExpr* ire_ge_flag2_or |
| 10874 | = binop(Iop_Or32, mkexpr(irt_ge_flag2), |
| 10875 | binop(Iop_Sub32, mkU32(0), mkexpr(irt_ge_flag2))); |
| 10876 | IRExpr* ire_ge_flag3_or |
| 10877 | = binop(Iop_Or32, mkexpr(irt_ge_flag3), |
| 10878 | binop(Iop_Sub32, mkU32(0), mkexpr(irt_ge_flag3))); |
| 10879 | |
| 10880 | IRExpr* ire_ge_flags |
| 10881 | = binop( Iop_Or32, |
| 10882 | binop(Iop_Or32, |
| 10883 | binop(Iop_And32, |
| 10884 | binop(Iop_Sar32, ire_ge_flag0_or, mkU8(31)), |
| 10885 | mkU32(0x000000ff)), |
| 10886 | binop(Iop_And32, |
| 10887 | binop(Iop_Sar32, ire_ge_flag1_or, mkU8(31)), |
| 10888 | mkU32(0x0000ff00))), |
| 10889 | binop(Iop_Or32, |
| 10890 | binop(Iop_And32, |
| 10891 | binop(Iop_Sar32, ire_ge_flag2_or, mkU8(31)), |
| 10892 | mkU32(0x00ff0000)), |
| 10893 | binop(Iop_And32, |
| 10894 | binop(Iop_Sar32, ire_ge_flag3_or, mkU8(31)), |
| 10895 | mkU32(0xff000000))) ); |
| 10896 | |
| 10897 | IRExpr* ire_result |
| 10898 | = binop(Iop_Or32, |
| 10899 | binop(Iop_And32, |
| 10900 | isT ? getIRegT(regN) : getIRegA(regN), |
| 10901 | ire_ge_flags ), |
| 10902 | binop(Iop_And32, |
| 10903 | isT ? getIRegT(regM) : getIRegA(regM), |
| 10904 | unop(Iop_Not32, ire_ge_flags))); |
| 10905 | |
| 10906 | if (isT) |
| 10907 | putIRegT( regD, ire_result, condT ); |
| 10908 | else |
| 10909 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 10910 | |
| 10911 | DIP("sel%s r%u, r%u, r%u\n", nCC(conq), regD, regN, regM ); |
| 10912 | return True; |
| 10913 | } |
| 10914 | /* fall through */ |
| 10915 | } |
| 10916 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 10917 | /* ----------------- uxtab16<c> Rd,Rn,Rm{,rot} ------------------ */ |
| 10918 | { |
| 10919 | UInt regD = 99, regN = 99, regM = 99, rotate = 99; |
| 10920 | Bool gate = False; |
| 10921 | |
| 10922 | if (isT) { |
| 10923 | if (INSNT0(15,4) == 0xFA3 && (INSNT1(15,0) & 0xF0C0) == 0xF080) { |
| 10924 | regN = INSNT0(3,0); |
| 10925 | regD = INSNT1(11,8); |
| 10926 | regM = INSNT1(3,0); |
| 10927 | rotate = INSNT1(5,4); |
| 10928 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 10929 | gate = True; |
| 10930 | } |
| 10931 | } else { |
| 10932 | if (INSNA(27,20) == BITS8(0,1,1,0,1,1,0,0) && |
| 10933 | INSNA(9,4) == BITS6(0,0,0,1,1,1) ) { |
| 10934 | regD = INSNA(15,12); |
| 10935 | regN = INSNA(19,16); |
| 10936 | regM = INSNA(3,0); |
| 10937 | rotate = INSNA(11,10); |
| 10938 | if (regD != 15 && regN != 15 && regM != 15) |
| 10939 | gate = True; |
| 10940 | } |
| 10941 | } |
| 10942 | |
| 10943 | if (gate) { |
| 10944 | IRTemp irt_regN = newTemp(Ity_I32); |
| 10945 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 10946 | |
| 10947 | IRTemp irt_regM = newTemp(Ity_I32); |
| 10948 | assign( irt_regM, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 10949 | |
| 10950 | IRTemp irt_rot = newTemp(Ity_I32); |
| 10951 | assign( irt_rot, binop(Iop_And32, |
| 10952 | genROR32(irt_regM, 8 * rotate), |
| 10953 | mkU32(0x00FF00FF)) ); |
| 10954 | |
| 10955 | IRExpr* resLo |
| 10956 | = binop(Iop_And32, |
| 10957 | binop(Iop_Add32, mkexpr(irt_regN), mkexpr(irt_rot)), |
| 10958 | mkU32(0x0000FFFF)); |
| 10959 | |
| 10960 | IRExpr* resHi |
| 10961 | = binop(Iop_Add32, |
| 10962 | binop(Iop_And32, mkexpr(irt_regN), mkU32(0xFFFF0000)), |
| 10963 | binop(Iop_And32, mkexpr(irt_rot), mkU32(0xFFFF0000))); |
| 10964 | |
| 10965 | IRExpr* ire_result |
| 10966 | = binop( Iop_Or32, resHi, resLo ); |
| 10967 | |
| 10968 | if (isT) |
| 10969 | putIRegT( regD, ire_result, condT ); |
| 10970 | else |
| 10971 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 10972 | |
| 10973 | DIP( "uxtab16%s r%u, r%u, r%u, ROR #%u\n", |
| 10974 | nCC(conq), regD, regN, regM, 8 * rotate ); |
| 10975 | return True; |
| 10976 | } |
| 10977 | /* fall through */ |
| 10978 | } |
| 10979 | |
sewardj | 310d6b2 | 2010-10-18 16:29:40 +0000 | [diff] [blame] | 10980 | /* --------------- usad8 Rd,Rn,Rm ---------------- */ |
| 10981 | /* --------------- usada8 Rd,Rn,Rm,Ra ---------------- */ |
| 10982 | { |
| 10983 | UInt rD = 99, rN = 99, rM = 99, rA = 99; |
| 10984 | Bool gate = False; |
| 10985 | |
| 10986 | if (isT) { |
| 10987 | if (INSNT0(15,4) == 0xFB7 && INSNT1(7,4) == BITS4(0,0,0,0)) { |
| 10988 | rN = INSNT0(3,0); |
| 10989 | rA = INSNT1(15,12); |
| 10990 | rD = INSNT1(11,8); |
| 10991 | rM = INSNT1(3,0); |
| 10992 | if (!isBadRegT(rD) && !isBadRegT(rN) && !isBadRegT(rM) && rA != 13) |
| 10993 | gate = True; |
| 10994 | } |
| 10995 | } else { |
| 10996 | if (INSNA(27,20) == BITS8(0,1,1,1,1,0,0,0) && |
| 10997 | INSNA(7,4) == BITS4(0,0,0,1) ) { |
| 10998 | rD = INSNA(19,16); |
| 10999 | rA = INSNA(15,12); |
| 11000 | rM = INSNA(11,8); |
| 11001 | rN = INSNA(3,0); |
| 11002 | if (rD != 15 && rN != 15 && rM != 15 /* but rA can be 15 */) |
| 11003 | gate = True; |
| 11004 | } |
| 11005 | } |
| 11006 | /* We allow rA == 15, to denote the usad8 (no accumulator) case. */ |
| 11007 | |
| 11008 | if (gate) { |
| 11009 | IRExpr* rNe = isT ? getIRegT(rN) : getIRegA(rN); |
| 11010 | IRExpr* rMe = isT ? getIRegT(rM) : getIRegA(rM); |
| 11011 | IRExpr* rAe = rA == 15 ? mkU32(0) |
| 11012 | : (isT ? getIRegT(rA) : getIRegA(rA)); |
| 11013 | IRExpr* res = binop(Iop_Add32, |
| 11014 | binop(Iop_Sad8Ux4, rNe, rMe), |
| 11015 | rAe); |
| 11016 | if (isT) |
| 11017 | putIRegT( rD, res, condT ); |
| 11018 | else |
| 11019 | putIRegA( rD, res, condT, Ijk_Boring ); |
| 11020 | |
| 11021 | if (rA == 15) { |
| 11022 | DIP( "usad8%s r%u, r%u, r%u\n", |
| 11023 | nCC(conq), rD, rN, rM ); |
| 11024 | } else { |
| 11025 | DIP( "usada8%s r%u, r%u, r%u, r%u\n", |
| 11026 | nCC(conq), rD, rN, rM, rA ); |
| 11027 | } |
| 11028 | return True; |
| 11029 | } |
| 11030 | /* fall through */ |
| 11031 | } |
| 11032 | |
sewardj | 44ce46d | 2012-07-11 13:19:10 +0000 | [diff] [blame] | 11033 | /* ------------------ qadd<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 11034 | { |
| 11035 | UInt regD = 99, regN = 99, regM = 99; |
| 11036 | Bool gate = False; |
| 11037 | |
| 11038 | if (isT) { |
| 11039 | if (INSNT0(15,4) == 0xFA8 && (INSNT1(15,0) & 0xF0F0) == 0xF080) { |
| 11040 | regN = INSNT0(3,0); |
| 11041 | regD = INSNT1(11,8); |
| 11042 | regM = INSNT1(3,0); |
| 11043 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11044 | gate = True; |
| 11045 | } |
| 11046 | } else { |
| 11047 | if (INSNA(27,20) == BITS8(0,0,0,1,0,0,0,0) && |
| 11048 | INSNA(11,8) == BITS4(0,0,0,0) && |
| 11049 | INSNA(7,4) == BITS4(0,1,0,1)) { |
| 11050 | regD = INSNA(15,12); |
| 11051 | regN = INSNA(19,16); |
| 11052 | regM = INSNA(3,0); |
| 11053 | if (regD != 15 && regN != 15 && regM != 15) |
| 11054 | gate = True; |
| 11055 | } |
| 11056 | } |
| 11057 | |
| 11058 | if (gate) { |
| 11059 | IRTemp rNt = newTemp(Ity_I32); |
| 11060 | IRTemp rMt = newTemp(Ity_I32); |
| 11061 | IRTemp res_q = newTemp(Ity_I32); |
| 11062 | |
| 11063 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11064 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11065 | |
| 11066 | assign(res_q, binop(Iop_QAdd32S, mkexpr(rMt), mkexpr(rNt))); |
| 11067 | if (isT) |
| 11068 | putIRegT( regD, mkexpr(res_q), condT ); |
| 11069 | else |
| 11070 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 11071 | |
| 11072 | or_into_QFLAG32( |
| 11073 | signed_overflow_after_Add32( |
| 11074 | binop(Iop_Add32, mkexpr(rMt), mkexpr(rNt)), rMt, rNt), |
| 11075 | condT |
| 11076 | ); |
| 11077 | |
| 11078 | DIP("qadd%s r%u, r%u, r%u\n", nCC(conq),regD,regM,regN); |
| 11079 | return True; |
| 11080 | } |
| 11081 | /* fall through */ |
| 11082 | } |
| 11083 | |
sewardj | 5a34b8b | 2012-08-23 20:56:17 +0000 | [diff] [blame] | 11084 | /* ------------------ qdadd<c> <Rd>,<Rm>,<Rn> ------------------- */ |
| 11085 | { |
| 11086 | UInt regD = 99, regN = 99, regM = 99; |
| 11087 | Bool gate = False; |
| 11088 | |
| 11089 | if (isT) { |
| 11090 | if (INSNT0(15,4) == 0xFA8 && (INSNT1(15,0) & 0xF0F0) == 0xF090) { |
| 11091 | regN = INSNT0(3,0); |
| 11092 | regD = INSNT1(11,8); |
| 11093 | regM = INSNT1(3,0); |
| 11094 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11095 | gate = True; |
| 11096 | } |
| 11097 | } else { |
| 11098 | if (INSNA(27,20) == BITS8(0,0,0,1,0,1,0,0) && |
| 11099 | INSNA(11,8) == BITS4(0,0,0,0) && |
| 11100 | INSNA(7,4) == BITS4(0,1,0,1)) { |
| 11101 | regD = INSNA(15,12); |
| 11102 | regN = INSNA(19,16); |
| 11103 | regM = INSNA(3,0); |
| 11104 | if (regD != 15 && regN != 15 && regM != 15) |
| 11105 | gate = True; |
| 11106 | } |
| 11107 | } |
| 11108 | |
| 11109 | if (gate) { |
| 11110 | IRTemp rNt = newTemp(Ity_I32); |
| 11111 | IRTemp rMt = newTemp(Ity_I32); |
| 11112 | IRTemp rN_d = newTemp(Ity_I32); |
| 11113 | IRTemp res_q = newTemp(Ity_I32); |
| 11114 | |
| 11115 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11116 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11117 | |
| 11118 | or_into_QFLAG32( |
| 11119 | signed_overflow_after_Add32( |
| 11120 | binop(Iop_Add32, mkexpr(rNt), mkexpr(rNt)), rNt, rNt), |
| 11121 | condT |
| 11122 | ); |
| 11123 | |
| 11124 | assign(rN_d, binop(Iop_QAdd32S, mkexpr(rNt), mkexpr(rNt))); |
| 11125 | assign(res_q, binop(Iop_QAdd32S, mkexpr(rMt), mkexpr(rN_d))); |
| 11126 | if (isT) |
| 11127 | putIRegT( regD, mkexpr(res_q), condT ); |
| 11128 | else |
| 11129 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 11130 | |
| 11131 | or_into_QFLAG32( |
| 11132 | signed_overflow_after_Add32( |
| 11133 | binop(Iop_Add32, mkexpr(rMt), mkexpr(rN_d)), rMt, rN_d), |
| 11134 | condT |
| 11135 | ); |
| 11136 | |
| 11137 | DIP("qdadd%s r%u, r%u, r%u\n", nCC(conq),regD,regM,regN); |
| 11138 | return True; |
| 11139 | } |
| 11140 | /* fall through */ |
| 11141 | } |
| 11142 | |
sewardj | 44ce46d | 2012-07-11 13:19:10 +0000 | [diff] [blame] | 11143 | /* ------------------ qsub<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 11144 | { |
| 11145 | UInt regD = 99, regN = 99, regM = 99; |
| 11146 | Bool gate = False; |
| 11147 | |
| 11148 | if (isT) { |
| 11149 | if (INSNT0(15,4) == 0xFA8 && (INSNT1(15,0) & 0xF0F0) == 0xF0A0) { |
| 11150 | regN = INSNT0(3,0); |
| 11151 | regD = INSNT1(11,8); |
| 11152 | regM = INSNT1(3,0); |
| 11153 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11154 | gate = True; |
| 11155 | } |
| 11156 | } else { |
| 11157 | if (INSNA(27,20) == BITS8(0,0,0,1,0,0,1,0) && |
| 11158 | INSNA(11,8) == BITS4(0,0,0,0) && |
| 11159 | INSNA(7,4) == BITS4(0,1,0,1)) { |
| 11160 | regD = INSNA(15,12); |
| 11161 | regN = INSNA(19,16); |
| 11162 | regM = INSNA(3,0); |
| 11163 | if (regD != 15 && regN != 15 && regM != 15) |
| 11164 | gate = True; |
| 11165 | } |
| 11166 | } |
| 11167 | |
| 11168 | if (gate) { |
| 11169 | IRTemp rNt = newTemp(Ity_I32); |
| 11170 | IRTemp rMt = newTemp(Ity_I32); |
| 11171 | IRTemp res_q = newTemp(Ity_I32); |
| 11172 | |
| 11173 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11174 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11175 | |
| 11176 | assign(res_q, binop(Iop_QSub32S, mkexpr(rMt), mkexpr(rNt))); |
| 11177 | if (isT) |
| 11178 | putIRegT( regD, mkexpr(res_q), condT ); |
| 11179 | else |
| 11180 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 11181 | |
| 11182 | or_into_QFLAG32( |
| 11183 | signed_overflow_after_Sub32( |
| 11184 | binop(Iop_Sub32, mkexpr(rMt), mkexpr(rNt)), rMt, rNt), |
| 11185 | condT |
| 11186 | ); |
| 11187 | |
| 11188 | DIP("qsub%s r%u, r%u, r%u\n", nCC(conq),regD,regM,regN); |
| 11189 | return True; |
| 11190 | } |
| 11191 | /* fall through */ |
| 11192 | } |
| 11193 | |
sewardj | 5a34b8b | 2012-08-23 20:56:17 +0000 | [diff] [blame] | 11194 | /* ------------------ qdsub<c> <Rd>,<Rm>,<Rn> ------------------- */ |
| 11195 | { |
| 11196 | UInt regD = 99, regN = 99, regM = 99; |
| 11197 | Bool gate = False; |
| 11198 | |
| 11199 | if (isT) { |
| 11200 | if (INSNT0(15,4) == 0xFA8 && (INSNT1(15,0) & 0xF0F0) == 0xF0B0) { |
| 11201 | regN = INSNT0(3,0); |
| 11202 | regD = INSNT1(11,8); |
| 11203 | regM = INSNT1(3,0); |
| 11204 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11205 | gate = True; |
| 11206 | } |
| 11207 | } else { |
| 11208 | if (INSNA(27,20) == BITS8(0,0,0,1,0,1,1,0) && |
| 11209 | INSNA(11,8) == BITS4(0,0,0,0) && |
| 11210 | INSNA(7,4) == BITS4(0,1,0,1)) { |
| 11211 | regD = INSNA(15,12); |
| 11212 | regN = INSNA(19,16); |
| 11213 | regM = INSNA(3,0); |
| 11214 | if (regD != 15 && regN != 15 && regM != 15) |
| 11215 | gate = True; |
| 11216 | } |
| 11217 | } |
| 11218 | |
| 11219 | if (gate) { |
| 11220 | IRTemp rNt = newTemp(Ity_I32); |
| 11221 | IRTemp rMt = newTemp(Ity_I32); |
| 11222 | IRTemp rN_d = newTemp(Ity_I32); |
| 11223 | IRTemp res_q = newTemp(Ity_I32); |
| 11224 | |
| 11225 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11226 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11227 | |
| 11228 | or_into_QFLAG32( |
| 11229 | signed_overflow_after_Add32( |
| 11230 | binop(Iop_Add32, mkexpr(rNt), mkexpr(rNt)), rNt, rNt), |
| 11231 | condT |
| 11232 | ); |
| 11233 | |
| 11234 | assign(rN_d, binop(Iop_QAdd32S, mkexpr(rNt), mkexpr(rNt))); |
| 11235 | assign(res_q, binop(Iop_QSub32S, mkexpr(rMt), mkexpr(rN_d))); |
| 11236 | if (isT) |
| 11237 | putIRegT( regD, mkexpr(res_q), condT ); |
| 11238 | else |
| 11239 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 11240 | |
| 11241 | or_into_QFLAG32( |
| 11242 | signed_overflow_after_Sub32( |
| 11243 | binop(Iop_Sub32, mkexpr(rMt), mkexpr(rN_d)), rMt, rN_d), |
| 11244 | condT |
| 11245 | ); |
| 11246 | |
| 11247 | DIP("qdsub%s r%u, r%u, r%u\n", nCC(conq),regD,regM,regN); |
| 11248 | return True; |
| 11249 | } |
| 11250 | /* fall through */ |
| 11251 | } |
| 11252 | |
sewardj | 32dd538 | 2012-09-17 15:27:58 +0000 | [diff] [blame] | 11253 | /* ------------------ uqsub16<c> <Rd>,<Rn>,<Rm> ------------------ */ |
| 11254 | { |
| 11255 | UInt regD = 99, regN = 99, regM = 99; |
| 11256 | Bool gate = False; |
| 11257 | |
| 11258 | if (isT) { |
| 11259 | if (INSNT0(15,4) == 0xFAD && (INSNT1(15,0) & 0xF0F0) == 0xF050) { |
| 11260 | regN = INSNT0(3,0); |
| 11261 | regD = INSNT1(11,8); |
| 11262 | regM = INSNT1(3,0); |
| 11263 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11264 | gate = True; |
| 11265 | } |
| 11266 | } else { |
| 11267 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,1,0) && |
| 11268 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 11269 | INSNA(7,4) == BITS4(0,1,1,1)) { |
| 11270 | regD = INSNA(15,12); |
| 11271 | regN = INSNA(19,16); |
| 11272 | regM = INSNA(3,0); |
| 11273 | if (regD != 15 && regN != 15 && regM != 15) |
| 11274 | gate = True; |
| 11275 | } |
| 11276 | } |
| 11277 | |
| 11278 | if (gate) { |
| 11279 | IRTemp rNt = newTemp(Ity_I32); |
| 11280 | IRTemp rMt = newTemp(Ity_I32); |
| 11281 | IRTemp res_q = newTemp(Ity_I32); |
| 11282 | |
| 11283 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11284 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11285 | |
| 11286 | assign(res_q, binop(Iop_QSub16Ux2, mkexpr(rNt), mkexpr(rMt))); |
| 11287 | if (isT) |
| 11288 | putIRegT( regD, mkexpr(res_q), condT ); |
| 11289 | else |
| 11290 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 11291 | |
| 11292 | DIP("uqsub16%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 11293 | return True; |
| 11294 | } |
| 11295 | /* fall through */ |
| 11296 | } |
| 11297 | |
| 11298 | /* ----------------- shadd16<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 11299 | { |
| 11300 | UInt regD = 99, regN = 99, regM = 99; |
| 11301 | Bool gate = False; |
| 11302 | |
| 11303 | if (isT) { |
| 11304 | if (INSNT0(15,4) == 0xFA9 && (INSNT1(15,0) & 0xF0F0) == 0xF020) { |
| 11305 | regN = INSNT0(3,0); |
| 11306 | regD = INSNT1(11,8); |
| 11307 | regM = INSNT1(3,0); |
| 11308 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11309 | gate = True; |
| 11310 | } |
| 11311 | } else { |
| 11312 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,1,1) && |
| 11313 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 11314 | INSNA(7,4) == BITS4(0,0,0,1)) { |
| 11315 | regD = INSNA(15,12); |
| 11316 | regN = INSNA(19,16); |
| 11317 | regM = INSNA(3,0); |
| 11318 | if (regD != 15 && regN != 15 && regM != 15) |
| 11319 | gate = True; |
| 11320 | } |
| 11321 | } |
| 11322 | |
| 11323 | if (gate) { |
| 11324 | IRTemp rNt = newTemp(Ity_I32); |
| 11325 | IRTemp rMt = newTemp(Ity_I32); |
| 11326 | IRTemp res_q = newTemp(Ity_I32); |
| 11327 | |
| 11328 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11329 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11330 | |
| 11331 | assign(res_q, binop(Iop_HAdd16Sx2, mkexpr(rNt), mkexpr(rMt))); |
| 11332 | if (isT) |
| 11333 | putIRegT( regD, mkexpr(res_q), condT ); |
| 11334 | else |
| 11335 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 11336 | |
| 11337 | DIP("shadd16%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 11338 | return True; |
| 11339 | } |
| 11340 | /* fall through */ |
| 11341 | } |
| 11342 | |
| 11343 | /* ----------------- uhsub8<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 11344 | { |
| 11345 | UInt regD = 99, regN = 99, regM = 99; |
| 11346 | Bool gate = False; |
| 11347 | |
| 11348 | if (isT) { |
| 11349 | if (INSNT0(15,4) == 0xFAC && (INSNT1(15,0) & 0xF0F0) == 0xF060) { |
| 11350 | regN = INSNT0(3,0); |
| 11351 | regD = INSNT1(11,8); |
| 11352 | regM = INSNT1(3,0); |
| 11353 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11354 | gate = True; |
| 11355 | } |
| 11356 | } else { |
| 11357 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,1,1) && |
| 11358 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 11359 | INSNA(7,4) == BITS4(1,1,1,1)) { |
| 11360 | regD = INSNA(15,12); |
| 11361 | regN = INSNA(19,16); |
| 11362 | regM = INSNA(3,0); |
| 11363 | if (regD != 15 && regN != 15 && regM != 15) |
| 11364 | gate = True; |
| 11365 | } |
| 11366 | } |
| 11367 | |
| 11368 | if (gate) { |
| 11369 | IRTemp rNt = newTemp(Ity_I32); |
| 11370 | IRTemp rMt = newTemp(Ity_I32); |
| 11371 | IRTemp res_q = newTemp(Ity_I32); |
| 11372 | |
| 11373 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11374 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11375 | |
| 11376 | assign(res_q, binop(Iop_HSub8Ux4, mkexpr(rNt), mkexpr(rMt))); |
| 11377 | if (isT) |
| 11378 | putIRegT( regD, mkexpr(res_q), condT ); |
| 11379 | else |
| 11380 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 11381 | |
| 11382 | DIP("uhsub8%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 11383 | return True; |
| 11384 | } |
| 11385 | /* fall through */ |
| 11386 | } |
| 11387 | |
| 11388 | /* ----------------- uhsub16<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 11389 | { |
| 11390 | UInt regD = 99, regN = 99, regM = 99; |
| 11391 | Bool gate = False; |
| 11392 | |
| 11393 | if (isT) { |
| 11394 | if (INSNT0(15,4) == 0xFAD && (INSNT1(15,0) & 0xF0F0) == 0xF060) { |
| 11395 | regN = INSNT0(3,0); |
| 11396 | regD = INSNT1(11,8); |
| 11397 | regM = INSNT1(3,0); |
| 11398 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11399 | gate = True; |
| 11400 | } |
| 11401 | } else { |
| 11402 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,1,1) && |
| 11403 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 11404 | INSNA(7,4) == BITS4(0,1,1,1)) { |
| 11405 | regD = INSNA(15,12); |
| 11406 | regN = INSNA(19,16); |
| 11407 | regM = INSNA(3,0); |
| 11408 | if (regD != 15 && regN != 15 && regM != 15) |
| 11409 | gate = True; |
| 11410 | } |
| 11411 | } |
| 11412 | |
| 11413 | if (gate) { |
| 11414 | IRTemp rNt = newTemp(Ity_I32); |
| 11415 | IRTemp rMt = newTemp(Ity_I32); |
| 11416 | IRTemp res_q = newTemp(Ity_I32); |
| 11417 | |
| 11418 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11419 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11420 | |
| 11421 | assign(res_q, binop(Iop_HSub16Ux2, mkexpr(rNt), mkexpr(rMt))); |
| 11422 | if (isT) |
| 11423 | putIRegT( regD, mkexpr(res_q), condT ); |
| 11424 | else |
| 11425 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 11426 | |
| 11427 | DIP("uhsub16%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 11428 | return True; |
| 11429 | } |
| 11430 | /* fall through */ |
| 11431 | } |
| 11432 | |
sewardj | a325ee1 | 2013-07-04 20:35:42 +0000 | [diff] [blame] | 11433 | /* ------------------ uqadd16<c> <Rd>,<Rn>,<Rm> ------------------ */ |
| 11434 | { |
| 11435 | UInt regD = 99, regN = 99, regM = 99; |
| 11436 | Bool gate = False; |
| 11437 | |
| 11438 | if (isT) { |
| 11439 | if (INSNT0(15,4) == 0xFA9 && (INSNT1(15,0) & 0xF0F0) == 0xF050) { |
| 11440 | regN = INSNT0(3,0); |
| 11441 | regD = INSNT1(11,8); |
| 11442 | regM = INSNT1(3,0); |
| 11443 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11444 | gate = True; |
| 11445 | } |
| 11446 | } else { |
| 11447 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,1,0) && |
| 11448 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 11449 | INSNA(7,4) == BITS4(0,0,0,1)) { |
| 11450 | regD = INSNA(15,12); |
| 11451 | regN = INSNA(19,16); |
| 11452 | regM = INSNA(3,0); |
| 11453 | if (regD != 15 && regN != 15 && regM != 15) |
| 11454 | gate = True; |
| 11455 | } |
| 11456 | } |
| 11457 | |
| 11458 | if (gate) { |
| 11459 | IRTemp rNt = newTemp(Ity_I32); |
| 11460 | IRTemp rMt = newTemp(Ity_I32); |
| 11461 | IRTemp res_q = newTemp(Ity_I32); |
| 11462 | |
| 11463 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11464 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11465 | |
| 11466 | assign(res_q, binop(Iop_QAdd16Ux2, mkexpr(rNt), mkexpr(rMt))); |
| 11467 | if (isT) |
| 11468 | putIRegT( regD, mkexpr(res_q), condT ); |
| 11469 | else |
| 11470 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 11471 | |
| 11472 | DIP("uqadd16%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 11473 | return True; |
| 11474 | } |
| 11475 | /* fall through */ |
| 11476 | } |
| 11477 | |
| 11478 | /* ------------------- uqsax<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 11479 | { |
| 11480 | UInt regD = 99, regN = 99, regM = 99; |
| 11481 | Bool gate = False; |
| 11482 | |
| 11483 | if (isT) { |
| 11484 | if (INSNT0(15,4) == 0xFAE && (INSNT1(15,0) & 0xF0F0) == 0xF050) { |
| 11485 | regN = INSNT0(3,0); |
| 11486 | regD = INSNT1(11,8); |
| 11487 | regM = INSNT1(3,0); |
| 11488 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11489 | gate = True; |
| 11490 | } |
| 11491 | } else { |
| 11492 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,1,0) && |
| 11493 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 11494 | INSNA(7,4) == BITS4(0,1,0,1)) { |
| 11495 | regD = INSNA(15,12); |
| 11496 | regN = INSNA(19,16); |
| 11497 | regM = INSNA(3,0); |
| 11498 | if (regD != 15 && regN != 15 && regM != 15) |
| 11499 | gate = True; |
| 11500 | } |
| 11501 | } |
| 11502 | |
| 11503 | if (gate) { |
| 11504 | IRTemp irt_regN = newTemp(Ity_I32); |
| 11505 | IRTemp irt_regM = newTemp(Ity_I32); |
| 11506 | IRTemp irt_sum = newTemp(Ity_I32); |
| 11507 | IRTemp irt_diff = newTemp(Ity_I32); |
| 11508 | IRTemp irt_sum_res = newTemp(Ity_I32); |
| 11509 | IRTemp irt_diff_res = newTemp(Ity_I32); |
| 11510 | |
| 11511 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11512 | assign( irt_regM, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11513 | |
| 11514 | assign( irt_diff, |
| 11515 | binop( Iop_Sub32, |
| 11516 | binop( Iop_Shr32, mkexpr(irt_regN), mkU8(16) ), |
| 11517 | binop( Iop_Shr32, |
| 11518 | binop(Iop_Shl32, mkexpr(irt_regM), mkU8(16)), |
| 11519 | mkU8(16) ) ) ); |
| 11520 | armUnsignedSatQ( &irt_diff_res, NULL, irt_diff, 0x10); |
| 11521 | |
| 11522 | assign( irt_sum, |
| 11523 | binop( Iop_Add32, |
| 11524 | binop( Iop_Shr32, |
| 11525 | binop( Iop_Shl32, mkexpr(irt_regN), mkU8(16) ), |
| 11526 | mkU8(16) ), |
| 11527 | binop( Iop_Shr32, mkexpr(irt_regM), mkU8(16) )) ); |
| 11528 | armUnsignedSatQ( &irt_sum_res, NULL, irt_sum, 0x10 ); |
| 11529 | |
| 11530 | IRExpr* ire_result = binop( Iop_Or32, |
| 11531 | binop( Iop_Shl32, mkexpr(irt_diff_res), |
| 11532 | mkU8(16) ), |
| 11533 | binop( Iop_And32, mkexpr(irt_sum_res), |
| 11534 | mkU32(0xFFFF)) ); |
| 11535 | |
| 11536 | if (isT) |
| 11537 | putIRegT( regD, ire_result, condT ); |
| 11538 | else |
| 11539 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 11540 | |
| 11541 | DIP( "uqsax%s r%u, r%u, r%u\n", nCC(conq), regD, regN, regM ); |
| 11542 | return True; |
| 11543 | } |
| 11544 | /* fall through */ |
| 11545 | } |
| 11546 | |
| 11547 | /* ------------------- uqasx<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 11548 | { |
| 11549 | UInt regD = 99, regN = 99, regM = 99; |
| 11550 | Bool gate = False; |
| 11551 | |
| 11552 | if (isT) { |
| 11553 | if (INSNT0(15,4) == 0xFAA && (INSNT1(15,0) & 0xF0F0) == 0xF050) { |
| 11554 | regN = INSNT0(3,0); |
| 11555 | regD = INSNT1(11,8); |
| 11556 | regM = INSNT1(3,0); |
| 11557 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11558 | gate = True; |
| 11559 | } |
| 11560 | } else { |
| 11561 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,1,0) && |
| 11562 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 11563 | INSNA(7,4) == BITS4(0,0,1,1)) { |
| 11564 | regD = INSNA(15,12); |
| 11565 | regN = INSNA(19,16); |
| 11566 | regM = INSNA(3,0); |
| 11567 | if (regD != 15 && regN != 15 && regM != 15) |
| 11568 | gate = True; |
| 11569 | } |
| 11570 | } |
| 11571 | |
| 11572 | if (gate) { |
| 11573 | IRTemp irt_regN = newTemp(Ity_I32); |
| 11574 | IRTemp irt_regM = newTemp(Ity_I32); |
| 11575 | IRTemp irt_sum = newTemp(Ity_I32); |
| 11576 | IRTemp irt_diff = newTemp(Ity_I32); |
| 11577 | IRTemp irt_res_sum = newTemp(Ity_I32); |
| 11578 | IRTemp irt_res_diff = newTemp(Ity_I32); |
| 11579 | |
| 11580 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11581 | assign( irt_regM, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11582 | |
| 11583 | assign( irt_diff, |
| 11584 | binop( Iop_Sub32, |
| 11585 | binop( Iop_Shr32, |
| 11586 | binop( Iop_Shl32, mkexpr(irt_regN), mkU8(16) ), |
| 11587 | mkU8(16) ), |
| 11588 | binop( Iop_Shr32, mkexpr(irt_regM), mkU8(16) ) ) ); |
| 11589 | armUnsignedSatQ( &irt_res_diff, NULL, irt_diff, 0x10 ); |
| 11590 | |
| 11591 | assign( irt_sum, |
| 11592 | binop( Iop_Add32, |
| 11593 | binop( Iop_Shr32, mkexpr(irt_regN), mkU8(16) ), |
| 11594 | binop( Iop_Shr32, |
| 11595 | binop( Iop_Shl32, mkexpr(irt_regM), mkU8(16) ), |
| 11596 | mkU8(16) ) ) ); |
| 11597 | armUnsignedSatQ( &irt_res_sum, NULL, irt_sum, 0x10 ); |
| 11598 | |
| 11599 | IRExpr* ire_result |
| 11600 | = binop( Iop_Or32, |
| 11601 | binop( Iop_Shl32, mkexpr(irt_res_sum), mkU8(16) ), |
| 11602 | binop( Iop_And32, mkexpr(irt_res_diff), mkU32(0xFFFF) ) ); |
| 11603 | |
| 11604 | if (isT) |
| 11605 | putIRegT( regD, ire_result, condT ); |
| 11606 | else |
| 11607 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 11608 | |
| 11609 | DIP( "uqasx%s r%u, r%u, r%u\n", nCC(conq), regD, regN, regM ); |
| 11610 | return True; |
| 11611 | } |
| 11612 | /* fall through */ |
| 11613 | } |
| 11614 | |
| 11615 | /* ------------------- usax<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 11616 | { |
| 11617 | UInt regD = 99, regN = 99, regM = 99; |
| 11618 | Bool gate = False; |
| 11619 | |
| 11620 | if (isT) { |
| 11621 | if (INSNT0(15,4) == 0xFAE && (INSNT1(15,0) & 0xF0F0) == 0xF040) { |
| 11622 | regN = INSNT0(3,0); |
| 11623 | regD = INSNT1(11,8); |
| 11624 | regM = INSNT1(3,0); |
| 11625 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11626 | gate = True; |
| 11627 | } |
| 11628 | } else { |
| 11629 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,0,1) && |
| 11630 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 11631 | INSNA(7,4) == BITS4(0,1,0,1)) { |
| 11632 | regD = INSNA(15,12); |
| 11633 | regN = INSNA(19,16); |
| 11634 | regM = INSNA(3,0); |
| 11635 | if (regD != 15 && regN != 15 && regM != 15) |
| 11636 | gate = True; |
| 11637 | } |
| 11638 | } |
| 11639 | |
| 11640 | if (gate) { |
| 11641 | IRTemp irt_regN = newTemp(Ity_I32); |
| 11642 | IRTemp irt_regM = newTemp(Ity_I32); |
| 11643 | IRTemp irt_sum = newTemp(Ity_I32); |
| 11644 | IRTemp irt_diff = newTemp(Ity_I32); |
| 11645 | |
| 11646 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11647 | assign( irt_regM, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11648 | |
| 11649 | assign( irt_sum, |
| 11650 | binop( Iop_Add32, |
| 11651 | unop( Iop_16Uto32, |
| 11652 | unop( Iop_32to16, mkexpr(irt_regN) ) |
| 11653 | ), |
| 11654 | binop( Iop_Shr32, mkexpr(irt_regM), mkU8(16) ) ) ); |
| 11655 | |
| 11656 | assign( irt_diff, |
| 11657 | binop( Iop_Sub32, |
| 11658 | binop( Iop_Shr32, mkexpr(irt_regN), mkU8(16) ), |
| 11659 | unop( Iop_16Uto32, |
| 11660 | unop( Iop_32to16, mkexpr(irt_regM) ) |
| 11661 | ) |
| 11662 | ) |
| 11663 | ); |
| 11664 | |
| 11665 | IRExpr* ire_result |
| 11666 | = binop( Iop_Or32, |
| 11667 | binop( Iop_Shl32, mkexpr(irt_diff), mkU8(16) ), |
| 11668 | binop( Iop_And32, mkexpr(irt_sum), mkU32(0xFFFF) ) ); |
| 11669 | |
| 11670 | IRTemp ge10 = newTemp(Ity_I32); |
| 11671 | assign( ge10, IRExpr_ITE( binop( Iop_CmpLE32U, |
| 11672 | mkU32(0x10000), mkexpr(irt_sum) ), |
| 11673 | mkU32(1), mkU32(0) ) ); |
| 11674 | put_GEFLAG32( 0, 0, mkexpr(ge10), condT ); |
| 11675 | put_GEFLAG32( 1, 0, mkexpr(ge10), condT ); |
| 11676 | |
| 11677 | IRTemp ge32 = newTemp(Ity_I32); |
| 11678 | assign(ge32, unop(Iop_Not32, mkexpr(irt_diff))); |
| 11679 | put_GEFLAG32( 2, 31, mkexpr(ge32), condT ); |
| 11680 | put_GEFLAG32( 3, 31, mkexpr(ge32), condT ); |
| 11681 | |
| 11682 | if (isT) |
| 11683 | putIRegT( regD, ire_result, condT ); |
| 11684 | else |
| 11685 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 11686 | |
| 11687 | DIP( "usax%s r%u, r%u, r%u\n", nCC(conq), regD, regN, regM ); |
| 11688 | return True; |
| 11689 | } |
| 11690 | /* fall through */ |
| 11691 | } |
| 11692 | |
| 11693 | /* ------------------- uasx<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 11694 | { |
| 11695 | UInt regD = 99, regN = 99, regM = 99; |
| 11696 | Bool gate = False; |
| 11697 | |
| 11698 | if (isT) { |
| 11699 | if (INSNT0(15,4) == 0xFAA && (INSNT1(15,0) & 0xF0F0) == 0xF040) { |
| 11700 | regN = INSNT0(3,0); |
| 11701 | regD = INSNT1(11,8); |
| 11702 | regM = INSNT1(3,0); |
| 11703 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11704 | gate = True; |
| 11705 | } |
| 11706 | } else { |
| 11707 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,0,1) && |
| 11708 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 11709 | INSNA(7,4) == BITS4(0,0,1,1)) { |
| 11710 | regD = INSNA(15,12); |
| 11711 | regN = INSNA(19,16); |
| 11712 | regM = INSNA(3,0); |
| 11713 | if (regD != 15 && regN != 15 && regM != 15) |
| 11714 | gate = True; |
| 11715 | } |
| 11716 | } |
| 11717 | |
| 11718 | if (gate) { |
| 11719 | IRTemp irt_regN = newTemp(Ity_I32); |
| 11720 | IRTemp irt_regM = newTemp(Ity_I32); |
| 11721 | IRTemp irt_sum = newTemp(Ity_I32); |
| 11722 | IRTemp irt_diff = newTemp(Ity_I32); |
| 11723 | |
| 11724 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11725 | assign( irt_regM, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11726 | |
| 11727 | assign( irt_diff, |
| 11728 | binop( Iop_Sub32, |
| 11729 | unop( Iop_16Uto32, |
| 11730 | unop( Iop_32to16, mkexpr(irt_regN) ) |
| 11731 | ), |
| 11732 | binop( Iop_Shr32, mkexpr(irt_regM), mkU8(16) ) ) ); |
| 11733 | |
| 11734 | assign( irt_sum, |
| 11735 | binop( Iop_Add32, |
| 11736 | binop( Iop_Shr32, mkexpr(irt_regN), mkU8(16) ), |
| 11737 | unop( Iop_16Uto32, |
| 11738 | unop( Iop_32to16, mkexpr(irt_regM) ) |
| 11739 | ) ) ); |
| 11740 | |
| 11741 | IRExpr* ire_result |
| 11742 | = binop( Iop_Or32, |
| 11743 | binop( Iop_Shl32, mkexpr(irt_sum), mkU8(16) ), |
| 11744 | binop( Iop_And32, mkexpr(irt_diff), mkU32(0xFFFF) ) ); |
| 11745 | |
| 11746 | IRTemp ge10 = newTemp(Ity_I32); |
| 11747 | assign(ge10, unop(Iop_Not32, mkexpr(irt_diff))); |
| 11748 | put_GEFLAG32( 0, 31, mkexpr(ge10), condT ); |
| 11749 | put_GEFLAG32( 1, 31, mkexpr(ge10), condT ); |
| 11750 | |
| 11751 | IRTemp ge32 = newTemp(Ity_I32); |
| 11752 | assign( ge32, IRExpr_ITE( binop( Iop_CmpLE32U, |
| 11753 | mkU32(0x10000), mkexpr(irt_sum) ), |
| 11754 | mkU32(1), mkU32(0) ) ); |
| 11755 | put_GEFLAG32( 2, 0, mkexpr(ge32), condT ); |
| 11756 | put_GEFLAG32( 3, 0, mkexpr(ge32), condT ); |
| 11757 | |
| 11758 | if (isT) |
| 11759 | putIRegT( regD, ire_result, condT ); |
| 11760 | else |
| 11761 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 11762 | |
| 11763 | DIP( "uasx%s r%u, r%u, r%u\n", nCC(conq), regD, regN, regM ); |
| 11764 | return True; |
| 11765 | } |
| 11766 | /* fall through */ |
| 11767 | } |
| 11768 | |
| 11769 | /* ------------------- ssax<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 11770 | { |
| 11771 | UInt regD = 99, regN = 99, regM = 99; |
| 11772 | Bool gate = False; |
| 11773 | |
| 11774 | if (isT) { |
| 11775 | if (INSNT0(15,4) == 0xFAE && (INSNT1(15,0) & 0xF0F0) == 0xF000) { |
| 11776 | regN = INSNT0(3,0); |
| 11777 | regD = INSNT1(11,8); |
| 11778 | regM = INSNT1(3,0); |
| 11779 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11780 | gate = True; |
| 11781 | } |
| 11782 | } else { |
| 11783 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,0,1) && |
| 11784 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 11785 | INSNA(7,4) == BITS4(0,1,0,1)) { |
| 11786 | regD = INSNA(15,12); |
| 11787 | regN = INSNA(19,16); |
| 11788 | regM = INSNA(3,0); |
| 11789 | if (regD != 15 && regN != 15 && regM != 15) |
| 11790 | gate = True; |
| 11791 | } |
| 11792 | } |
| 11793 | |
| 11794 | if (gate) { |
| 11795 | IRTemp irt_regN = newTemp(Ity_I32); |
| 11796 | IRTemp irt_regM = newTemp(Ity_I32); |
| 11797 | IRTemp irt_sum = newTemp(Ity_I32); |
| 11798 | IRTemp irt_diff = newTemp(Ity_I32); |
| 11799 | |
| 11800 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11801 | assign( irt_regM, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11802 | |
| 11803 | assign( irt_sum, |
| 11804 | binop( Iop_Add32, |
| 11805 | binop( Iop_Sar32, |
| 11806 | binop( Iop_Shl32, mkexpr(irt_regN), mkU8(16) ), |
| 11807 | mkU8(16) ), |
| 11808 | binop( Iop_Sar32, mkexpr(irt_regM), mkU8(16) ) ) ); |
| 11809 | |
| 11810 | assign( irt_diff, |
| 11811 | binop( Iop_Sub32, |
| 11812 | binop( Iop_Sar32, mkexpr(irt_regN), mkU8(16) ), |
| 11813 | binop( Iop_Sar32, |
| 11814 | binop( Iop_Shl32, mkexpr(irt_regM), mkU8(16) ), |
| 11815 | mkU8(16) ) ) ); |
| 11816 | |
| 11817 | IRExpr* ire_result |
| 11818 | = binop( Iop_Or32, |
| 11819 | binop( Iop_Shl32, mkexpr(irt_diff), mkU8(16) ), |
| 11820 | binop( Iop_And32, mkexpr(irt_sum), mkU32(0xFFFF) ) ); |
| 11821 | |
| 11822 | IRTemp ge10 = newTemp(Ity_I32); |
| 11823 | assign(ge10, unop(Iop_Not32, mkexpr(irt_sum))); |
| 11824 | put_GEFLAG32( 0, 31, mkexpr(ge10), condT ); |
| 11825 | put_GEFLAG32( 1, 31, mkexpr(ge10), condT ); |
| 11826 | |
| 11827 | IRTemp ge32 = newTemp(Ity_I32); |
| 11828 | assign(ge32, unop(Iop_Not32, mkexpr(irt_diff))); |
| 11829 | put_GEFLAG32( 2, 31, mkexpr(ge32), condT ); |
| 11830 | put_GEFLAG32( 3, 31, mkexpr(ge32), condT ); |
| 11831 | |
| 11832 | if (isT) |
| 11833 | putIRegT( regD, ire_result, condT ); |
| 11834 | else |
| 11835 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 11836 | |
| 11837 | DIP( "ssax%s r%u, r%u, r%u\n", nCC(conq), regD, regN, regM ); |
| 11838 | return True; |
| 11839 | } |
| 11840 | /* fall through */ |
| 11841 | } |
| 11842 | |
| 11843 | /* ----------------- shsub8<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 11844 | { |
| 11845 | UInt regD = 99, regN = 99, regM = 99; |
| 11846 | Bool gate = False; |
| 11847 | |
| 11848 | if (isT) { |
| 11849 | if (INSNT0(15,4) == 0xFAC && (INSNT1(15,0) & 0xF0F0) == 0xF020) { |
| 11850 | regN = INSNT0(3,0); |
| 11851 | regD = INSNT1(11,8); |
| 11852 | regM = INSNT1(3,0); |
| 11853 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11854 | gate = True; |
| 11855 | } |
| 11856 | } else { |
| 11857 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,1,1) && |
| 11858 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 11859 | INSNA(7,4) == BITS4(1,1,1,1)) { |
| 11860 | regD = INSNA(15,12); |
| 11861 | regN = INSNA(19,16); |
| 11862 | regM = INSNA(3,0); |
| 11863 | if (regD != 15 && regN != 15 && regM != 15) |
| 11864 | gate = True; |
| 11865 | } |
| 11866 | } |
| 11867 | |
| 11868 | if (gate) { |
| 11869 | IRTemp rNt = newTemp(Ity_I32); |
| 11870 | IRTemp rMt = newTemp(Ity_I32); |
| 11871 | IRTemp res_q = newTemp(Ity_I32); |
| 11872 | |
| 11873 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11874 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11875 | |
| 11876 | assign(res_q, binop(Iop_HSub8Sx4, mkexpr(rNt), mkexpr(rMt))); |
| 11877 | if (isT) |
| 11878 | putIRegT( regD, mkexpr(res_q), condT ); |
| 11879 | else |
| 11880 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 11881 | |
| 11882 | DIP("shsub8%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 11883 | return True; |
| 11884 | } |
| 11885 | /* fall through */ |
| 11886 | } |
| 11887 | |
| 11888 | /* ----------------- sxtab16<c> Rd,Rn,Rm{,rot} ------------------ */ |
| 11889 | { |
| 11890 | UInt regD = 99, regN = 99, regM = 99, rotate = 99; |
| 11891 | Bool gate = False; |
| 11892 | |
| 11893 | if (isT) { |
| 11894 | if (INSNT0(15,4) == 0xFA2 && (INSNT1(15,0) & 0xF0C0) == 0xF080) { |
| 11895 | regN = INSNT0(3,0); |
| 11896 | regD = INSNT1(11,8); |
| 11897 | regM = INSNT1(3,0); |
| 11898 | rotate = INSNT1(5,4); |
| 11899 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11900 | gate = True; |
| 11901 | } |
| 11902 | } else { |
| 11903 | if (INSNA(27,20) == BITS8(0,1,1,0,1,0,0,0) && |
| 11904 | INSNA(9,4) == BITS6(0,0,0,1,1,1) ) { |
| 11905 | regD = INSNA(15,12); |
| 11906 | regN = INSNA(19,16); |
| 11907 | regM = INSNA(3,0); |
| 11908 | rotate = INSNA(11,10); |
| 11909 | if (regD != 15 && regN != 15 && regM != 15) |
| 11910 | gate = True; |
| 11911 | } |
| 11912 | } |
| 11913 | |
| 11914 | if (gate) { |
| 11915 | IRTemp irt_regN = newTemp(Ity_I32); |
| 11916 | assign( irt_regN, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11917 | |
| 11918 | IRTemp irt_regM = newTemp(Ity_I32); |
| 11919 | assign( irt_regM, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11920 | |
| 11921 | IRTemp irt_rot = newTemp(Ity_I32); |
| 11922 | assign( irt_rot, genROR32(irt_regM, 8 * rotate) ); |
| 11923 | |
| 11924 | /* FIXME Maybe we can write this arithmetic in shorter form. */ |
| 11925 | IRExpr* resLo |
| 11926 | = binop(Iop_And32, |
| 11927 | binop(Iop_Add32, |
| 11928 | mkexpr(irt_regN), |
| 11929 | unop(Iop_16Uto32, |
| 11930 | unop(Iop_8Sto16, |
| 11931 | unop(Iop_32to8, mkexpr(irt_rot))))), |
| 11932 | mkU32(0x0000FFFF)); |
| 11933 | |
| 11934 | IRExpr* resHi |
| 11935 | = binop(Iop_And32, |
| 11936 | binop(Iop_Add32, |
| 11937 | mkexpr(irt_regN), |
| 11938 | binop(Iop_Shl32, |
| 11939 | unop(Iop_16Uto32, |
| 11940 | unop(Iop_8Sto16, |
| 11941 | unop(Iop_32to8, |
| 11942 | binop(Iop_Shr32, |
| 11943 | mkexpr(irt_rot), |
| 11944 | mkU8(16))))), |
| 11945 | mkU8(16))), |
| 11946 | mkU32(0xFFFF0000)); |
| 11947 | |
| 11948 | IRExpr* ire_result |
| 11949 | = binop( Iop_Or32, resHi, resLo ); |
| 11950 | |
| 11951 | if (isT) |
| 11952 | putIRegT( regD, ire_result, condT ); |
| 11953 | else |
| 11954 | putIRegA( regD, ire_result, condT, Ijk_Boring ); |
| 11955 | |
| 11956 | DIP( "sxtab16%s r%u, r%u, r%u, ROR #%u\n", |
| 11957 | nCC(conq), regD, regN, regM, 8 * rotate ); |
| 11958 | return True; |
| 11959 | } |
| 11960 | /* fall through */ |
| 11961 | } |
| 11962 | |
| 11963 | /* ----------------- shasx<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 11964 | { |
| 11965 | UInt regD = 99, regN = 99, regM = 99; |
| 11966 | Bool gate = False; |
| 11967 | |
| 11968 | if (isT) { |
| 11969 | if (INSNT0(15,4) == 0xFAA && (INSNT1(15,0) & 0xF0F0) == 0xF020) { |
| 11970 | regN = INSNT0(3,0); |
| 11971 | regD = INSNT1(11,8); |
| 11972 | regM = INSNT1(3,0); |
| 11973 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 11974 | gate = True; |
| 11975 | } |
| 11976 | } else { |
| 11977 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,1,1) && |
| 11978 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 11979 | INSNA(7,4) == BITS4(0,0,1,1)) { |
| 11980 | regD = INSNA(15,12); |
| 11981 | regN = INSNA(19,16); |
| 11982 | regM = INSNA(3,0); |
| 11983 | if (regD != 15 && regN != 15 && regM != 15) |
| 11984 | gate = True; |
| 11985 | } |
| 11986 | } |
| 11987 | |
| 11988 | if (gate) { |
| 11989 | IRTemp rNt = newTemp(Ity_I32); |
| 11990 | IRTemp rMt = newTemp(Ity_I32); |
| 11991 | IRTemp irt_diff = newTemp(Ity_I32); |
| 11992 | IRTemp irt_sum = newTemp(Ity_I32); |
| 11993 | IRTemp res_q = newTemp(Ity_I32); |
| 11994 | |
| 11995 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 11996 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 11997 | |
| 11998 | assign( irt_diff, |
| 11999 | binop(Iop_Sub32, |
| 12000 | unop(Iop_16Sto32, |
| 12001 | unop(Iop_32to16, |
| 12002 | mkexpr(rNt) |
| 12003 | ) |
| 12004 | ), |
| 12005 | unop(Iop_16Sto32, |
| 12006 | unop(Iop_32to16, |
| 12007 | binop(Iop_Shr32, |
| 12008 | mkexpr(rMt), mkU8(16) |
| 12009 | ) |
| 12010 | ) |
| 12011 | ) |
| 12012 | ) |
| 12013 | ); |
| 12014 | |
| 12015 | assign( irt_sum, |
| 12016 | binop(Iop_Add32, |
| 12017 | unop(Iop_16Sto32, |
| 12018 | unop(Iop_32to16, |
| 12019 | binop(Iop_Shr32, |
| 12020 | mkexpr(rNt), mkU8(16) |
| 12021 | ) |
| 12022 | ) |
| 12023 | ), |
| 12024 | unop(Iop_16Sto32, |
| 12025 | unop(Iop_32to16, mkexpr(rMt) |
| 12026 | ) |
| 12027 | ) |
| 12028 | ) |
| 12029 | ); |
| 12030 | |
| 12031 | assign( res_q, |
| 12032 | binop(Iop_Or32, |
| 12033 | unop(Iop_16Uto32, |
| 12034 | unop(Iop_32to16, |
| 12035 | binop(Iop_Shr32, |
| 12036 | mkexpr(irt_diff), mkU8(1) |
| 12037 | ) |
| 12038 | ) |
| 12039 | ), |
| 12040 | binop(Iop_Shl32, |
| 12041 | binop(Iop_Shr32, |
| 12042 | mkexpr(irt_sum), mkU8(1) |
| 12043 | ), |
| 12044 | mkU8(16) |
| 12045 | ) |
| 12046 | ) |
| 12047 | ); |
| 12048 | |
| 12049 | if (isT) |
| 12050 | putIRegT( regD, mkexpr(res_q), condT ); |
| 12051 | else |
| 12052 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 12053 | |
| 12054 | DIP("shasx%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 12055 | return True; |
| 12056 | } |
| 12057 | /* fall through */ |
| 12058 | } |
| 12059 | |
| 12060 | /* ----------------- uhasx<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 12061 | { |
| 12062 | UInt regD = 99, regN = 99, regM = 99; |
| 12063 | Bool gate = False; |
| 12064 | |
| 12065 | if (isT) { |
| 12066 | if (INSNT0(15,4) == 0xFAA && (INSNT1(15,0) & 0xF0F0) == 0xF060) { |
| 12067 | regN = INSNT0(3,0); |
| 12068 | regD = INSNT1(11,8); |
| 12069 | regM = INSNT1(3,0); |
| 12070 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 12071 | gate = True; |
| 12072 | } |
| 12073 | } else { |
| 12074 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,1,1) && |
| 12075 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 12076 | INSNA(7,4) == BITS4(0,0,1,1)) { |
| 12077 | regD = INSNA(15,12); |
| 12078 | regN = INSNA(19,16); |
| 12079 | regM = INSNA(3,0); |
| 12080 | if (regD != 15 && regN != 15 && regM != 15) |
| 12081 | gate = True; |
| 12082 | } |
| 12083 | } |
| 12084 | |
| 12085 | if (gate) { |
| 12086 | IRTemp rNt = newTemp(Ity_I32); |
| 12087 | IRTemp rMt = newTemp(Ity_I32); |
| 12088 | IRTemp irt_diff = newTemp(Ity_I32); |
| 12089 | IRTemp irt_sum = newTemp(Ity_I32); |
| 12090 | IRTemp res_q = newTemp(Ity_I32); |
| 12091 | |
| 12092 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 12093 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 12094 | |
| 12095 | assign( irt_diff, |
| 12096 | binop(Iop_Sub32, |
| 12097 | unop(Iop_16Uto32, |
| 12098 | unop(Iop_32to16, |
| 12099 | mkexpr(rNt) |
| 12100 | ) |
| 12101 | ), |
| 12102 | unop(Iop_16Uto32, |
| 12103 | unop(Iop_32to16, |
| 12104 | binop(Iop_Shr32, |
| 12105 | mkexpr(rMt), mkU8(16) |
| 12106 | ) |
| 12107 | ) |
| 12108 | ) |
| 12109 | ) |
| 12110 | ); |
| 12111 | |
| 12112 | assign( irt_sum, |
| 12113 | binop(Iop_Add32, |
| 12114 | unop(Iop_16Uto32, |
| 12115 | unop(Iop_32to16, |
| 12116 | binop(Iop_Shr32, |
| 12117 | mkexpr(rNt), mkU8(16) |
| 12118 | ) |
| 12119 | ) |
| 12120 | ), |
| 12121 | unop(Iop_16Uto32, |
| 12122 | unop(Iop_32to16, mkexpr(rMt) |
| 12123 | ) |
| 12124 | ) |
| 12125 | ) |
| 12126 | ); |
| 12127 | |
| 12128 | assign( res_q, |
| 12129 | binop(Iop_Or32, |
| 12130 | unop(Iop_16Uto32, |
| 12131 | unop(Iop_32to16, |
| 12132 | binop(Iop_Shr32, |
| 12133 | mkexpr(irt_diff), mkU8(1) |
| 12134 | ) |
| 12135 | ) |
| 12136 | ), |
| 12137 | binop(Iop_Shl32, |
| 12138 | binop(Iop_Shr32, |
| 12139 | mkexpr(irt_sum), mkU8(1) |
| 12140 | ), |
| 12141 | mkU8(16) |
| 12142 | ) |
| 12143 | ) |
| 12144 | ); |
| 12145 | |
| 12146 | if (isT) |
| 12147 | putIRegT( regD, mkexpr(res_q), condT ); |
| 12148 | else |
| 12149 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 12150 | |
| 12151 | DIP("uhasx%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 12152 | return True; |
| 12153 | } |
| 12154 | /* fall through */ |
| 12155 | } |
| 12156 | |
| 12157 | /* ----------------- shsax<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 12158 | { |
| 12159 | UInt regD = 99, regN = 99, regM = 99; |
| 12160 | Bool gate = False; |
| 12161 | |
| 12162 | if (isT) { |
| 12163 | if (INSNT0(15,4) == 0xFAE && (INSNT1(15,0) & 0xF0F0) == 0xF020) { |
| 12164 | regN = INSNT0(3,0); |
| 12165 | regD = INSNT1(11,8); |
| 12166 | regM = INSNT1(3,0); |
| 12167 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 12168 | gate = True; |
| 12169 | } |
| 12170 | } else { |
| 12171 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,1,1) && |
| 12172 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 12173 | INSNA(7,4) == BITS4(0,1,0,1)) { |
| 12174 | regD = INSNA(15,12); |
| 12175 | regN = INSNA(19,16); |
| 12176 | regM = INSNA(3,0); |
| 12177 | if (regD != 15 && regN != 15 && regM != 15) |
| 12178 | gate = True; |
| 12179 | } |
| 12180 | } |
| 12181 | |
| 12182 | if (gate) { |
| 12183 | IRTemp rNt = newTemp(Ity_I32); |
| 12184 | IRTemp rMt = newTemp(Ity_I32); |
| 12185 | IRTemp irt_diff = newTemp(Ity_I32); |
| 12186 | IRTemp irt_sum = newTemp(Ity_I32); |
| 12187 | IRTemp res_q = newTemp(Ity_I32); |
| 12188 | |
| 12189 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 12190 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 12191 | |
| 12192 | assign( irt_sum, |
| 12193 | binop(Iop_Add32, |
| 12194 | unop(Iop_16Sto32, |
| 12195 | unop(Iop_32to16, |
| 12196 | mkexpr(rNt) |
| 12197 | ) |
| 12198 | ), |
| 12199 | unop(Iop_16Sto32, |
| 12200 | unop(Iop_32to16, |
| 12201 | binop(Iop_Shr32, |
| 12202 | mkexpr(rMt), mkU8(16) |
| 12203 | ) |
| 12204 | ) |
| 12205 | ) |
| 12206 | ) |
| 12207 | ); |
| 12208 | |
| 12209 | assign( irt_diff, |
| 12210 | binop(Iop_Sub32, |
| 12211 | unop(Iop_16Sto32, |
| 12212 | unop(Iop_32to16, |
| 12213 | binop(Iop_Shr32, |
| 12214 | mkexpr(rNt), mkU8(16) |
| 12215 | ) |
| 12216 | ) |
| 12217 | ), |
| 12218 | unop(Iop_16Sto32, |
| 12219 | unop(Iop_32to16, mkexpr(rMt) |
| 12220 | ) |
| 12221 | ) |
| 12222 | ) |
| 12223 | ); |
| 12224 | |
| 12225 | assign( res_q, |
| 12226 | binop(Iop_Or32, |
| 12227 | unop(Iop_16Uto32, |
| 12228 | unop(Iop_32to16, |
| 12229 | binop(Iop_Shr32, |
| 12230 | mkexpr(irt_sum), mkU8(1) |
| 12231 | ) |
| 12232 | ) |
| 12233 | ), |
| 12234 | binop(Iop_Shl32, |
| 12235 | binop(Iop_Shr32, |
| 12236 | mkexpr(irt_diff), mkU8(1) |
| 12237 | ), |
| 12238 | mkU8(16) |
| 12239 | ) |
| 12240 | ) |
| 12241 | ); |
| 12242 | |
| 12243 | if (isT) |
| 12244 | putIRegT( regD, mkexpr(res_q), condT ); |
| 12245 | else |
| 12246 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 12247 | |
| 12248 | DIP("shsax%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 12249 | return True; |
| 12250 | } |
| 12251 | /* fall through */ |
| 12252 | } |
| 12253 | |
| 12254 | /* ----------------- uhsax<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 12255 | { |
| 12256 | UInt regD = 99, regN = 99, regM = 99; |
| 12257 | Bool gate = False; |
| 12258 | |
| 12259 | if (isT) { |
| 12260 | if (INSNT0(15,4) == 0xFAE && (INSNT1(15,0) & 0xF0F0) == 0xF060) { |
| 12261 | regN = INSNT0(3,0); |
| 12262 | regD = INSNT1(11,8); |
| 12263 | regM = INSNT1(3,0); |
| 12264 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 12265 | gate = True; |
| 12266 | } |
| 12267 | } else { |
| 12268 | if (INSNA(27,20) == BITS8(0,1,1,0,0,1,1,1) && |
| 12269 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 12270 | INSNA(7,4) == BITS4(0,1,0,1)) { |
| 12271 | regD = INSNA(15,12); |
| 12272 | regN = INSNA(19,16); |
| 12273 | regM = INSNA(3,0); |
| 12274 | if (regD != 15 && regN != 15 && regM != 15) |
| 12275 | gate = True; |
| 12276 | } |
| 12277 | } |
| 12278 | |
| 12279 | if (gate) { |
| 12280 | IRTemp rNt = newTemp(Ity_I32); |
| 12281 | IRTemp rMt = newTemp(Ity_I32); |
| 12282 | IRTemp irt_diff = newTemp(Ity_I32); |
| 12283 | IRTemp irt_sum = newTemp(Ity_I32); |
| 12284 | IRTemp res_q = newTemp(Ity_I32); |
| 12285 | |
| 12286 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 12287 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 12288 | |
| 12289 | assign( irt_sum, |
| 12290 | binop(Iop_Add32, |
| 12291 | unop(Iop_16Uto32, |
| 12292 | unop(Iop_32to16, |
| 12293 | mkexpr(rNt) |
| 12294 | ) |
| 12295 | ), |
| 12296 | unop(Iop_16Uto32, |
| 12297 | unop(Iop_32to16, |
| 12298 | binop(Iop_Shr32, |
| 12299 | mkexpr(rMt), mkU8(16) |
| 12300 | ) |
| 12301 | ) |
| 12302 | ) |
| 12303 | ) |
| 12304 | ); |
| 12305 | |
| 12306 | assign( irt_diff, |
| 12307 | binop(Iop_Sub32, |
| 12308 | unop(Iop_16Uto32, |
| 12309 | unop(Iop_32to16, |
| 12310 | binop(Iop_Shr32, |
| 12311 | mkexpr(rNt), mkU8(16) |
| 12312 | ) |
| 12313 | ) |
| 12314 | ), |
| 12315 | unop(Iop_16Uto32, |
| 12316 | unop(Iop_32to16, mkexpr(rMt) |
| 12317 | ) |
| 12318 | ) |
| 12319 | ) |
| 12320 | ); |
| 12321 | |
| 12322 | assign( res_q, |
| 12323 | binop(Iop_Or32, |
| 12324 | unop(Iop_16Uto32, |
| 12325 | unop(Iop_32to16, |
| 12326 | binop(Iop_Shr32, |
| 12327 | mkexpr(irt_sum), mkU8(1) |
| 12328 | ) |
| 12329 | ) |
| 12330 | ), |
| 12331 | binop(Iop_Shl32, |
| 12332 | binop(Iop_Shr32, |
| 12333 | mkexpr(irt_diff), mkU8(1) |
| 12334 | ), |
| 12335 | mkU8(16) |
| 12336 | ) |
| 12337 | ) |
| 12338 | ); |
| 12339 | |
| 12340 | if (isT) |
| 12341 | putIRegT( regD, mkexpr(res_q), condT ); |
| 12342 | else |
| 12343 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 12344 | |
| 12345 | DIP("uhsax%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 12346 | return True; |
| 12347 | } |
| 12348 | /* fall through */ |
| 12349 | } |
| 12350 | |
| 12351 | /* ----------------- shsub16<c> <Rd>,<Rn>,<Rm> ------------------- */ |
| 12352 | { |
| 12353 | UInt regD = 99, regN = 99, regM = 99; |
| 12354 | Bool gate = False; |
| 12355 | |
| 12356 | if (isT) { |
| 12357 | if (INSNT0(15,4) == 0xFAD && (INSNT1(15,0) & 0xF0F0) == 0xF020) { |
| 12358 | regN = INSNT0(3,0); |
| 12359 | regD = INSNT1(11,8); |
| 12360 | regM = INSNT1(3,0); |
| 12361 | if (!isBadRegT(regD) && !isBadRegT(regN) && !isBadRegT(regM)) |
| 12362 | gate = True; |
| 12363 | } |
| 12364 | } else { |
| 12365 | if (INSNA(27,20) == BITS8(0,1,1,0,0,0,1,1) && |
| 12366 | INSNA(11,8) == BITS4(1,1,1,1) && |
| 12367 | INSNA(7,4) == BITS4(0,1,1,1)) { |
| 12368 | regD = INSNA(15,12); |
| 12369 | regN = INSNA(19,16); |
| 12370 | regM = INSNA(3,0); |
| 12371 | if (regD != 15 && regN != 15 && regM != 15) |
| 12372 | gate = True; |
| 12373 | } |
| 12374 | } |
| 12375 | |
| 12376 | if (gate) { |
| 12377 | IRTemp rNt = newTemp(Ity_I32); |
| 12378 | IRTemp rMt = newTemp(Ity_I32); |
| 12379 | IRTemp res_q = newTemp(Ity_I32); |
| 12380 | |
| 12381 | assign( rNt, isT ? getIRegT(regN) : getIRegA(regN) ); |
| 12382 | assign( rMt, isT ? getIRegT(regM) : getIRegA(regM) ); |
| 12383 | |
| 12384 | assign(res_q, binop(Iop_HSub16Sx2, mkexpr(rNt), mkexpr(rMt))); |
| 12385 | if (isT) |
| 12386 | putIRegT( regD, mkexpr(res_q), condT ); |
| 12387 | else |
| 12388 | putIRegA( regD, mkexpr(res_q), condT, Ijk_Boring ); |
| 12389 | |
| 12390 | DIP("shsub16%s r%u, r%u, r%u\n", nCC(conq),regD,regN,regM); |
| 12391 | return True; |
| 12392 | } |
| 12393 | /* fall through */ |
| 12394 | } |
| 12395 | |
sewardj | 4ccd258 | 2013-09-29 19:48:44 +0000 | [diff] [blame] | 12396 | /* ----------------- smmls{r}<c> <Rd>,<Rn>,<Rm>,<Ra> ------------------- */ |
| 12397 | { |
| 12398 | UInt rD = 99, rN = 99, rM = 99, rA = 99; |
| 12399 | Bool round = False; |
| 12400 | Bool gate = False; |
| 12401 | |
| 12402 | if (isT) { |
| 12403 | if (INSNT0(15,7) == BITS9(1,1,1,1,1,0,1,1,0) |
| 12404 | && INSNT0(6,4) == BITS3(1,1,0) |
| 12405 | && INSNT1(7,5) == BITS3(0,0,0)) { |
| 12406 | round = INSNT1(4,4); |
| 12407 | rA = INSNT1(15,12); |
| 12408 | rD = INSNT1(11,8); |
| 12409 | rM = INSNT1(3,0); |
| 12410 | rN = INSNT0(3,0); |
| 12411 | if (!isBadRegT(rD) |
| 12412 | && !isBadRegT(rN) && !isBadRegT(rM) && !isBadRegT(rA)) |
| 12413 | gate = True; |
| 12414 | } |
| 12415 | } else { |
| 12416 | if (INSNA(27,20) == BITS8(0,1,1,1,0,1,0,1) |
| 12417 | && INSNA(15,12) != BITS4(1,1,1,1) |
| 12418 | && (INSNA(7,4) & BITS4(1,1,0,1)) == BITS4(1,1,0,1)) { |
| 12419 | round = INSNA(5,5); |
| 12420 | rD = INSNA(19,16); |
| 12421 | rA = INSNA(15,12); |
| 12422 | rM = INSNA(11,8); |
| 12423 | rN = INSNA(3,0); |
| 12424 | if (rD != 15 && rM != 15 && rN != 15) |
| 12425 | gate = True; |
| 12426 | } |
| 12427 | } |
| 12428 | if (gate) { |
| 12429 | IRTemp irt_rA = newTemp(Ity_I32); |
| 12430 | IRTemp irt_rN = newTemp(Ity_I32); |
| 12431 | IRTemp irt_rM = newTemp(Ity_I32); |
| 12432 | assign( irt_rA, isT ? getIRegT(rA) : getIRegA(rA) ); |
| 12433 | assign( irt_rN, isT ? getIRegT(rN) : getIRegA(rN) ); |
| 12434 | assign( irt_rM, isT ? getIRegT(rM) : getIRegA(rM) ); |
| 12435 | IRExpr* res |
| 12436 | = unop(Iop_64HIto32, |
| 12437 | binop(Iop_Add64, |
| 12438 | binop(Iop_Sub64, |
| 12439 | binop(Iop_32HLto64, mkexpr(irt_rA), mkU32(0)), |
| 12440 | binop(Iop_MullS32, mkexpr(irt_rN), mkexpr(irt_rM))), |
| 12441 | mkU64(round ? 0x80000000ULL : 0ULL))); |
| 12442 | if (isT) |
| 12443 | putIRegT( rD, res, condT ); |
| 12444 | else |
| 12445 | putIRegA(rD, res, condT, Ijk_Boring); |
| 12446 | DIP("smmls%s%s r%u, r%u, r%u, r%u\n", |
| 12447 | round ? "r" : "", nCC(conq), rD, rN, rM, rA); |
| 12448 | return True; |
| 12449 | } |
| 12450 | /* fall through */ |
| 12451 | } |
| 12452 | |
| 12453 | /* -------------- smlald{x}<c> <RdLo>,<RdHi>,<Rn>,<Rm> ---------------- */ |
| 12454 | { |
| 12455 | UInt rN = 99, rDlo = 99, rDhi = 99, rM = 99; |
| 12456 | Bool m_swap = False; |
| 12457 | Bool gate = False; |
| 12458 | |
| 12459 | if (isT) { |
| 12460 | if (INSNT0(15,4) == 0xFBC && |
| 12461 | (INSNT1(7,4) & BITS4(1,1,1,0)) == BITS4(1,1,0,0)) { |
| 12462 | rN = INSNT0(3,0); |
| 12463 | rDlo = INSNT1(15,12); |
| 12464 | rDhi = INSNT1(11,8); |
| 12465 | rM = INSNT1(3,0); |
| 12466 | m_swap = (INSNT1(4,4) & 1) == 1; |
| 12467 | if (!isBadRegT(rDlo) && !isBadRegT(rDhi) && !isBadRegT(rN) |
| 12468 | && !isBadRegT(rM) && rDhi != rDlo) |
| 12469 | gate = True; |
| 12470 | } |
| 12471 | } else { |
| 12472 | if (INSNA(27,20) == BITS8(0,1,1,1,0,1,0,0) |
| 12473 | && (INSNA(7,4) & BITS4(1,1,0,1)) == BITS4(0,0,0,1)) { |
| 12474 | rN = INSNA(3,0); |
| 12475 | rDlo = INSNA(15,12); |
| 12476 | rDhi = INSNA(19,16); |
| 12477 | rM = INSNA(11,8); |
| 12478 | m_swap = ( INSNA(5,5) & 1 ) == 1; |
| 12479 | if (rDlo != 15 && rDhi != 15 |
| 12480 | && rN != 15 && rM != 15 && rDlo != rDhi) |
| 12481 | gate = True; |
| 12482 | } |
| 12483 | } |
| 12484 | |
| 12485 | if (gate) { |
| 12486 | IRTemp irt_rM = newTemp(Ity_I32); |
| 12487 | IRTemp irt_rN = newTemp(Ity_I32); |
| 12488 | IRTemp irt_rDhi = newTemp(Ity_I32); |
| 12489 | IRTemp irt_rDlo = newTemp(Ity_I32); |
| 12490 | IRTemp op_2 = newTemp(Ity_I32); |
| 12491 | IRTemp pr_1 = newTemp(Ity_I64); |
| 12492 | IRTemp pr_2 = newTemp(Ity_I64); |
| 12493 | IRTemp result = newTemp(Ity_I64); |
| 12494 | IRTemp resHi = newTemp(Ity_I32); |
| 12495 | IRTemp resLo = newTemp(Ity_I32); |
| 12496 | assign( irt_rM, isT ? getIRegT(rM) : getIRegA(rM)); |
| 12497 | assign( irt_rN, isT ? getIRegT(rN) : getIRegA(rN)); |
| 12498 | assign( irt_rDhi, isT ? getIRegT(rDhi) : getIRegA(rDhi)); |
| 12499 | assign( irt_rDlo, isT ? getIRegT(rDlo) : getIRegA(rDlo)); |
| 12500 | assign( op_2, genROR32(irt_rM, m_swap ? 16 : 0) ); |
| 12501 | assign( pr_1, binop(Iop_MullS32, |
| 12502 | unop(Iop_16Sto32, |
| 12503 | unop(Iop_32to16, mkexpr(irt_rN)) |
| 12504 | ), |
| 12505 | unop(Iop_16Sto32, |
| 12506 | unop(Iop_32to16, mkexpr(op_2)) |
| 12507 | ) |
| 12508 | ) |
| 12509 | ); |
| 12510 | assign( pr_2, binop(Iop_MullS32, |
| 12511 | binop(Iop_Sar32, mkexpr(irt_rN), mkU8(16)), |
| 12512 | binop(Iop_Sar32, mkexpr(op_2), mkU8(16)) |
| 12513 | ) |
| 12514 | ); |
| 12515 | assign( result, binop(Iop_Add64, |
| 12516 | binop(Iop_Add64, |
| 12517 | mkexpr(pr_1), |
| 12518 | mkexpr(pr_2) |
| 12519 | ), |
| 12520 | binop(Iop_32HLto64, |
| 12521 | mkexpr(irt_rDhi), |
| 12522 | mkexpr(irt_rDlo) |
| 12523 | ) |
| 12524 | ) |
| 12525 | ); |
| 12526 | assign( resHi, unop(Iop_64HIto32, mkexpr(result)) ); |
| 12527 | assign( resLo, unop(Iop_64to32, mkexpr(result)) ); |
| 12528 | if (isT) { |
| 12529 | putIRegT( rDhi, mkexpr(resHi), condT ); |
| 12530 | putIRegT( rDlo, mkexpr(resLo), condT ); |
| 12531 | } else { |
| 12532 | putIRegA( rDhi, mkexpr(resHi), condT, Ijk_Boring ); |
| 12533 | putIRegA( rDlo, mkexpr(resLo), condT, Ijk_Boring ); |
| 12534 | } |
| 12535 | DIP("smlald%c%s r%u, r%u, r%u, r%u\n", |
| 12536 | m_swap ? 'x' : ' ', nCC(conq), rDlo, rDhi, rN, rM); |
| 12537 | return True; |
| 12538 | } |
| 12539 | /* fall through */ |
| 12540 | } |
| 12541 | |
| 12542 | /* -------------- smlsld{x}<c> <RdLo>,<RdHi>,<Rn>,<Rm> ---------------- */ |
| 12543 | { |
| 12544 | UInt rN = 99, rDlo = 99, rDhi = 99, rM = 99; |
| 12545 | Bool m_swap = False; |
| 12546 | Bool gate = False; |
| 12547 | |
| 12548 | if (isT) { |
| 12549 | if ((INSNT0(15,4) == 0xFBD && |
| 12550 | (INSNT1(7,4) & BITS4(1,1,1,0)) == BITS4(1,1,0,0))) { |
| 12551 | rN = INSNT0(3,0); |
| 12552 | rDlo = INSNT1(15,12); |
| 12553 | rDhi = INSNT1(11,8); |
| 12554 | rM = INSNT1(3,0); |
| 12555 | m_swap = (INSNT1(4,4) & 1) == 1; |
| 12556 | if (!isBadRegT(rDlo) && !isBadRegT(rDhi) && !isBadRegT(rN) && |
| 12557 | !isBadRegT(rM) && rDhi != rDlo) |
| 12558 | gate = True; |
| 12559 | } |
| 12560 | } else { |
| 12561 | if (INSNA(27,20) == BITS8(0,1,1,1,0,1,0,0) && |
| 12562 | (INSNA(7,4) & BITS4(1,1,0,1)) == BITS4(0,1,0,1)) { |
| 12563 | rN = INSNA(3,0); |
| 12564 | rDlo = INSNA(15,12); |
| 12565 | rDhi = INSNA(19,16); |
| 12566 | rM = INSNA(11,8); |
| 12567 | m_swap = (INSNA(5,5) & 1) == 1; |
| 12568 | if (rDlo != 15 && rDhi != 15 && |
| 12569 | rN != 15 && rM != 15 && rDlo != rDhi) |
| 12570 | gate = True; |
| 12571 | } |
| 12572 | } |
| 12573 | if (gate) { |
| 12574 | IRTemp irt_rM = newTemp(Ity_I32); |
| 12575 | IRTemp irt_rN = newTemp(Ity_I32); |
| 12576 | IRTemp irt_rDhi = newTemp(Ity_I32); |
| 12577 | IRTemp irt_rDlo = newTemp(Ity_I32); |
| 12578 | IRTemp op_2 = newTemp(Ity_I32); |
| 12579 | IRTemp pr_1 = newTemp(Ity_I64); |
| 12580 | IRTemp pr_2 = newTemp(Ity_I64); |
| 12581 | IRTemp result = newTemp(Ity_I64); |
| 12582 | IRTemp resHi = newTemp(Ity_I32); |
| 12583 | IRTemp resLo = newTemp(Ity_I32); |
| 12584 | assign( irt_rM, isT ? getIRegT(rM) : getIRegA(rM) ); |
| 12585 | assign( irt_rN, isT ? getIRegT(rN) : getIRegA(rN) ); |
| 12586 | assign( irt_rDhi, isT ? getIRegT(rDhi) : getIRegA(rDhi) ); |
| 12587 | assign( irt_rDlo, isT ? getIRegT(rDlo) : getIRegA(rDlo) ); |
| 12588 | assign( op_2, genROR32(irt_rM, m_swap ? 16 : 0) ); |
| 12589 | assign( pr_1, binop(Iop_MullS32, |
| 12590 | unop(Iop_16Sto32, |
| 12591 | unop(Iop_32to16, mkexpr(irt_rN)) |
| 12592 | ), |
| 12593 | unop(Iop_16Sto32, |
| 12594 | unop(Iop_32to16, mkexpr(op_2)) |
| 12595 | ) |
| 12596 | ) |
| 12597 | ); |
| 12598 | assign( pr_2, binop(Iop_MullS32, |
| 12599 | binop(Iop_Sar32, mkexpr(irt_rN), mkU8(16)), |
| 12600 | binop(Iop_Sar32, mkexpr(op_2), mkU8(16)) |
| 12601 | ) |
| 12602 | ); |
| 12603 | assign( result, binop(Iop_Add64, |
| 12604 | binop(Iop_Sub64, |
| 12605 | mkexpr(pr_1), |
| 12606 | mkexpr(pr_2) |
| 12607 | ), |
| 12608 | binop(Iop_32HLto64, |
| 12609 | mkexpr(irt_rDhi), |
| 12610 | mkexpr(irt_rDlo) |
| 12611 | ) |
| 12612 | ) |
| 12613 | ); |
| 12614 | assign( resHi, unop(Iop_64HIto32, mkexpr(result)) ); |
| 12615 | assign( resLo, unop(Iop_64to32, mkexpr(result)) ); |
| 12616 | if (isT) { |
| 12617 | putIRegT( rDhi, mkexpr(resHi), condT ); |
| 12618 | putIRegT( rDlo, mkexpr(resLo), condT ); |
| 12619 | } else { |
| 12620 | putIRegA( rDhi, mkexpr(resHi), condT, Ijk_Boring ); |
| 12621 | putIRegA( rDlo, mkexpr(resLo), condT, Ijk_Boring ); |
| 12622 | } |
| 12623 | DIP("smlsld%c%s r%u, r%u, r%u, r%u\n", |
| 12624 | m_swap ? 'x' : ' ', nCC(conq), rDlo, rDhi, rN, rM); |
| 12625 | return True; |
| 12626 | } |
| 12627 | /* fall through */ |
| 12628 | } |
| 12629 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 12630 | /* ---------- Doesn't match anything. ---------- */ |
| 12631 | return False; |
| 12632 | |
| 12633 | # undef INSNA |
| 12634 | # undef INSNT0 |
| 12635 | # undef INSNT1 |
| 12636 | } |
| 12637 | |
| 12638 | |
| 12639 | /*------------------------------------------------------------*/ |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 12640 | /*--- V8 instructions ---*/ |
| 12641 | /*------------------------------------------------------------*/ |
| 12642 | |
| 12643 | /* Break a V128-bit value up into four 32-bit ints. */ |
| 12644 | |
| 12645 | static void breakupV128to32s ( IRTemp t128, |
| 12646 | /*OUTs*/ |
| 12647 | IRTemp* t3, IRTemp* t2, |
| 12648 | IRTemp* t1, IRTemp* t0 ) |
| 12649 | { |
| 12650 | IRTemp hi64 = newTemp(Ity_I64); |
| 12651 | IRTemp lo64 = newTemp(Ity_I64); |
| 12652 | assign( hi64, unop(Iop_V128HIto64, mkexpr(t128)) ); |
| 12653 | assign( lo64, unop(Iop_V128to64, mkexpr(t128)) ); |
| 12654 | |
| 12655 | vassert(t0 && *t0 == IRTemp_INVALID); |
| 12656 | vassert(t1 && *t1 == IRTemp_INVALID); |
| 12657 | vassert(t2 && *t2 == IRTemp_INVALID); |
| 12658 | vassert(t3 && *t3 == IRTemp_INVALID); |
| 12659 | |
| 12660 | *t0 = newTemp(Ity_I32); |
| 12661 | *t1 = newTemp(Ity_I32); |
| 12662 | *t2 = newTemp(Ity_I32); |
| 12663 | *t3 = newTemp(Ity_I32); |
| 12664 | assign( *t0, unop(Iop_64to32, mkexpr(lo64)) ); |
| 12665 | assign( *t1, unop(Iop_64HIto32, mkexpr(lo64)) ); |
| 12666 | assign( *t2, unop(Iop_64to32, mkexpr(hi64)) ); |
| 12667 | assign( *t3, unop(Iop_64HIto32, mkexpr(hi64)) ); |
| 12668 | } |
| 12669 | |
| 12670 | |
| 12671 | /* Both ARM and Thumb */ |
| 12672 | |
| 12673 | /* Translate a V8 instruction. If successful, returns True and *dres |
| 12674 | may or may not be updated. If unsuccessful, returns False and |
| 12675 | doesn't change *dres nor create any IR. |
| 12676 | |
| 12677 | The Thumb and ARM encodings are potentially different. In both |
| 12678 | ARM and Thumb mode, the caller must pass the entire 32 bits of |
| 12679 | the instruction. Callers may pass any instruction; this function |
| 12680 | ignores anything it doesn't recognise. |
| 12681 | |
| 12682 | Caller must supply an IRTemp 'condT' holding the gating condition, |
| 12683 | or IRTemp_INVALID indicating the insn is always executed. |
| 12684 | |
| 12685 | If we are decoding an ARM instruction which is in the NV space |
| 12686 | then it is expected that condT will be IRTemp_INVALID, and that is |
| 12687 | asserted for. That condition is ensured by the logic near the top |
| 12688 | of disInstr_ARM_WRK, that sets up condT. |
| 12689 | |
| 12690 | When decoding for Thumb, the caller must pass the ITState pre/post |
| 12691 | this instruction, so that we can generate a SIGILL in the cases where |
| 12692 | the instruction may not be in an IT block. When decoding for ARM, |
| 12693 | both of these must be IRTemp_INVALID. |
| 12694 | |
| 12695 | Finally, the caller must indicate whether this occurs in ARM or in |
| 12696 | Thumb code. |
| 12697 | */ |
| 12698 | static Bool decode_V8_instruction ( |
| 12699 | /*MOD*/DisResult* dres, |
| 12700 | UInt insnv8, |
| 12701 | IRTemp condT, |
| 12702 | Bool isT, |
| 12703 | IRTemp old_itstate, |
| 12704 | IRTemp new_itstate |
| 12705 | ) |
| 12706 | { |
| 12707 | # define INSN(_bMax,_bMin) SLICE_UInt(insnv8, (_bMax), (_bMin)) |
| 12708 | |
| 12709 | if (isT) { |
| 12710 | vassert(old_itstate != IRTemp_INVALID); |
| 12711 | vassert(new_itstate != IRTemp_INVALID); |
| 12712 | } else { |
| 12713 | vassert(old_itstate == IRTemp_INVALID); |
| 12714 | vassert(new_itstate == IRTemp_INVALID); |
| 12715 | } |
| 12716 | |
| 12717 | /* ARMCondcode 'conq' is only used for debug printing and for no other |
| 12718 | purpose. For ARM, this is simply the top 4 bits of the instruction. |
| 12719 | For Thumb, the condition is not (really) known until run time, and so |
| 12720 | we set it to ARMCondAL in order that printing of these instructions |
| 12721 | does not show any condition. */ |
| 12722 | ARMCondcode conq; |
| 12723 | if (isT) { |
| 12724 | conq = ARMCondAL; |
| 12725 | } else { |
| 12726 | conq = (ARMCondcode)INSN(31,28); |
| 12727 | if (conq == ARMCondNV || conq == ARMCondAL) { |
| 12728 | vassert(condT == IRTemp_INVALID); |
| 12729 | } else { |
| 12730 | vassert(condT != IRTemp_INVALID); |
| 12731 | } |
| 12732 | vassert(conq >= ARMCondEQ && conq <= ARMCondNV); |
| 12733 | } |
| 12734 | |
| 12735 | /* ----------- {AESD, AESE, AESMC, AESIMC}.8 q_q ----------- */ |
| 12736 | /* 31 27 23 21 19 17 15 11 7 3 |
| 12737 | T1: 1111 1111 1 D 11 sz 00 d 0011 00 M 0 m AESE Qd, Qm |
| 12738 | A1: 1111 0011 1 D 11 sz 00 d 0011 00 M 0 m AESE Qd, Qm |
| 12739 | |
| 12740 | T1: 1111 1111 1 D 11 sz 00 d 0011 01 M 0 m AESD Qd, Qm |
| 12741 | A1: 1111 0011 1 D 11 sz 00 d 0011 01 M 0 m AESD Qd, Qm |
| 12742 | |
| 12743 | T1: 1111 1111 1 D 11 sz 00 d 0011 10 M 0 m AESMC Qd, Qm |
| 12744 | A1: 1111 0011 1 D 11 sz 00 d 0011 10 M 0 m AESMC Qd, Qm |
| 12745 | |
| 12746 | T1: 1111 1111 1 D 11 sz 00 d 0011 11 M 0 m AESIMC Qd, Qm |
| 12747 | A1: 1111 0011 1 D 11 sz 00 d 0011 11 M 0 m AESIMC Qd, Qm |
| 12748 | |
| 12749 | sz must be 00 |
| 12750 | ARM encoding is in NV space. |
| 12751 | In Thumb mode, we must not be in an IT block. |
| 12752 | */ |
| 12753 | { |
| 12754 | UInt regD = 99, regM = 99, opc = 4/*invalid*/; |
| 12755 | Bool gate = True; |
| 12756 | |
| 12757 | UInt high9 = isT ? BITS9(1,1,1,1,1,1,1,1,1) : BITS9(1,1,1,1,0,0,1,1,1); |
| 12758 | if (INSN(31,23) == high9 && INSN(21,16) == BITS6(1,1,0,0,0,0) |
| 12759 | && INSN(11,8) == BITS4(0,0,1,1) && INSN(4,4) == 0) { |
| 12760 | UInt bitD = INSN(22,22); |
| 12761 | UInt fldD = INSN(15,12); |
| 12762 | UInt bitM = INSN(5,5); |
| 12763 | UInt fldM = INSN(3,0); |
| 12764 | opc = INSN(7,6); |
| 12765 | regD = (bitD << 4) | fldD; |
| 12766 | regM = (bitM << 4) | fldM; |
| 12767 | } |
| 12768 | if ((regD & 1) == 1 || (regM & 1) == 1) |
| 12769 | gate = False; |
| 12770 | |
| 12771 | if (gate) { |
| 12772 | if (isT) { |
| 12773 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 12774 | } |
| 12775 | /* In ARM mode, this is statically unconditional. In Thumb mode, |
| 12776 | this must be dynamically unconditional, and we've SIGILLd if not. |
| 12777 | In either case we can create unconditional IR. */ |
| 12778 | IRTemp op1 = newTemp(Ity_V128); |
| 12779 | IRTemp op2 = newTemp(Ity_V128); |
| 12780 | IRTemp src = newTemp(Ity_V128); |
| 12781 | IRTemp res = newTemp(Ity_V128); |
| 12782 | assign(op1, getQReg(regD >> 1)); |
| 12783 | assign(op2, getQReg(regM >> 1)); |
| 12784 | assign(src, opc == BITS2(0,0) || opc == BITS2(0,1) |
| 12785 | ? binop(Iop_XorV128, mkexpr(op1), mkexpr(op2)) |
| 12786 | : mkexpr(op2)); |
| 12787 | |
| 12788 | void* helpers[4] |
| 12789 | = { &armg_dirtyhelper_AESE, &armg_dirtyhelper_AESD, |
| 12790 | &armg_dirtyhelper_AESMC, &armg_dirtyhelper_AESIMC }; |
| 12791 | const HChar* hNames[4] |
| 12792 | = { "armg_dirtyhelper_AESE", "armg_dirtyhelper_AESD", |
| 12793 | "armg_dirtyhelper_AESMC", "armg_dirtyhelper_AESIMC" }; |
| 12794 | const HChar* iNames[4] |
| 12795 | = { "aese", "aesd", "aesmc", "aesimc" }; |
| 12796 | |
| 12797 | vassert(opc >= 0 && opc <= 3); |
| 12798 | void* helper = helpers[opc]; |
| 12799 | const HChar* hname = hNames[opc]; |
| 12800 | |
| 12801 | IRTemp w32_3, w32_2, w32_1, w32_0; |
| 12802 | w32_3 = w32_2 = w32_1 = w32_0 = IRTemp_INVALID; |
| 12803 | breakupV128to32s( src, &w32_3, &w32_2, &w32_1, &w32_0 ); |
| 12804 | |
| 12805 | IRDirty* di |
| 12806 | = unsafeIRDirty_1_N( res, 0/*regparms*/, hname, helper, |
| 12807 | mkIRExprVec_5( |
| 12808 | IRExpr_VECRET(), |
| 12809 | mkexpr(w32_3), mkexpr(w32_2), |
| 12810 | mkexpr(w32_1), mkexpr(w32_0)) ); |
| 12811 | stmt(IRStmt_Dirty(di)); |
| 12812 | |
| 12813 | putQReg(regD >> 1, mkexpr(res), IRTemp_INVALID); |
| 12814 | DIP("%s.8 q%d, q%d\n", iNames[opc], regD >> 1, regM >> 1); |
| 12815 | return True; |
| 12816 | } |
| 12817 | /* fall through */ |
| 12818 | } |
| 12819 | |
| 12820 | /* ----------- SHA 3-reg insns q_q_q ----------- */ |
| 12821 | /* |
| 12822 | 31 27 23 19 15 11 7 3 |
| 12823 | T1: 1110 1111 0 D 00 n d 1100 N Q M 0 m SHA1C Qd, Qn, Qm ix=0 |
| 12824 | A1: 1111 0010 ---------------------------- |
| 12825 | |
| 12826 | T1: 1110 1111 0 D 01 n d 1100 N Q M 0 m SHA1P Qd, Qn, Qm ix=1 |
| 12827 | A1: 1111 0010 ---------------------------- |
| 12828 | |
| 12829 | T1: 1110 1111 0 D 10 n d 1100 N Q M 0 m SHA1M Qd, Qn, Qm ix=2 |
| 12830 | A1: 1111 0010 ---------------------------- |
| 12831 | |
| 12832 | T1: 1110 1111 0 D 11 n d 1100 N Q M 0 m SHA1SU0 Qd, Qn, Qm ix=3 |
| 12833 | A1: 1111 0010 ---------------------------- |
| 12834 | (that's a complete set of 4, based on insn[21,20]) |
| 12835 | |
| 12836 | T1: 1111 1111 0 D 00 n d 1100 N Q M 0 m SHA256H Qd, Qn, Qm ix=4 |
| 12837 | A1: 1111 0011 ---------------------------- |
| 12838 | |
| 12839 | T1: 1111 1111 0 D 01 n d 1100 N Q M 0 m SHA256H2 Qd, Qn, Qm ix=5 |
| 12840 | A1: 1111 0011 ---------------------------- |
| 12841 | |
| 12842 | T1: 1111 1111 0 D 10 n d 1100 N Q M 0 m SHA256SU1 Qd, Qn, Qm ix=6 |
| 12843 | A1: 1111 0011 ---------------------------- |
| 12844 | (3/4 of a complete set of 4, based on insn[21,20]) |
| 12845 | |
| 12846 | Q must be 1. Same comments about conditionalisation as for the AES |
| 12847 | group above apply. |
| 12848 | */ |
| 12849 | { |
| 12850 | UInt ix = 8; /* invalid */ |
| 12851 | Bool gate = False; |
| 12852 | |
| 12853 | UInt hi9_sha1 = isT ? BITS9(1,1,1,0,1,1,1,1,0) |
| 12854 | : BITS9(1,1,1,1,0,0,1,0,0); |
| 12855 | UInt hi9_sha256 = isT ? BITS9(1,1,1,1,1,1,1,1,0) |
| 12856 | : BITS9(1,1,1,1,0,0,1,1,0); |
| 12857 | if ((INSN(31,23) == hi9_sha1 || INSN(31,23) == hi9_sha256) |
| 12858 | && INSN(11,8) == BITS4(1,1,0,0) |
| 12859 | && INSN(6,6) == 1 && INSN(4,4) == 0) { |
| 12860 | ix = INSN(21,20); |
| 12861 | if (INSN(31,23) == hi9_sha256) |
| 12862 | ix |= 4; |
| 12863 | if (ix < 7) |
| 12864 | gate = True; |
| 12865 | } |
| 12866 | |
| 12867 | UInt regN = (INSN(7,7) << 4) | INSN(19,16); |
| 12868 | UInt regD = (INSN(22,22) << 4) | INSN(15,12); |
| 12869 | UInt regM = (INSN(5,5) << 4) | INSN(3,0); |
| 12870 | if ((regD & 1) == 1 || (regM & 1) == 1 || (regN & 1) == 1) |
| 12871 | gate = False; |
| 12872 | |
| 12873 | if (gate) { |
| 12874 | vassert(ix >= 0 && ix < 7); |
| 12875 | const HChar* inames[7] |
| 12876 | = { "sha1c", "sha1p", "sha1m", "sha1su0", |
| 12877 | "sha256h", "sha256h2", "sha256su1" }; |
| 12878 | void(*helpers[7])(V128*,UInt,UInt,UInt,UInt,UInt,UInt, |
| 12879 | UInt,UInt,UInt,UInt,UInt,UInt) |
| 12880 | = { &armg_dirtyhelper_SHA1C, &armg_dirtyhelper_SHA1P, |
| 12881 | &armg_dirtyhelper_SHA1M, &armg_dirtyhelper_SHA1SU0, |
| 12882 | &armg_dirtyhelper_SHA256H, &armg_dirtyhelper_SHA256H2, |
| 12883 | &armg_dirtyhelper_SHA256SU1 }; |
| 12884 | const HChar* hnames[7] |
| 12885 | = { "armg_dirtyhelper_SHA1C", "armg_dirtyhelper_SHA1P", |
| 12886 | "armg_dirtyhelper_SHA1M", "armg_dirtyhelper_SHA1SU0", |
| 12887 | "armg_dirtyhelper_SHA256H", "armg_dirtyhelper_SHA256H2", |
| 12888 | "armg_dirtyhelper_SHA256SU1" }; |
| 12889 | |
| 12890 | /* This is a really lame way to implement this, even worse than |
| 12891 | the arm64 version. But at least it works. */ |
| 12892 | |
| 12893 | if (isT) { |
| 12894 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 12895 | } |
| 12896 | |
| 12897 | IRTemp vD = newTemp(Ity_V128); |
| 12898 | IRTemp vN = newTemp(Ity_V128); |
| 12899 | IRTemp vM = newTemp(Ity_V128); |
| 12900 | assign(vD, getQReg(regD >> 1)); |
| 12901 | assign(vN, getQReg(regN >> 1)); |
| 12902 | assign(vM, getQReg(regM >> 1)); |
| 12903 | |
| 12904 | IRTemp d32_3, d32_2, d32_1, d32_0; |
| 12905 | d32_3 = d32_2 = d32_1 = d32_0 = IRTemp_INVALID; |
| 12906 | breakupV128to32s( vD, &d32_3, &d32_2, &d32_1, &d32_0 ); |
| 12907 | |
| 12908 | IRTemp n32_3_pre, n32_2_pre, n32_1_pre, n32_0_pre; |
| 12909 | n32_3_pre = n32_2_pre = n32_1_pre = n32_0_pre = IRTemp_INVALID; |
| 12910 | breakupV128to32s( vN, &n32_3_pre, &n32_2_pre, &n32_1_pre, &n32_0_pre ); |
| 12911 | |
| 12912 | IRTemp m32_3, m32_2, m32_1, m32_0; |
| 12913 | m32_3 = m32_2 = m32_1 = m32_0 = IRTemp_INVALID; |
| 12914 | breakupV128to32s( vM, &m32_3, &m32_2, &m32_1, &m32_0 ); |
| 12915 | |
| 12916 | IRTemp n32_3 = newTemp(Ity_I32); |
| 12917 | IRTemp n32_2 = newTemp(Ity_I32); |
| 12918 | IRTemp n32_1 = newTemp(Ity_I32); |
| 12919 | IRTemp n32_0 = newTemp(Ity_I32); |
| 12920 | |
| 12921 | /* Mask off any bits of the N register operand that aren't actually |
| 12922 | needed, so that Memcheck doesn't complain unnecessarily. */ |
| 12923 | switch (ix) { |
| 12924 | case 0: case 1: case 2: |
| 12925 | assign(n32_3, mkU32(0)); |
| 12926 | assign(n32_2, mkU32(0)); |
| 12927 | assign(n32_1, mkU32(0)); |
| 12928 | assign(n32_0, mkexpr(n32_0_pre)); |
| 12929 | break; |
| 12930 | case 3: case 4: case 5: case 6: |
| 12931 | assign(n32_3, mkexpr(n32_3_pre)); |
| 12932 | assign(n32_2, mkexpr(n32_2_pre)); |
| 12933 | assign(n32_1, mkexpr(n32_1_pre)); |
| 12934 | assign(n32_0, mkexpr(n32_0_pre)); |
| 12935 | break; |
| 12936 | default: |
| 12937 | vassert(0); |
| 12938 | } |
| 12939 | |
| 12940 | IRExpr** argvec |
| 12941 | = mkIRExprVec_13( |
| 12942 | IRExpr_VECRET(), |
| 12943 | mkexpr(d32_3), mkexpr(d32_2), mkexpr(d32_1), mkexpr(d32_0), |
| 12944 | mkexpr(n32_3), mkexpr(n32_2), mkexpr(n32_1), mkexpr(n32_0), |
| 12945 | mkexpr(m32_3), mkexpr(m32_2), mkexpr(m32_1), mkexpr(m32_0) |
| 12946 | ); |
| 12947 | |
| 12948 | IRTemp res = newTemp(Ity_V128); |
| 12949 | IRDirty* di = unsafeIRDirty_1_N( res, 0/*regparms*/, |
| 12950 | hnames[ix], helpers[ix], argvec ); |
| 12951 | stmt(IRStmt_Dirty(di)); |
| 12952 | putQReg(regD >> 1, mkexpr(res), IRTemp_INVALID); |
| 12953 | |
| 12954 | DIP("%s.8 q%u, q%u, q%u\n", |
| 12955 | inames[ix], regD >> 1, regN >> 1, regM >> 1); |
| 12956 | return True; |
| 12957 | } |
| 12958 | /* fall through */ |
| 12959 | } |
| 12960 | |
| 12961 | /* ----------- SHA1SU1, SHA256SU0 ----------- */ |
| 12962 | /* |
| 12963 | 31 27 23 21 19 15 11 7 3 |
| 12964 | T1: 1111 1111 1 D 11 1010 d 0011 10 M 0 m SHA1SU1 Qd, Qm |
| 12965 | A1: 1111 0011 ---------------------------- |
| 12966 | |
| 12967 | T1: 1111 1111 1 D 11 1010 d 0011 11 M 0 m SHA256SU0 Qd, Qm |
| 12968 | A1: 1111 0011 ---------------------------- |
| 12969 | |
| 12970 | Same comments about conditionalisation as for the AES group above apply. |
| 12971 | */ |
| 12972 | { |
| 12973 | Bool gate = False; |
| 12974 | |
| 12975 | UInt hi9 = isT ? BITS9(1,1,1,1,1,1,1,1,1) : BITS9(1,1,1,1,0,0,1,1,1); |
| 12976 | if (INSN(31,23) == hi9 && INSN(21,16) == BITS6(1,1,1,0,1,0) |
| 12977 | && INSN(11,7) == BITS5(0,0,1,1,1) && INSN(4,4) == 0) { |
| 12978 | gate = True; |
| 12979 | } |
| 12980 | |
| 12981 | UInt regD = (INSN(22,22) << 4) | INSN(15,12); |
| 12982 | UInt regM = (INSN(5,5) << 4) | INSN(3,0); |
| 12983 | if ((regD & 1) == 1 || (regM & 1) == 1) |
| 12984 | gate = False; |
| 12985 | |
| 12986 | Bool is_1SU1 = INSN(6,6) == 0; |
| 12987 | |
| 12988 | if (gate) { |
| 12989 | const HChar* iname |
| 12990 | = is_1SU1 ? "sha1su1" : "sha256su0"; |
| 12991 | void (*helper)(V128*,UInt,UInt,UInt,UInt,UInt,UInt,UInt,UInt) |
| 12992 | = is_1SU1 ? &armg_dirtyhelper_SHA1SU1 |
| 12993 | : *armg_dirtyhelper_SHA256SU0; |
| 12994 | const HChar* hname |
| 12995 | = is_1SU1 ? "armg_dirtyhelper_SHA1SU1" |
| 12996 | : "armg_dirtyhelper_SHA256SU0"; |
| 12997 | |
| 12998 | if (isT) { |
| 12999 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 13000 | } |
| 13001 | |
| 13002 | IRTemp vD = newTemp(Ity_V128); |
| 13003 | IRTemp vM = newTemp(Ity_V128); |
| 13004 | assign(vD, getQReg(regD >> 1)); |
| 13005 | assign(vM, getQReg(regM >> 1)); |
| 13006 | |
| 13007 | IRTemp d32_3, d32_2, d32_1, d32_0; |
| 13008 | d32_3 = d32_2 = d32_1 = d32_0 = IRTemp_INVALID; |
| 13009 | breakupV128to32s( vD, &d32_3, &d32_2, &d32_1, &d32_0 ); |
| 13010 | |
| 13011 | IRTemp m32_3, m32_2, m32_1, m32_0; |
| 13012 | m32_3 = m32_2 = m32_1 = m32_0 = IRTemp_INVALID; |
| 13013 | breakupV128to32s( vM, &m32_3, &m32_2, &m32_1, &m32_0 ); |
| 13014 | |
| 13015 | IRExpr** argvec |
| 13016 | = mkIRExprVec_9( |
| 13017 | IRExpr_VECRET(), |
| 13018 | mkexpr(d32_3), mkexpr(d32_2), mkexpr(d32_1), mkexpr(d32_0), |
| 13019 | mkexpr(m32_3), mkexpr(m32_2), mkexpr(m32_1), mkexpr(m32_0) |
| 13020 | ); |
| 13021 | |
| 13022 | IRTemp res = newTemp(Ity_V128); |
| 13023 | IRDirty* di = unsafeIRDirty_1_N( res, 0/*regparms*/, |
| 13024 | hname, helper, argvec ); |
| 13025 | stmt(IRStmt_Dirty(di)); |
| 13026 | putQReg(regD >> 1, mkexpr(res), IRTemp_INVALID); |
| 13027 | |
| 13028 | DIP("%s.8 q%u, q%u\n", iname, regD >> 1, regM >> 1); |
| 13029 | return True; |
| 13030 | } |
| 13031 | /* fall through */ |
| 13032 | } |
| 13033 | |
| 13034 | /* ----------- SHA1H ----------- */ |
| 13035 | /* |
| 13036 | 31 27 23 21 19 15 11 7 3 |
| 13037 | T1: 1111 1111 1 D 11 1001 d 0010 11 M 0 m SHA1H Qd, Qm |
| 13038 | A1: 1111 0011 ---------------------------- |
| 13039 | |
| 13040 | Same comments about conditionalisation as for the AES group above apply. |
| 13041 | */ |
| 13042 | { |
| 13043 | Bool gate = False; |
| 13044 | |
| 13045 | UInt hi9 = isT ? BITS9(1,1,1,1,1,1,1,1,1) : BITS9(1,1,1,1,0,0,1,1,1); |
| 13046 | if (INSN(31,23) == hi9 && INSN(21,16) == BITS6(1,1,1,0,0,1) |
| 13047 | && INSN(11,6) == BITS6(0,0,1,0,1,1) && INSN(4,4) == 0) { |
| 13048 | gate = True; |
| 13049 | } |
| 13050 | |
| 13051 | UInt regD = (INSN(22,22) << 4) | INSN(15,12); |
| 13052 | UInt regM = (INSN(5,5) << 4) | INSN(3,0); |
| 13053 | if ((regD & 1) == 1 || (regM & 1) == 1) |
| 13054 | gate = False; |
| 13055 | |
| 13056 | if (gate) { |
| 13057 | const HChar* iname = "sha1h"; |
| 13058 | void (*helper)(V128*,UInt,UInt,UInt,UInt) = &armg_dirtyhelper_SHA1H; |
| 13059 | const HChar* hname = "armg_dirtyhelper_SHA1H"; |
| 13060 | |
| 13061 | if (isT) { |
| 13062 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 13063 | } |
| 13064 | |
| 13065 | IRTemp vM = newTemp(Ity_V128); |
| 13066 | assign(vM, getQReg(regM >> 1)); |
| 13067 | |
| 13068 | IRTemp m32_3, m32_2, m32_1, m32_0; |
| 13069 | m32_3 = m32_2 = m32_1 = m32_0 = IRTemp_INVALID; |
| 13070 | breakupV128to32s( vM, &m32_3, &m32_2, &m32_1, &m32_0 ); |
| 13071 | /* m32_3, m32_2, m32_1 are just abandoned. No harm; iropt will |
| 13072 | remove them. */ |
| 13073 | |
| 13074 | IRExpr* zero = mkU32(0); |
| 13075 | IRExpr** argvec = mkIRExprVec_5(IRExpr_VECRET(), |
| 13076 | zero, zero, zero, mkexpr(m32_0)); |
| 13077 | |
| 13078 | IRTemp res = newTemp(Ity_V128); |
| 13079 | IRDirty* di = unsafeIRDirty_1_N( res, 0/*regparms*/, |
| 13080 | hname, helper, argvec ); |
| 13081 | stmt(IRStmt_Dirty(di)); |
| 13082 | putQReg(regD >> 1, mkexpr(res), IRTemp_INVALID); |
| 13083 | |
| 13084 | DIP("%s.8 q%u, q%u\n", iname, regD >> 1, regM >> 1); |
| 13085 | return True; |
| 13086 | } |
| 13087 | /* fall through */ |
| 13088 | } |
| 13089 | |
| 13090 | /* ----------- VMULL.P64 ----------- */ |
| 13091 | /* |
| 13092 | 31 27 23 21 19 15 11 7 3 |
| 13093 | T2: 1110 1111 1 D 10 n d 1110 N 0 M 0 m |
| 13094 | A2: 1111 0010 ------------------------- |
| 13095 | |
| 13096 | The ARM documentation is pretty difficult to follow here. |
| 13097 | Same comments about conditionalisation as for the AES group above apply. |
| 13098 | */ |
| 13099 | { |
| 13100 | Bool gate = False; |
| 13101 | |
| 13102 | UInt hi9 = isT ? BITS9(1,1,1,0,1,1,1,1,1) : BITS9(1,1,1,1,0,0,1,0,1); |
| 13103 | if (INSN(31,23) == hi9 && INSN(21,20) == BITS2(1,0) |
| 13104 | && INSN(11,8) == BITS4(1,1,1,0) |
| 13105 | && INSN(6,6) == 0 && INSN(4,4) == 0) { |
| 13106 | gate = True; |
| 13107 | } |
| 13108 | |
| 13109 | UInt regN = (INSN(7,7) << 4) | INSN(19,16); |
| 13110 | UInt regD = (INSN(22,22) << 4) | INSN(15,12); |
| 13111 | UInt regM = (INSN(5,5) << 4) | INSN(3,0); |
| 13112 | |
| 13113 | if ((regD & 1) == 1) |
| 13114 | gate = False; |
| 13115 | |
| 13116 | if (gate) { |
| 13117 | const HChar* iname = "vmull"; |
| 13118 | void (*helper)(V128*,UInt,UInt,UInt,UInt) = &armg_dirtyhelper_VMULLP64; |
| 13119 | const HChar* hname = "armg_dirtyhelper_VMULLP64"; |
| 13120 | |
| 13121 | if (isT) { |
| 13122 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 13123 | } |
| 13124 | |
| 13125 | IRTemp srcN = newTemp(Ity_I64); |
| 13126 | IRTemp srcM = newTemp(Ity_I64); |
| 13127 | assign(srcN, getDRegI64(regN)); |
| 13128 | assign(srcM, getDRegI64(regM)); |
| 13129 | |
| 13130 | IRExpr** argvec = mkIRExprVec_5(IRExpr_VECRET(), |
| 13131 | unop(Iop_64HIto32, mkexpr(srcN)), |
| 13132 | unop(Iop_64to32, mkexpr(srcN)), |
| 13133 | unop(Iop_64HIto32, mkexpr(srcM)), |
| 13134 | unop(Iop_64to32, mkexpr(srcM))); |
| 13135 | |
| 13136 | IRTemp res = newTemp(Ity_V128); |
| 13137 | IRDirty* di = unsafeIRDirty_1_N( res, 0/*regparms*/, |
| 13138 | hname, helper, argvec ); |
| 13139 | stmt(IRStmt_Dirty(di)); |
| 13140 | putQReg(regD >> 1, mkexpr(res), IRTemp_INVALID); |
| 13141 | |
| 13142 | DIP("%s.p64 q%u, q%u, w%u\n", iname, regD >> 1, regN, regM); |
| 13143 | return True; |
| 13144 | } |
| 13145 | /* fall through */ |
| 13146 | } |
| 13147 | |
| 13148 | /* ----------- LDA{,B,H}, STL{,B,H} ----------- */ |
| 13149 | /* 31 27 23 19 15 11 7 3 |
| 13150 | A1: cond 0001 1001 n t 1100 1001 1111 LDA Rt, [Rn] |
| 13151 | A1: cond 0001 1111 n t 1100 1001 1111 LDAH Rt, [Rn] |
| 13152 | A1: cond 0001 1101 n t 1100 1001 1111 LDAB Rt, [Rn] |
| 13153 | |
| 13154 | A1: cond 0001 1000 n 1111 1100 1001 t STL Rt, [Rn] |
| 13155 | A1: cond 0001 1110 n 1111 1100 1001 t STLH Rt, [Rn] |
| 13156 | A1: cond 0001 1100 n 1111 1100 1001 t STLB Rt, [Rn] |
| 13157 | |
| 13158 | T1: 1110 1000 1101 n t 1111 1010 1111 LDA Rt, [Rn] |
| 13159 | T1: 1110 1000 1101 n t 1111 1001 1111 LDAH Rt, [Rn] |
| 13160 | T1: 1110 1000 1101 n t 1111 1000 1111 LDAB Rt, [Rn] |
| 13161 | |
| 13162 | T1: 1110 1000 1100 n t 1111 1010 1111 STL Rt, [Rn] |
| 13163 | T1: 1110 1000 1100 n t 1111 1001 1111 STLH Rt, [Rn] |
| 13164 | T1: 1110 1000 1100 n t 1111 1000 1111 STLB Rt, [Rn] |
| 13165 | */ |
| 13166 | { |
| 13167 | UInt nn = 16; // invalid |
| 13168 | UInt tt = 16; // invalid |
| 13169 | UInt szBlg2 = 4; // invalid |
| 13170 | Bool isLoad = False; |
| 13171 | Bool gate = False; |
| 13172 | if (isT) { |
| 13173 | if (INSN(31,21) == BITS11(1,1,1,0,1,0,0,0,1,1,0) |
| 13174 | && INSN(11,6) == BITS6(1,1,1,1,1,0) |
| 13175 | && INSN(3,0) == BITS4(1,1,1,1)) { |
| 13176 | nn = INSN(19,16); |
| 13177 | tt = INSN(15,12); |
| 13178 | isLoad = INSN(20,20) == 1; |
| 13179 | szBlg2 = INSN(5,4); // 00:B 01:H 10:W 11:invalid |
| 13180 | gate = szBlg2 != BITS2(1,1) && tt != 15 && nn != 15; |
| 13181 | } |
| 13182 | } else { |
| 13183 | if (INSN(27,23) == BITS5(0,0,0,1,1) && INSN(20,20) == 1 |
| 13184 | && INSN(11,0) == BITS12(1,1,0,0,1,0,0,1,1,1,1,1)) { |
| 13185 | nn = INSN(19,16); |
| 13186 | tt = INSN(15,12); |
| 13187 | isLoad = True; |
| 13188 | szBlg2 = INSN(22,21); // 10:B 11:H 00:W 01:invalid |
| 13189 | gate = szBlg2 != BITS2(0,1) && tt != 15 && nn != 15; |
| 13190 | } |
| 13191 | else |
| 13192 | if (INSN(27,23) == BITS5(0,0,0,1,1) && INSN(20,20) == 0 |
| 13193 | && INSN(15,4) == BITS12(1,1,1,1,1,1,0,0,1,0,0,1)) { |
| 13194 | nn = INSN(19,16); |
| 13195 | tt = INSN(3,0); |
| 13196 | isLoad = False; |
| 13197 | szBlg2 = INSN(22,21); // 10:B 11:H 00:W 01:invalid |
| 13198 | gate = szBlg2 != BITS2(0,1) && tt != 15 && nn != 15; |
| 13199 | } |
| 13200 | if (gate) { |
| 13201 | // Rearrange szBlg2 bits to be the same as the Thumb case |
| 13202 | switch (szBlg2) { |
| 13203 | case 2: szBlg2 = 0; break; |
| 13204 | case 3: szBlg2 = 1; break; |
| 13205 | case 0: szBlg2 = 2; break; |
| 13206 | default: /*NOTREACHED*/vassert(0); |
| 13207 | } |
| 13208 | } |
| 13209 | } |
| 13210 | // For both encodings, the instruction is guarded by condT, which |
| 13211 | // is passed in by the caller. Note that the the loads and stores |
| 13212 | // are conditional, so we don't have to truncate the IRSB at this |
| 13213 | // point, but the fence is unconditional. There's no way to |
| 13214 | // represent a conditional fence without a side exit, but it |
| 13215 | // doesn't matter from a correctness standpoint that it is |
| 13216 | // unconditional -- it just loses a bit of performance in the |
| 13217 | // case where the condition doesn't hold. |
| 13218 | if (gate) { |
| 13219 | vassert(szBlg2 <= 2 && nn <= 14 && tt <= 14); |
| 13220 | IRExpr* ea = llGetIReg(nn); |
| 13221 | if (isLoad) { |
| 13222 | static IRLoadGOp cvt[3] |
| 13223 | = { ILGop_8Uto32, ILGop_16Uto32, ILGop_Ident32 }; |
| 13224 | IRTemp data = newTemp(Ity_I32); |
| 13225 | loadGuardedLE(data, cvt[szBlg2], ea, mkU32(0)/*alt*/, condT); |
| 13226 | if (isT) { |
| 13227 | putIRegT(tt, mkexpr(data), condT); |
| 13228 | } else { |
| 13229 | putIRegA(tt, mkexpr(data), condT, Ijk_INVALID); |
| 13230 | } |
| 13231 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 13232 | } else { |
| 13233 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 13234 | IRExpr* data = llGetIReg(tt); |
| 13235 | switch (szBlg2) { |
| 13236 | case 0: data = unop(Iop_32to8, data); break; |
| 13237 | case 1: data = unop(Iop_32to16, data); break; |
| 13238 | case 2: break; |
| 13239 | default: vassert(0); |
| 13240 | } |
| 13241 | storeGuardedLE(ea, data, condT); |
| 13242 | } |
| 13243 | const HChar* ldNames[3] = { "ldab", "ldah", "lda" }; |
| 13244 | const HChar* stNames[3] = { "stlb", "stlh", "stl" }; |
| 13245 | DIP("%s r%u, [r%u]", (isLoad ? ldNames : stNames)[szBlg2], tt, nn); |
| 13246 | return True; |
| 13247 | } |
| 13248 | /* else fall through */ |
| 13249 | } |
| 13250 | |
| 13251 | /* ----------- LDAEX{,B,H,D}, STLEX{,B,H,D} ----------- */ |
| 13252 | /* 31 27 23 19 15 11 7 3 |
| 13253 | A1: cond 0001 1101 n t 1110 1001 1111 LDAEXB Rt, [Rn] |
| 13254 | A1: cond 0001 1111 n t 1110 1001 1111 LDAEXH Rt, [Rn] |
| 13255 | A1: cond 0001 1001 n t 1110 1001 1111 LDAEX Rt, [Rn] |
| 13256 | A1: cond 0001 1011 n t 1110 1001 1111 LDAEXD Rt, Rt+1, [Rn] |
| 13257 | |
| 13258 | A1: cond 0001 1100 n d 1110 1001 t STLEXB Rd, Rt, [Rn] |
| 13259 | A1: cond 0001 1110 n d 1110 1001 t STLEXH Rd, Rt, [Rn] |
| 13260 | A1: cond 0001 1000 n d 1110 1001 t STLEX Rd, Rt, [Rn] |
| 13261 | A1: cond 0001 1010 n d 1110 1001 t STLEXD Rd, Rt, Rt+1, [Rn] |
| 13262 | |
| 13263 | 31 28 24 19 15 11 7 3 |
| 13264 | T1: 111 0100 01101 n t 1111 1100 1111 LDAEXB Rt, [Rn] |
| 13265 | T1: 111 0100 01101 n t 1111 1101 1111 LDAEXH Rt, [Rn] |
| 13266 | T1: 111 0100 01101 n t 1111 1110 1111 LDAEX Rt, [Rn] |
| 13267 | T1: 111 0100 01101 n t t2 1111 1111 LDAEXD Rt, Rt2, [Rn] |
| 13268 | |
| 13269 | T1: 111 0100 01100 n t 1111 1100 d STLEXB Rd, Rt, [Rn] |
| 13270 | T1: 111 0100 01100 n t 1111 1101 d STLEXH Rd, Rt, [Rn] |
| 13271 | T1: 111 0100 01100 n t 1111 1110 d STLEX Rd, Rt, [Rn] |
| 13272 | T1: 111 0100 01100 n t t2 1111 d STLEXD Rd, Rt, Rt2, [Rn] |
| 13273 | */ |
| 13274 | { |
| 13275 | UInt nn = 16; // invalid |
| 13276 | UInt tt = 16; // invalid |
| 13277 | UInt tt2 = 16; // invalid |
| 13278 | UInt dd = 16; // invalid |
| 13279 | UInt szBlg2 = 4; // invalid |
| 13280 | Bool isLoad = False; |
| 13281 | Bool gate = False; |
| 13282 | if (isT) { |
| 13283 | if (INSN(31,21) == BITS11(1,1,1,0,1,0,0,0,1,1,0) |
| 13284 | && INSN(7,6) == BITS2(1,1)) { |
| 13285 | isLoad = INSN(20,20) == 1; |
| 13286 | nn = INSN(19,16); |
| 13287 | tt = INSN(15,12); |
| 13288 | tt2 = INSN(11,8); |
| 13289 | szBlg2 = INSN(5,4); |
| 13290 | dd = INSN(3,0); |
| 13291 | gate = True; |
| 13292 | if (szBlg2 < BITS2(1,1) && tt2 != BITS4(1,1,1,1)) gate = False; |
| 13293 | if (isLoad && dd != BITS4(1,1,1,1)) gate = False; |
| 13294 | // re-set not-used register values to invalid |
| 13295 | if (szBlg2 < BITS2(1,1)) tt2 = 16; |
| 13296 | if (isLoad) dd = 16; |
| 13297 | } |
| 13298 | } else { |
| 13299 | /* ARM encoding. Do the load and store cases separately as |
| 13300 | the register numbers are in different places and a combined decode |
| 13301 | is too confusing. */ |
| 13302 | if (INSN(27,23) == BITS5(0,0,0,1,1) && INSN(20,20) == 1 |
| 13303 | && INSN(11,0) == BITS12(1,1,1,0,1,0,0,1,1,1,1,1)) { |
| 13304 | szBlg2 = INSN(22,21); |
| 13305 | isLoad = True; |
| 13306 | nn = INSN(19,16); |
| 13307 | tt = INSN(15,12); |
| 13308 | gate = True; |
| 13309 | } |
| 13310 | else |
| 13311 | if (INSN(27,23) == BITS5(0,0,0,1,1) && INSN(20,20) == 0 |
| 13312 | && INSN(11,4) == BITS8(1,1,1,0,1,0,0,1)) { |
| 13313 | szBlg2 = INSN(22,21); |
| 13314 | isLoad = False; |
| 13315 | nn = INSN(19,16); |
| 13316 | dd = INSN(15,12); |
| 13317 | tt = INSN(3,0); |
| 13318 | gate = True; |
| 13319 | } |
| 13320 | if (gate) { |
| 13321 | // Rearrange szBlg2 bits to be the same as the Thumb case |
| 13322 | switch (szBlg2) { |
| 13323 | case 2: szBlg2 = 0; break; |
| 13324 | case 3: szBlg2 = 1; break; |
| 13325 | case 0: szBlg2 = 2; break; |
| 13326 | case 1: szBlg2 = 3; break; |
| 13327 | default: /*NOTREACHED*/vassert(0); |
| 13328 | } |
| 13329 | } |
| 13330 | } |
| 13331 | // Perform further checks on register numbers |
| 13332 | if (gate) { |
| 13333 | /**/ if (isT && isLoad) { |
| 13334 | // Thumb load |
| 13335 | if (szBlg2 < 3) { |
| 13336 | if (! (tt != 13 && tt != 15 && nn != 15)) gate = False; |
| 13337 | } else { |
| 13338 | if (! (tt != 13 && tt != 15 && tt2 != 13 && tt2 != 15 |
| 13339 | && tt != tt2 && nn != 15)) gate = False; |
| 13340 | } |
| 13341 | } |
| 13342 | else if (isT && !isLoad) { |
| 13343 | // Thumb store |
| 13344 | if (szBlg2 < 3) { |
| 13345 | if (! (dd != 13 && dd != 15 && tt != 13 && tt != 15 |
| 13346 | && nn != 15 && dd != nn && dd != tt)) gate = False; |
| 13347 | } else { |
| 13348 | if (! (dd != 13 && dd != 15 && tt != 13 && tt != 15 |
| 13349 | && tt2 != 13 && tt2 != 15 && nn != 15 && dd != nn |
| 13350 | && dd != tt && dd != tt2)) gate = False; |
| 13351 | } |
| 13352 | } |
| 13353 | else if (!isT && isLoad) { |
| 13354 | // ARM Load |
| 13355 | if (szBlg2 < 3) { |
| 13356 | if (! (tt != 15 && nn != 15)) gate = False; |
| 13357 | } else { |
| 13358 | if (! ((tt & 1) == 0 && tt != 14 && nn != 15)) gate = False; |
| 13359 | vassert(tt2 == 16/*invalid*/); |
| 13360 | tt2 = tt + 1; |
| 13361 | } |
| 13362 | } |
| 13363 | else if (!isT && !isLoad) { |
| 13364 | // ARM Store |
| 13365 | if (szBlg2 < 3) { |
| 13366 | if (! (dd != 15 && tt != 15 && nn != 15 |
| 13367 | && dd != nn && dd != tt)) gate = False; |
| 13368 | } else { |
| 13369 | if (! (dd != 15 && (tt & 1) == 0 && tt != 14 && nn != 15 |
| 13370 | && dd != nn && dd != tt && dd != tt+1)) gate = False; |
| 13371 | vassert(tt2 == 16/*invalid*/); |
| 13372 | tt2 = tt + 1; |
| 13373 | } |
| 13374 | } |
| 13375 | else /*NOTREACHED*/vassert(0); |
| 13376 | } |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 13377 | if (gate) { |
| 13378 | // Paranoia .. |
| 13379 | vassert(szBlg2 <= 3); |
| 13380 | if (szBlg2 < 3) { vassert(tt2 == 16/*invalid*/); } |
| 13381 | else { vassert(tt2 <= 14); } |
| 13382 | if (isLoad) { vassert(dd == 16/*invalid*/); } |
| 13383 | else { vassert(dd <= 14); } |
| 13384 | } |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 13385 | // If we're still good even after all that, generate the IR. |
| 13386 | if (gate) { |
| 13387 | /* First, go unconditional. Staying in-line is too complex. */ |
| 13388 | if (isT) { |
| 13389 | vassert(condT != IRTemp_INVALID); |
| 13390 | mk_skip_over_T32_if_cond_is_false( condT ); |
| 13391 | } else { |
| 13392 | if (condT != IRTemp_INVALID) { |
| 13393 | mk_skip_over_A32_if_cond_is_false( condT ); |
| 13394 | condT = IRTemp_INVALID; |
| 13395 | } |
| 13396 | } |
| 13397 | /* Now the load or store. */ |
| 13398 | IRType ty = Ity_INVALID; /* the type of the transferred data */ |
| 13399 | const HChar* nm = NULL; |
| 13400 | switch (szBlg2) { |
| 13401 | case 0: nm = "b"; ty = Ity_I8; break; |
| 13402 | case 1: nm = "h"; ty = Ity_I16; break; |
| 13403 | case 2: nm = ""; ty = Ity_I32; break; |
| 13404 | case 3: nm = "d"; ty = Ity_I64; break; |
| 13405 | default: vassert(0); |
| 13406 | } |
| 13407 | IRExpr* ea = isT ? getIRegT(nn) : getIRegA(nn); |
| 13408 | if (isLoad) { |
| 13409 | // LOAD. Transaction, then fence. |
| 13410 | IROp widen = Iop_INVALID; |
| 13411 | switch (szBlg2) { |
| 13412 | case 0: widen = Iop_8Uto32; break; |
| 13413 | case 1: widen = Iop_16Uto32; break; |
| 13414 | case 2: case 3: break; |
| 13415 | default: vassert(0); |
| 13416 | } |
| 13417 | IRTemp res = newTemp(ty); |
| 13418 | // FIXME: assumes little-endian guest |
| 13419 | stmt( IRStmt_LLSC(Iend_LE, res, ea, NULL/*this is a load*/) ); |
| 13420 | |
| 13421 | # define PUT_IREG(_nnz, _eez) \ |
| 13422 | do { vassert((_nnz) <= 14); /* no writes to the PC */ \ |
| 13423 | if (isT) { putIRegT((_nnz), (_eez), IRTemp_INVALID); } \ |
| 13424 | else { putIRegA((_nnz), (_eez), \ |
| 13425 | IRTemp_INVALID, Ijk_Boring); } } while(0) |
| 13426 | if (ty == Ity_I64) { |
| 13427 | // FIXME: assumes little-endian guest |
| 13428 | PUT_IREG(tt, unop(Iop_64to32, mkexpr(res))); |
| 13429 | PUT_IREG(tt2, unop(Iop_64HIto32, mkexpr(res))); |
| 13430 | } else { |
| 13431 | PUT_IREG(tt, widen == Iop_INVALID |
| 13432 | ? mkexpr(res) : unop(widen, mkexpr(res))); |
| 13433 | } |
| 13434 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 13435 | if (ty == Ity_I64) { |
| 13436 | DIP("ldrex%s%s r%u, r%u, [r%u]\n", |
| 13437 | nm, isT ? "" : nCC(conq), tt, tt2, nn); |
| 13438 | } else { |
| 13439 | DIP("ldrex%s%s r%u, [r%u]\n", nm, isT ? "" : nCC(conq), tt, nn); |
| 13440 | } |
| 13441 | # undef PUT_IREG |
| 13442 | } else { |
| 13443 | // STORE. Fence, then transaction. |
| 13444 | IRTemp resSC1, resSC32, data; |
| 13445 | IROp narrow = Iop_INVALID; |
| 13446 | switch (szBlg2) { |
| 13447 | case 0: narrow = Iop_32to8; break; |
| 13448 | case 1: narrow = Iop_32to16; break; |
| 13449 | case 2: case 3: break; |
| 13450 | default: vassert(0); |
| 13451 | } |
| 13452 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 13453 | data = newTemp(ty); |
| 13454 | # define GET_IREG(_nnz) (isT ? getIRegT(_nnz) : getIRegA(_nnz)) |
| 13455 | assign(data, |
| 13456 | ty == Ity_I64 |
| 13457 | // FIXME: assumes little-endian guest |
| 13458 | ? binop(Iop_32HLto64, GET_IREG(tt2), GET_IREG(tt)) |
| 13459 | : narrow == Iop_INVALID |
| 13460 | ? GET_IREG(tt) |
| 13461 | : unop(narrow, GET_IREG(tt))); |
| 13462 | # undef GET_IREG |
| 13463 | resSC1 = newTemp(Ity_I1); |
| 13464 | // FIXME: assumes little-endian guest |
| 13465 | stmt( IRStmt_LLSC(Iend_LE, resSC1, ea, mkexpr(data)) ); |
| 13466 | |
| 13467 | /* Set rDD to 1 on failure, 0 on success. Currently we have |
| 13468 | resSC1 == 0 on failure, 1 on success. */ |
| 13469 | resSC32 = newTemp(Ity_I32); |
| 13470 | assign(resSC32, |
| 13471 | unop(Iop_1Uto32, unop(Iop_Not1, mkexpr(resSC1)))); |
| 13472 | vassert(dd <= 14); /* no writes to the PC */ |
| 13473 | if (isT) { |
| 13474 | putIRegT(dd, mkexpr(resSC32), IRTemp_INVALID); |
| 13475 | } else { |
| 13476 | putIRegA(dd, mkexpr(resSC32), IRTemp_INVALID, Ijk_Boring); |
| 13477 | } |
| 13478 | if (ty == Ity_I64) { |
| 13479 | DIP("strex%s%s r%u, r%u, r%u, [r%u]\n", |
| 13480 | nm, isT ? "" : nCC(conq), dd, tt, tt2, nn); |
| 13481 | } else { |
| 13482 | DIP("strex%s%s r%u, r%u, [r%u]\n", |
| 13483 | nm, isT ? "" : nCC(conq), dd, tt, nn); |
| 13484 | } |
| 13485 | } /* if (isLoad) */ |
| 13486 | return True; |
| 13487 | } /* if (gate) */ |
| 13488 | /* else fall through */ |
| 13489 | } |
| 13490 | |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 13491 | /* ----------- VSEL<c>.F64 d_d_d, VSEL<c>.F32 s_s_s ----------- */ |
| 13492 | /* 31 27 22 21 19 15 11 8 7 6 5 4 3 |
| 13493 | T1/A1: 1111 11100 D cc n d 101 1 N 0 M 0 m VSEL<c>.F64 Dd, Dn, Dm |
| 13494 | T1/A1: 1111 11100 D cc n d 101 0 N 0 M 0 m VSEL<c>.F32 Sd, Sn, Sm |
| 13495 | |
| 13496 | ARM encoding is in NV space. |
| 13497 | In Thumb mode, we must not be in an IT block. |
| 13498 | */ |
| 13499 | if (INSN(31,23) == BITS9(1,1,1,1,1,1,1,0,0) && INSN(11,9) == BITS3(1,0,1) |
| 13500 | && INSN(6,6) == 0 && INSN(4,4) == 0) { |
| 13501 | UInt bit_D = INSN(22,22); |
| 13502 | UInt fld_cc = INSN(21,20); |
| 13503 | UInt fld_n = INSN(19,16); |
| 13504 | UInt fld_d = INSN(15,12); |
| 13505 | Bool isF64 = INSN(8,8) == 1; |
| 13506 | UInt bit_N = INSN(7,7); |
| 13507 | UInt bit_M = INSN(5,5); |
| 13508 | UInt fld_m = INSN(3,0); |
| 13509 | |
| 13510 | UInt dd = isF64 ? ((bit_D << 4) | fld_d) : ((fld_d << 1) | bit_D); |
| 13511 | UInt nn = isF64 ? ((bit_N << 4) | fld_n) : ((fld_n << 1) | bit_N); |
| 13512 | UInt mm = isF64 ? ((bit_M << 4) | fld_m) : ((fld_m << 1) | bit_M); |
| 13513 | |
| 13514 | UInt cc_1 = (fld_cc >> 1) & 1; |
| 13515 | UInt cc_0 = (fld_cc >> 0) & 1; |
| 13516 | UInt cond = (fld_cc << 2) | ((cc_1 ^ cc_0) << 1) | 0; |
| 13517 | |
| 13518 | if (isT) { |
| 13519 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 13520 | } |
| 13521 | /* In ARM mode, this is statically unconditional. In Thumb mode, |
| 13522 | this must be dynamically unconditional, and we've SIGILLd if not. |
| 13523 | In either case we can create unconditional IR. */ |
| 13524 | |
| 13525 | IRTemp guard = newTemp(Ity_I32); |
| 13526 | assign(guard, mk_armg_calculate_condition(cond)); |
| 13527 | IRExpr* srcN = (isF64 ? llGetDReg : llGetFReg)(nn); |
| 13528 | IRExpr* srcM = (isF64 ? llGetDReg : llGetFReg)(mm); |
| 13529 | IRExpr* res = IRExpr_ITE(unop(Iop_32to1, mkexpr(guard)), srcN, srcM); |
| 13530 | (isF64 ? llPutDReg : llPutFReg)(dd, res); |
| 13531 | |
| 13532 | UChar rch = isF64 ? 'd' : 'f'; |
| 13533 | DIP("vsel%s.%s %c%u, %c%u, %c%u\n", |
| 13534 | nCC(cond), isF64 ? "f64" : "f32", rch, dd, rch, nn, rch, mm); |
| 13535 | return True; |
| 13536 | } |
| 13537 | |
| 13538 | /* -------- VRINT{A,N,P,M}.F64 d_d, VRINT{A,N,P,M}.F32 s_s -------- */ |
| 13539 | /* 31 22 21 17 15 11 8 7 5 4 3 |
| 13540 | T1/A1: 111111101 D 1110 rm Vd 101 1 01 M 0 Vm VRINT{A,N,P,M}.F64 Dd, Dm |
| 13541 | T1/A1: 111111101 D 1110 rm Vd 101 0 01 M 0 Vm VRINT{A,N,P,M}.F32 Sd, Sm |
| 13542 | |
| 13543 | ARM encoding is in NV space. |
| 13544 | In Thumb mode, we must not be in an IT block. |
| 13545 | */ |
| 13546 | if (INSN(31,23) == BITS9(1,1,1,1,1,1,1,0,1) |
| 13547 | && INSN(21,18) == BITS4(1,1,1,0) && INSN(11,9) == BITS3(1,0,1) |
| 13548 | && INSN(7,6) == BITS2(0,1) && INSN(4,4) == 0) { |
| 13549 | UInt bit_D = INSN(22,22); |
| 13550 | UInt fld_rm = INSN(17,16); |
| 13551 | UInt fld_d = INSN(15,12); |
| 13552 | Bool isF64 = INSN(8,8) == 1; |
| 13553 | UInt bit_M = INSN(5,5); |
| 13554 | UInt fld_m = INSN(3,0); |
| 13555 | |
| 13556 | UInt dd = isF64 ? ((bit_D << 4) | fld_d) : ((fld_d << 1) | bit_D); |
| 13557 | UInt mm = isF64 ? ((bit_M << 4) | fld_m) : ((fld_m << 1) | bit_M); |
| 13558 | |
| 13559 | if (isT) { |
| 13560 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 13561 | } |
| 13562 | /* In ARM mode, this is statically unconditional. In Thumb mode, |
| 13563 | this must be dynamically unconditional, and we've SIGILLd if not. |
| 13564 | In either case we can create unconditional IR. */ |
| 13565 | |
| 13566 | UChar c = '?'; |
| 13567 | IRRoundingMode rm = Irrm_NEAREST; |
| 13568 | switch (fld_rm) { |
| 13569 | /* The use of NEAREST for both the 'a' and 'n' cases is a bit of a |
| 13570 | kludge since it doesn't take into account the nearest-even vs |
| 13571 | nearest-away semantics. */ |
| 13572 | case BITS2(0,0): c = 'a'; rm = Irrm_NEAREST; break; |
| 13573 | case BITS2(0,1): c = 'n'; rm = Irrm_NEAREST; break; |
| 13574 | case BITS2(1,0): c = 'p'; rm = Irrm_PosINF; break; |
| 13575 | case BITS2(1,1): c = 'm'; rm = Irrm_NegINF; break; |
| 13576 | default: vassert(0); |
| 13577 | } |
| 13578 | |
| 13579 | IRExpr* srcM = (isF64 ? llGetDReg : llGetFReg)(mm); |
| 13580 | IRExpr* res = binop(isF64 ? Iop_RoundF64toInt : Iop_RoundF32toInt, |
| 13581 | mkU32((UInt)rm), srcM); |
| 13582 | (isF64 ? llPutDReg : llPutFReg)(dd, res); |
| 13583 | |
| 13584 | UChar rch = isF64 ? 'd' : 'f'; |
| 13585 | DIP("vrint%c.%s.%s %c%u, %c%u\n", |
| 13586 | c, isF64 ? "f64" : "f32", isF64 ? "f64" : "f32", rch, dd, rch, mm); |
| 13587 | return True; |
| 13588 | } |
| 13589 | |
| 13590 | /* -------- VRINT{Z,R}.F64.F64 d_d, VRINT{Z,R}.F32.F32 s_s -------- */ |
| 13591 | /* 31 27 22 21 15 11 7 6 5 4 3 |
| 13592 | T1: 1110 11101 D 110110 Vd 1011 op 1 M 0 Vm VRINT<r><c>.F64.F64 Dd, Dm |
| 13593 | A1: cond 11101 D 110110 Vd 1011 op 1 M 0 Vm |
| 13594 | |
| 13595 | T1: 1110 11101 D 110110 Vd 1010 op 1 M 0 Vm VRINT<r><c>.F32.F32 Sd, Sm |
| 13596 | A1: cond 11101 D 110110 Vd 1010 op 1 M 0 Vm |
| 13597 | |
| 13598 | In contrast to the VRINT variants just above, this can be conditional. |
| 13599 | */ |
| 13600 | if ((isT ? (INSN(31,28) == BITS4(1,1,1,0)) : True) |
| 13601 | && INSN(27,23) == BITS5(1,1,1,0,1) && INSN(21,16) == BITS6(1,1,0,1,1,0) |
| 13602 | && INSN(11,9) == BITS3(1,0,1) && INSN(6,6) == 1 && INSN(4,4) == 0) { |
| 13603 | UInt bit_D = INSN(22,22); |
| 13604 | UInt fld_Vd = INSN(15,12); |
| 13605 | Bool isF64 = INSN(8,8) == 1; |
| 13606 | Bool rToZero = INSN(7,7) == 1; |
| 13607 | UInt bit_M = INSN(5,5); |
| 13608 | UInt fld_Vm = INSN(3,0); |
| 13609 | UInt dd = isF64 ? ((bit_D << 4) | fld_Vd) : ((fld_Vd << 1) | bit_D); |
| 13610 | UInt mm = isF64 ? ((bit_M << 4) | fld_Vm) : ((fld_Vm << 1) | bit_M); |
| 13611 | |
| 13612 | if (isT) vassert(condT != IRTemp_INVALID); |
| 13613 | IRType ty = isF64 ? Ity_F64 : Ity_F32; |
| 13614 | IRTemp src = newTemp(ty); |
| 13615 | IRTemp res = newTemp(ty); |
| 13616 | assign(src, (isF64 ? getDReg : getFReg)(mm)); |
| 13617 | |
| 13618 | IRTemp rm = newTemp(Ity_I32); |
| 13619 | assign(rm, rToZero ? mkU32(Irrm_ZERO) |
| 13620 | : mkexpr(mk_get_IR_rounding_mode())); |
| 13621 | assign(res, binop(isF64 ? Iop_RoundF64toInt : Iop_RoundF32toInt, |
| 13622 | mkexpr(rm), mkexpr(src))); |
| 13623 | (isF64 ? putDReg : putFReg)(dd, mkexpr(res), condT); |
| 13624 | |
| 13625 | UChar rch = isF64 ? 'd' : 'f'; |
| 13626 | DIP("vrint%c.%s.%s %c%u, %c%u\n", |
| 13627 | rToZero ? 'z' : 'r', |
| 13628 | isF64 ? "f64" : "f32", isF64 ? "f64" : "f32", rch, dd, rch, mm); |
| 13629 | return True; |
| 13630 | } |
| 13631 | |
| 13632 | /* ----------- VCVT{A,N,P,M}{.S32,.U32}{.F64,.F32} ----------- */ |
| 13633 | /* 31 27 22 21 17 15 11 8 7 6 5 4 3 |
| 13634 | T1/A1: 1111 11101 D 1111 rm Vd 101 sz op 1 M 0 Vm |
| 13635 | VCVT{A,N,P,M}{.S32,.U32}.F64 Sd, Dm |
| 13636 | VCVT{A,N,P,M}{.S32,.U32}.F32 Sd, Sm |
| 13637 | |
| 13638 | ARM encoding is in NV space. |
| 13639 | In Thumb mode, we must not be in an IT block. |
| 13640 | */ |
| 13641 | if (INSN(31,23) == BITS9(1,1,1,1,1,1,1,0,1) && INSN(21,18) == BITS4(1,1,1,1) |
| 13642 | && INSN(11,9) == BITS3(1,0,1) && INSN(6,6) == 1 && INSN(4,4) == 0) { |
| 13643 | UInt bit_D = INSN(22,22); |
| 13644 | UInt fld_rm = INSN(17,16); |
| 13645 | UInt fld_Vd = INSN(15,12); |
| 13646 | Bool isF64 = INSN(8,8) == 1; |
| 13647 | Bool isU = INSN(7,7) == 0; |
| 13648 | UInt bit_M = INSN(5,5); |
| 13649 | UInt fld_Vm = INSN(3,0); |
| 13650 | |
| 13651 | UInt dd = (fld_Vd << 1) | bit_D; |
| 13652 | UInt mm = isF64 ? ((bit_M << 4) | fld_Vm) : ((fld_Vm << 1) | bit_M); |
| 13653 | |
| 13654 | if (isT) { |
| 13655 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 13656 | } |
| 13657 | /* In ARM mode, this is statically unconditional. In Thumb mode, |
| 13658 | this must be dynamically unconditional, and we've SIGILLd if not. |
| 13659 | In either case we can create unconditional IR. */ |
| 13660 | |
| 13661 | UChar c = '?'; |
| 13662 | IRRoundingMode rm = Irrm_NEAREST; |
| 13663 | switch (fld_rm) { |
| 13664 | /* The use of NEAREST for both the 'a' and 'n' cases is a bit of a |
| 13665 | kludge since it doesn't take into account the nearest-even vs |
| 13666 | nearest-away semantics. */ |
| 13667 | case BITS2(0,0): c = 'a'; rm = Irrm_NEAREST; break; |
| 13668 | case BITS2(0,1): c = 'n'; rm = Irrm_NEAREST; break; |
| 13669 | case BITS2(1,0): c = 'p'; rm = Irrm_PosINF; break; |
| 13670 | case BITS2(1,1): c = 'm'; rm = Irrm_NegINF; break; |
| 13671 | default: vassert(0); |
| 13672 | } |
| 13673 | |
| 13674 | IRExpr* srcM = (isF64 ? llGetDReg : llGetFReg)(mm); |
| 13675 | IRTemp res = newTemp(Ity_I32); |
| 13676 | |
| 13677 | /* The arm back end doesn't support use of Iop_F32toI32U or |
| 13678 | Iop_F32toI32S, so for those cases we widen the F32 to F64 |
| 13679 | and then follow the F64 route. */ |
| 13680 | if (!isF64) { |
| 13681 | srcM = unop(Iop_F32toF64, srcM); |
| 13682 | } |
| 13683 | assign(res, binop(isU ? Iop_F64toI32U : Iop_F64toI32S, |
| 13684 | mkU32((UInt)rm), srcM)); |
| 13685 | |
| 13686 | llPutFReg(dd, unop(Iop_ReinterpI32asF32, mkexpr(res))); |
| 13687 | |
| 13688 | UChar rch = isF64 ? 'd' : 'f'; |
| 13689 | DIP("vcvt%c.%s.%s %c%u, %c%u\n", |
| 13690 | c, isU ? "u32" : "s32", isF64 ? "f64" : "f32", 's', dd, rch, mm); |
| 13691 | return True; |
| 13692 | } |
| 13693 | |
| 13694 | /* ----------- V{MAX,MIN}NM{.F64 d_d_d, .F32 s_s_s} ----------- */ |
| 13695 | /* 31 27 22 21 19 15 11 8 7 6 5 4 3 |
| 13696 | 1111 11101 D 00 Vn Vd 101 1 N op M 0 Vm V{MIN,MAX}NM.F64 Dd, Dn, Dm |
| 13697 | 1111 11101 D 00 Vn Vd 101 0 N op M 0 Vm V{MIN,MAX}NM.F32 Sd, Sn, Sm |
| 13698 | |
| 13699 | ARM encoding is in NV space. |
| 13700 | In Thumb mode, we must not be in an IT block. |
| 13701 | */ |
| 13702 | if (INSN(31,23) == BITS9(1,1,1,1,1,1,1,0,1) && INSN(21,20) == BITS2(0,0) |
| 13703 | && INSN(11,9) == BITS3(1,0,1) && INSN(4,4) == 0) { |
| 13704 | UInt bit_D = INSN(22,22); |
| 13705 | UInt fld_Vn = INSN(19,16); |
| 13706 | UInt fld_Vd = INSN(15,12); |
| 13707 | Bool isF64 = INSN(8,8) == 1; |
| 13708 | UInt bit_N = INSN(7,7); |
| 13709 | Bool isMAX = INSN(6,6) == 0; |
| 13710 | UInt bit_M = INSN(5,5); |
| 13711 | UInt fld_Vm = INSN(3,0); |
| 13712 | |
| 13713 | UInt dd = isF64 ? ((bit_D << 4) | fld_Vd) : ((fld_Vd << 1) | bit_D); |
| 13714 | UInt nn = isF64 ? ((bit_N << 4) | fld_Vn) : ((fld_Vn << 1) | bit_N); |
| 13715 | UInt mm = isF64 ? ((bit_M << 4) | fld_Vm) : ((fld_Vm << 1) | bit_M); |
| 13716 | |
| 13717 | if (isT) { |
| 13718 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 13719 | } |
| 13720 | /* In ARM mode, this is statically unconditional. In Thumb mode, |
| 13721 | this must be dynamically unconditional, and we've SIGILLd if not. |
| 13722 | In either case we can create unconditional IR. */ |
| 13723 | |
| 13724 | IROp op = isF64 ? (isMAX ? Iop_MaxNumF64 : Iop_MinNumF64) |
| 13725 | : (isMAX ? Iop_MaxNumF32 : Iop_MinNumF32); |
| 13726 | IRExpr* srcN = (isF64 ? llGetDReg : llGetFReg)(nn); |
| 13727 | IRExpr* srcM = (isF64 ? llGetDReg : llGetFReg)(mm); |
| 13728 | IRExpr* res = binop(op, srcN, srcM); |
| 13729 | (isF64 ? llPutDReg : llPutFReg)(dd, res); |
| 13730 | |
| 13731 | UChar rch = isF64 ? 'd' : 'f'; |
| 13732 | DIP("v%snm.%s %c%u, %c%u, %c%u\n", |
| 13733 | isMAX ? "max" : "min", isF64 ? "f64" : "f32", |
| 13734 | rch, dd, rch, nn, rch, mm); |
| 13735 | return True; |
| 13736 | } |
| 13737 | |
| 13738 | /* ----------- VRINTX.F64.F64 d_d, VRINTX.F32.F32 s_s ----------- */ |
| 13739 | /* 31 27 22 21 15 11 8 7 5 4 3 |
| 13740 | T1: 1110 11101 D 110111 Vd 101 1 01 M 0 Vm VRINTX<c>.F64.F64 Dd, Dm |
| 13741 | A1: cond 11101 D 110111 Vd 101 1 01 M 0 Vm |
| 13742 | |
| 13743 | T1: 1110 11101 D 110111 Vd 101 0 01 M 0 Vm VRINTX<c>.F32.F32 Dd, Dm |
| 13744 | A1: cond 11101 D 110111 Vd 101 0 01 M 0 Vm |
| 13745 | |
| 13746 | Like VRINT{Z,R}{.F64.F64, .F32.F32} just above, this can be conditional. |
| 13747 | This produces the same code as the VRINTR case since we ignore the |
| 13748 | requirement to signal inexactness. |
| 13749 | */ |
| 13750 | if ((isT ? (INSN(31,28) == BITS4(1,1,1,0)) : True) |
| 13751 | && INSN(27,23) == BITS5(1,1,1,0,1) && INSN(21,16) == BITS6(1,1,0,1,1,1) |
| 13752 | && INSN(11,9) == BITS3(1,0,1) && INSN(7,6) == BITS2(0,1) |
| 13753 | && INSN(4,4) == 0) { |
| 13754 | UInt bit_D = INSN(22,22); |
| 13755 | UInt fld_Vd = INSN(15,12); |
| 13756 | Bool isF64 = INSN(8,8) == 1; |
| 13757 | UInt bit_M = INSN(5,5); |
| 13758 | UInt fld_Vm = INSN(3,0); |
| 13759 | UInt dd = isF64 ? ((bit_D << 4) | fld_Vd) : ((fld_Vd << 1) | bit_D); |
| 13760 | UInt mm = isF64 ? ((bit_M << 4) | fld_Vm) : ((fld_Vm << 1) | bit_M); |
| 13761 | |
| 13762 | if (isT) vassert(condT != IRTemp_INVALID); |
| 13763 | IRType ty = isF64 ? Ity_F64 : Ity_F32; |
| 13764 | IRTemp src = newTemp(ty); |
| 13765 | IRTemp res = newTemp(ty); |
| 13766 | assign(src, (isF64 ? getDReg : getFReg)(mm)); |
| 13767 | |
| 13768 | IRTemp rm = newTemp(Ity_I32); |
| 13769 | assign(rm, mkexpr(mk_get_IR_rounding_mode())); |
| 13770 | assign(res, binop(isF64 ? Iop_RoundF64toInt : Iop_RoundF32toInt, |
| 13771 | mkexpr(rm), mkexpr(src))); |
| 13772 | (isF64 ? putDReg : putFReg)(dd, mkexpr(res), condT); |
| 13773 | |
| 13774 | UChar rch = isF64 ? 'd' : 'f'; |
| 13775 | DIP("vrint%c.%s.%s %c%u, %c%u\n", |
| 13776 | 'x', |
| 13777 | isF64 ? "f64" : "f32", isF64 ? "f64" : "f32", rch, dd, rch, mm); |
| 13778 | return True; |
| 13779 | } |
| 13780 | |
| 13781 | /* ----------- V{MAX,MIN}NM{.F32 d_d_d, .F32 q_q_q} ----------- */ |
| 13782 | /* 31 27 22 21 20 19 15 11 7 6 5 4 3 |
| 13783 | T1: 1111 11110 D op 0 Vn Vd 1111 N 1 M 1 Vm V{MIN,MAX}NM.F32 Qd,Qn,Qm |
| 13784 | A1: 1111 00110 D op 0 Vn Vd 1111 N 1 M 1 Vm |
| 13785 | |
| 13786 | T1: 1111 11110 D op 0 Vn Vd 1111 N 0 M 1 Vm V{MIN,MAX}NM.F32 Dd,Dn,Dm |
| 13787 | A1: 1111 00110 D op 0 Vn Vd 1111 N 0 M 1 Vm |
| 13788 | |
| 13789 | ARM encoding is in NV space. |
| 13790 | In Thumb mode, we must not be in an IT block. |
| 13791 | */ |
| 13792 | if (INSN(31,23) == (isT ? BITS9(1,1,1,1,1,1,1,1,0) |
| 13793 | : BITS9(1,1,1,1,0,0,1,1,0)) |
| 13794 | && INSN(20,20) == 0 && INSN(11,8) == BITS4(1,1,1,1) && INSN(4,4) == 1) { |
| 13795 | UInt bit_D = INSN(22,22); |
| 13796 | Bool isMax = INSN(21,21) == 0; |
| 13797 | UInt fld_Vn = INSN(19,16); |
| 13798 | UInt fld_Vd = INSN(15,12); |
| 13799 | UInt bit_N = INSN(7,7); |
| 13800 | Bool isQ = INSN(6,6) == 1; |
| 13801 | UInt bit_M = INSN(5,5); |
| 13802 | UInt fld_Vm = INSN(3,0); |
| 13803 | |
| 13804 | /* dd, nn, mm are D-register numbers. */ |
| 13805 | UInt dd = (bit_D << 4) | fld_Vd; |
| 13806 | UInt nn = (bit_N << 4) | fld_Vn; |
| 13807 | UInt mm = (bit_M << 4) | fld_Vm; |
| 13808 | |
| 13809 | if (! (isQ && ((dd & 1) == 1 || (nn & 1) == 1 || (mm & 1) == 1))) { |
| 13810 | /* Do this piecewise on f regs. This is a bit tricky |
| 13811 | though because we are dealing with the full 16 x Q == 32 x D |
| 13812 | register set, so the implied F reg numbers are 0 to 63. But |
| 13813 | ll{Get,Put}FReg only allow the 0 .. 31 as those are the only |
| 13814 | architected F regs. */ |
| 13815 | UInt ddF = dd << 1; |
| 13816 | UInt nnF = nn << 1; |
| 13817 | UInt mmF = mm << 1; |
| 13818 | |
| 13819 | if (isT) { |
| 13820 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 13821 | } |
| 13822 | /* In ARM mode, this is statically unconditional. In Thumb mode, |
| 13823 | this must be dynamically unconditional, and we've SIGILLd if not. |
| 13824 | In either case we can create unconditional IR. */ |
| 13825 | |
| 13826 | IROp op = isMax ? Iop_MaxNumF32 : Iop_MinNumF32; |
| 13827 | |
| 13828 | IRTemp r0 = newTemp(Ity_F32); |
| 13829 | IRTemp r1 = newTemp(Ity_F32); |
| 13830 | IRTemp r2 = isQ ? newTemp(Ity_F32) : IRTemp_INVALID; |
| 13831 | IRTemp r3 = isQ ? newTemp(Ity_F32) : IRTemp_INVALID; |
| 13832 | |
| 13833 | assign(r0, binop(op, llGetFReg_up_to_64(nnF+0), |
| 13834 | llGetFReg_up_to_64(mmF+0))); |
| 13835 | assign(r1, binop(op, llGetFReg_up_to_64(nnF+1), |
| 13836 | llGetFReg_up_to_64(mmF+1))); |
| 13837 | if (isQ) { |
| 13838 | assign(r2, binop(op, llGetFReg_up_to_64(nnF+2), |
| 13839 | llGetFReg_up_to_64(mmF+2))); |
| 13840 | assign(r3, binop(op, llGetFReg_up_to_64(nnF+3), |
| 13841 | llGetFReg_up_to_64(mmF+3))); |
| 13842 | } |
| 13843 | llPutFReg_up_to_64(ddF+0, mkexpr(r0)); |
| 13844 | llPutFReg_up_to_64(ddF+1, mkexpr(r1)); |
| 13845 | if (isQ) { |
| 13846 | llPutFReg_up_to_64(ddF+2, mkexpr(r2)); |
| 13847 | llPutFReg_up_to_64(ddF+3, mkexpr(r3)); |
| 13848 | } |
| 13849 | |
| 13850 | HChar rch = isQ ? 'q' : 'd'; |
| 13851 | UInt sh = isQ ? 1 : 0; |
| 13852 | DIP("v%snm.f32 %c%u, %c%u, %c%u\n", |
| 13853 | isMax ? "max" : "min", rch, |
| 13854 | dd >> sh, rch, nn >> sh, rch, mm >> sh); |
| 13855 | return True; |
| 13856 | } |
| 13857 | /* else fall through */ |
| 13858 | } |
| 13859 | |
| 13860 | /* ----------- VCVT{A,N,P,M}{.F32 d_d, .F32 q_q} ----------- */ |
| 13861 | /* 31 27 22 21 15 11 9 7 6 5 4 3 |
| 13862 | T1: 1111 11111 D 111011 Vd 00 rm op Q M 0 Vm |
| 13863 | A1: 1111 00111 D 111011 Vd 00 rm op Q M 0 Vm |
| 13864 | |
| 13865 | ARM encoding is in NV space. |
| 13866 | In Thumb mode, we must not be in an IT block. |
| 13867 | */ |
| 13868 | if (INSN(31,23) == (isT ? BITS9(1,1,1,1,1,1,1,1,1) |
| 13869 | : BITS9(1,1,1,1,0,0,1,1,1)) |
| 13870 | && INSN(21,16) == BITS6(1,1,1,0,1,1) && INSN(11,10) == BITS2(0,0) |
| 13871 | && INSN(4,4) == 0) { |
| 13872 | UInt bit_D = INSN(22,22); |
| 13873 | UInt fld_Vd = INSN(15,12); |
| 13874 | UInt fld_rm = INSN(9,8); |
| 13875 | Bool isU = INSN(7,7) == 1; |
| 13876 | Bool isQ = INSN(6,6) == 1; |
| 13877 | UInt bit_M = INSN(5,5); |
| 13878 | UInt fld_Vm = INSN(3,0); |
| 13879 | |
| 13880 | /* dd, nn, mm are D-register numbers. */ |
| 13881 | UInt dd = (bit_D << 4) | fld_Vd; |
| 13882 | UInt mm = (bit_M << 4) | fld_Vm; |
| 13883 | |
| 13884 | if (! (isQ && ((dd & 1) == 1 || (mm & 1) == 1))) { |
| 13885 | /* Do this piecewise on f regs. */ |
| 13886 | UInt ddF = dd << 1; |
| 13887 | UInt mmF = mm << 1; |
| 13888 | |
| 13889 | if (isT) { |
| 13890 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 13891 | } |
| 13892 | /* In ARM mode, this is statically unconditional. In Thumb mode, |
| 13893 | this must be dynamically unconditional, and we've SIGILLd if not. |
| 13894 | In either case we can create unconditional IR. */ |
| 13895 | |
| 13896 | UChar cvtc = '?'; |
| 13897 | IRRoundingMode rm = Irrm_NEAREST; |
| 13898 | switch (fld_rm) { |
| 13899 | /* The use of NEAREST for both the 'a' and 'n' cases is a bit of a |
| 13900 | kludge since it doesn't take into account the nearest-even vs |
| 13901 | nearest-away semantics. */ |
| 13902 | case BITS2(0,0): cvtc = 'a'; rm = Irrm_NEAREST; break; |
| 13903 | case BITS2(0,1): cvtc = 'n'; rm = Irrm_NEAREST; break; |
| 13904 | case BITS2(1,0): cvtc = 'p'; rm = Irrm_PosINF; break; |
| 13905 | case BITS2(1,1): cvtc = 'm'; rm = Irrm_NegINF; break; |
| 13906 | default: vassert(0); |
| 13907 | } |
| 13908 | |
| 13909 | IROp cvt = isU ? Iop_F64toI32U : Iop_F64toI32S; |
| 13910 | |
| 13911 | IRTemp r0 = newTemp(Ity_F32); |
| 13912 | IRTemp r1 = newTemp(Ity_F32); |
| 13913 | IRTemp r2 = isQ ? newTemp(Ity_F32) : IRTemp_INVALID; |
| 13914 | IRTemp r3 = isQ ? newTemp(Ity_F32) : IRTemp_INVALID; |
| 13915 | |
| 13916 | IRExpr* rmE = mkU32((UInt)rm); |
| 13917 | |
| 13918 | assign(r0, unop(Iop_ReinterpI32asF32, |
| 13919 | binop(cvt, rmE, unop(Iop_F32toF64, |
| 13920 | llGetFReg_up_to_64(mmF+0))))); |
| 13921 | assign(r1, unop(Iop_ReinterpI32asF32, |
| 13922 | binop(cvt, rmE, unop(Iop_F32toF64, |
| 13923 | llGetFReg_up_to_64(mmF+1))))); |
| 13924 | if (isQ) { |
| 13925 | assign(r2, unop(Iop_ReinterpI32asF32, |
| 13926 | binop(cvt, rmE, unop(Iop_F32toF64, |
| 13927 | llGetFReg_up_to_64(mmF+2))))); |
| 13928 | assign(r3, unop(Iop_ReinterpI32asF32, |
| 13929 | binop(cvt, rmE, unop(Iop_F32toF64, |
| 13930 | llGetFReg_up_to_64(mmF+3))))); |
| 13931 | } |
| 13932 | |
| 13933 | llPutFReg_up_to_64(ddF+0, mkexpr(r0)); |
| 13934 | llPutFReg_up_to_64(ddF+1, mkexpr(r1)); |
| 13935 | if (isQ) { |
| 13936 | llPutFReg_up_to_64(ddF+2, mkexpr(r2)); |
| 13937 | llPutFReg_up_to_64(ddF+3, mkexpr(r3)); |
| 13938 | } |
| 13939 | |
| 13940 | HChar rch = isQ ? 'q' : 'd'; |
| 13941 | UInt sh = isQ ? 1 : 0; |
| 13942 | DIP("vcvt%c.%c32.f32 %c%u, %c%u\n", |
| 13943 | cvtc, isU ? 'u' : 's', rch, dd >> sh, rch, mm >> sh); |
| 13944 | return True; |
| 13945 | } |
| 13946 | /* else fall through */ |
| 13947 | } |
| 13948 | |
| 13949 | /* ----------- VRINT{A,N,P,M,X,Z}{.F32 d_d, .F32 q_q} ----------- */ |
| 13950 | /* 31 27 22 21 15 11 9 6 5 4 3 |
| 13951 | T1: 1111 11111 D 111010 Vd 01 op Q M 0 Vm |
| 13952 | A1: 1111 00111 D 111010 Vd 01 op Q M 0 Vm |
| 13953 | |
| 13954 | ARM encoding is in NV space. |
| 13955 | In Thumb mode, we must not be in an IT block. |
| 13956 | */ |
| 13957 | if (INSN(31,23) == (isT ? BITS9(1,1,1,1,1,1,1,1,1) |
| 13958 | : BITS9(1,1,1,1,0,0,1,1,1)) |
| 13959 | && INSN(21,16) == BITS6(1,1,1,0,1,0) && INSN(11,10) == BITS2(0,1) |
| 13960 | && INSN(4,4) == 0) { |
| 13961 | UInt bit_D = INSN(22,22); |
| 13962 | UInt fld_Vd = INSN(15,12); |
| 13963 | UInt fld_op = INSN(9,7); |
| 13964 | Bool isQ = INSN(6,6) == 1; |
| 13965 | UInt bit_M = INSN(5,5); |
| 13966 | UInt fld_Vm = INSN(3,0); |
| 13967 | |
| 13968 | /* dd, nn, mm are D-register numbers. */ |
| 13969 | UInt dd = (bit_D << 4) | fld_Vd; |
| 13970 | UInt mm = (bit_M << 4) | fld_Vm; |
| 13971 | |
| 13972 | if (! (fld_op == BITS3(1,0,0) || fld_op == BITS3(1,1,0)) |
| 13973 | && ! (isQ && ((dd & 1) == 1 || (mm & 1) == 1))) { |
| 13974 | /* Do this piecewise on f regs. */ |
| 13975 | UInt ddF = dd << 1; |
| 13976 | UInt mmF = mm << 1; |
| 13977 | |
| 13978 | if (isT) { |
| 13979 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 13980 | } |
| 13981 | /* In ARM mode, this is statically unconditional. In Thumb mode, |
| 13982 | this must be dynamically unconditional, and we've SIGILLd if not. |
| 13983 | In either case we can create unconditional IR. */ |
| 13984 | |
| 13985 | UChar cvtc = '?'; |
| 13986 | IRRoundingMode rm = Irrm_NEAREST; |
| 13987 | switch (fld_op) { |
| 13988 | /* Various kludges: |
| 13989 | - The use of NEAREST for both the 'a' and 'n' cases, |
| 13990 | since it doesn't take into account the nearest-even vs |
| 13991 | nearest-away semantics. |
| 13992 | - For the 'x' case, we don't signal inexactness. |
| 13993 | */ |
| 13994 | case BITS3(0,1,0): cvtc = 'a'; rm = Irrm_NEAREST; break; |
| 13995 | case BITS3(0,0,0): cvtc = 'n'; rm = Irrm_NEAREST; break; |
| 13996 | case BITS3(1,1,1): cvtc = 'p'; rm = Irrm_PosINF; break; |
| 13997 | case BITS3(1,0,1): cvtc = 'm'; rm = Irrm_NegINF; break; |
| 13998 | case BITS3(0,1,1): cvtc = 'z'; rm = Irrm_ZERO; break; |
| 13999 | case BITS3(0,0,1): cvtc = 'x'; rm = Irrm_NEAREST; break; |
| 14000 | case BITS3(1,0,0): |
| 14001 | case BITS3(1,1,0): |
| 14002 | default: vassert(0); |
| 14003 | } |
| 14004 | |
| 14005 | IRTemp r0 = newTemp(Ity_F32); |
| 14006 | IRTemp r1 = newTemp(Ity_F32); |
| 14007 | IRTemp r2 = isQ ? newTemp(Ity_F32) : IRTemp_INVALID; |
| 14008 | IRTemp r3 = isQ ? newTemp(Ity_F32) : IRTemp_INVALID; |
| 14009 | |
| 14010 | IRExpr* rmE = mkU32((UInt)rm); |
| 14011 | IROp rnd = Iop_RoundF32toInt; |
| 14012 | |
| 14013 | assign(r0, binop(rnd, rmE, llGetFReg_up_to_64(mmF+0))); |
| 14014 | assign(r1, binop(rnd, rmE, llGetFReg_up_to_64(mmF+1))); |
| 14015 | if (isQ) { |
| 14016 | assign(r2, binop(rnd, rmE, llGetFReg_up_to_64(mmF+2))); |
| 14017 | assign(r3, binop(rnd, rmE, llGetFReg_up_to_64(mmF+3))); |
| 14018 | } |
| 14019 | |
| 14020 | llPutFReg_up_to_64(ddF+0, mkexpr(r0)); |
| 14021 | llPutFReg_up_to_64(ddF+1, mkexpr(r1)); |
| 14022 | if (isQ) { |
| 14023 | llPutFReg_up_to_64(ddF+2, mkexpr(r2)); |
| 14024 | llPutFReg_up_to_64(ddF+3, mkexpr(r3)); |
| 14025 | } |
| 14026 | |
| 14027 | HChar rch = isQ ? 'q' : 'd'; |
| 14028 | UInt sh = isQ ? 1 : 0; |
| 14029 | DIP("vrint%c.f32.f32 %c%u, %c%u\n", |
| 14030 | cvtc, rch, dd >> sh, rch, mm >> sh); |
| 14031 | return True; |
| 14032 | } |
| 14033 | /* else fall through */ |
| 14034 | } |
| 14035 | |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 14036 | /* ---------- Doesn't match anything. ---------- */ |
| 14037 | return False; |
| 14038 | |
| 14039 | # undef INSN |
| 14040 | } |
| 14041 | |
| 14042 | |
| 14043 | /*------------------------------------------------------------*/ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14044 | /*--- LDMxx/STMxx helper (both ARM and Thumb32) ---*/ |
| 14045 | /*------------------------------------------------------------*/ |
| 14046 | |
| 14047 | /* Generate IR for LDMxx and STMxx. This is complex. Assumes it's |
| 14048 | unconditional, so the caller must produce a jump-around before |
| 14049 | calling this, if the insn is to be conditional. Caller is |
| 14050 | responsible for all validation of parameters. For LDMxx, if PC is |
| 14051 | amongst the values loaded, caller is also responsible for |
| 14052 | generating the jump. */ |
| 14053 | static void mk_ldm_stm ( Bool arm, /* True: ARM, False: Thumb */ |
| 14054 | UInt rN, /* base reg */ |
| 14055 | UInt bINC, /* 1: inc, 0: dec */ |
| 14056 | UInt bBEFORE, /* 1: inc/dec before, 0: after */ |
| 14057 | UInt bW, /* 1: writeback to Rn */ |
| 14058 | UInt bL, /* 1: load, 0: store */ |
| 14059 | UInt regList ) |
| 14060 | { |
| 14061 | Int i, r, m, nRegs; |
sewardj | f580065 | 2011-10-14 15:44:00 +0000 | [diff] [blame] | 14062 | IRTemp jk = Ijk_Boring; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14063 | |
| 14064 | /* Get hold of the old Rn value. We might need to write its value |
| 14065 | to memory during a store, and if it's also the writeback |
| 14066 | register then we need to get its value now. We can't treat it |
| 14067 | exactly like the other registers we're going to transfer, |
| 14068 | because for xxMDA and xxMDB writeback forms, the generated IR |
| 14069 | updates Rn in the guest state before any transfers take place. |
| 14070 | We have to do this as per comments below, in order that if Rn is |
| 14071 | the stack pointer then it always has a value is below or equal |
| 14072 | to any of the transfer addresses. Ick. */ |
| 14073 | IRTemp oldRnT = newTemp(Ity_I32); |
| 14074 | assign(oldRnT, arm ? getIRegA(rN) : getIRegT(rN)); |
| 14075 | |
| 14076 | IRTemp anchorT = newTemp(Ity_I32); |
| 14077 | /* The old (Addison-Wesley) ARM ARM seems to say that LDMxx/STMxx |
| 14078 | ignore the bottom two bits of the address. However, Cortex-A8 |
| 14079 | doesn't seem to care. Hence: */ |
| 14080 | /* No .. don't force alignment .. */ |
| 14081 | /* assign(anchorT, binop(Iop_And32, mkexpr(oldRnT), mkU32(~3U))); */ |
| 14082 | /* Instead, use the potentially misaligned address directly. */ |
| 14083 | assign(anchorT, mkexpr(oldRnT)); |
| 14084 | |
| 14085 | IROp opADDorSUB = bINC ? Iop_Add32 : Iop_Sub32; |
| 14086 | // bINC == 1: xxMIA, xxMIB |
| 14087 | // bINC == 0: xxMDA, xxMDB |
| 14088 | |
| 14089 | // For xxMDA and xxMDB, update Rn first if necessary. We have |
| 14090 | // to do this first so that, for the common idiom of the transfers |
| 14091 | // faulting because we're pushing stuff onto a stack and the stack |
| 14092 | // is growing down onto allocate-on-fault pages (as Valgrind simulates), |
| 14093 | // we need to have the SP up-to-date "covering" (pointing below) the |
| 14094 | // transfer area. For the same reason, if we are doing xxMIA or xxMIB, |
| 14095 | // do the transfer first, and then update rN afterwards. |
| 14096 | nRegs = 0; |
| 14097 | for (i = 0; i < 16; i++) { |
| 14098 | if ((regList & (1 << i)) != 0) |
| 14099 | nRegs++; |
| 14100 | } |
| 14101 | if (bW == 1 && !bINC) { |
| 14102 | IRExpr* e = binop(opADDorSUB, mkexpr(oldRnT), mkU32(4*nRegs)); |
| 14103 | if (arm) |
| 14104 | putIRegA( rN, e, IRTemp_INVALID, Ijk_Boring ); |
| 14105 | else |
| 14106 | putIRegT( rN, e, IRTemp_INVALID ); |
| 14107 | } |
| 14108 | |
| 14109 | // Make up a list of the registers to transfer, and their offsets |
| 14110 | // in memory relative to the anchor. If the base reg (Rn) is part |
| 14111 | // of the transfer, then do it last for a load and first for a store. |
| 14112 | UInt xReg[16], xOff[16]; |
| 14113 | Int nX = 0; |
| 14114 | m = 0; |
| 14115 | for (i = 0; i < 16; i++) { |
| 14116 | r = bINC ? i : (15-i); |
| 14117 | if (0 == (regList & (1<<r))) |
| 14118 | continue; |
| 14119 | if (bBEFORE) |
| 14120 | m++; |
| 14121 | /* paranoia: check we aren't transferring the writeback |
| 14122 | register during a load. Should be assured by decode-point |
| 14123 | check above. */ |
| 14124 | if (bW == 1 && bL == 1) |
| 14125 | vassert(r != rN); |
| 14126 | |
| 14127 | xOff[nX] = 4 * m; |
| 14128 | xReg[nX] = r; |
| 14129 | nX++; |
| 14130 | |
| 14131 | if (!bBEFORE) |
| 14132 | m++; |
| 14133 | } |
| 14134 | vassert(m == nRegs); |
| 14135 | vassert(nX == nRegs); |
| 14136 | vassert(nX <= 16); |
| 14137 | |
| 14138 | if (bW == 0 && (regList & (1<<rN)) != 0) { |
| 14139 | /* Non-writeback, and basereg is to be transferred. Do its |
| 14140 | transfer last for a load and first for a store. Requires |
| 14141 | reordering xOff/xReg. */ |
| 14142 | if (0) { |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 14143 | vex_printf("\nREG_LIST_PRE: (rN=%u)\n", rN); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14144 | for (i = 0; i < nX; i++) |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 14145 | vex_printf("reg %u off %u\n", xReg[i], xOff[i]); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14146 | vex_printf("\n"); |
| 14147 | } |
| 14148 | |
| 14149 | vassert(nX > 0); |
| 14150 | for (i = 0; i < nX; i++) { |
| 14151 | if (xReg[i] == rN) |
| 14152 | break; |
| 14153 | } |
| 14154 | vassert(i < nX); /* else we didn't find it! */ |
| 14155 | UInt tReg = xReg[i]; |
| 14156 | UInt tOff = xOff[i]; |
| 14157 | if (bL == 1) { |
| 14158 | /* load; make this transfer happen last */ |
| 14159 | if (i < nX-1) { |
| 14160 | for (m = i+1; m < nX; m++) { |
| 14161 | xReg[m-1] = xReg[m]; |
| 14162 | xOff[m-1] = xOff[m]; |
| 14163 | } |
| 14164 | vassert(m == nX); |
| 14165 | xReg[m-1] = tReg; |
| 14166 | xOff[m-1] = tOff; |
| 14167 | } |
| 14168 | } else { |
| 14169 | /* store; make this transfer happen first */ |
| 14170 | if (i > 0) { |
| 14171 | for (m = i-1; m >= 0; m--) { |
| 14172 | xReg[m+1] = xReg[m]; |
| 14173 | xOff[m+1] = xOff[m]; |
| 14174 | } |
| 14175 | vassert(m == -1); |
| 14176 | xReg[0] = tReg; |
| 14177 | xOff[0] = tOff; |
| 14178 | } |
| 14179 | } |
| 14180 | |
| 14181 | if (0) { |
| 14182 | vex_printf("REG_LIST_POST:\n"); |
| 14183 | for (i = 0; i < nX; i++) |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 14184 | vex_printf("reg %u off %u\n", xReg[i], xOff[i]); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14185 | vex_printf("\n"); |
| 14186 | } |
| 14187 | } |
| 14188 | |
sewardj | f580065 | 2011-10-14 15:44:00 +0000 | [diff] [blame] | 14189 | /* According to the Cortex A8 TRM Sec. 5.2.1, LDM(1) with r13 as the base |
| 14190 | register and PC in the register list is a return for purposes of branch |
| 14191 | prediction. |
| 14192 | The ARM ARM Sec. C9.10.1 further specifies that writeback must be enabled |
| 14193 | to be counted in event 0x0E (Procedure return).*/ |
| 14194 | if (rN == 13 && bL == 1 && bINC && !bBEFORE && bW == 1) { |
| 14195 | jk = Ijk_Ret; |
| 14196 | } |
| 14197 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14198 | /* Actually generate the transfers */ |
| 14199 | for (i = 0; i < nX; i++) { |
| 14200 | r = xReg[i]; |
| 14201 | if (bL == 1) { |
| 14202 | IRExpr* e = loadLE(Ity_I32, |
| 14203 | binop(opADDorSUB, mkexpr(anchorT), |
| 14204 | mkU32(xOff[i]))); |
| 14205 | if (arm) { |
sewardj | f580065 | 2011-10-14 15:44:00 +0000 | [diff] [blame] | 14206 | putIRegA( r, e, IRTemp_INVALID, jk ); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14207 | } else { |
| 14208 | // no: putIRegT( r, e, IRTemp_INVALID ); |
| 14209 | // putIRegT refuses to write to R15. But that might happen. |
| 14210 | // Since this is uncond, and we need to be able to |
| 14211 | // write the PC, just use the low level put: |
| 14212 | llPutIReg( r, e ); |
| 14213 | } |
| 14214 | } else { |
| 14215 | /* if we're storing Rn, make sure we use the correct |
| 14216 | value, as per extensive comments above */ |
| 14217 | storeLE( binop(opADDorSUB, mkexpr(anchorT), mkU32(xOff[i])), |
| 14218 | r == rN ? mkexpr(oldRnT) |
| 14219 | : (arm ? getIRegA(r) : getIRegT(r) ) ); |
| 14220 | } |
| 14221 | } |
| 14222 | |
| 14223 | // If we are doing xxMIA or xxMIB, |
| 14224 | // do the transfer first, and then update rN afterwards. |
| 14225 | if (bW == 1 && bINC) { |
| 14226 | IRExpr* e = binop(opADDorSUB, mkexpr(oldRnT), mkU32(4*nRegs)); |
| 14227 | if (arm) |
| 14228 | putIRegA( rN, e, IRTemp_INVALID, Ijk_Boring ); |
| 14229 | else |
| 14230 | putIRegT( rN, e, IRTemp_INVALID ); |
| 14231 | } |
| 14232 | } |
| 14233 | |
| 14234 | |
| 14235 | /*------------------------------------------------------------*/ |
| 14236 | /*--- VFP (CP 10 and 11) instructions ---*/ |
| 14237 | /*------------------------------------------------------------*/ |
| 14238 | |
| 14239 | /* Both ARM and Thumb */ |
| 14240 | |
| 14241 | /* Translate a CP10 or CP11 instruction. If successful, returns |
| 14242 | True and *dres may or may not be updated. If failure, returns |
| 14243 | False and doesn't change *dres nor create any IR. |
| 14244 | |
| 14245 | The ARM and Thumb encodings are identical for the low 28 bits of |
| 14246 | the insn (yay!) and that's what the caller must supply, iow, imm28 |
| 14247 | has the top 4 bits masked out. Caller is responsible for |
| 14248 | determining whether the masked-out bits are valid for a CP10/11 |
| 14249 | insn. The rules for the top 4 bits are: |
| 14250 | |
| 14251 | ARM: 0000 to 1110 allowed, and this is the gating condition. |
| 14252 | 1111 (NV) is not allowed. |
| 14253 | |
| 14254 | Thumb: must be 1110. The gating condition is taken from |
| 14255 | ITSTATE in the normal way. |
| 14256 | |
| 14257 | Conditionalisation: |
| 14258 | |
| 14259 | Caller must supply an IRTemp 'condT' holding the gating condition, |
| 14260 | or IRTemp_INVALID indicating the insn is always executed. |
| 14261 | |
| 14262 | Caller must also supply an ARMCondcode 'cond'. This is only used |
| 14263 | for debug printing, no other purpose. For ARM, this is simply the |
| 14264 | top 4 bits of the original instruction. For Thumb, the condition |
| 14265 | is not (really) known until run time, and so ARMCondAL should be |
| 14266 | passed, only so that printing of these instructions does not show |
| 14267 | any condition. |
| 14268 | |
| 14269 | Finally, the caller must indicate whether this occurs in ARM or |
| 14270 | Thumb code. |
| 14271 | */ |
| 14272 | static Bool decode_CP10_CP11_instruction ( |
| 14273 | /*MOD*/DisResult* dres, |
| 14274 | UInt insn28, |
| 14275 | IRTemp condT, |
| 14276 | ARMCondcode conq, |
| 14277 | Bool isT |
| 14278 | ) |
| 14279 | { |
| 14280 | # define INSN(_bMax,_bMin) SLICE_UInt(insn28, (_bMax), (_bMin)) |
| 14281 | |
| 14282 | vassert(INSN(31,28) == BITS4(0,0,0,0)); // caller's obligation |
| 14283 | |
| 14284 | if (isT) { |
| 14285 | vassert(conq == ARMCondAL); |
| 14286 | } else { |
| 14287 | vassert(conq >= ARMCondEQ && conq <= ARMCondAL); |
| 14288 | } |
| 14289 | |
| 14290 | /* ----------------------------------------------------------- */ |
| 14291 | /* -- VFP instructions -- double precision (mostly) -- */ |
| 14292 | /* ----------------------------------------------------------- */ |
| 14293 | |
| 14294 | /* --------------------- fldmx, fstmx --------------------- */ |
| 14295 | /* |
| 14296 | 31 27 23 19 15 11 7 0 |
| 14297 | P U WL |
| 14298 | C4-100, C5-26 1 FSTMX cond 1100 1000 Rn Dd 1011 offset |
| 14299 | C4-100, C5-28 2 FSTMIAX cond 1100 1010 Rn Dd 1011 offset |
| 14300 | C4-100, C5-30 3 FSTMDBX cond 1101 0010 Rn Dd 1011 offset |
| 14301 | |
| 14302 | C4-42, C5-26 1 FLDMX cond 1100 1001 Rn Dd 1011 offset |
| 14303 | C4-42, C5-28 2 FLDMIAX cond 1100 1011 Rn Dd 1011 offset |
| 14304 | C4-42, C5-30 3 FLDMDBX cond 1101 0011 Rn Dd 1011 offset |
| 14305 | |
| 14306 | Regs transferred: Dd .. D(d + (offset-3)/2) |
| 14307 | offset must be odd, must not imply a reg > 15 |
| 14308 | IA/DB: Rn is changed by (4 + 8 x # regs transferred) |
| 14309 | |
| 14310 | case coding: |
| 14311 | 1 at-Rn (access at Rn) |
| 14312 | 2 ia-Rn (access at Rn, then Rn += 4+8n) |
| 14313 | 3 db-Rn (Rn -= 4+8n, then access at Rn) |
| 14314 | */ |
| 14315 | if (BITS8(1,1,0,0,0,0,0,0) == (INSN(27,20) & BITS8(1,1,1,0,0,0,0,0)) |
| 14316 | && INSN(11,8) == BITS4(1,0,1,1)) { |
| 14317 | UInt bP = (insn28 >> 24) & 1; |
| 14318 | UInt bU = (insn28 >> 23) & 1; |
| 14319 | UInt bW = (insn28 >> 21) & 1; |
| 14320 | UInt bL = (insn28 >> 20) & 1; |
| 14321 | UInt offset = (insn28 >> 0) & 0xFF; |
| 14322 | UInt rN = INSN(19,16); |
| 14323 | UInt dD = (INSN(22,22) << 4) | INSN(15,12); |
| 14324 | UInt nRegs = (offset - 1) / 2; |
| 14325 | UInt summary = 0; |
| 14326 | Int i; |
| 14327 | |
| 14328 | /**/ if (bP == 0 && bU == 1 && bW == 0) { |
| 14329 | summary = 1; |
| 14330 | } |
| 14331 | else if (bP == 0 && bU == 1 && bW == 1) { |
| 14332 | summary = 2; |
| 14333 | } |
| 14334 | else if (bP == 1 && bU == 0 && bW == 1) { |
| 14335 | summary = 3; |
| 14336 | } |
| 14337 | else goto after_vfp_fldmx_fstmx; |
| 14338 | |
| 14339 | /* no writebacks to r15 allowed. No use of r15 in thumb mode. */ |
| 14340 | if (rN == 15 && (summary == 2 || summary == 3 || isT)) |
| 14341 | goto after_vfp_fldmx_fstmx; |
| 14342 | |
| 14343 | /* offset must be odd, and specify at least one register */ |
| 14344 | if (0 == (offset & 1) || offset < 3) |
| 14345 | goto after_vfp_fldmx_fstmx; |
| 14346 | |
| 14347 | /* can't transfer regs after D15 */ |
| 14348 | if (dD + nRegs - 1 >= 32) |
| 14349 | goto after_vfp_fldmx_fstmx; |
| 14350 | |
| 14351 | /* Now, we can't do a conditional load or store, since that very |
| 14352 | likely will generate an exception. So we have to take a side |
| 14353 | exit at this point if the condition is false. */ |
| 14354 | if (condT != IRTemp_INVALID) { |
| 14355 | if (isT) |
| 14356 | mk_skip_over_T32_if_cond_is_false( condT ); |
| 14357 | else |
| 14358 | mk_skip_over_A32_if_cond_is_false( condT ); |
| 14359 | condT = IRTemp_INVALID; |
| 14360 | } |
| 14361 | /* Ok, now we're unconditional. Do the load or store. */ |
| 14362 | |
| 14363 | /* get the old Rn value */ |
| 14364 | IRTemp rnT = newTemp(Ity_I32); |
| 14365 | assign(rnT, align4if(isT ? getIRegT(rN) : getIRegA(rN), |
| 14366 | rN == 15)); |
| 14367 | |
| 14368 | /* make a new value for Rn, post-insn */ |
| 14369 | IRTemp rnTnew = IRTemp_INVALID; |
| 14370 | if (summary == 2 || summary == 3) { |
| 14371 | rnTnew = newTemp(Ity_I32); |
| 14372 | assign(rnTnew, binop(summary == 2 ? Iop_Add32 : Iop_Sub32, |
| 14373 | mkexpr(rnT), |
| 14374 | mkU32(4 + 8 * nRegs))); |
| 14375 | } |
| 14376 | |
| 14377 | /* decide on the base transfer address */ |
| 14378 | IRTemp taT = newTemp(Ity_I32); |
| 14379 | assign(taT, summary == 3 ? mkexpr(rnTnew) : mkexpr(rnT)); |
| 14380 | |
| 14381 | /* update Rn if necessary -- in case 3, we're moving it down, so |
| 14382 | update before any memory reference, in order to keep Memcheck |
| 14383 | and V's stack-extending logic (on linux) happy */ |
| 14384 | if (summary == 3) { |
| 14385 | if (isT) |
| 14386 | putIRegT(rN, mkexpr(rnTnew), IRTemp_INVALID); |
| 14387 | else |
| 14388 | putIRegA(rN, mkexpr(rnTnew), IRTemp_INVALID, Ijk_Boring); |
| 14389 | } |
| 14390 | |
| 14391 | /* generate the transfers */ |
| 14392 | for (i = 0; i < nRegs; i++) { |
| 14393 | IRExpr* addr = binop(Iop_Add32, mkexpr(taT), mkU32(8*i)); |
| 14394 | if (bL) { |
| 14395 | putDReg(dD + i, loadLE(Ity_F64, addr), IRTemp_INVALID); |
| 14396 | } else { |
| 14397 | storeLE(addr, getDReg(dD + i)); |
| 14398 | } |
| 14399 | } |
| 14400 | |
| 14401 | /* update Rn if necessary -- in case 2, we're moving it up, so |
| 14402 | update after any memory reference, in order to keep Memcheck |
| 14403 | and V's stack-extending logic (on linux) happy */ |
| 14404 | if (summary == 2) { |
| 14405 | if (isT) |
| 14406 | putIRegT(rN, mkexpr(rnTnew), IRTemp_INVALID); |
| 14407 | else |
| 14408 | putIRegA(rN, mkexpr(rnTnew), IRTemp_INVALID, Ijk_Boring); |
| 14409 | } |
| 14410 | |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 14411 | const HChar* nm = bL==1 ? "ld" : "st"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14412 | switch (summary) { |
| 14413 | case 1: DIP("f%smx%s r%u, {d%u-d%u}\n", |
| 14414 | nm, nCC(conq), rN, dD, dD + nRegs - 1); |
| 14415 | break; |
| 14416 | case 2: DIP("f%smiax%s r%u!, {d%u-d%u}\n", |
| 14417 | nm, nCC(conq), rN, dD, dD + nRegs - 1); |
| 14418 | break; |
| 14419 | case 3: DIP("f%smdbx%s r%u!, {d%u-d%u}\n", |
| 14420 | nm, nCC(conq), rN, dD, dD + nRegs - 1); |
| 14421 | break; |
| 14422 | default: vassert(0); |
| 14423 | } |
| 14424 | |
| 14425 | goto decode_success_vfp; |
| 14426 | /* FIXME alignment constraints? */ |
| 14427 | } |
| 14428 | |
| 14429 | after_vfp_fldmx_fstmx: |
| 14430 | |
| 14431 | /* --------------------- fldmd, fstmd --------------------- */ |
| 14432 | /* |
| 14433 | 31 27 23 19 15 11 7 0 |
| 14434 | P U WL |
| 14435 | C4-96, C5-26 1 FSTMD cond 1100 1000 Rn Dd 1011 offset |
| 14436 | C4-96, C5-28 2 FSTMDIA cond 1100 1010 Rn Dd 1011 offset |
| 14437 | C4-96, C5-30 3 FSTMDDB cond 1101 0010 Rn Dd 1011 offset |
| 14438 | |
| 14439 | C4-38, C5-26 1 FLDMD cond 1100 1001 Rn Dd 1011 offset |
| 14440 | C4-38, C5-28 2 FLDMIAD cond 1100 1011 Rn Dd 1011 offset |
| 14441 | C4-38, C5-30 3 FLDMDBD cond 1101 0011 Rn Dd 1011 offset |
| 14442 | |
| 14443 | Regs transferred: Dd .. D(d + (offset-2)/2) |
| 14444 | offset must be even, must not imply a reg > 15 |
| 14445 | IA/DB: Rn is changed by (8 x # regs transferred) |
| 14446 | |
| 14447 | case coding: |
| 14448 | 1 at-Rn (access at Rn) |
| 14449 | 2 ia-Rn (access at Rn, then Rn += 8n) |
| 14450 | 3 db-Rn (Rn -= 8n, then access at Rn) |
| 14451 | */ |
| 14452 | if (BITS8(1,1,0,0,0,0,0,0) == (INSN(27,20) & BITS8(1,1,1,0,0,0,0,0)) |
| 14453 | && INSN(11,8) == BITS4(1,0,1,1)) { |
| 14454 | UInt bP = (insn28 >> 24) & 1; |
| 14455 | UInt bU = (insn28 >> 23) & 1; |
| 14456 | UInt bW = (insn28 >> 21) & 1; |
| 14457 | UInt bL = (insn28 >> 20) & 1; |
| 14458 | UInt offset = (insn28 >> 0) & 0xFF; |
| 14459 | UInt rN = INSN(19,16); |
| 14460 | UInt dD = (INSN(22,22) << 4) | INSN(15,12); |
| 14461 | UInt nRegs = offset / 2; |
| 14462 | UInt summary = 0; |
| 14463 | Int i; |
| 14464 | |
| 14465 | /**/ if (bP == 0 && bU == 1 && bW == 0) { |
| 14466 | summary = 1; |
| 14467 | } |
| 14468 | else if (bP == 0 && bU == 1 && bW == 1) { |
| 14469 | summary = 2; |
| 14470 | } |
| 14471 | else if (bP == 1 && bU == 0 && bW == 1) { |
| 14472 | summary = 3; |
| 14473 | } |
| 14474 | else goto after_vfp_fldmd_fstmd; |
| 14475 | |
| 14476 | /* no writebacks to r15 allowed. No use of r15 in thumb mode. */ |
| 14477 | if (rN == 15 && (summary == 2 || summary == 3 || isT)) |
| 14478 | goto after_vfp_fldmd_fstmd; |
| 14479 | |
| 14480 | /* offset must be even, and specify at least one register */ |
| 14481 | if (1 == (offset & 1) || offset < 2) |
| 14482 | goto after_vfp_fldmd_fstmd; |
| 14483 | |
| 14484 | /* can't transfer regs after D15 */ |
| 14485 | if (dD + nRegs - 1 >= 32) |
| 14486 | goto after_vfp_fldmd_fstmd; |
| 14487 | |
| 14488 | /* Now, we can't do a conditional load or store, since that very |
| 14489 | likely will generate an exception. So we have to take a side |
| 14490 | exit at this point if the condition is false. */ |
| 14491 | if (condT != IRTemp_INVALID) { |
| 14492 | if (isT) |
| 14493 | mk_skip_over_T32_if_cond_is_false( condT ); |
| 14494 | else |
| 14495 | mk_skip_over_A32_if_cond_is_false( condT ); |
| 14496 | condT = IRTemp_INVALID; |
| 14497 | } |
| 14498 | /* Ok, now we're unconditional. Do the load or store. */ |
| 14499 | |
| 14500 | /* get the old Rn value */ |
| 14501 | IRTemp rnT = newTemp(Ity_I32); |
| 14502 | assign(rnT, align4if(isT ? getIRegT(rN) : getIRegA(rN), |
| 14503 | rN == 15)); |
| 14504 | |
| 14505 | /* make a new value for Rn, post-insn */ |
| 14506 | IRTemp rnTnew = IRTemp_INVALID; |
| 14507 | if (summary == 2 || summary == 3) { |
| 14508 | rnTnew = newTemp(Ity_I32); |
| 14509 | assign(rnTnew, binop(summary == 2 ? Iop_Add32 : Iop_Sub32, |
| 14510 | mkexpr(rnT), |
| 14511 | mkU32(8 * nRegs))); |
| 14512 | } |
| 14513 | |
| 14514 | /* decide on the base transfer address */ |
| 14515 | IRTemp taT = newTemp(Ity_I32); |
| 14516 | assign(taT, summary == 3 ? mkexpr(rnTnew) : mkexpr(rnT)); |
| 14517 | |
| 14518 | /* update Rn if necessary -- in case 3, we're moving it down, so |
| 14519 | update before any memory reference, in order to keep Memcheck |
| 14520 | and V's stack-extending logic (on linux) happy */ |
| 14521 | if (summary == 3) { |
| 14522 | if (isT) |
| 14523 | putIRegT(rN, mkexpr(rnTnew), IRTemp_INVALID); |
| 14524 | else |
| 14525 | putIRegA(rN, mkexpr(rnTnew), IRTemp_INVALID, Ijk_Boring); |
| 14526 | } |
| 14527 | |
| 14528 | /* generate the transfers */ |
| 14529 | for (i = 0; i < nRegs; i++) { |
| 14530 | IRExpr* addr = binop(Iop_Add32, mkexpr(taT), mkU32(8*i)); |
| 14531 | if (bL) { |
| 14532 | putDReg(dD + i, loadLE(Ity_F64, addr), IRTemp_INVALID); |
| 14533 | } else { |
| 14534 | storeLE(addr, getDReg(dD + i)); |
| 14535 | } |
| 14536 | } |
| 14537 | |
| 14538 | /* update Rn if necessary -- in case 2, we're moving it up, so |
| 14539 | update after any memory reference, in order to keep Memcheck |
| 14540 | and V's stack-extending logic (on linux) happy */ |
| 14541 | if (summary == 2) { |
| 14542 | if (isT) |
| 14543 | putIRegT(rN, mkexpr(rnTnew), IRTemp_INVALID); |
| 14544 | else |
| 14545 | putIRegA(rN, mkexpr(rnTnew), IRTemp_INVALID, Ijk_Boring); |
| 14546 | } |
| 14547 | |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 14548 | const HChar* nm = bL==1 ? "ld" : "st"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14549 | switch (summary) { |
| 14550 | case 1: DIP("f%smd%s r%u, {d%u-d%u}\n", |
| 14551 | nm, nCC(conq), rN, dD, dD + nRegs - 1); |
| 14552 | break; |
| 14553 | case 2: DIP("f%smiad%s r%u!, {d%u-d%u}\n", |
| 14554 | nm, nCC(conq), rN, dD, dD + nRegs - 1); |
| 14555 | break; |
| 14556 | case 3: DIP("f%smdbd%s r%u!, {d%u-d%u}\n", |
| 14557 | nm, nCC(conq), rN, dD, dD + nRegs - 1); |
| 14558 | break; |
| 14559 | default: vassert(0); |
| 14560 | } |
| 14561 | |
| 14562 | goto decode_success_vfp; |
| 14563 | /* FIXME alignment constraints? */ |
| 14564 | } |
| 14565 | |
| 14566 | after_vfp_fldmd_fstmd: |
| 14567 | |
| 14568 | /* ------------------- fmrx, fmxr ------------------- */ |
| 14569 | if (BITS8(1,1,1,0,1,1,1,1) == INSN(27,20) |
| 14570 | && BITS4(1,0,1,0) == INSN(11,8) |
| 14571 | && BITS8(0,0,0,1,0,0,0,0) == (insn28 & 0xFF)) { |
| 14572 | UInt rD = INSN(15,12); |
| 14573 | UInt reg = INSN(19,16); |
| 14574 | if (reg == BITS4(0,0,0,1)) { |
| 14575 | if (rD == 15) { |
| 14576 | IRTemp nzcvT = newTemp(Ity_I32); |
| 14577 | /* When rD is 15, we are copying the top 4 bits of FPSCR |
| 14578 | into CPSR. That is, set the flags thunk to COPY and |
| 14579 | install FPSCR[31:28] as the value to copy. */ |
| 14580 | assign(nzcvT, binop(Iop_And32, |
| 14581 | IRExpr_Get(OFFB_FPSCR, Ity_I32), |
| 14582 | mkU32(0xF0000000))); |
| 14583 | setFlags_D1(ARMG_CC_OP_COPY, nzcvT, condT); |
| 14584 | DIP("fmstat%s\n", nCC(conq)); |
| 14585 | } else { |
| 14586 | /* Otherwise, merely transfer FPSCR to r0 .. r14. */ |
| 14587 | IRExpr* e = IRExpr_Get(OFFB_FPSCR, Ity_I32); |
| 14588 | if (isT) |
| 14589 | putIRegT(rD, e, condT); |
| 14590 | else |
| 14591 | putIRegA(rD, e, condT, Ijk_Boring); |
| 14592 | DIP("fmrx%s r%u, fpscr\n", nCC(conq), rD); |
| 14593 | } |
| 14594 | goto decode_success_vfp; |
| 14595 | } |
| 14596 | /* fall through */ |
| 14597 | } |
| 14598 | |
| 14599 | if (BITS8(1,1,1,0,1,1,1,0) == INSN(27,20) |
| 14600 | && BITS4(1,0,1,0) == INSN(11,8) |
| 14601 | && BITS8(0,0,0,1,0,0,0,0) == (insn28 & 0xFF)) { |
| 14602 | UInt rD = INSN(15,12); |
| 14603 | UInt reg = INSN(19,16); |
| 14604 | if (reg == BITS4(0,0,0,1)) { |
| 14605 | putMiscReg32(OFFB_FPSCR, |
| 14606 | isT ? getIRegT(rD) : getIRegA(rD), condT); |
| 14607 | DIP("fmxr%s fpscr, r%u\n", nCC(conq), rD); |
| 14608 | goto decode_success_vfp; |
| 14609 | } |
| 14610 | /* fall through */ |
| 14611 | } |
| 14612 | |
| 14613 | /* --------------------- vmov --------------------- */ |
| 14614 | // VMOV dM, rD, rN |
| 14615 | if (0x0C400B10 == (insn28 & 0x0FF00FD0)) { |
| 14616 | UInt dM = INSN(3,0) | (INSN(5,5) << 4); |
| 14617 | UInt rD = INSN(15,12); /* lo32 */ |
| 14618 | UInt rN = INSN(19,16); /* hi32 */ |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14619 | if (rD == 15 || rN == 15 || (isT && (rD == 13 || rN == 13))) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14620 | /* fall through */ |
| 14621 | } else { |
| 14622 | putDReg(dM, |
| 14623 | unop(Iop_ReinterpI64asF64, |
| 14624 | binop(Iop_32HLto64, |
| 14625 | isT ? getIRegT(rN) : getIRegA(rN), |
| 14626 | isT ? getIRegT(rD) : getIRegA(rD))), |
| 14627 | condT); |
| 14628 | DIP("vmov%s d%u, r%u, r%u\n", nCC(conq), dM, rD, rN); |
| 14629 | goto decode_success_vfp; |
| 14630 | } |
| 14631 | /* fall through */ |
| 14632 | } |
| 14633 | |
| 14634 | // VMOV rD, rN, dM |
| 14635 | if (0x0C500B10 == (insn28 & 0x0FF00FD0)) { |
| 14636 | UInt dM = INSN(3,0) | (INSN(5,5) << 4); |
| 14637 | UInt rD = INSN(15,12); /* lo32 */ |
| 14638 | UInt rN = INSN(19,16); /* hi32 */ |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14639 | if (rD == 15 || rN == 15 || (isT && (rD == 13 || rN == 13)) |
| 14640 | || rD == rN) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14641 | /* fall through */ |
| 14642 | } else { |
| 14643 | IRTemp i64 = newTemp(Ity_I64); |
| 14644 | assign(i64, unop(Iop_ReinterpF64asI64, getDReg(dM))); |
| 14645 | IRExpr* hi32 = unop(Iop_64HIto32, mkexpr(i64)); |
| 14646 | IRExpr* lo32 = unop(Iop_64to32, mkexpr(i64)); |
| 14647 | if (isT) { |
| 14648 | putIRegT(rN, hi32, condT); |
| 14649 | putIRegT(rD, lo32, condT); |
| 14650 | } else { |
| 14651 | putIRegA(rN, hi32, condT, Ijk_Boring); |
| 14652 | putIRegA(rD, lo32, condT, Ijk_Boring); |
| 14653 | } |
| 14654 | DIP("vmov%s r%u, r%u, d%u\n", nCC(conq), rD, rN, dM); |
| 14655 | goto decode_success_vfp; |
| 14656 | } |
| 14657 | /* fall through */ |
| 14658 | } |
| 14659 | |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14660 | // VMOV sD, sD+1, rN, rM |
| 14661 | if (0x0C400A10 == (insn28 & 0x0FF00FD0)) { |
| 14662 | UInt sD = (INSN(3,0) << 1) | INSN(5,5); |
| 14663 | UInt rN = INSN(15,12); |
| 14664 | UInt rM = INSN(19,16); |
| 14665 | if (rM == 15 || rN == 15 || (isT && (rM == 13 || rN == 13)) |
| 14666 | || sD == 31) { |
| 14667 | /* fall through */ |
| 14668 | } else { |
| 14669 | putFReg(sD, |
| 14670 | unop(Iop_ReinterpI32asF32, isT ? getIRegT(rN) : getIRegA(rN)), |
| 14671 | condT); |
| 14672 | putFReg(sD+1, |
| 14673 | unop(Iop_ReinterpI32asF32, isT ? getIRegT(rM) : getIRegA(rM)), |
| 14674 | condT); |
| 14675 | DIP("vmov%s, s%u, s%u, r%u, r%u\n", |
| 14676 | nCC(conq), sD, sD + 1, rN, rM); |
| 14677 | goto decode_success_vfp; |
| 14678 | } |
| 14679 | } |
| 14680 | |
| 14681 | // VMOV rN, rM, sD, sD+1 |
| 14682 | if (0x0C500A10 == (insn28 & 0x0FF00FD0)) { |
| 14683 | UInt sD = (INSN(3,0) << 1) | INSN(5,5); |
| 14684 | UInt rN = INSN(15,12); |
| 14685 | UInt rM = INSN(19,16); |
| 14686 | if (rM == 15 || rN == 15 || (isT && (rM == 13 || rN == 13)) |
| 14687 | || sD == 31 || rN == rM) { |
| 14688 | /* fall through */ |
| 14689 | } else { |
| 14690 | IRExpr* res0 = unop(Iop_ReinterpF32asI32, getFReg(sD)); |
| 14691 | IRExpr* res1 = unop(Iop_ReinterpF32asI32, getFReg(sD+1)); |
| 14692 | if (isT) { |
| 14693 | putIRegT(rN, res0, condT); |
| 14694 | putIRegT(rM, res1, condT); |
| 14695 | } else { |
| 14696 | putIRegA(rN, res0, condT, Ijk_Boring); |
| 14697 | putIRegA(rM, res1, condT, Ijk_Boring); |
| 14698 | } |
| 14699 | DIP("vmov%s, r%u, r%u, s%u, s%u\n", |
| 14700 | nCC(conq), rN, rM, sD, sD + 1); |
| 14701 | goto decode_success_vfp; |
| 14702 | } |
| 14703 | } |
| 14704 | |
| 14705 | // VMOV rD[x], rT (ARM core register to scalar) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14706 | if (0x0E000B10 == (insn28 & 0x0F900F1F)) { |
| 14707 | UInt rD = (INSN(7,7) << 4) | INSN(19,16); |
| 14708 | UInt rT = INSN(15,12); |
| 14709 | UInt opc = (INSN(22,21) << 2) | INSN(6,5); |
| 14710 | UInt index; |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14711 | if (rT == 15 || (isT && rT == 13)) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14712 | /* fall through */ |
| 14713 | } else { |
| 14714 | if ((opc & BITS4(1,0,0,0)) == BITS4(1,0,0,0)) { |
| 14715 | index = opc & 7; |
| 14716 | putDRegI64(rD, triop(Iop_SetElem8x8, |
| 14717 | getDRegI64(rD), |
| 14718 | mkU8(index), |
| 14719 | unop(Iop_32to8, |
| 14720 | isT ? getIRegT(rT) : getIRegA(rT))), |
| 14721 | condT); |
| 14722 | DIP("vmov%s.8 d%u[%u], r%u\n", nCC(conq), rD, index, rT); |
| 14723 | goto decode_success_vfp; |
| 14724 | } |
| 14725 | else if ((opc & BITS4(1,0,0,1)) == BITS4(0,0,0,1)) { |
| 14726 | index = (opc >> 1) & 3; |
| 14727 | putDRegI64(rD, triop(Iop_SetElem16x4, |
| 14728 | getDRegI64(rD), |
| 14729 | mkU8(index), |
| 14730 | unop(Iop_32to16, |
| 14731 | isT ? getIRegT(rT) : getIRegA(rT))), |
| 14732 | condT); |
| 14733 | DIP("vmov%s.16 d%u[%u], r%u\n", nCC(conq), rD, index, rT); |
| 14734 | goto decode_success_vfp; |
| 14735 | } |
| 14736 | else if ((opc & BITS4(1,0,1,1)) == BITS4(0,0,0,0)) { |
| 14737 | index = (opc >> 2) & 1; |
| 14738 | putDRegI64(rD, triop(Iop_SetElem32x2, |
| 14739 | getDRegI64(rD), |
| 14740 | mkU8(index), |
| 14741 | isT ? getIRegT(rT) : getIRegA(rT)), |
| 14742 | condT); |
| 14743 | DIP("vmov%s.32 d%u[%u], r%u\n", nCC(conq), rD, index, rT); |
| 14744 | goto decode_success_vfp; |
| 14745 | } else { |
| 14746 | /* fall through */ |
| 14747 | } |
| 14748 | } |
| 14749 | } |
| 14750 | |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14751 | // VMOV (scalar to ARM core register) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14752 | // VMOV rT, rD[x] |
| 14753 | if (0x0E100B10 == (insn28 & 0x0F100F1F)) { |
| 14754 | UInt rN = (INSN(7,7) << 4) | INSN(19,16); |
| 14755 | UInt rT = INSN(15,12); |
| 14756 | UInt U = INSN(23,23); |
| 14757 | UInt opc = (INSN(22,21) << 2) | INSN(6,5); |
| 14758 | UInt index; |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14759 | if (rT == 15 || (isT && rT == 13)) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14760 | /* fall through */ |
| 14761 | } else { |
| 14762 | if ((opc & BITS4(1,0,0,0)) == BITS4(1,0,0,0)) { |
| 14763 | index = opc & 7; |
| 14764 | IRExpr* e = unop(U ? Iop_8Uto32 : Iop_8Sto32, |
| 14765 | binop(Iop_GetElem8x8, |
| 14766 | getDRegI64(rN), |
| 14767 | mkU8(index))); |
| 14768 | if (isT) |
| 14769 | putIRegT(rT, e, condT); |
| 14770 | else |
| 14771 | putIRegA(rT, e, condT, Ijk_Boring); |
| 14772 | DIP("vmov%s.%c8 r%u, d%u[%u]\n", nCC(conq), U ? 'u' : 's', |
| 14773 | rT, rN, index); |
| 14774 | goto decode_success_vfp; |
| 14775 | } |
| 14776 | else if ((opc & BITS4(1,0,0,1)) == BITS4(0,0,0,1)) { |
| 14777 | index = (opc >> 1) & 3; |
| 14778 | IRExpr* e = unop(U ? Iop_16Uto32 : Iop_16Sto32, |
| 14779 | binop(Iop_GetElem16x4, |
| 14780 | getDRegI64(rN), |
| 14781 | mkU8(index))); |
| 14782 | if (isT) |
| 14783 | putIRegT(rT, e, condT); |
| 14784 | else |
| 14785 | putIRegA(rT, e, condT, Ijk_Boring); |
| 14786 | DIP("vmov%s.%c16 r%u, d%u[%u]\n", nCC(conq), U ? 'u' : 's', |
| 14787 | rT, rN, index); |
| 14788 | goto decode_success_vfp; |
| 14789 | } |
| 14790 | else if ((opc & BITS4(1,0,1,1)) == BITS4(0,0,0,0) && U == 0) { |
| 14791 | index = (opc >> 2) & 1; |
| 14792 | IRExpr* e = binop(Iop_GetElem32x2, getDRegI64(rN), mkU8(index)); |
| 14793 | if (isT) |
| 14794 | putIRegT(rT, e, condT); |
| 14795 | else |
| 14796 | putIRegA(rT, e, condT, Ijk_Boring); |
| 14797 | DIP("vmov%s.32 r%u, d%u[%u]\n", nCC(conq), rT, rN, index); |
| 14798 | goto decode_success_vfp; |
| 14799 | } else { |
| 14800 | /* fall through */ |
| 14801 | } |
| 14802 | } |
| 14803 | } |
| 14804 | |
| 14805 | // VMOV.F32 sD, #imm |
| 14806 | // FCONSTS sD, #imm |
| 14807 | if (BITS8(1,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
| 14808 | && BITS4(0,0,0,0) == INSN(7,4) && INSN(11,8) == BITS4(1,0,1,0)) { |
| 14809 | UInt rD = (INSN(15,12) << 1) | INSN(22,22); |
| 14810 | UInt imm8 = (INSN(19,16) << 4) | INSN(3,0); |
| 14811 | UInt b = (imm8 >> 6) & 1; |
| 14812 | UInt imm; |
| 14813 | imm = (BITS8((imm8 >> 7) & 1,(~b) & 1,b,b,b,b,b,(imm8 >> 5) & 1) << 8) |
| 14814 | | ((imm8 & 0x1f) << 3); |
| 14815 | imm <<= 16; |
| 14816 | putFReg(rD, unop(Iop_ReinterpI32asF32, mkU32(imm)), condT); |
| 14817 | DIP("fconsts%s s%u #%u", nCC(conq), rD, imm8); |
| 14818 | goto decode_success_vfp; |
| 14819 | } |
| 14820 | |
| 14821 | // VMOV.F64 dD, #imm |
| 14822 | // FCONSTD dD, #imm |
| 14823 | if (BITS8(1,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
| 14824 | && BITS4(0,0,0,0) == INSN(7,4) && INSN(11,8) == BITS4(1,0,1,1)) { |
| 14825 | UInt rD = INSN(15,12) | (INSN(22,22) << 4); |
| 14826 | UInt imm8 = (INSN(19,16) << 4) | INSN(3,0); |
| 14827 | UInt b = (imm8 >> 6) & 1; |
| 14828 | ULong imm; |
| 14829 | imm = (BITS8((imm8 >> 7) & 1,(~b) & 1,b,b,b,b,b,b) << 8) |
| 14830 | | BITS8(b,b,0,0,0,0,0,0) | (imm8 & 0x3f); |
| 14831 | imm <<= 48; |
| 14832 | putDReg(rD, unop(Iop_ReinterpI64asF64, mkU64(imm)), condT); |
| 14833 | DIP("fconstd%s d%u #%u", nCC(conq), rD, imm8); |
| 14834 | goto decode_success_vfp; |
| 14835 | } |
| 14836 | |
| 14837 | /* ---------------------- vdup ------------------------- */ |
| 14838 | // VDUP dD, rT |
| 14839 | // VDUP qD, rT |
| 14840 | if (BITS8(1,1,1,0,1,0,0,0) == (INSN(27,20) & BITS8(1,1,1,1,1,0,0,1)) |
| 14841 | && BITS4(1,0,1,1) == INSN(11,8) && INSN(6,6) == 0 && INSN(4,4) == 1) { |
| 14842 | UInt rD = (INSN(7,7) << 4) | INSN(19,16); |
| 14843 | UInt rT = INSN(15,12); |
| 14844 | UInt Q = INSN(21,21); |
| 14845 | UInt size = (INSN(22,22) << 1) | INSN(5,5); |
sewardj | 4d47547 | 2011-02-02 13:35:01 +0000 | [diff] [blame] | 14846 | if (rT == 15 || (isT && rT == 13) || size == 3 || (Q && (rD & 1))) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14847 | /* fall through */ |
| 14848 | } else { |
| 14849 | IRExpr* e = isT ? getIRegT(rT) : getIRegA(rT); |
| 14850 | if (Q) { |
| 14851 | rD >>= 1; |
| 14852 | switch (size) { |
| 14853 | case 0: |
| 14854 | putQReg(rD, unop(Iop_Dup32x4, e), condT); |
| 14855 | break; |
| 14856 | case 1: |
| 14857 | putQReg(rD, unop(Iop_Dup16x8, unop(Iop_32to16, e)), |
| 14858 | condT); |
| 14859 | break; |
| 14860 | case 2: |
| 14861 | putQReg(rD, unop(Iop_Dup8x16, unop(Iop_32to8, e)), |
| 14862 | condT); |
| 14863 | break; |
| 14864 | default: |
| 14865 | vassert(0); |
| 14866 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 14867 | DIP("vdup.%d q%u, r%u\n", 32 / (1<<size), rD, rT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14868 | } else { |
| 14869 | switch (size) { |
| 14870 | case 0: |
| 14871 | putDRegI64(rD, unop(Iop_Dup32x2, e), condT); |
| 14872 | break; |
| 14873 | case 1: |
| 14874 | putDRegI64(rD, unop(Iop_Dup16x4, unop(Iop_32to16, e)), |
| 14875 | condT); |
| 14876 | break; |
| 14877 | case 2: |
| 14878 | putDRegI64(rD, unop(Iop_Dup8x8, unop(Iop_32to8, e)), |
| 14879 | condT); |
| 14880 | break; |
| 14881 | default: |
| 14882 | vassert(0); |
| 14883 | } |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 14884 | DIP("vdup.%d d%u, r%u\n", 32 / (1<<size), rD, rT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14885 | } |
| 14886 | goto decode_success_vfp; |
| 14887 | } |
| 14888 | } |
| 14889 | |
| 14890 | /* --------------------- f{ld,st}d --------------------- */ |
| 14891 | // FLDD, FSTD |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14892 | if (BITS8(1,1,0,1,0,0,0,0) == (INSN(27,20) & BITS8(1,1,1,1,0,0,1,0)) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14893 | && BITS4(1,0,1,1) == INSN(11,8)) { |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14894 | UInt dD = INSN(15,12) | (INSN(22,22) << 4); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14895 | UInt rN = INSN(19,16); |
| 14896 | UInt offset = (insn28 & 0xFF) << 2; |
| 14897 | UInt bU = (insn28 >> 23) & 1; /* 1: +offset 0: -offset */ |
| 14898 | UInt bL = (insn28 >> 20) & 1; /* 1: load 0: store */ |
| 14899 | /* make unconditional */ |
| 14900 | if (condT != IRTemp_INVALID) { |
| 14901 | if (isT) |
| 14902 | mk_skip_over_T32_if_cond_is_false( condT ); |
| 14903 | else |
| 14904 | mk_skip_over_A32_if_cond_is_false( condT ); |
| 14905 | condT = IRTemp_INVALID; |
| 14906 | } |
| 14907 | IRTemp ea = newTemp(Ity_I32); |
| 14908 | assign(ea, binop(bU ? Iop_Add32 : Iop_Sub32, |
| 14909 | align4if(isT ? getIRegT(rN) : getIRegA(rN), |
| 14910 | rN == 15), |
| 14911 | mkU32(offset))); |
| 14912 | if (bL) { |
| 14913 | putDReg(dD, loadLE(Ity_F64,mkexpr(ea)), IRTemp_INVALID); |
| 14914 | } else { |
| 14915 | storeLE(mkexpr(ea), getDReg(dD)); |
| 14916 | } |
| 14917 | DIP("f%sd%s d%u, [r%u, %c#%u]\n", |
| 14918 | bL ? "ld" : "st", nCC(conq), dD, rN, |
| 14919 | bU ? '+' : '-', offset); |
| 14920 | goto decode_success_vfp; |
| 14921 | } |
| 14922 | |
| 14923 | /* --------------------- dp insns (D) --------------------- */ |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14924 | if (BITS8(1,1,1,0,0,0,0,0) == (INSN(27,20) & BITS8(1,1,1,1,0,0,0,0)) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14925 | && BITS4(1,0,1,1) == INSN(11,8) |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14926 | && BITS4(0,0,0,0) == (INSN(7,4) & BITS4(0,0,0,1))) { |
| 14927 | UInt dM = INSN(3,0) | (INSN(5,5) << 4); /* argR */ |
| 14928 | UInt dD = INSN(15,12) | (INSN(22,22) << 4); /* dst/acc */ |
| 14929 | UInt dN = INSN(19,16) | (INSN(7,7) << 4); /* argL */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14930 | UInt bP = (insn28 >> 23) & 1; |
| 14931 | UInt bQ = (insn28 >> 21) & 1; |
| 14932 | UInt bR = (insn28 >> 20) & 1; |
| 14933 | UInt bS = (insn28 >> 6) & 1; |
| 14934 | UInt opc = (bP << 3) | (bQ << 2) | (bR << 1) | bS; |
| 14935 | IRExpr* rm = get_FAKE_roundingmode(); /* XXXROUNDINGFIXME */ |
| 14936 | switch (opc) { |
| 14937 | case BITS4(0,0,0,0): /* MAC: d + n * m */ |
| 14938 | putDReg(dD, triop(Iop_AddF64, rm, |
| 14939 | getDReg(dD), |
| 14940 | triop(Iop_MulF64, rm, getDReg(dN), |
| 14941 | getDReg(dM))), |
| 14942 | condT); |
| 14943 | DIP("fmacd%s d%u, d%u, d%u\n", nCC(conq), dD, dN, dM); |
| 14944 | goto decode_success_vfp; |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14945 | case BITS4(0,0,0,1): /* NMAC: d + -(n * m) */ |
| 14946 | putDReg(dD, triop(Iop_AddF64, rm, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14947 | getDReg(dD), |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14948 | unop(Iop_NegF64, |
| 14949 | triop(Iop_MulF64, rm, getDReg(dN), |
| 14950 | getDReg(dM)))), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14951 | condT); |
| 14952 | DIP("fnmacd%s d%u, d%u, d%u\n", nCC(conq), dD, dN, dM); |
| 14953 | goto decode_success_vfp; |
| 14954 | case BITS4(0,0,1,0): /* MSC: - d + n * m */ |
| 14955 | putDReg(dD, triop(Iop_AddF64, rm, |
| 14956 | unop(Iop_NegF64, getDReg(dD)), |
| 14957 | triop(Iop_MulF64, rm, getDReg(dN), |
| 14958 | getDReg(dM))), |
| 14959 | condT); |
| 14960 | DIP("fmscd%s d%u, d%u, d%u\n", nCC(conq), dD, dN, dM); |
| 14961 | goto decode_success_vfp; |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14962 | case BITS4(0,0,1,1): /* NMSC: - d + -(n * m) */ |
| 14963 | putDReg(dD, triop(Iop_AddF64, rm, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14964 | unop(Iop_NegF64, getDReg(dD)), |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 14965 | unop(Iop_NegF64, |
| 14966 | triop(Iop_MulF64, rm, getDReg(dN), |
| 14967 | getDReg(dM)))), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 14968 | condT); |
| 14969 | DIP("fnmscd%s d%u, d%u, d%u\n", nCC(conq), dD, dN, dM); |
| 14970 | goto decode_success_vfp; |
| 14971 | case BITS4(0,1,0,0): /* MUL: n * m */ |
| 14972 | putDReg(dD, triop(Iop_MulF64, rm, getDReg(dN), getDReg(dM)), |
| 14973 | condT); |
| 14974 | DIP("fmuld%s d%u, d%u, d%u\n", nCC(conq), dD, dN, dM); |
| 14975 | goto decode_success_vfp; |
| 14976 | case BITS4(0,1,0,1): /* NMUL: - n * m */ |
| 14977 | putDReg(dD, unop(Iop_NegF64, |
| 14978 | triop(Iop_MulF64, rm, getDReg(dN), |
| 14979 | getDReg(dM))), |
| 14980 | condT); |
| 14981 | DIP("fnmuld%s d%u, d%u, d%u\n", nCC(conq), dD, dN, dM); |
| 14982 | goto decode_success_vfp; |
| 14983 | case BITS4(0,1,1,0): /* ADD: n + m */ |
| 14984 | putDReg(dD, triop(Iop_AddF64, rm, getDReg(dN), getDReg(dM)), |
| 14985 | condT); |
| 14986 | DIP("faddd%s d%u, d%u, d%u\n", nCC(conq), dD, dN, dM); |
| 14987 | goto decode_success_vfp; |
| 14988 | case BITS4(0,1,1,1): /* SUB: n - m */ |
| 14989 | putDReg(dD, triop(Iop_SubF64, rm, getDReg(dN), getDReg(dM)), |
| 14990 | condT); |
| 14991 | DIP("fsubd%s d%u, d%u, d%u\n", nCC(conq), dD, dN, dM); |
| 14992 | goto decode_success_vfp; |
| 14993 | case BITS4(1,0,0,0): /* DIV: n / m */ |
| 14994 | putDReg(dD, triop(Iop_DivF64, rm, getDReg(dN), getDReg(dM)), |
| 14995 | condT); |
| 14996 | DIP("fdivd%s d%u, d%u, d%u\n", nCC(conq), dD, dN, dM); |
| 14997 | goto decode_success_vfp; |
sewardj | d6d13b3 | 2014-11-19 16:07:28 +0000 | [diff] [blame] | 14998 | case BITS4(1,0,1,0): /* VNFMS: -(d - n * m) (fused) */ |
| 14999 | /* XXXROUNDINGFIXME look up ARM reference for fused |
| 15000 | multiply-add rounding */ |
| 15001 | putDReg(dD, triop(Iop_AddF64, rm, |
| 15002 | unop(Iop_NegF64, getDReg(dD)), |
| 15003 | triop(Iop_MulF64, rm, |
| 15004 | getDReg(dN), |
| 15005 | getDReg(dM))), |
| 15006 | condT); |
| 15007 | DIP("vfnmsd%s d%u, d%u, d%u\n", nCC(conq), dD, dN, dM); |
| 15008 | goto decode_success_vfp; |
| 15009 | case BITS4(1,0,1,1): /* VNFMA: -(d + n * m) (fused) */ |
| 15010 | /* XXXROUNDINGFIXME look up ARM reference for fused |
| 15011 | multiply-add rounding */ |
| 15012 | putDReg(dD, triop(Iop_AddF64, rm, |
| 15013 | unop(Iop_NegF64, getDReg(dD)), |
| 15014 | triop(Iop_MulF64, rm, |
| 15015 | unop(Iop_NegF64, getDReg(dN)), |
| 15016 | getDReg(dM))), |
| 15017 | condT); |
| 15018 | DIP("vfnmad%s d%u, d%u, d%u\n", nCC(conq), dD, dN, dM); |
| 15019 | goto decode_success_vfp; |
sewardj | 7bfbbe9 | 2014-05-14 23:38:23 +0000 | [diff] [blame] | 15020 | case BITS4(1,1,0,0): /* VFMA: d + n * m (fused) */ |
| 15021 | /* XXXROUNDINGFIXME look up ARM reference for fused |
| 15022 | multiply-add rounding */ |
| 15023 | putDReg(dD, triop(Iop_AddF64, rm, |
| 15024 | getDReg(dD), |
| 15025 | triop(Iop_MulF64, rm, getDReg(dN), |
| 15026 | getDReg(dM))), |
| 15027 | condT); |
| 15028 | DIP("vfmad%s d%u, d%u, d%u\n", nCC(conq), dD, dN, dM); |
| 15029 | goto decode_success_vfp; |
| 15030 | case BITS4(1,1,0,1): /* VFMS: d + (-n * m) (fused) */ |
| 15031 | /* XXXROUNDINGFIXME look up ARM reference for fused |
| 15032 | multiply-add rounding */ |
| 15033 | putDReg(dD, triop(Iop_AddF64, rm, |
| 15034 | getDReg(dD), |
| 15035 | triop(Iop_MulF64, rm, |
| 15036 | unop(Iop_NegF64, getDReg(dN)), |
| 15037 | getDReg(dM))), |
| 15038 | condT); |
| 15039 | DIP("vfmsd%s d%u, d%u, d%u\n", nCC(conq), dD, dN, dM); |
| 15040 | goto decode_success_vfp; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15041 | default: |
| 15042 | break; |
| 15043 | } |
| 15044 | } |
| 15045 | |
| 15046 | /* --------------------- compares (D) --------------------- */ |
| 15047 | /* 31 27 23 19 15 11 7 3 |
| 15048 | 28 24 20 16 12 8 4 0 |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15049 | FCMPD cond 1110 1D11 0100 Dd 1011 0100 Dm |
| 15050 | FCMPED cond 1110 1D11 0100 Dd 1011 1100 Dm |
| 15051 | FCMPZD cond 1110 1D11 0101 Dd 1011 0100 0000 |
| 15052 | FCMPZED cond 1110 1D11 0101 Dd 1011 1100 0000 |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15053 | Z N |
| 15054 | |
| 15055 | Z=0 Compare Dd vs Dm and set FPSCR 31:28 accordingly |
| 15056 | Z=1 Compare Dd vs zero |
| 15057 | |
| 15058 | N=1 generates Invalid Operation exn if either arg is any kind of NaN |
| 15059 | N=0 generates Invalid Operation exn if either arg is a signalling NaN |
| 15060 | (Not that we pay any attention to N here) |
| 15061 | */ |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15062 | if (BITS8(1,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15063 | && BITS4(0,1,0,0) == (INSN(19,16) & BITS4(1,1,1,0)) |
| 15064 | && BITS4(1,0,1,1) == INSN(11,8) |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15065 | && BITS4(0,1,0,0) == (INSN(7,4) & BITS4(0,1,0,1))) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15066 | UInt bZ = (insn28 >> 16) & 1; |
| 15067 | UInt bN = (insn28 >> 7) & 1; |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15068 | UInt dD = INSN(15,12) | (INSN(22,22) << 4); |
| 15069 | UInt dM = INSN(3,0) | (INSN(5,5) << 4); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15070 | if (bZ && INSN(3,0) != 0) { |
| 15071 | /* does not decode; fall through */ |
| 15072 | } else { |
| 15073 | IRTemp argL = newTemp(Ity_F64); |
| 15074 | IRTemp argR = newTemp(Ity_F64); |
| 15075 | IRTemp irRes = newTemp(Ity_I32); |
| 15076 | assign(argL, getDReg(dD)); |
| 15077 | assign(argR, bZ ? IRExpr_Const(IRConst_F64i(0)) : getDReg(dM)); |
| 15078 | assign(irRes, binop(Iop_CmpF64, mkexpr(argL), mkexpr(argR))); |
| 15079 | |
| 15080 | IRTemp nzcv = IRTemp_INVALID; |
| 15081 | IRTemp oldFPSCR = newTemp(Ity_I32); |
| 15082 | IRTemp newFPSCR = newTemp(Ity_I32); |
| 15083 | |
| 15084 | /* This is where the fun starts. We have to convert 'irRes' |
| 15085 | from an IR-convention return result (IRCmpF64Result) to an |
| 15086 | ARM-encoded (N,Z,C,V) group. The final result is in the |
| 15087 | bottom 4 bits of 'nzcv'. */ |
| 15088 | /* Map compare result from IR to ARM(nzcv) */ |
| 15089 | /* |
| 15090 | FP cmp result | IR | ARM(nzcv) |
| 15091 | -------------------------------- |
| 15092 | UN 0x45 0011 |
| 15093 | LT 0x01 1000 |
| 15094 | GT 0x00 0010 |
| 15095 | EQ 0x40 0110 |
| 15096 | */ |
| 15097 | nzcv = mk_convert_IRCmpF64Result_to_NZCV(irRes); |
| 15098 | |
| 15099 | /* And update FPSCR accordingly */ |
| 15100 | assign(oldFPSCR, IRExpr_Get(OFFB_FPSCR, Ity_I32)); |
| 15101 | assign(newFPSCR, |
| 15102 | binop(Iop_Or32, |
| 15103 | binop(Iop_And32, mkexpr(oldFPSCR), mkU32(0x0FFFFFFF)), |
| 15104 | binop(Iop_Shl32, mkexpr(nzcv), mkU8(28)))); |
| 15105 | |
| 15106 | putMiscReg32(OFFB_FPSCR, mkexpr(newFPSCR), condT); |
| 15107 | |
| 15108 | if (bZ) { |
| 15109 | DIP("fcmpz%sd%s d%u\n", bN ? "e" : "", nCC(conq), dD); |
| 15110 | } else { |
| 15111 | DIP("fcmp%sd%s d%u, d%u\n", bN ? "e" : "", nCC(conq), dD, dM); |
| 15112 | } |
| 15113 | goto decode_success_vfp; |
| 15114 | } |
| 15115 | /* fall through */ |
| 15116 | } |
| 15117 | |
| 15118 | /* --------------------- unary (D) --------------------- */ |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15119 | if (BITS8(1,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15120 | && BITS4(0,0,0,0) == (INSN(19,16) & BITS4(1,1,1,0)) |
| 15121 | && BITS4(1,0,1,1) == INSN(11,8) |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15122 | && BITS4(0,1,0,0) == (INSN(7,4) & BITS4(0,1,0,1))) { |
| 15123 | UInt dD = INSN(15,12) | (INSN(22,22) << 4); |
| 15124 | UInt dM = INSN(3,0) | (INSN(5,5) << 4); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15125 | UInt b16 = (insn28 >> 16) & 1; |
| 15126 | UInt b7 = (insn28 >> 7) & 1; |
| 15127 | /**/ if (b16 == 0 && b7 == 0) { |
| 15128 | // FCPYD |
| 15129 | putDReg(dD, getDReg(dM), condT); |
| 15130 | DIP("fcpyd%s d%u, d%u\n", nCC(conq), dD, dM); |
| 15131 | goto decode_success_vfp; |
| 15132 | } |
| 15133 | else if (b16 == 0 && b7 == 1) { |
| 15134 | // FABSD |
| 15135 | putDReg(dD, unop(Iop_AbsF64, getDReg(dM)), condT); |
| 15136 | DIP("fabsd%s d%u, d%u\n", nCC(conq), dD, dM); |
| 15137 | goto decode_success_vfp; |
| 15138 | } |
| 15139 | else if (b16 == 1 && b7 == 0) { |
| 15140 | // FNEGD |
| 15141 | putDReg(dD, unop(Iop_NegF64, getDReg(dM)), condT); |
| 15142 | DIP("fnegd%s d%u, d%u\n", nCC(conq), dD, dM); |
| 15143 | goto decode_success_vfp; |
| 15144 | } |
| 15145 | else if (b16 == 1 && b7 == 1) { |
| 15146 | // FSQRTD |
| 15147 | IRExpr* rm = get_FAKE_roundingmode(); /* XXXROUNDINGFIXME */ |
| 15148 | putDReg(dD, binop(Iop_SqrtF64, rm, getDReg(dM)), condT); |
| 15149 | DIP("fsqrtd%s d%u, d%u\n", nCC(conq), dD, dM); |
| 15150 | goto decode_success_vfp; |
| 15151 | } |
| 15152 | else |
| 15153 | vassert(0); |
| 15154 | |
| 15155 | /* fall through */ |
| 15156 | } |
| 15157 | |
| 15158 | /* ----------------- I <-> D conversions ----------------- */ |
| 15159 | |
| 15160 | // F{S,U}ITOD dD, fM |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15161 | if (BITS8(1,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15162 | && BITS4(1,0,0,0) == (INSN(19,16) & BITS4(1,1,1,1)) |
| 15163 | && BITS4(1,0,1,1) == INSN(11,8) |
| 15164 | && BITS4(0,1,0,0) == (INSN(7,4) & BITS4(0,1,0,1))) { |
| 15165 | UInt bM = (insn28 >> 5) & 1; |
| 15166 | UInt fM = (INSN(3,0) << 1) | bM; |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15167 | UInt dD = INSN(15,12) | (INSN(22,22) << 4); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15168 | UInt syned = (insn28 >> 7) & 1; |
| 15169 | if (syned) { |
| 15170 | // FSITOD |
| 15171 | putDReg(dD, unop(Iop_I32StoF64, |
| 15172 | unop(Iop_ReinterpF32asI32, getFReg(fM))), |
| 15173 | condT); |
| 15174 | DIP("fsitod%s d%u, s%u\n", nCC(conq), dD, fM); |
| 15175 | } else { |
| 15176 | // FUITOD |
| 15177 | putDReg(dD, unop(Iop_I32UtoF64, |
| 15178 | unop(Iop_ReinterpF32asI32, getFReg(fM))), |
| 15179 | condT); |
| 15180 | DIP("fuitod%s d%u, s%u\n", nCC(conq), dD, fM); |
| 15181 | } |
| 15182 | goto decode_success_vfp; |
| 15183 | } |
| 15184 | |
| 15185 | // FTO{S,U}ID fD, dM |
| 15186 | if (BITS8(1,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
| 15187 | && BITS4(1,1,0,0) == (INSN(19,16) & BITS4(1,1,1,0)) |
| 15188 | && BITS4(1,0,1,1) == INSN(11,8) |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15189 | && BITS4(0,1,0,0) == (INSN(7,4) & BITS4(0,1,0,1))) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15190 | UInt bD = (insn28 >> 22) & 1; |
| 15191 | UInt fD = (INSN(15,12) << 1) | bD; |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15192 | UInt dM = INSN(3,0) | (INSN(5,5) << 4); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15193 | UInt bZ = (insn28 >> 7) & 1; |
| 15194 | UInt syned = (insn28 >> 16) & 1; |
| 15195 | IRTemp rmode = newTemp(Ity_I32); |
| 15196 | assign(rmode, bZ ? mkU32(Irrm_ZERO) |
| 15197 | : mkexpr(mk_get_IR_rounding_mode())); |
| 15198 | if (syned) { |
| 15199 | // FTOSID |
| 15200 | putFReg(fD, unop(Iop_ReinterpI32asF32, |
| 15201 | binop(Iop_F64toI32S, mkexpr(rmode), |
| 15202 | getDReg(dM))), |
| 15203 | condT); |
| 15204 | DIP("ftosi%sd%s s%u, d%u\n", bZ ? "z" : "", |
| 15205 | nCC(conq), fD, dM); |
| 15206 | } else { |
| 15207 | // FTOUID |
| 15208 | putFReg(fD, unop(Iop_ReinterpI32asF32, |
| 15209 | binop(Iop_F64toI32U, mkexpr(rmode), |
| 15210 | getDReg(dM))), |
| 15211 | condT); |
| 15212 | DIP("ftoui%sd%s s%u, d%u\n", bZ ? "z" : "", |
| 15213 | nCC(conq), fD, dM); |
| 15214 | } |
| 15215 | goto decode_success_vfp; |
| 15216 | } |
| 15217 | |
| 15218 | /* ----------------------------------------------------------- */ |
| 15219 | /* -- VFP instructions -- single precision -- */ |
| 15220 | /* ----------------------------------------------------------- */ |
| 15221 | |
| 15222 | /* --------------------- fldms, fstms --------------------- */ |
| 15223 | /* |
| 15224 | 31 27 23 19 15 11 7 0 |
| 15225 | P UDWL |
| 15226 | C4-98, C5-26 1 FSTMD cond 1100 1x00 Rn Fd 1010 offset |
| 15227 | C4-98, C5-28 2 FSTMDIA cond 1100 1x10 Rn Fd 1010 offset |
| 15228 | C4-98, C5-30 3 FSTMDDB cond 1101 0x10 Rn Fd 1010 offset |
| 15229 | |
| 15230 | C4-40, C5-26 1 FLDMD cond 1100 1x01 Rn Fd 1010 offset |
| 15231 | C4-40, C5-26 2 FLDMIAD cond 1100 1x11 Rn Fd 1010 offset |
| 15232 | C4-40, C5-26 3 FLDMDBD cond 1101 0x11 Rn Fd 1010 offset |
| 15233 | |
| 15234 | Regs transferred: F(Fd:D) .. F(Fd:d + offset) |
| 15235 | offset must not imply a reg > 15 |
| 15236 | IA/DB: Rn is changed by (4 x # regs transferred) |
| 15237 | |
| 15238 | case coding: |
| 15239 | 1 at-Rn (access at Rn) |
| 15240 | 2 ia-Rn (access at Rn, then Rn += 4n) |
| 15241 | 3 db-Rn (Rn -= 4n, then access at Rn) |
| 15242 | */ |
| 15243 | if (BITS8(1,1,0,0,0,0,0,0) == (INSN(27,20) & BITS8(1,1,1,0,0,0,0,0)) |
| 15244 | && INSN(11,8) == BITS4(1,0,1,0)) { |
| 15245 | UInt bP = (insn28 >> 24) & 1; |
| 15246 | UInt bU = (insn28 >> 23) & 1; |
| 15247 | UInt bW = (insn28 >> 21) & 1; |
| 15248 | UInt bL = (insn28 >> 20) & 1; |
| 15249 | UInt bD = (insn28 >> 22) & 1; |
| 15250 | UInt offset = (insn28 >> 0) & 0xFF; |
| 15251 | UInt rN = INSN(19,16); |
| 15252 | UInt fD = (INSN(15,12) << 1) | bD; |
| 15253 | UInt nRegs = offset; |
| 15254 | UInt summary = 0; |
| 15255 | Int i; |
| 15256 | |
| 15257 | /**/ if (bP == 0 && bU == 1 && bW == 0) { |
| 15258 | summary = 1; |
| 15259 | } |
| 15260 | else if (bP == 0 && bU == 1 && bW == 1) { |
| 15261 | summary = 2; |
| 15262 | } |
| 15263 | else if (bP == 1 && bU == 0 && bW == 1) { |
| 15264 | summary = 3; |
| 15265 | } |
| 15266 | else goto after_vfp_fldms_fstms; |
| 15267 | |
| 15268 | /* no writebacks to r15 allowed. No use of r15 in thumb mode. */ |
| 15269 | if (rN == 15 && (summary == 2 || summary == 3 || isT)) |
| 15270 | goto after_vfp_fldms_fstms; |
| 15271 | |
| 15272 | /* offset must specify at least one register */ |
| 15273 | if (offset < 1) |
| 15274 | goto after_vfp_fldms_fstms; |
| 15275 | |
| 15276 | /* can't transfer regs after S31 */ |
| 15277 | if (fD + nRegs - 1 >= 32) |
| 15278 | goto after_vfp_fldms_fstms; |
| 15279 | |
| 15280 | /* Now, we can't do a conditional load or store, since that very |
| 15281 | likely will generate an exception. So we have to take a side |
| 15282 | exit at this point if the condition is false. */ |
| 15283 | if (condT != IRTemp_INVALID) { |
| 15284 | if (isT) |
| 15285 | mk_skip_over_T32_if_cond_is_false( condT ); |
| 15286 | else |
| 15287 | mk_skip_over_A32_if_cond_is_false( condT ); |
| 15288 | condT = IRTemp_INVALID; |
| 15289 | } |
| 15290 | /* Ok, now we're unconditional. Do the load or store. */ |
| 15291 | |
| 15292 | /* get the old Rn value */ |
| 15293 | IRTemp rnT = newTemp(Ity_I32); |
| 15294 | assign(rnT, align4if(isT ? getIRegT(rN) : getIRegA(rN), |
| 15295 | rN == 15)); |
| 15296 | |
| 15297 | /* make a new value for Rn, post-insn */ |
| 15298 | IRTemp rnTnew = IRTemp_INVALID; |
| 15299 | if (summary == 2 || summary == 3) { |
| 15300 | rnTnew = newTemp(Ity_I32); |
| 15301 | assign(rnTnew, binop(summary == 2 ? Iop_Add32 : Iop_Sub32, |
| 15302 | mkexpr(rnT), |
| 15303 | mkU32(4 * nRegs))); |
| 15304 | } |
| 15305 | |
| 15306 | /* decide on the base transfer address */ |
| 15307 | IRTemp taT = newTemp(Ity_I32); |
| 15308 | assign(taT, summary == 3 ? mkexpr(rnTnew) : mkexpr(rnT)); |
| 15309 | |
| 15310 | /* update Rn if necessary -- in case 3, we're moving it down, so |
| 15311 | update before any memory reference, in order to keep Memcheck |
| 15312 | and V's stack-extending logic (on linux) happy */ |
| 15313 | if (summary == 3) { |
| 15314 | if (isT) |
| 15315 | putIRegT(rN, mkexpr(rnTnew), IRTemp_INVALID); |
| 15316 | else |
| 15317 | putIRegA(rN, mkexpr(rnTnew), IRTemp_INVALID, Ijk_Boring); |
| 15318 | } |
| 15319 | |
| 15320 | /* generate the transfers */ |
| 15321 | for (i = 0; i < nRegs; i++) { |
| 15322 | IRExpr* addr = binop(Iop_Add32, mkexpr(taT), mkU32(4*i)); |
| 15323 | if (bL) { |
| 15324 | putFReg(fD + i, loadLE(Ity_F32, addr), IRTemp_INVALID); |
| 15325 | } else { |
| 15326 | storeLE(addr, getFReg(fD + i)); |
| 15327 | } |
| 15328 | } |
| 15329 | |
| 15330 | /* update Rn if necessary -- in case 2, we're moving it up, so |
| 15331 | update after any memory reference, in order to keep Memcheck |
| 15332 | and V's stack-extending logic (on linux) happy */ |
| 15333 | if (summary == 2) { |
| 15334 | if (isT) |
| 15335 | putIRegT(rN, mkexpr(rnTnew), IRTemp_INVALID); |
| 15336 | else |
| 15337 | putIRegA(rN, mkexpr(rnTnew), IRTemp_INVALID, Ijk_Boring); |
| 15338 | } |
| 15339 | |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 15340 | const HChar* nm = bL==1 ? "ld" : "st"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15341 | switch (summary) { |
| 15342 | case 1: DIP("f%sms%s r%u, {s%u-s%u}\n", |
| 15343 | nm, nCC(conq), rN, fD, fD + nRegs - 1); |
| 15344 | break; |
| 15345 | case 2: DIP("f%smias%s r%u!, {s%u-s%u}\n", |
| 15346 | nm, nCC(conq), rN, fD, fD + nRegs - 1); |
| 15347 | break; |
| 15348 | case 3: DIP("f%smdbs%s r%u!, {s%u-s%u}\n", |
| 15349 | nm, nCC(conq), rN, fD, fD + nRegs - 1); |
| 15350 | break; |
| 15351 | default: vassert(0); |
| 15352 | } |
| 15353 | |
| 15354 | goto decode_success_vfp; |
| 15355 | /* FIXME alignment constraints? */ |
| 15356 | } |
| 15357 | |
| 15358 | after_vfp_fldms_fstms: |
| 15359 | |
| 15360 | /* --------------------- fmsr, fmrs --------------------- */ |
| 15361 | if (BITS8(1,1,1,0,0,0,0,0) == (INSN(27,20) & BITS8(1,1,1,1,1,1,1,0)) |
| 15362 | && BITS4(1,0,1,0) == INSN(11,8) |
| 15363 | && BITS4(0,0,0,0) == INSN(3,0) |
| 15364 | && BITS4(0,0,0,1) == (INSN(7,4) & BITS4(0,1,1,1))) { |
| 15365 | UInt rD = INSN(15,12); |
| 15366 | UInt b7 = (insn28 >> 7) & 1; |
| 15367 | UInt fN = (INSN(19,16) << 1) | b7; |
| 15368 | UInt b20 = (insn28 >> 20) & 1; |
| 15369 | if (rD == 15) { |
| 15370 | /* fall through */ |
| 15371 | /* Let's assume that no sane person would want to do |
| 15372 | floating-point transfers to or from the program counter, |
| 15373 | and simply decline to decode the instruction. The ARM ARM |
| 15374 | doesn't seem to explicitly disallow this case, though. */ |
| 15375 | } else { |
| 15376 | if (b20) { |
| 15377 | IRExpr* res = unop(Iop_ReinterpF32asI32, getFReg(fN)); |
| 15378 | if (isT) |
| 15379 | putIRegT(rD, res, condT); |
| 15380 | else |
| 15381 | putIRegA(rD, res, condT, Ijk_Boring); |
| 15382 | DIP("fmrs%s r%u, s%u\n", nCC(conq), rD, fN); |
| 15383 | } else { |
| 15384 | putFReg(fN, unop(Iop_ReinterpI32asF32, |
| 15385 | isT ? getIRegT(rD) : getIRegA(rD)), |
| 15386 | condT); |
| 15387 | DIP("fmsr%s s%u, r%u\n", nCC(conq), fN, rD); |
| 15388 | } |
| 15389 | goto decode_success_vfp; |
| 15390 | } |
| 15391 | /* fall through */ |
| 15392 | } |
| 15393 | |
| 15394 | /* --------------------- f{ld,st}s --------------------- */ |
| 15395 | // FLDS, FSTS |
| 15396 | if (BITS8(1,1,0,1,0,0,0,0) == (INSN(27,20) & BITS8(1,1,1,1,0,0,1,0)) |
| 15397 | && BITS4(1,0,1,0) == INSN(11,8)) { |
| 15398 | UInt bD = (insn28 >> 22) & 1; |
| 15399 | UInt fD = (INSN(15,12) << 1) | bD; |
| 15400 | UInt rN = INSN(19,16); |
| 15401 | UInt offset = (insn28 & 0xFF) << 2; |
| 15402 | UInt bU = (insn28 >> 23) & 1; /* 1: +offset 0: -offset */ |
| 15403 | UInt bL = (insn28 >> 20) & 1; /* 1: load 0: store */ |
| 15404 | /* make unconditional */ |
| 15405 | if (condT != IRTemp_INVALID) { |
| 15406 | if (isT) |
| 15407 | mk_skip_over_T32_if_cond_is_false( condT ); |
| 15408 | else |
| 15409 | mk_skip_over_A32_if_cond_is_false( condT ); |
| 15410 | condT = IRTemp_INVALID; |
| 15411 | } |
| 15412 | IRTemp ea = newTemp(Ity_I32); |
| 15413 | assign(ea, binop(bU ? Iop_Add32 : Iop_Sub32, |
| 15414 | align4if(isT ? getIRegT(rN) : getIRegA(rN), |
| 15415 | rN == 15), |
| 15416 | mkU32(offset))); |
| 15417 | if (bL) { |
| 15418 | putFReg(fD, loadLE(Ity_F32,mkexpr(ea)), IRTemp_INVALID); |
| 15419 | } else { |
| 15420 | storeLE(mkexpr(ea), getFReg(fD)); |
| 15421 | } |
| 15422 | DIP("f%ss%s s%u, [r%u, %c#%u]\n", |
| 15423 | bL ? "ld" : "st", nCC(conq), fD, rN, |
| 15424 | bU ? '+' : '-', offset); |
| 15425 | goto decode_success_vfp; |
| 15426 | } |
| 15427 | |
| 15428 | /* --------------------- dp insns (F) --------------------- */ |
| 15429 | if (BITS8(1,1,1,0,0,0,0,0) == (INSN(27,20) & BITS8(1,1,1,1,0,0,0,0)) |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15430 | && BITS4(1,0,1,0) == (INSN(11,8) & BITS4(1,1,1,0)) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15431 | && BITS4(0,0,0,0) == (INSN(7,4) & BITS4(0,0,0,1))) { |
| 15432 | UInt bM = (insn28 >> 5) & 1; |
| 15433 | UInt bD = (insn28 >> 22) & 1; |
| 15434 | UInt bN = (insn28 >> 7) & 1; |
| 15435 | UInt fM = (INSN(3,0) << 1) | bM; /* argR */ |
| 15436 | UInt fD = (INSN(15,12) << 1) | bD; /* dst/acc */ |
| 15437 | UInt fN = (INSN(19,16) << 1) | bN; /* argL */ |
| 15438 | UInt bP = (insn28 >> 23) & 1; |
| 15439 | UInt bQ = (insn28 >> 21) & 1; |
| 15440 | UInt bR = (insn28 >> 20) & 1; |
| 15441 | UInt bS = (insn28 >> 6) & 1; |
| 15442 | UInt opc = (bP << 3) | (bQ << 2) | (bR << 1) | bS; |
| 15443 | IRExpr* rm = get_FAKE_roundingmode(); /* XXXROUNDINGFIXME */ |
| 15444 | switch (opc) { |
| 15445 | case BITS4(0,0,0,0): /* MAC: d + n * m */ |
| 15446 | putFReg(fD, triop(Iop_AddF32, rm, |
| 15447 | getFReg(fD), |
| 15448 | triop(Iop_MulF32, rm, getFReg(fN), getFReg(fM))), |
| 15449 | condT); |
| 15450 | DIP("fmacs%s s%u, s%u, s%u\n", nCC(conq), fD, fN, fM); |
| 15451 | goto decode_success_vfp; |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15452 | case BITS4(0,0,0,1): /* NMAC: d + -(n * m) */ |
| 15453 | putFReg(fD, triop(Iop_AddF32, rm, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15454 | getFReg(fD), |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15455 | unop(Iop_NegF32, |
| 15456 | triop(Iop_MulF32, rm, getFReg(fN), |
| 15457 | getFReg(fM)))), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15458 | condT); |
| 15459 | DIP("fnmacs%s s%u, s%u, s%u\n", nCC(conq), fD, fN, fM); |
| 15460 | goto decode_success_vfp; |
| 15461 | case BITS4(0,0,1,0): /* MSC: - d + n * m */ |
| 15462 | putFReg(fD, triop(Iop_AddF32, rm, |
| 15463 | unop(Iop_NegF32, getFReg(fD)), |
| 15464 | triop(Iop_MulF32, rm, getFReg(fN), getFReg(fM))), |
| 15465 | condT); |
| 15466 | DIP("fmscs%s s%u, s%u, s%u\n", nCC(conq), fD, fN, fM); |
| 15467 | goto decode_success_vfp; |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15468 | case BITS4(0,0,1,1): /* NMSC: - d + -(n * m) */ |
| 15469 | putFReg(fD, triop(Iop_AddF32, rm, |
| 15470 | unop(Iop_NegF32, getFReg(fD)), |
| 15471 | unop(Iop_NegF32, |
| 15472 | triop(Iop_MulF32, rm, |
| 15473 | getFReg(fN), |
| 15474 | getFReg(fM)))), |
| 15475 | condT); |
| 15476 | DIP("fnmscs%s s%u, s%u, s%u\n", nCC(conq), fD, fN, fM); |
| 15477 | goto decode_success_vfp; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15478 | case BITS4(0,1,0,0): /* MUL: n * m */ |
| 15479 | putFReg(fD, triop(Iop_MulF32, rm, getFReg(fN), getFReg(fM)), |
| 15480 | condT); |
| 15481 | DIP("fmuls%s s%u, s%u, s%u\n", nCC(conq), fD, fN, fM); |
| 15482 | goto decode_success_vfp; |
| 15483 | case BITS4(0,1,0,1): /* NMUL: - n * m */ |
| 15484 | putFReg(fD, unop(Iop_NegF32, |
| 15485 | triop(Iop_MulF32, rm, getFReg(fN), |
| 15486 | getFReg(fM))), |
| 15487 | condT); |
| 15488 | DIP("fnmuls%s s%u, s%u, s%u\n", nCC(conq), fD, fN, fM); |
| 15489 | goto decode_success_vfp; |
| 15490 | case BITS4(0,1,1,0): /* ADD: n + m */ |
| 15491 | putFReg(fD, triop(Iop_AddF32, rm, getFReg(fN), getFReg(fM)), |
| 15492 | condT); |
| 15493 | DIP("fadds%s s%u, s%u, s%u\n", nCC(conq), fD, fN, fM); |
| 15494 | goto decode_success_vfp; |
| 15495 | case BITS4(0,1,1,1): /* SUB: n - m */ |
| 15496 | putFReg(fD, triop(Iop_SubF32, rm, getFReg(fN), getFReg(fM)), |
| 15497 | condT); |
| 15498 | DIP("fsubs%s s%u, s%u, s%u\n", nCC(conq), fD, fN, fM); |
| 15499 | goto decode_success_vfp; |
| 15500 | case BITS4(1,0,0,0): /* DIV: n / m */ |
| 15501 | putFReg(fD, triop(Iop_DivF32, rm, getFReg(fN), getFReg(fM)), |
| 15502 | condT); |
| 15503 | DIP("fdivs%s s%u, s%u, s%u\n", nCC(conq), fD, fN, fM); |
| 15504 | goto decode_success_vfp; |
sewardj | d6d13b3 | 2014-11-19 16:07:28 +0000 | [diff] [blame] | 15505 | case BITS4(1,0,1,0): /* VNFMS: -(d - n * m) (fused) */ |
| 15506 | /* XXXROUNDINGFIXME look up ARM reference for fused |
| 15507 | multiply-add rounding */ |
| 15508 | putFReg(fD, triop(Iop_AddF32, rm, |
| 15509 | unop(Iop_NegF32, getFReg(fD)), |
| 15510 | triop(Iop_MulF32, rm, |
| 15511 | getFReg(fN), |
| 15512 | getFReg(fM))), |
| 15513 | condT); |
| 15514 | DIP("vfnmss%s s%u, s%u, s%u\n", nCC(conq), fD, fN, fM); |
| 15515 | goto decode_success_vfp; |
| 15516 | case BITS4(1,0,1,1): /* VNFMA: -(d + n * m) (fused) */ |
| 15517 | /* XXXROUNDINGFIXME look up ARM reference for fused |
| 15518 | multiply-add rounding */ |
| 15519 | putFReg(fD, triop(Iop_AddF32, rm, |
| 15520 | unop(Iop_NegF32, getFReg(fD)), |
| 15521 | triop(Iop_MulF32, rm, |
| 15522 | unop(Iop_NegF32, getFReg(fN)), |
| 15523 | getFReg(fM))), |
| 15524 | condT); |
| 15525 | DIP("vfnmas%s s%u, s%u, s%u\n", nCC(conq), fD, fN, fM); |
| 15526 | goto decode_success_vfp; |
sewardj | 7bfbbe9 | 2014-05-14 23:38:23 +0000 | [diff] [blame] | 15527 | case BITS4(1,1,0,0): /* VFMA: d + n * m (fused) */ |
| 15528 | /* XXXROUNDINGFIXME look up ARM reference for fused |
| 15529 | multiply-add rounding */ |
| 15530 | putFReg(fD, triop(Iop_AddF32, rm, |
| 15531 | getFReg(fD), |
| 15532 | triop(Iop_MulF32, rm, getFReg(fN), |
| 15533 | getFReg(fM))), |
| 15534 | condT); |
| 15535 | DIP("vfmas%s s%u, s%u, s%u\n", nCC(conq), fD, fN, fM); |
| 15536 | goto decode_success_vfp; |
| 15537 | case BITS4(1,1,0,1): /* VFMS: d + (-n * m) (fused) */ |
| 15538 | /* XXXROUNDINGFIXME look up ARM reference for fused |
| 15539 | multiply-add rounding */ |
| 15540 | putFReg(fD, triop(Iop_AddF32, rm, |
| 15541 | getFReg(fD), |
| 15542 | triop(Iop_MulF32, rm, |
| 15543 | unop(Iop_NegF32, getFReg(fN)), |
| 15544 | getFReg(fM))), |
| 15545 | condT); |
| 15546 | DIP("vfmss%s s%u, s%u, s%u\n", nCC(conq), fD, fN, fM); |
| 15547 | goto decode_success_vfp; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15548 | default: |
| 15549 | break; |
| 15550 | } |
| 15551 | } |
| 15552 | |
| 15553 | /* --------------------- compares (S) --------------------- */ |
| 15554 | /* 31 27 23 19 15 11 7 3 |
| 15555 | 28 24 20 16 12 8 4 0 |
| 15556 | FCMPS cond 1110 1D11 0100 Fd 1010 01M0 Fm |
| 15557 | FCMPES cond 1110 1D11 0100 Fd 1010 11M0 Fm |
| 15558 | FCMPZS cond 1110 1D11 0101 Fd 1010 0100 0000 |
| 15559 | FCMPZED cond 1110 1D11 0101 Fd 1010 1100 0000 |
| 15560 | Z N |
| 15561 | |
| 15562 | Z=0 Compare Fd:D vs Fm:M and set FPSCR 31:28 accordingly |
| 15563 | Z=1 Compare Fd:D vs zero |
| 15564 | |
| 15565 | N=1 generates Invalid Operation exn if either arg is any kind of NaN |
| 15566 | N=0 generates Invalid Operation exn if either arg is a signalling NaN |
| 15567 | (Not that we pay any attention to N here) |
| 15568 | */ |
| 15569 | if (BITS8(1,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
| 15570 | && BITS4(0,1,0,0) == (INSN(19,16) & BITS4(1,1,1,0)) |
| 15571 | && BITS4(1,0,1,0) == INSN(11,8) |
| 15572 | && BITS4(0,1,0,0) == (INSN(7,4) & BITS4(0,1,0,1))) { |
| 15573 | UInt bZ = (insn28 >> 16) & 1; |
| 15574 | UInt bN = (insn28 >> 7) & 1; |
| 15575 | UInt bD = (insn28 >> 22) & 1; |
| 15576 | UInt bM = (insn28 >> 5) & 1; |
| 15577 | UInt fD = (INSN(15,12) << 1) | bD; |
| 15578 | UInt fM = (INSN(3,0) << 1) | bM; |
| 15579 | if (bZ && (INSN(3,0) != 0 || (INSN(7,4) & 3) != 0)) { |
| 15580 | /* does not decode; fall through */ |
| 15581 | } else { |
| 15582 | IRTemp argL = newTemp(Ity_F64); |
| 15583 | IRTemp argR = newTemp(Ity_F64); |
| 15584 | IRTemp irRes = newTemp(Ity_I32); |
| 15585 | |
| 15586 | assign(argL, unop(Iop_F32toF64, getFReg(fD))); |
| 15587 | assign(argR, bZ ? IRExpr_Const(IRConst_F64i(0)) |
| 15588 | : unop(Iop_F32toF64, getFReg(fM))); |
| 15589 | assign(irRes, binop(Iop_CmpF64, mkexpr(argL), mkexpr(argR))); |
| 15590 | |
| 15591 | IRTemp nzcv = IRTemp_INVALID; |
| 15592 | IRTemp oldFPSCR = newTemp(Ity_I32); |
| 15593 | IRTemp newFPSCR = newTemp(Ity_I32); |
| 15594 | |
| 15595 | /* This is where the fun starts. We have to convert 'irRes' |
| 15596 | from an IR-convention return result (IRCmpF64Result) to an |
| 15597 | ARM-encoded (N,Z,C,V) group. The final result is in the |
| 15598 | bottom 4 bits of 'nzcv'. */ |
| 15599 | /* Map compare result from IR to ARM(nzcv) */ |
| 15600 | /* |
| 15601 | FP cmp result | IR | ARM(nzcv) |
| 15602 | -------------------------------- |
| 15603 | UN 0x45 0011 |
| 15604 | LT 0x01 1000 |
| 15605 | GT 0x00 0010 |
| 15606 | EQ 0x40 0110 |
| 15607 | */ |
| 15608 | nzcv = mk_convert_IRCmpF64Result_to_NZCV(irRes); |
| 15609 | |
| 15610 | /* And update FPSCR accordingly */ |
| 15611 | assign(oldFPSCR, IRExpr_Get(OFFB_FPSCR, Ity_I32)); |
| 15612 | assign(newFPSCR, |
| 15613 | binop(Iop_Or32, |
| 15614 | binop(Iop_And32, mkexpr(oldFPSCR), mkU32(0x0FFFFFFF)), |
| 15615 | binop(Iop_Shl32, mkexpr(nzcv), mkU8(28)))); |
| 15616 | |
| 15617 | putMiscReg32(OFFB_FPSCR, mkexpr(newFPSCR), condT); |
| 15618 | |
| 15619 | if (bZ) { |
| 15620 | DIP("fcmpz%ss%s s%u\n", bN ? "e" : "", nCC(conq), fD); |
| 15621 | } else { |
| 15622 | DIP("fcmp%ss%s s%u, s%u\n", bN ? "e" : "", |
| 15623 | nCC(conq), fD, fM); |
| 15624 | } |
| 15625 | goto decode_success_vfp; |
| 15626 | } |
| 15627 | /* fall through */ |
| 15628 | } |
| 15629 | |
| 15630 | /* --------------------- unary (S) --------------------- */ |
| 15631 | if (BITS8(1,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
| 15632 | && BITS4(0,0,0,0) == (INSN(19,16) & BITS4(1,1,1,0)) |
| 15633 | && BITS4(1,0,1,0) == INSN(11,8) |
| 15634 | && BITS4(0,1,0,0) == (INSN(7,4) & BITS4(0,1,0,1))) { |
| 15635 | UInt bD = (insn28 >> 22) & 1; |
| 15636 | UInt bM = (insn28 >> 5) & 1; |
| 15637 | UInt fD = (INSN(15,12) << 1) | bD; |
| 15638 | UInt fM = (INSN(3,0) << 1) | bM; |
| 15639 | UInt b16 = (insn28 >> 16) & 1; |
| 15640 | UInt b7 = (insn28 >> 7) & 1; |
| 15641 | /**/ if (b16 == 0 && b7 == 0) { |
| 15642 | // FCPYS |
| 15643 | putFReg(fD, getFReg(fM), condT); |
| 15644 | DIP("fcpys%s s%u, s%u\n", nCC(conq), fD, fM); |
| 15645 | goto decode_success_vfp; |
| 15646 | } |
| 15647 | else if (b16 == 0 && b7 == 1) { |
| 15648 | // FABSS |
| 15649 | putFReg(fD, unop(Iop_AbsF32, getFReg(fM)), condT); |
| 15650 | DIP("fabss%s s%u, s%u\n", nCC(conq), fD, fM); |
| 15651 | goto decode_success_vfp; |
| 15652 | } |
| 15653 | else if (b16 == 1 && b7 == 0) { |
| 15654 | // FNEGS |
| 15655 | putFReg(fD, unop(Iop_NegF32, getFReg(fM)), condT); |
| 15656 | DIP("fnegs%s s%u, s%u\n", nCC(conq), fD, fM); |
| 15657 | goto decode_success_vfp; |
| 15658 | } |
| 15659 | else if (b16 == 1 && b7 == 1) { |
| 15660 | // FSQRTS |
| 15661 | IRExpr* rm = get_FAKE_roundingmode(); /* XXXROUNDINGFIXME */ |
| 15662 | putFReg(fD, binop(Iop_SqrtF32, rm, getFReg(fM)), condT); |
| 15663 | DIP("fsqrts%s s%u, s%u\n", nCC(conq), fD, fM); |
| 15664 | goto decode_success_vfp; |
| 15665 | } |
| 15666 | else |
| 15667 | vassert(0); |
| 15668 | |
| 15669 | /* fall through */ |
| 15670 | } |
| 15671 | |
| 15672 | /* ----------------- I <-> S conversions ----------------- */ |
| 15673 | |
| 15674 | // F{S,U}ITOS fD, fM |
| 15675 | /* These are more complex than FSITOD/FUITOD. In the D cases, a 32 |
| 15676 | bit int will always fit within the 53 bit mantissa, so there's |
| 15677 | no possibility of a loss of precision, but that's obviously not |
| 15678 | the case here. Hence this case possibly requires rounding, and |
| 15679 | so it drags in the current rounding mode. */ |
| 15680 | if (BITS8(1,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15681 | && BITS4(1,0,0,0) == INSN(19,16) |
| 15682 | && BITS4(1,0,1,0) == (INSN(11,8) & BITS4(1,1,1,0)) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15683 | && BITS4(0,1,0,0) == (INSN(7,4) & BITS4(0,1,0,1))) { |
| 15684 | UInt bM = (insn28 >> 5) & 1; |
| 15685 | UInt bD = (insn28 >> 22) & 1; |
| 15686 | UInt fM = (INSN(3,0) << 1) | bM; |
| 15687 | UInt fD = (INSN(15,12) << 1) | bD; |
| 15688 | UInt syned = (insn28 >> 7) & 1; |
| 15689 | IRTemp rmode = newTemp(Ity_I32); |
| 15690 | assign(rmode, mkexpr(mk_get_IR_rounding_mode())); |
| 15691 | if (syned) { |
| 15692 | // FSITOS |
| 15693 | putFReg(fD, binop(Iop_F64toF32, |
| 15694 | mkexpr(rmode), |
| 15695 | unop(Iop_I32StoF64, |
| 15696 | unop(Iop_ReinterpF32asI32, getFReg(fM)))), |
| 15697 | condT); |
| 15698 | DIP("fsitos%s s%u, s%u\n", nCC(conq), fD, fM); |
| 15699 | } else { |
| 15700 | // FUITOS |
| 15701 | putFReg(fD, binop(Iop_F64toF32, |
| 15702 | mkexpr(rmode), |
| 15703 | unop(Iop_I32UtoF64, |
| 15704 | unop(Iop_ReinterpF32asI32, getFReg(fM)))), |
| 15705 | condT); |
| 15706 | DIP("fuitos%s s%u, s%u\n", nCC(conq), fD, fM); |
| 15707 | } |
| 15708 | goto decode_success_vfp; |
| 15709 | } |
| 15710 | |
| 15711 | // FTO{S,U}IS fD, fM |
| 15712 | if (BITS8(1,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
| 15713 | && BITS4(1,1,0,0) == (INSN(19,16) & BITS4(1,1,1,0)) |
| 15714 | && BITS4(1,0,1,0) == INSN(11,8) |
| 15715 | && BITS4(0,1,0,0) == (INSN(7,4) & BITS4(0,1,0,1))) { |
| 15716 | UInt bM = (insn28 >> 5) & 1; |
| 15717 | UInt bD = (insn28 >> 22) & 1; |
| 15718 | UInt fD = (INSN(15,12) << 1) | bD; |
| 15719 | UInt fM = (INSN(3,0) << 1) | bM; |
| 15720 | UInt bZ = (insn28 >> 7) & 1; |
| 15721 | UInt syned = (insn28 >> 16) & 1; |
| 15722 | IRTemp rmode = newTemp(Ity_I32); |
| 15723 | assign(rmode, bZ ? mkU32(Irrm_ZERO) |
| 15724 | : mkexpr(mk_get_IR_rounding_mode())); |
| 15725 | if (syned) { |
| 15726 | // FTOSIS |
| 15727 | putFReg(fD, unop(Iop_ReinterpI32asF32, |
| 15728 | binop(Iop_F64toI32S, mkexpr(rmode), |
| 15729 | unop(Iop_F32toF64, getFReg(fM)))), |
| 15730 | condT); |
| 15731 | DIP("ftosi%ss%s s%u, d%u\n", bZ ? "z" : "", |
| 15732 | nCC(conq), fD, fM); |
| 15733 | goto decode_success_vfp; |
| 15734 | } else { |
| 15735 | // FTOUIS |
| 15736 | putFReg(fD, unop(Iop_ReinterpI32asF32, |
| 15737 | binop(Iop_F64toI32U, mkexpr(rmode), |
| 15738 | unop(Iop_F32toF64, getFReg(fM)))), |
| 15739 | condT); |
| 15740 | DIP("ftoui%ss%s s%u, d%u\n", bZ ? "z" : "", |
| 15741 | nCC(conq), fD, fM); |
| 15742 | goto decode_success_vfp; |
| 15743 | } |
| 15744 | } |
| 15745 | |
| 15746 | /* ----------------- S <-> D conversions ----------------- */ |
| 15747 | |
| 15748 | // FCVTDS |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15749 | if (BITS8(1,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15750 | && BITS4(0,1,1,1) == INSN(19,16) |
| 15751 | && BITS4(1,0,1,0) == INSN(11,8) |
| 15752 | && BITS4(1,1,0,0) == (INSN(7,4) & BITS4(1,1,0,1))) { |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15753 | UInt dD = INSN(15,12) | (INSN(22,22) << 4); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15754 | UInt bM = (insn28 >> 5) & 1; |
| 15755 | UInt fM = (INSN(3,0) << 1) | bM; |
| 15756 | putDReg(dD, unop(Iop_F32toF64, getFReg(fM)), condT); |
| 15757 | DIP("fcvtds%s d%u, s%u\n", nCC(conq), dD, fM); |
| 15758 | goto decode_success_vfp; |
| 15759 | } |
| 15760 | |
| 15761 | // FCVTSD |
| 15762 | if (BITS8(1,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
| 15763 | && BITS4(0,1,1,1) == INSN(19,16) |
| 15764 | && BITS4(1,0,1,1) == INSN(11,8) |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15765 | && BITS4(1,1,0,0) == (INSN(7,4) & BITS4(1,1,0,1))) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15766 | UInt bD = (insn28 >> 22) & 1; |
| 15767 | UInt fD = (INSN(15,12) << 1) | bD; |
sewardj | 93ba93f | 2010-09-22 16:15:50 +0000 | [diff] [blame] | 15768 | UInt dM = INSN(3,0) | (INSN(5,5) << 4); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15769 | IRTemp rmode = newTemp(Ity_I32); |
| 15770 | assign(rmode, mkexpr(mk_get_IR_rounding_mode())); |
| 15771 | putFReg(fD, binop(Iop_F64toF32, mkexpr(rmode), getDReg(dM)), |
| 15772 | condT); |
| 15773 | DIP("fcvtsd%s s%u, d%u\n", nCC(conq), fD, dM); |
| 15774 | goto decode_success_vfp; |
| 15775 | } |
| 15776 | |
sewardj | 9fe0cc7 | 2012-08-01 20:05:42 +0000 | [diff] [blame] | 15777 | /* --------------- VCVT fixed<->floating, VFP --------------- */ |
| 15778 | /* 31 27 23 19 15 11 7 3 |
| 15779 | 28 24 20 16 12 8 4 0 |
| 15780 | |
| 15781 | cond 1110 1D11 1p1U Vd 101f x1i0 imm4 |
| 15782 | |
| 15783 | VCVT<c>.<Td>.F64 <Dd>, <Dd>, #fbits |
| 15784 | VCVT<c>.<Td>.F32 <Dd>, <Dd>, #fbits |
| 15785 | VCVT<c>.F64.<Td> <Dd>, <Dd>, #fbits |
| 15786 | VCVT<c>.F32.<Td> <Dd>, <Dd>, #fbits |
| 15787 | are of this form. We only handle a subset of the cases though. |
| 15788 | */ |
| 15789 | if (BITS8(1,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
| 15790 | && BITS4(1,0,1,0) == (INSN(19,16) & BITS4(1,0,1,0)) |
| 15791 | && BITS3(1,0,1) == INSN(11,9) |
| 15792 | && BITS3(1,0,0) == (INSN(6,4) & BITS3(1,0,1))) { |
sewardj | 6b7bdec | 2013-03-05 14:26:22 +0000 | [diff] [blame] | 15793 | UInt bD = INSN(22,22); |
| 15794 | UInt bOP = INSN(18,18); |
| 15795 | UInt bU = INSN(16,16); |
| 15796 | UInt Vd = INSN(15,12); |
| 15797 | UInt bSF = INSN(8,8); |
| 15798 | UInt bSX = INSN(7,7); |
| 15799 | UInt bI = INSN(5,5); |
| 15800 | UInt imm4 = INSN(3,0); |
| 15801 | Bool to_fixed = bOP == 1; |
| 15802 | Bool dp_op = bSF == 1; |
| 15803 | Bool unsyned = bU == 1; |
| 15804 | UInt size = bSX == 0 ? 16 : 32; |
| 15805 | Int frac_bits = size - ((imm4 << 1) | bI); |
| 15806 | UInt d = dp_op ? ((bD << 4) | Vd) : ((Vd << 1) | bD); |
sewardj | f3ae738 | 2015-08-17 08:20:55 +0000 | [diff] [blame] | 15807 | |
| 15808 | IRExpr* rm = mkU32(Irrm_NEAREST); |
| 15809 | IRTemp scale = newTemp(Ity_F64); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 15810 | assign(scale, unop(Iop_I32UtoF64, mkU32( ((UInt)1) << (frac_bits-1) ))); |
sewardj | f3ae738 | 2015-08-17 08:20:55 +0000 | [diff] [blame] | 15811 | |
sewardj | 6b7bdec | 2013-03-05 14:26:22 +0000 | [diff] [blame] | 15812 | if (frac_bits >= 1 && frac_bits <= 32 && !to_fixed && !dp_op |
| 15813 | && size == 32) { |
sewardj | 9fe0cc7 | 2012-08-01 20:05:42 +0000 | [diff] [blame] | 15814 | /* VCVT.F32.{S,U}32 S[d], S[d], #frac_bits */ |
| 15815 | /* This generates really horrible code. We could potentially |
| 15816 | do much better. */ |
| 15817 | IRTemp rmode = newTemp(Ity_I32); |
sewardj | 6b7bdec | 2013-03-05 14:26:22 +0000 | [diff] [blame] | 15818 | assign(rmode, mkU32(Irrm_NEAREST)); // per the spec |
sewardj | 9fe0cc7 | 2012-08-01 20:05:42 +0000 | [diff] [blame] | 15819 | IRTemp src32 = newTemp(Ity_I32); |
| 15820 | assign(src32, unop(Iop_ReinterpF32asI32, getFReg(d))); |
| 15821 | IRExpr* as_F64 = unop( unsyned ? Iop_I32UtoF64 : Iop_I32StoF64, |
| 15822 | mkexpr(src32 ) ); |
sewardj | 9fe0cc7 | 2012-08-01 20:05:42 +0000 | [diff] [blame] | 15823 | IRExpr* resF64 = triop(Iop_DivF64, |
| 15824 | rm, as_F64, |
sewardj | 6b7bdec | 2013-03-05 14:26:22 +0000 | [diff] [blame] | 15825 | triop(Iop_AddF64, rm, mkexpr(scale), |
| 15826 | mkexpr(scale))); |
sewardj | 9fe0cc7 | 2012-08-01 20:05:42 +0000 | [diff] [blame] | 15827 | IRExpr* resF32 = binop(Iop_F64toF32, mkexpr(rmode), resF64); |
| 15828 | putFReg(d, resF32, condT); |
| 15829 | DIP("vcvt.f32.%c32, s%u, s%u, #%d\n", |
| 15830 | unsyned ? 'u' : 's', d, d, frac_bits); |
| 15831 | goto decode_success_vfp; |
| 15832 | } |
sewardj | 6b7bdec | 2013-03-05 14:26:22 +0000 | [diff] [blame] | 15833 | if (frac_bits >= 1 && frac_bits <= 32 && !to_fixed && dp_op |
| 15834 | && size == 32) { |
| 15835 | /* VCVT.F64.{S,U}32 D[d], D[d], #frac_bits */ |
| 15836 | /* This generates really horrible code. We could potentially |
| 15837 | do much better. */ |
sewardj | 6b7bdec | 2013-03-05 14:26:22 +0000 | [diff] [blame] | 15838 | IRTemp src32 = newTemp(Ity_I32); |
sewardj | a39fb0c | 2013-03-22 13:28:50 +0000 | [diff] [blame] | 15839 | assign(src32, unop(Iop_64to32, getDRegI64(d))); |
sewardj | 6b7bdec | 2013-03-05 14:26:22 +0000 | [diff] [blame] | 15840 | IRExpr* as_F64 = unop( unsyned ? Iop_I32UtoF64 : Iop_I32StoF64, |
| 15841 | mkexpr(src32 ) ); |
sewardj | 6b7bdec | 2013-03-05 14:26:22 +0000 | [diff] [blame] | 15842 | IRExpr* resF64 = triop(Iop_DivF64, |
| 15843 | rm, as_F64, |
| 15844 | triop(Iop_AddF64, rm, mkexpr(scale), |
| 15845 | mkexpr(scale))); |
| 15846 | putDReg(d, resF64, condT); |
| 15847 | DIP("vcvt.f64.%c32, d%u, d%u, #%d\n", |
| 15848 | unsyned ? 'u' : 's', d, d, frac_bits); |
| 15849 | goto decode_success_vfp; |
| 15850 | } |
sewardj | ade5552 | 2013-03-06 08:34:04 +0000 | [diff] [blame] | 15851 | if (frac_bits >= 1 && frac_bits <= 32 && to_fixed && dp_op |
| 15852 | && size == 32) { |
| 15853 | /* VCVT.{S,U}32.F64 D[d], D[d], #frac_bits */ |
| 15854 | IRTemp srcF64 = newTemp(Ity_F64); |
| 15855 | assign(srcF64, getDReg(d)); |
sewardj | ade5552 | 2013-03-06 08:34:04 +0000 | [diff] [blame] | 15856 | IRTemp scaledF64 = newTemp(Ity_F64); |
sewardj | ade5552 | 2013-03-06 08:34:04 +0000 | [diff] [blame] | 15857 | assign(scaledF64, triop(Iop_MulF64, |
| 15858 | rm, mkexpr(srcF64), |
| 15859 | triop(Iop_AddF64, rm, mkexpr(scale), |
| 15860 | mkexpr(scale)))); |
| 15861 | IRTemp rmode = newTemp(Ity_I32); |
| 15862 | assign(rmode, mkU32(Irrm_ZERO)); // as per the spec |
| 15863 | IRTemp asI32 = newTemp(Ity_I32); |
| 15864 | assign(asI32, binop(unsyned ? Iop_F64toI32U : Iop_F64toI32S, |
| 15865 | mkexpr(rmode), mkexpr(scaledF64))); |
| 15866 | putDRegI64(d, unop(unsyned ? Iop_32Uto64 : Iop_32Sto64, |
| 15867 | mkexpr(asI32)), condT); |
sewardj | f3ae738 | 2015-08-17 08:20:55 +0000 | [diff] [blame] | 15868 | |
| 15869 | DIP("vcvt.%c32.f64, d%u, d%u, #%d\n", |
| 15870 | unsyned ? 'u' : 's', d, d, frac_bits); |
| 15871 | goto decode_success_vfp; |
| 15872 | } |
| 15873 | if (frac_bits >= 1 && frac_bits <= 32 && to_fixed && !dp_op |
| 15874 | && size == 32) { |
| 15875 | /* VCVT.{S,U}32.F32 S[d], S[d], #frac_bits */ |
| 15876 | IRTemp srcF32 = newTemp(Ity_F32); |
| 15877 | assign(srcF32, getFReg(d)); |
| 15878 | IRTemp scaledF64 = newTemp(Ity_F64); |
| 15879 | assign(scaledF64, triop(Iop_MulF64, |
| 15880 | rm, unop(Iop_F32toF64, mkexpr(srcF32)), |
| 15881 | triop(Iop_AddF64, rm, mkexpr(scale), |
| 15882 | mkexpr(scale)))); |
| 15883 | IRTemp rmode = newTemp(Ity_I32); |
| 15884 | assign(rmode, mkU32(Irrm_ZERO)); // as per the spec |
| 15885 | IRTemp asI32 = newTemp(Ity_I32); |
| 15886 | assign(asI32, binop(unsyned ? Iop_F64toI32U : Iop_F64toI32S, |
| 15887 | mkexpr(rmode), mkexpr(scaledF64))); |
| 15888 | putFReg(d, unop(Iop_ReinterpI32asF32, mkexpr(asI32)), condT); |
| 15889 | DIP("vcvt.%c32.f32, d%u, d%u, #%d\n", |
| 15890 | unsyned ? 'u' : 's', d, d, frac_bits); |
sewardj | ade5552 | 2013-03-06 08:34:04 +0000 | [diff] [blame] | 15891 | goto decode_success_vfp; |
| 15892 | } |
sewardj | 9fe0cc7 | 2012-08-01 20:05:42 +0000 | [diff] [blame] | 15893 | /* fall through */ |
| 15894 | } |
| 15895 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15896 | /* FAILURE */ |
| 15897 | return False; |
| 15898 | |
| 15899 | decode_success_vfp: |
| 15900 | /* Check that any accepted insn really is a CP10 or CP11 insn, iow, |
| 15901 | assert that we aren't accepting, in this fn, insns that actually |
| 15902 | should be handled somewhere else. */ |
| 15903 | vassert(INSN(11,9) == BITS3(1,0,1)); // 11:8 = 1010 or 1011 |
| 15904 | return True; |
| 15905 | |
| 15906 | # undef INSN |
| 15907 | } |
| 15908 | |
| 15909 | |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 15910 | /*------------------------------------------------------------*/ |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 15911 | /*--- Instructions in NV (never) space ---*/ |
| 15912 | /*------------------------------------------------------------*/ |
| 15913 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15914 | /* ARM only */ |
| 15915 | /* Translate a NV space instruction. If successful, returns True and |
| 15916 | *dres may or may not be updated. If failure, returns False and |
| 15917 | doesn't change *dres nor create any IR. |
| 15918 | |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 15919 | Note that all NEON instructions (in ARM mode) up to and including |
| 15920 | ARMv7, but not later, are handled through here, since they are all |
| 15921 | in NV space. |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15922 | */ |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 15923 | static Bool decode_NV_instruction_ARMv7_and_below |
| 15924 | ( /*MOD*/DisResult* dres, |
florian | cacba8e | 2014-12-15 18:58:07 +0000 | [diff] [blame] | 15925 | const VexArchInfo* archinfo, |
sewardj | 1fce8de | 2010-09-09 07:27:24 +0000 | [diff] [blame] | 15926 | UInt insn ) |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 15927 | { |
| 15928 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 15929 | # define INSN_COND SLICE_UInt(insn, 31, 28) |
| 15930 | |
| 15931 | HChar dis_buf[128]; |
| 15932 | |
| 15933 | // Should only be called for NV instructions |
| 15934 | vassert(BITS4(1,1,1,1) == INSN_COND); |
| 15935 | |
sewardj | 0657e9a | 2014-07-03 11:01:38 +0000 | [diff] [blame] | 15936 | /* ------------------------ pld{w} ------------------------ */ |
| 15937 | if (BITS8(0,1,0,1, 0,0, 0,1) == (INSN(27,20) & BITS8(1,1,1,1, 0,0, 1,1)) |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 15938 | && BITS4(1,1,1,1) == INSN(15,12)) { |
| 15939 | UInt rN = INSN(19,16); |
| 15940 | UInt imm12 = INSN(11,0); |
| 15941 | UInt bU = INSN(23,23); |
sewardj | 0657e9a | 2014-07-03 11:01:38 +0000 | [diff] [blame] | 15942 | UInt bR = INSN(22,22); |
| 15943 | DIP("pld%c [r%u, #%c%u]\n", bR ? ' ' : 'w', rN, bU ? '+' : '-', imm12); |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 15944 | return True; |
| 15945 | } |
| 15946 | |
sewardj | fe1c2aa | 2014-07-03 12:27:21 +0000 | [diff] [blame] | 15947 | if (BITS8(0,1,1,1, 0,0, 0,1) == (INSN(27,20) & BITS8(1,1,1,1, 0,0, 1,1)) |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 15948 | && BITS4(1,1,1,1) == INSN(15,12) |
| 15949 | && 0 == INSN(4,4)) { |
| 15950 | UInt rN = INSN(19,16); |
| 15951 | UInt rM = INSN(3,0); |
| 15952 | UInt imm5 = INSN(11,7); |
| 15953 | UInt sh2 = INSN(6,5); |
| 15954 | UInt bU = INSN(23,23); |
sewardj | fe1c2aa | 2014-07-03 12:27:21 +0000 | [diff] [blame] | 15955 | UInt bR = INSN(22,22); |
| 15956 | if (rM != 15 && (rN != 15 || bR)) { |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 15957 | IRExpr* eaE = mk_EA_reg_plusminus_shifted_reg(rN, bU, rM, |
| 15958 | sh2, imm5, dis_buf); |
| 15959 | IRTemp eaT = newTemp(Ity_I32); |
| 15960 | /* Bind eaE to a temp merely for debugging-vex purposes, so we |
| 15961 | can check it's a plausible decoding. It will get removed |
| 15962 | by iropt a little later on. */ |
| 15963 | vassert(eaE); |
| 15964 | assign(eaT, eaE); |
sewardj | fe1c2aa | 2014-07-03 12:27:21 +0000 | [diff] [blame] | 15965 | DIP("pld%c %s\n", bR ? ' ' : 'w', dis_buf); |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 15966 | return True; |
| 15967 | } |
| 15968 | /* fall through */ |
| 15969 | } |
| 15970 | |
sewardj | 4c3839e | 2010-08-31 09:31:06 +0000 | [diff] [blame] | 15971 | /* ------------------------ pli ------------------------ */ |
| 15972 | if (BITS8(0,1,0,0, 0, 1,0,1) == (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) |
| 15973 | && BITS4(1,1,1,1) == INSN(15,12)) { |
| 15974 | UInt rN = INSN(19,16); |
| 15975 | UInt imm12 = INSN(11,0); |
| 15976 | UInt bU = INSN(23,23); |
| 15977 | DIP("pli [r%u, #%c%u]\n", rN, bU ? '+' : '-', imm12); |
| 15978 | return True; |
| 15979 | } |
| 15980 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15981 | /* --------------------- Interworking branches --------------------- */ |
| 15982 | |
| 15983 | // BLX (1), viz, unconditional branch and link to R15+simm24 |
| 15984 | // and set CPSR.T = 1, that is, switch to Thumb mode |
| 15985 | if (INSN(31,25) == BITS7(1,1,1,1,1,0,1)) { |
| 15986 | UInt bitH = INSN(24,24); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 15987 | UInt uimm24 = INSN(23,0); uimm24 <<= 8; |
| 15988 | Int simm24 = (Int)uimm24; simm24 >>= 8; |
| 15989 | simm24 = (((UInt)simm24) << 2) + (bitH << 1); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 15990 | /* Now this is a bit tricky. Since we're decoding an ARM insn, |
| 15991 | it is implies that CPSR.T == 0. Hence the current insn's |
| 15992 | address is guaranteed to be of the form X--(30)--X00. So, no |
| 15993 | need to mask any bits off it. But need to set the lowest bit |
| 15994 | to 1 to denote we're in Thumb mode after this, since |
| 15995 | guest_R15T has CPSR.T as the lowest bit. And we can't chase |
| 15996 | into the call, so end the block at this point. */ |
| 15997 | UInt dst = guest_R15_curr_instr_notENC + 8 + (simm24 | 1); |
| 15998 | putIRegA( 14, mkU32(guest_R15_curr_instr_notENC + 4), |
| 15999 | IRTemp_INVALID/*because AL*/, Ijk_Boring ); |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 16000 | llPutIReg(15, mkU32(dst)); |
| 16001 | dres->jk_StopHere = Ijk_Call; |
| 16002 | dres->whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16003 | DIP("blx 0x%x (and switch to Thumb mode)\n", dst - 1); |
| 16004 | return True; |
| 16005 | } |
| 16006 | |
sewardj | 412098c | 2010-05-04 08:48:43 +0000 | [diff] [blame] | 16007 | /* ------------------- v7 barrier insns ------------------- */ |
| 16008 | switch (insn) { |
| 16009 | case 0xF57FF06F: /* ISB */ |
| 16010 | stmt( IRStmt_MBE(Imbe_Fence) ); |
| 16011 | DIP("ISB\n"); |
| 16012 | return True; |
sewardj | e407ced | 2011-05-03 14:57:59 +0000 | [diff] [blame] | 16013 | case 0xF57FF04F: /* DSB sy */ |
| 16014 | case 0xF57FF04E: /* DSB st */ |
| 16015 | case 0xF57FF04B: /* DSB ish */ |
| 16016 | case 0xF57FF04A: /* DSB ishst */ |
| 16017 | case 0xF57FF047: /* DSB nsh */ |
| 16018 | case 0xF57FF046: /* DSB nshst */ |
| 16019 | case 0xF57FF043: /* DSB osh */ |
| 16020 | case 0xF57FF042: /* DSB oshst */ |
sewardj | 412098c | 2010-05-04 08:48:43 +0000 | [diff] [blame] | 16021 | stmt( IRStmt_MBE(Imbe_Fence) ); |
| 16022 | DIP("DSB\n"); |
| 16023 | return True; |
sewardj | e407ced | 2011-05-03 14:57:59 +0000 | [diff] [blame] | 16024 | case 0xF57FF05F: /* DMB sy */ |
| 16025 | case 0xF57FF05E: /* DMB st */ |
| 16026 | case 0xF57FF05B: /* DMB ish */ |
| 16027 | case 0xF57FF05A: /* DMB ishst */ |
| 16028 | case 0xF57FF057: /* DMB nsh */ |
| 16029 | case 0xF57FF056: /* DMB nshst */ |
| 16030 | case 0xF57FF053: /* DMB osh */ |
| 16031 | case 0xF57FF052: /* DMB oshst */ |
sewardj | 412098c | 2010-05-04 08:48:43 +0000 | [diff] [blame] | 16032 | stmt( IRStmt_MBE(Imbe_Fence) ); |
| 16033 | DIP("DMB\n"); |
| 16034 | return True; |
| 16035 | default: |
| 16036 | break; |
| 16037 | } |
| 16038 | |
sewardj | 6d615ba | 2011-09-26 16:19:43 +0000 | [diff] [blame] | 16039 | /* ------------------- CLREX ------------------ */ |
| 16040 | if (insn == 0xF57FF01F) { |
| 16041 | /* AFAICS, this simply cancels a (all?) reservations made by a |
| 16042 | (any?) preceding LDREX(es). Arrange to hand it through to |
| 16043 | the back end. */ |
| 16044 | stmt( IRStmt_MBE(Imbe_CancelReservation) ); |
| 16045 | DIP("clrex\n"); |
| 16046 | return True; |
| 16047 | } |
| 16048 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16049 | /* ------------------- NEON ------------------- */ |
sewardj | 1fce8de | 2010-09-09 07:27:24 +0000 | [diff] [blame] | 16050 | if (archinfo->hwcaps & VEX_HWCAPS_ARM_NEON) { |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 16051 | Bool ok_neon = decode_NEON_instruction_ARMv7_and_below( |
sewardj | 1fce8de | 2010-09-09 07:27:24 +0000 | [diff] [blame] | 16052 | dres, insn, IRTemp_INVALID/*unconditional*/, |
| 16053 | False/*!isT*/ |
| 16054 | ); |
| 16055 | if (ok_neon) |
| 16056 | return True; |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 16057 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16058 | |
| 16059 | // unrecognised |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 16060 | return False; |
| 16061 | |
| 16062 | # undef INSN_COND |
| 16063 | # undef INSN |
| 16064 | } |
| 16065 | |
| 16066 | |
| 16067 | /*------------------------------------------------------------*/ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16068 | /*--- Disassemble a single ARM instruction ---*/ |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 16069 | /*------------------------------------------------------------*/ |
| 16070 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16071 | /* Disassemble a single ARM instruction into IR. The instruction is |
| 16072 | located in host memory at guest_instr, and has (decoded) guest IP |
| 16073 | of guest_R15_curr_instr_notENC, which will have been set before the |
| 16074 | call here. */ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16075 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16076 | static |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16077 | DisResult disInstr_ARM_WRK ( |
florian | beac530 | 2014-12-31 12:09:38 +0000 | [diff] [blame] | 16078 | Bool (*resteerOkFn) ( /*opaque*/void*, Addr ), |
sewardj | 82f5688 | 2010-01-17 09:36:11 +0000 | [diff] [blame] | 16079 | Bool resteerCisOk, |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16080 | void* callback_opaque, |
florian | 8462d11 | 2014-09-24 15:18:09 +0000 | [diff] [blame] | 16081 | const UChar* guest_instr, |
florian | cacba8e | 2014-12-15 18:58:07 +0000 | [diff] [blame] | 16082 | const VexArchInfo* archinfo, |
| 16083 | const VexAbiInfo* abiinfo, |
sewardj | 442e51a | 2012-12-06 18:08:04 +0000 | [diff] [blame] | 16084 | Bool sigill_diag |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16085 | ) |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 16086 | { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16087 | // A macro to fish bits out of 'insn'. |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 16088 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 16089 | # define INSN_COND SLICE_UInt(insn, 31, 28) |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16090 | |
| 16091 | DisResult dres; |
| 16092 | UInt insn; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16093 | IRTemp condT; /* :: Ity_I32 */ |
| 16094 | UInt summary; |
| 16095 | HChar dis_buf[128]; // big enough to hold LDMIA etc text |
| 16096 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16097 | /* Set result defaults. */ |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 16098 | dres.whatNext = Dis_Continue; |
| 16099 | dres.len = 4; |
| 16100 | dres.continueAt = 0; |
| 16101 | dres.jk_StopHere = Ijk_INVALID; |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 16102 | dres.hint = Dis_HintNone; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16103 | |
| 16104 | /* Set default actions for post-insn handling of writes to r15, if |
| 16105 | required. */ |
| 16106 | r15written = False; |
| 16107 | r15guard = IRTemp_INVALID; /* unconditional */ |
| 16108 | r15kind = Ijk_Boring; |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 16109 | |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 16110 | /* At least this is simple on ARM: insns are all 4 bytes long, and |
| 16111 | 4-aligned. So just fish the whole thing out of memory right now |
| 16112 | and have done. */ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16113 | insn = getUIntLittleEndianly( guest_instr ); |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 16114 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16115 | if (0) vex_printf("insn: 0x%x\n", insn); |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 16116 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16117 | DIP("\t(arm) 0x%x: ", (UInt)guest_R15_curr_instr_notENC); |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 16118 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16119 | vassert(0 == (guest_R15_curr_instr_notENC & 3)); |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 16120 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16121 | /* ----------------------------------------------------------- */ |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 16122 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16123 | /* Spot "Special" instructions (see comment at top of file). */ |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 16124 | { |
florian | 8462d11 | 2014-09-24 15:18:09 +0000 | [diff] [blame] | 16125 | const UChar* code = guest_instr; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16126 | /* Spot the 16-byte preamble: |
sewardj | cca7194 | 2004-12-02 23:35:18 +0000 | [diff] [blame] | 16127 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16128 | e1a0c1ec mov r12, r12, ROR #3 |
| 16129 | e1a0c6ec mov r12, r12, ROR #13 |
| 16130 | e1a0ceec mov r12, r12, ROR #29 |
| 16131 | e1a0c9ec mov r12, r12, ROR #19 |
| 16132 | */ |
| 16133 | UInt word1 = 0xE1A0C1EC; |
| 16134 | UInt word2 = 0xE1A0C6EC; |
| 16135 | UInt word3 = 0xE1A0CEEC; |
| 16136 | UInt word4 = 0xE1A0C9EC; |
| 16137 | if (getUIntLittleEndianly(code+ 0) == word1 && |
| 16138 | getUIntLittleEndianly(code+ 4) == word2 && |
| 16139 | getUIntLittleEndianly(code+ 8) == word3 && |
| 16140 | getUIntLittleEndianly(code+12) == word4) { |
| 16141 | /* Got a "Special" instruction preamble. Which one is it? */ |
| 16142 | if (getUIntLittleEndianly(code+16) == 0xE18AA00A |
| 16143 | /* orr r10,r10,r10 */) { |
| 16144 | /* R3 = client_request ( R4 ) */ |
| 16145 | DIP("r3 = client_request ( %%r4 )\n"); |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 16146 | llPutIReg(15, mkU32( guest_R15_curr_instr_notENC + 20 )); |
| 16147 | dres.jk_StopHere = Ijk_ClientReq; |
| 16148 | dres.whatNext = Dis_StopHere; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16149 | goto decode_success; |
| 16150 | } |
| 16151 | else |
| 16152 | if (getUIntLittleEndianly(code+16) == 0xE18BB00B |
| 16153 | /* orr r11,r11,r11 */) { |
| 16154 | /* R3 = guest_NRADDR */ |
| 16155 | DIP("r3 = guest_NRADDR\n"); |
| 16156 | dres.len = 20; |
| 16157 | llPutIReg(3, IRExpr_Get( OFFB_NRADDR, Ity_I32 )); |
| 16158 | goto decode_success; |
| 16159 | } |
| 16160 | else |
| 16161 | if (getUIntLittleEndianly(code+16) == 0xE18CC00C |
| 16162 | /* orr r12,r12,r12 */) { |
| 16163 | /* branch-and-link-to-noredir R4 */ |
| 16164 | DIP("branch-and-link-to-noredir r4\n"); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16165 | llPutIReg(14, mkU32( guest_R15_curr_instr_notENC + 20) ); |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 16166 | llPutIReg(15, llGetIReg(4)); |
| 16167 | dres.jk_StopHere = Ijk_NoRedir; |
| 16168 | dres.whatNext = Dis_StopHere; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16169 | goto decode_success; |
| 16170 | } |
florian | 2245ce9 | 2012-08-28 16:49:30 +0000 | [diff] [blame] | 16171 | else |
sewardj | a0d8eb8 | 2012-09-03 21:48:42 +0000 | [diff] [blame] | 16172 | if (getUIntLittleEndianly(code+16) == 0xE1899009 |
| 16173 | /* orr r9,r9,r9 */) { |
florian | 2245ce9 | 2012-08-28 16:49:30 +0000 | [diff] [blame] | 16174 | /* IR injection */ |
| 16175 | DIP("IR injection\n"); |
florian | 2245ce9 | 2012-08-28 16:49:30 +0000 | [diff] [blame] | 16176 | vex_inject_ir(irsb, Iend_LE); |
florian | 2245ce9 | 2012-08-28 16:49:30 +0000 | [diff] [blame] | 16177 | // Invalidate the current insn. The reason is that the IRop we're |
| 16178 | // injecting here can change. In which case the translation has to |
| 16179 | // be redone. For ease of handling, we simply invalidate all the |
| 16180 | // time. |
sewardj | 05f5e01 | 2014-05-04 10:52:11 +0000 | [diff] [blame] | 16181 | stmt(IRStmt_Put(OFFB_CMSTART, mkU32(guest_R15_curr_instr_notENC))); |
| 16182 | stmt(IRStmt_Put(OFFB_CMLEN, mkU32(20))); |
florian | 2245ce9 | 2012-08-28 16:49:30 +0000 | [diff] [blame] | 16183 | llPutIReg(15, mkU32( guest_R15_curr_instr_notENC + 20 )); |
florian | 2245ce9 | 2012-08-28 16:49:30 +0000 | [diff] [blame] | 16184 | dres.whatNext = Dis_StopHere; |
sewardj | 05f5e01 | 2014-05-04 10:52:11 +0000 | [diff] [blame] | 16185 | dres.jk_StopHere = Ijk_InvalICache; |
florian | 2245ce9 | 2012-08-28 16:49:30 +0000 | [diff] [blame] | 16186 | goto decode_success; |
florian | 2245ce9 | 2012-08-28 16:49:30 +0000 | [diff] [blame] | 16187 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16188 | /* We don't know what it is. Set opc1/opc2 so decode_failure |
| 16189 | can print the insn following the Special-insn preamble. */ |
| 16190 | insn = getUIntLittleEndianly(code+16); |
| 16191 | goto decode_failure; |
| 16192 | /*NOTREACHED*/ |
| 16193 | } |
| 16194 | |
| 16195 | } |
| 16196 | |
| 16197 | /* ----------------------------------------------------------- */ |
| 16198 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16199 | /* Main ARM instruction decoder starts here. */ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16200 | |
| 16201 | /* Deal with the condition. Strategy is to merely generate a |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16202 | condition temporary at this point (or IRTemp_INVALID, meaning |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16203 | unconditional). We leave it to lower-level instruction decoders |
| 16204 | to decide whether they can generate straight-line code, or |
| 16205 | whether they must generate a side exit before the instruction. |
| 16206 | condT :: Ity_I32 and is always either zero or one. */ |
| 16207 | condT = IRTemp_INVALID; |
| 16208 | switch ( (ARMCondcode)INSN_COND ) { |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 16209 | case ARMCondNV: { |
| 16210 | // Illegal instruction prior to v5 (see ARM ARM A3-5), but |
| 16211 | // some cases are acceptable |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 16212 | Bool ok |
| 16213 | = decode_NV_instruction_ARMv7_and_below(&dres, archinfo, insn); |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 16214 | if (ok) |
| 16215 | goto decode_success; |
| 16216 | else |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 16217 | goto after_v7_decoder; |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 16218 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16219 | case ARMCondAL: // Always executed |
| 16220 | break; |
| 16221 | case ARMCondEQ: case ARMCondNE: case ARMCondHS: case ARMCondLO: |
| 16222 | case ARMCondMI: case ARMCondPL: case ARMCondVS: case ARMCondVC: |
| 16223 | case ARMCondHI: case ARMCondLS: case ARMCondGE: case ARMCondLT: |
| 16224 | case ARMCondGT: case ARMCondLE: |
| 16225 | condT = newTemp(Ity_I32); |
| 16226 | assign( condT, mk_armg_calculate_condition( INSN_COND )); |
| 16227 | break; |
| 16228 | } |
| 16229 | |
| 16230 | /* ----------------------------------------------------------- */ |
| 16231 | /* -- ARMv5 integer instructions -- */ |
| 16232 | /* ----------------------------------------------------------- */ |
| 16233 | |
| 16234 | /* ---------------- Data processing ops ------------------- */ |
| 16235 | |
| 16236 | if (0 == (INSN(27,20) & BITS8(1,1,0,0,0,0,0,0)) |
| 16237 | && !(INSN(25,25) == 0 && INSN(7,7) == 1 && INSN(4,4) == 1)) { |
| 16238 | IRTemp shop = IRTemp_INVALID; /* shifter operand */ |
| 16239 | IRTemp shco = IRTemp_INVALID; /* shifter carry out */ |
| 16240 | UInt rD = (insn >> 12) & 0xF; /* 15:12 */ |
| 16241 | UInt rN = (insn >> 16) & 0xF; /* 19:16 */ |
| 16242 | UInt bitS = (insn >> 20) & 1; /* 20:20 */ |
| 16243 | IRTemp rNt = IRTemp_INVALID; |
| 16244 | IRTemp res = IRTemp_INVALID; |
| 16245 | IRTemp oldV = IRTemp_INVALID; |
| 16246 | IRTemp oldC = IRTemp_INVALID; |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 16247 | const HChar* name = NULL; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16248 | IROp op = Iop_INVALID; |
| 16249 | Bool ok; |
| 16250 | |
| 16251 | switch (INSN(24,21)) { |
| 16252 | |
| 16253 | /* --------- ADD, SUB, AND, OR --------- */ |
| 16254 | case BITS4(0,1,0,0): /* ADD: Rd = Rn + shifter_operand */ |
| 16255 | name = "add"; op = Iop_Add32; goto rd_eq_rn_op_SO; |
| 16256 | case BITS4(0,0,1,0): /* SUB: Rd = Rn - shifter_operand */ |
| 16257 | name = "sub"; op = Iop_Sub32; goto rd_eq_rn_op_SO; |
| 16258 | case BITS4(0,0,1,1): /* RSB: Rd = shifter_operand - Rn */ |
| 16259 | name = "rsb"; op = Iop_Sub32; goto rd_eq_rn_op_SO; |
| 16260 | case BITS4(0,0,0,0): /* AND: Rd = Rn & shifter_operand */ |
| 16261 | name = "and"; op = Iop_And32; goto rd_eq_rn_op_SO; |
| 16262 | case BITS4(1,1,0,0): /* OR: Rd = Rn | shifter_operand */ |
| 16263 | name = "orr"; op = Iop_Or32; goto rd_eq_rn_op_SO; |
| 16264 | case BITS4(0,0,0,1): /* EOR: Rd = Rn ^ shifter_operand */ |
| 16265 | name = "eor"; op = Iop_Xor32; goto rd_eq_rn_op_SO; |
| 16266 | case BITS4(1,1,1,0): /* BIC: Rd = Rn & ~shifter_operand */ |
| 16267 | name = "bic"; op = Iop_And32; goto rd_eq_rn_op_SO; |
| 16268 | rd_eq_rn_op_SO: { |
| 16269 | Bool isRSB = False; |
| 16270 | Bool isBIC = False; |
| 16271 | switch (INSN(24,21)) { |
| 16272 | case BITS4(0,0,1,1): |
| 16273 | vassert(op == Iop_Sub32); isRSB = True; break; |
| 16274 | case BITS4(1,1,1,0): |
| 16275 | vassert(op == Iop_And32); isBIC = True; break; |
| 16276 | default: |
| 16277 | break; |
| 16278 | } |
| 16279 | rNt = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16280 | assign(rNt, getIRegA(rN)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16281 | ok = mk_shifter_operand( |
| 16282 | INSN(25,25), INSN(11,0), |
| 16283 | &shop, bitS ? &shco : NULL, dis_buf |
| 16284 | ); |
| 16285 | if (!ok) |
| 16286 | break; |
| 16287 | res = newTemp(Ity_I32); |
| 16288 | // compute the main result |
| 16289 | if (isRSB) { |
| 16290 | // reverse-subtract: shifter_operand - Rn |
| 16291 | vassert(op == Iop_Sub32); |
| 16292 | assign(res, binop(op, mkexpr(shop), mkexpr(rNt)) ); |
| 16293 | } else if (isBIC) { |
| 16294 | // andn: shifter_operand & ~Rn |
| 16295 | vassert(op == Iop_And32); |
| 16296 | assign(res, binop(op, mkexpr(rNt), |
| 16297 | unop(Iop_Not32, mkexpr(shop))) ); |
| 16298 | } else { |
| 16299 | // normal: Rn op shifter_operand |
| 16300 | assign(res, binop(op, mkexpr(rNt), mkexpr(shop)) ); |
| 16301 | } |
| 16302 | // but don't commit it until after we've finished |
| 16303 | // all necessary reads from the guest state |
| 16304 | if (bitS |
| 16305 | && (op == Iop_And32 || op == Iop_Or32 || op == Iop_Xor32)) { |
| 16306 | oldV = newTemp(Ity_I32); |
| 16307 | assign( oldV, mk_armg_calculate_flag_v() ); |
| 16308 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16309 | // can't safely read guest state after here |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16310 | // now safe to put the main result |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16311 | putIRegA( rD, mkexpr(res), condT, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16312 | // XXXX!! not safe to read any guest state after |
| 16313 | // this point (I think the code below doesn't do that). |
| 16314 | if (!bitS) |
| 16315 | vassert(shco == IRTemp_INVALID); |
| 16316 | /* Update the flags thunk if necessary */ |
| 16317 | if (bitS) { |
| 16318 | vassert(shco != IRTemp_INVALID); |
| 16319 | switch (op) { |
| 16320 | case Iop_Add32: |
| 16321 | setFlags_D1_D2( ARMG_CC_OP_ADD, rNt, shop, condT ); |
| 16322 | break; |
| 16323 | case Iop_Sub32: |
| 16324 | if (isRSB) { |
| 16325 | setFlags_D1_D2( ARMG_CC_OP_SUB, shop, rNt, condT ); |
| 16326 | } else { |
| 16327 | setFlags_D1_D2( ARMG_CC_OP_SUB, rNt, shop, condT ); |
| 16328 | } |
| 16329 | break; |
| 16330 | case Iop_And32: /* BIC and AND set the flags the same */ |
| 16331 | case Iop_Or32: |
| 16332 | case Iop_Xor32: |
| 16333 | // oldV has been read just above |
| 16334 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, |
| 16335 | res, shco, oldV, condT ); |
| 16336 | break; |
| 16337 | default: |
| 16338 | vassert(0); |
| 16339 | } |
| 16340 | } |
| 16341 | DIP("%s%s%s r%u, r%u, %s\n", |
| 16342 | name, nCC(INSN_COND), bitS ? "s" : "", rD, rN, dis_buf ); |
| 16343 | goto decode_success; |
| 16344 | } |
| 16345 | |
| 16346 | /* --------- MOV, MVN --------- */ |
| 16347 | case BITS4(1,1,0,1): /* MOV: Rd = shifter_operand */ |
| 16348 | case BITS4(1,1,1,1): { /* MVN: Rd = not(shifter_operand) */ |
| 16349 | Bool isMVN = INSN(24,21) == BITS4(1,1,1,1); |
sewardj | f580065 | 2011-10-14 15:44:00 +0000 | [diff] [blame] | 16350 | IRTemp jk = Ijk_Boring; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16351 | if (rN != 0) |
| 16352 | break; /* rN must be zero */ |
| 16353 | ok = mk_shifter_operand( |
| 16354 | INSN(25,25), INSN(11,0), |
| 16355 | &shop, bitS ? &shco : NULL, dis_buf |
| 16356 | ); |
| 16357 | if (!ok) |
| 16358 | break; |
| 16359 | res = newTemp(Ity_I32); |
| 16360 | assign( res, isMVN ? unop(Iop_Not32, mkexpr(shop)) |
| 16361 | : mkexpr(shop) ); |
| 16362 | if (bitS) { |
| 16363 | vassert(shco != IRTemp_INVALID); |
| 16364 | oldV = newTemp(Ity_I32); |
| 16365 | assign( oldV, mk_armg_calculate_flag_v() ); |
| 16366 | } else { |
| 16367 | vassert(shco == IRTemp_INVALID); |
| 16368 | } |
sewardj | f580065 | 2011-10-14 15:44:00 +0000 | [diff] [blame] | 16369 | /* According to the Cortex A8 TRM Sec. 5.2.1, MOV PC, r14 is a |
| 16370 | return for purposes of branch prediction. */ |
| 16371 | if (!isMVN && INSN(11,0) == 14) { |
| 16372 | jk = Ijk_Ret; |
| 16373 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16374 | // can't safely read guest state after here |
sewardj | f580065 | 2011-10-14 15:44:00 +0000 | [diff] [blame] | 16375 | putIRegA( rD, mkexpr(res), condT, jk ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16376 | /* Update the flags thunk if necessary */ |
| 16377 | if (bitS) { |
| 16378 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, |
| 16379 | res, shco, oldV, condT ); |
| 16380 | } |
| 16381 | DIP("%s%s%s r%u, %s\n", |
| 16382 | isMVN ? "mvn" : "mov", |
| 16383 | nCC(INSN_COND), bitS ? "s" : "", rD, dis_buf ); |
| 16384 | goto decode_success; |
| 16385 | } |
| 16386 | |
| 16387 | /* --------- CMP --------- */ |
| 16388 | case BITS4(1,0,1,0): /* CMP: (void) Rn - shifter_operand */ |
| 16389 | case BITS4(1,0,1,1): { /* CMN: (void) Rn + shifter_operand */ |
| 16390 | Bool isCMN = INSN(24,21) == BITS4(1,0,1,1); |
| 16391 | if (rD != 0) |
| 16392 | break; /* rD must be zero */ |
| 16393 | if (bitS == 0) |
| 16394 | break; /* if S (bit 20) is not set, it's not CMP/CMN */ |
| 16395 | rNt = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16396 | assign(rNt, getIRegA(rN)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16397 | ok = mk_shifter_operand( |
| 16398 | INSN(25,25), INSN(11,0), |
| 16399 | &shop, NULL, dis_buf |
| 16400 | ); |
| 16401 | if (!ok) |
| 16402 | break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16403 | // can't safely read guest state after here |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16404 | /* Update the flags thunk. */ |
| 16405 | setFlags_D1_D2( isCMN ? ARMG_CC_OP_ADD : ARMG_CC_OP_SUB, |
| 16406 | rNt, shop, condT ); |
| 16407 | DIP("%s%s r%u, %s\n", |
| 16408 | isCMN ? "cmn" : "cmp", |
| 16409 | nCC(INSN_COND), rN, dis_buf ); |
| 16410 | goto decode_success; |
| 16411 | } |
| 16412 | |
| 16413 | /* --------- TST --------- */ |
| 16414 | case BITS4(1,0,0,0): /* TST: (void) Rn & shifter_operand */ |
| 16415 | case BITS4(1,0,0,1): { /* TEQ: (void) Rn ^ shifter_operand */ |
| 16416 | Bool isTEQ = INSN(24,21) == BITS4(1,0,0,1); |
| 16417 | if (rD != 0) |
| 16418 | break; /* rD must be zero */ |
| 16419 | if (bitS == 0) |
| 16420 | break; /* if S (bit 20) is not set, it's not TST/TEQ */ |
| 16421 | rNt = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16422 | assign(rNt, getIRegA(rN)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16423 | ok = mk_shifter_operand( |
| 16424 | INSN(25,25), INSN(11,0), |
| 16425 | &shop, &shco, dis_buf |
| 16426 | ); |
| 16427 | if (!ok) |
| 16428 | break; |
| 16429 | /* Update the flags thunk. */ |
| 16430 | res = newTemp(Ity_I32); |
| 16431 | assign( res, binop(isTEQ ? Iop_Xor32 : Iop_And32, |
| 16432 | mkexpr(rNt), mkexpr(shop)) ); |
| 16433 | oldV = newTemp(Ity_I32); |
| 16434 | assign( oldV, mk_armg_calculate_flag_v() ); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16435 | // can't safely read guest state after here |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16436 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, |
| 16437 | res, shco, oldV, condT ); |
| 16438 | DIP("%s%s r%u, %s\n", |
| 16439 | isTEQ ? "teq" : "tst", |
| 16440 | nCC(INSN_COND), rN, dis_buf ); |
| 16441 | goto decode_success; |
| 16442 | } |
| 16443 | |
| 16444 | /* --------- ADC, SBC, RSC --------- */ |
| 16445 | case BITS4(0,1,0,1): /* ADC: Rd = Rn + shifter_operand + oldC */ |
| 16446 | name = "adc"; goto rd_eq_rn_op_SO_op_oldC; |
| 16447 | case BITS4(0,1,1,0): /* SBC: Rd = Rn - shifter_operand - (oldC ^ 1) */ |
| 16448 | name = "sbc"; goto rd_eq_rn_op_SO_op_oldC; |
| 16449 | case BITS4(0,1,1,1): /* RSC: Rd = shifter_operand - Rn - (oldC ^ 1) */ |
| 16450 | name = "rsc"; goto rd_eq_rn_op_SO_op_oldC; |
| 16451 | rd_eq_rn_op_SO_op_oldC: { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16452 | // FIXME: shco isn't used for anything. Get rid of it. |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16453 | rNt = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16454 | assign(rNt, getIRegA(rN)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16455 | ok = mk_shifter_operand( |
| 16456 | INSN(25,25), INSN(11,0), |
| 16457 | &shop, bitS ? &shco : NULL, dis_buf |
| 16458 | ); |
| 16459 | if (!ok) |
| 16460 | break; |
| 16461 | oldC = newTemp(Ity_I32); |
| 16462 | assign( oldC, mk_armg_calculate_flag_c() ); |
| 16463 | res = newTemp(Ity_I32); |
| 16464 | // compute the main result |
| 16465 | switch (INSN(24,21)) { |
| 16466 | case BITS4(0,1,0,1): /* ADC */ |
| 16467 | assign(res, |
| 16468 | binop(Iop_Add32, |
| 16469 | binop(Iop_Add32, mkexpr(rNt), mkexpr(shop)), |
| 16470 | mkexpr(oldC) )); |
| 16471 | break; |
| 16472 | case BITS4(0,1,1,0): /* SBC */ |
| 16473 | assign(res, |
| 16474 | binop(Iop_Sub32, |
| 16475 | binop(Iop_Sub32, mkexpr(rNt), mkexpr(shop)), |
| 16476 | binop(Iop_Xor32, mkexpr(oldC), mkU32(1)) )); |
| 16477 | break; |
| 16478 | case BITS4(0,1,1,1): /* RSC */ |
| 16479 | assign(res, |
| 16480 | binop(Iop_Sub32, |
| 16481 | binop(Iop_Sub32, mkexpr(shop), mkexpr(rNt)), |
| 16482 | binop(Iop_Xor32, mkexpr(oldC), mkU32(1)) )); |
| 16483 | break; |
| 16484 | default: |
| 16485 | vassert(0); |
| 16486 | } |
| 16487 | // but don't commit it until after we've finished |
| 16488 | // all necessary reads from the guest state |
| 16489 | // now safe to put the main result |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16490 | putIRegA( rD, mkexpr(res), condT, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16491 | // XXXX!! not safe to read any guest state after |
| 16492 | // this point (I think the code below doesn't do that). |
| 16493 | if (!bitS) |
| 16494 | vassert(shco == IRTemp_INVALID); |
| 16495 | /* Update the flags thunk if necessary */ |
| 16496 | if (bitS) { |
| 16497 | vassert(shco != IRTemp_INVALID); |
| 16498 | switch (INSN(24,21)) { |
| 16499 | case BITS4(0,1,0,1): /* ADC */ |
| 16500 | setFlags_D1_D2_ND( ARMG_CC_OP_ADC, |
| 16501 | rNt, shop, oldC, condT ); |
| 16502 | break; |
| 16503 | case BITS4(0,1,1,0): /* SBC */ |
| 16504 | setFlags_D1_D2_ND( ARMG_CC_OP_SBB, |
| 16505 | rNt, shop, oldC, condT ); |
| 16506 | break; |
| 16507 | case BITS4(0,1,1,1): /* RSC */ |
| 16508 | setFlags_D1_D2_ND( ARMG_CC_OP_SBB, |
| 16509 | shop, rNt, oldC, condT ); |
| 16510 | break; |
| 16511 | default: |
| 16512 | vassert(0); |
| 16513 | } |
| 16514 | } |
| 16515 | DIP("%s%s%s r%u, r%u, %s\n", |
| 16516 | name, nCC(INSN_COND), bitS ? "s" : "", rD, rN, dis_buf ); |
| 16517 | goto decode_success; |
| 16518 | } |
| 16519 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16520 | default: |
florian | c6c7a0d | 2013-01-23 17:17:22 +0000 | [diff] [blame] | 16521 | vassert(0); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16522 | } |
| 16523 | } /* if (0 == (INSN(27,20) & BITS8(1,1,0,0,0,0,0,0)) */ |
| 16524 | |
| 16525 | /* --------------------- Load/store (ubyte & word) -------- */ |
| 16526 | // LDR STR LDRB STRB |
| 16527 | /* 31 27 23 19 15 11 6 4 3 # highest bit |
| 16528 | 28 24 20 16 12 |
| 16529 | A5-20 1 | 16 cond 0101 UB0L Rn Rd imm12 |
| 16530 | A5-22 1 | 32 cond 0111 UBOL Rn Rd imm5 sh2 0 Rm |
| 16531 | A5-24 2 | 16 cond 0101 UB1L Rn Rd imm12 |
| 16532 | A5-26 2 | 32 cond 0111 UB1L Rn Rd imm5 sh2 0 Rm |
| 16533 | A5-28 3 | 16 cond 0100 UB0L Rn Rd imm12 |
| 16534 | A5-32 3 | 32 cond 0110 UB0L Rn Rd imm5 sh2 0 Rm |
| 16535 | */ |
| 16536 | /* case coding: |
| 16537 | 1 at-ea (access at ea) |
| 16538 | 2 at-ea-then-upd (access at ea, then Rn = ea) |
| 16539 | 3 at-Rn-then-upd (access at Rn, then Rn = ea) |
| 16540 | ea coding |
| 16541 | 16 Rn +/- imm12 |
| 16542 | 32 Rn +/- Rm sh2 imm5 |
| 16543 | */ |
| 16544 | /* Quickly skip over all of this for hopefully most instructions */ |
| 16545 | if ((INSN(27,24) & BITS4(1,1,0,0)) != BITS4(0,1,0,0)) |
| 16546 | goto after_load_store_ubyte_or_word; |
| 16547 | |
| 16548 | summary = 0; |
| 16549 | |
| 16550 | /**/ if (INSN(27,24) == BITS4(0,1,0,1) && INSN(21,21) == 0) { |
| 16551 | summary = 1 | 16; |
| 16552 | } |
| 16553 | else if (INSN(27,24) == BITS4(0,1,1,1) && INSN(21,21) == 0 |
| 16554 | && INSN(4,4) == 0) { |
| 16555 | summary = 1 | 32; |
| 16556 | } |
| 16557 | else if (INSN(27,24) == BITS4(0,1,0,1) && INSN(21,21) == 1) { |
| 16558 | summary = 2 | 16; |
| 16559 | } |
| 16560 | else if (INSN(27,24) == BITS4(0,1,1,1) && INSN(21,21) == 1 |
| 16561 | && INSN(4,4) == 0) { |
| 16562 | summary = 2 | 32; |
| 16563 | } |
| 16564 | else if (INSN(27,24) == BITS4(0,1,0,0) && INSN(21,21) == 0) { |
| 16565 | summary = 3 | 16; |
| 16566 | } |
| 16567 | else if (INSN(27,24) == BITS4(0,1,1,0) && INSN(21,21) == 0 |
| 16568 | && INSN(4,4) == 0) { |
| 16569 | summary = 3 | 32; |
| 16570 | } |
| 16571 | else goto after_load_store_ubyte_or_word; |
| 16572 | |
| 16573 | { UInt rN = (insn >> 16) & 0xF; /* 19:16 */ |
| 16574 | UInt rD = (insn >> 12) & 0xF; /* 15:12 */ |
| 16575 | UInt rM = (insn >> 0) & 0xF; /* 3:0 */ |
| 16576 | UInt bU = (insn >> 23) & 1; /* 23 */ |
| 16577 | UInt bB = (insn >> 22) & 1; /* 22 */ |
| 16578 | UInt bL = (insn >> 20) & 1; /* 20 */ |
| 16579 | UInt imm12 = (insn >> 0) & 0xFFF; /* 11:0 */ |
| 16580 | UInt imm5 = (insn >> 7) & 0x1F; /* 11:7 */ |
| 16581 | UInt sh2 = (insn >> 5) & 3; /* 6:5 */ |
| 16582 | |
| 16583 | /* Skip some invalid cases, which would lead to two competing |
| 16584 | updates to the same register, or which are otherwise |
| 16585 | disallowed by the spec. */ |
| 16586 | switch (summary) { |
| 16587 | case 1 | 16: |
| 16588 | break; |
| 16589 | case 1 | 32: |
| 16590 | if (rM == 15) goto after_load_store_ubyte_or_word; |
| 16591 | break; |
| 16592 | case 2 | 16: case 3 | 16: |
| 16593 | if (rN == 15) goto after_load_store_ubyte_or_word; |
| 16594 | if (bL == 1 && rN == rD) goto after_load_store_ubyte_or_word; |
| 16595 | break; |
| 16596 | case 2 | 32: case 3 | 32: |
| 16597 | if (rM == 15) goto after_load_store_ubyte_or_word; |
| 16598 | if (rN == 15) goto after_load_store_ubyte_or_word; |
| 16599 | if (rN == rM) goto after_load_store_ubyte_or_word; |
| 16600 | if (bL == 1 && rN == rD) goto after_load_store_ubyte_or_word; |
| 16601 | break; |
| 16602 | default: |
| 16603 | vassert(0); |
| 16604 | } |
| 16605 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16606 | /* compute the effective address. Bind it to a tmp since we |
| 16607 | may need to use it twice. */ |
| 16608 | IRExpr* eaE = NULL; |
| 16609 | switch (summary & 0xF0) { |
| 16610 | case 16: |
| 16611 | eaE = mk_EA_reg_plusminus_imm12( rN, bU, imm12, dis_buf ); |
| 16612 | break; |
| 16613 | case 32: |
| 16614 | eaE = mk_EA_reg_plusminus_shifted_reg( rN, bU, rM, sh2, imm5, |
| 16615 | dis_buf ); |
| 16616 | break; |
| 16617 | } |
| 16618 | vassert(eaE); |
| 16619 | IRTemp eaT = newTemp(Ity_I32); |
| 16620 | assign(eaT, eaE); |
| 16621 | |
| 16622 | /* get the old Rn value */ |
| 16623 | IRTemp rnT = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16624 | assign(rnT, getIRegA(rN)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16625 | |
| 16626 | /* decide on the transfer address */ |
| 16627 | IRTemp taT = IRTemp_INVALID; |
| 16628 | switch (summary & 0x0F) { |
| 16629 | case 1: case 2: taT = eaT; break; |
| 16630 | case 3: taT = rnT; break; |
| 16631 | } |
| 16632 | vassert(taT != IRTemp_INVALID); |
| 16633 | |
| 16634 | if (bL == 0) { |
| 16635 | /* Store. If necessary, update the base register before the |
| 16636 | store itself, so that the common idiom of "str rX, [sp, |
| 16637 | #-4]!" (store rX at sp-4, then do new sp = sp-4, a.k.a "push |
| 16638 | rX") doesn't cause Memcheck to complain that the access is |
| 16639 | below the stack pointer. Also, not updating sp before the |
| 16640 | store confuses Valgrind's dynamic stack-extending logic. So |
| 16641 | do it before the store. Hence we need to snarf the store |
| 16642 | data before doing the basereg update. */ |
| 16643 | |
| 16644 | /* get hold of the data to be stored */ |
| 16645 | IRTemp rDt = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16646 | assign(rDt, getIRegA(rD)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16647 | |
| 16648 | /* Update Rn if necessary. */ |
| 16649 | switch (summary & 0x0F) { |
| 16650 | case 2: case 3: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16651 | putIRegA( rN, mkexpr(eaT), condT, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16652 | break; |
| 16653 | } |
| 16654 | |
| 16655 | /* generate the transfer */ |
| 16656 | if (bB == 0) { // word store |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16657 | storeGuardedLE( mkexpr(taT), mkexpr(rDt), condT ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16658 | } else { // byte store |
| 16659 | vassert(bB == 1); |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16660 | storeGuardedLE( mkexpr(taT), unop(Iop_32to8, mkexpr(rDt)), condT ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16661 | } |
| 16662 | |
| 16663 | } else { |
| 16664 | /* Load */ |
| 16665 | vassert(bL == 1); |
| 16666 | |
| 16667 | /* generate the transfer */ |
| 16668 | if (bB == 0) { // word load |
sewardj | f580065 | 2011-10-14 15:44:00 +0000 | [diff] [blame] | 16669 | IRTemp jk = Ijk_Boring; |
| 16670 | /* According to the Cortex A8 TRM Sec. 5.2.1, LDR(1) with r13 as the |
| 16671 | base register and PC as the destination register is a return for |
| 16672 | purposes of branch prediction. |
| 16673 | The ARM ARM Sec. C9.10.1 further specifies that it must use a |
| 16674 | post-increment by immediate addressing mode to be counted in |
| 16675 | event 0x0E (Procedure return).*/ |
| 16676 | if (rN == 13 && summary == (3 | 16) && bB == 0) { |
| 16677 | jk = Ijk_Ret; |
| 16678 | } |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16679 | IRTemp tD = newTemp(Ity_I32); |
| 16680 | loadGuardedLE( tD, ILGop_Ident32, |
| 16681 | mkexpr(taT), llGetIReg(rD), condT ); |
| 16682 | /* "rD == 15 ? condT : IRTemp_INVALID": simply |
| 16683 | IRTemp_INVALID would be correct in all cases here, and |
| 16684 | for the non-r15 case it generates better code, by |
| 16685 | avoiding two tests of the cond (since it is already |
| 16686 | tested by loadGuardedLE). However, the logic at the end |
| 16687 | of this function, that deals with writes to r15, has an |
| 16688 | optimisation which depends on seeing whether or not the |
| 16689 | write is conditional. Hence in this particular case we |
| 16690 | let it "see" the guard condition. */ |
| 16691 | putIRegA( rD, mkexpr(tD), |
| 16692 | rD == 15 ? condT : IRTemp_INVALID, jk ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16693 | } else { // byte load |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16694 | vassert(bB == 1); |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16695 | IRTemp tD = newTemp(Ity_I32); |
| 16696 | loadGuardedLE( tD, ILGop_8Uto32, mkexpr(taT), llGetIReg(rD), condT ); |
| 16697 | /* No point in similar 3rd arg complexity here, since we |
| 16698 | can't sanely write anything to r15 like this. */ |
| 16699 | putIRegA( rD, mkexpr(tD), IRTemp_INVALID, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16700 | } |
| 16701 | |
| 16702 | /* Update Rn if necessary. */ |
| 16703 | switch (summary & 0x0F) { |
| 16704 | case 2: case 3: |
| 16705 | // should be assured by logic above: |
| 16706 | if (bL == 1) |
| 16707 | vassert(rD != rN); /* since we just wrote rD */ |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16708 | putIRegA( rN, mkexpr(eaT), condT, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16709 | break; |
| 16710 | } |
| 16711 | } |
| 16712 | |
| 16713 | switch (summary & 0x0F) { |
| 16714 | case 1: DIP("%sr%s%s r%u, %s\n", |
| 16715 | bL == 0 ? "st" : "ld", |
| 16716 | bB == 0 ? "" : "b", nCC(INSN_COND), rD, dis_buf); |
| 16717 | break; |
| 16718 | case 2: DIP("%sr%s%s r%u, %s! (at-EA-then-Rn=EA)\n", |
| 16719 | bL == 0 ? "st" : "ld", |
| 16720 | bB == 0 ? "" : "b", nCC(INSN_COND), rD, dis_buf); |
| 16721 | break; |
| 16722 | case 3: DIP("%sr%s%s r%u, %s! (at-Rn-then-Rn=EA)\n", |
| 16723 | bL == 0 ? "st" : "ld", |
| 16724 | bB == 0 ? "" : "b", nCC(INSN_COND), rD, dis_buf); |
| 16725 | break; |
| 16726 | default: vassert(0); |
| 16727 | } |
| 16728 | |
| 16729 | /* XXX deal with alignment constraints */ |
| 16730 | |
| 16731 | goto decode_success; |
| 16732 | |
| 16733 | /* Complications: |
| 16734 | |
| 16735 | For all loads: if the Amode specifies base register |
| 16736 | writeback, and the same register is specified for Rd and Rn, |
| 16737 | the results are UNPREDICTABLE. |
| 16738 | |
| 16739 | For all loads and stores: if R15 is written, branch to |
| 16740 | that address afterwards. |
| 16741 | |
| 16742 | STRB: straightforward |
| 16743 | LDRB: loaded data is zero extended |
| 16744 | STR: lowest 2 bits of address are ignored |
| 16745 | LDR: if the lowest 2 bits of the address are nonzero |
| 16746 | then the loaded value is rotated right by 8 * the lowest 2 bits |
| 16747 | */ |
| 16748 | } |
| 16749 | |
| 16750 | after_load_store_ubyte_or_word: |
| 16751 | |
| 16752 | /* --------------------- Load/store (sbyte & hword) -------- */ |
| 16753 | // LDRH LDRSH STRH LDRSB |
| 16754 | /* 31 27 23 19 15 11 7 3 # highest bit |
| 16755 | 28 24 20 16 12 8 4 0 |
| 16756 | A5-36 1 | 16 cond 0001 U10L Rn Rd im4h 1SH1 im4l |
| 16757 | A5-38 1 | 32 cond 0001 U00L Rn Rd 0000 1SH1 Rm |
| 16758 | A5-40 2 | 16 cond 0001 U11L Rn Rd im4h 1SH1 im4l |
| 16759 | A5-42 2 | 32 cond 0001 U01L Rn Rd 0000 1SH1 Rm |
| 16760 | A5-44 3 | 16 cond 0000 U10L Rn Rd im4h 1SH1 im4l |
| 16761 | A5-46 3 | 32 cond 0000 U00L Rn Rd 0000 1SH1 Rm |
| 16762 | */ |
| 16763 | /* case coding: |
| 16764 | 1 at-ea (access at ea) |
| 16765 | 2 at-ea-then-upd (access at ea, then Rn = ea) |
| 16766 | 3 at-Rn-then-upd (access at Rn, then Rn = ea) |
| 16767 | ea coding |
| 16768 | 16 Rn +/- imm8 |
| 16769 | 32 Rn +/- Rm |
| 16770 | */ |
| 16771 | /* Quickly skip over all of this for hopefully most instructions */ |
| 16772 | if ((INSN(27,24) & BITS4(1,1,1,0)) != BITS4(0,0,0,0)) |
| 16773 | goto after_load_store_sbyte_or_hword; |
| 16774 | |
| 16775 | /* Check the "1SH1" thing. */ |
| 16776 | if ((INSN(7,4) & BITS4(1,0,0,1)) != BITS4(1,0,0,1)) |
| 16777 | goto after_load_store_sbyte_or_hword; |
| 16778 | |
| 16779 | summary = 0; |
| 16780 | |
| 16781 | /**/ if (INSN(27,24) == BITS4(0,0,0,1) && INSN(22,21) == BITS2(1,0)) { |
| 16782 | summary = 1 | 16; |
| 16783 | } |
| 16784 | else if (INSN(27,24) == BITS4(0,0,0,1) && INSN(22,21) == BITS2(0,0)) { |
| 16785 | summary = 1 | 32; |
| 16786 | } |
| 16787 | else if (INSN(27,24) == BITS4(0,0,0,1) && INSN(22,21) == BITS2(1,1)) { |
| 16788 | summary = 2 | 16; |
| 16789 | } |
| 16790 | else if (INSN(27,24) == BITS4(0,0,0,1) && INSN(22,21) == BITS2(0,1)) { |
| 16791 | summary = 2 | 32; |
| 16792 | } |
| 16793 | else if (INSN(27,24) == BITS4(0,0,0,0) && INSN(22,21) == BITS2(1,0)) { |
| 16794 | summary = 3 | 16; |
| 16795 | } |
| 16796 | else if (INSN(27,24) == BITS4(0,0,0,0) && INSN(22,21) == BITS2(0,0)) { |
| 16797 | summary = 3 | 32; |
| 16798 | } |
| 16799 | else goto after_load_store_sbyte_or_hword; |
| 16800 | |
| 16801 | { UInt rN = (insn >> 16) & 0xF; /* 19:16 */ |
| 16802 | UInt rD = (insn >> 12) & 0xF; /* 15:12 */ |
| 16803 | UInt rM = (insn >> 0) & 0xF; /* 3:0 */ |
| 16804 | UInt bU = (insn >> 23) & 1; /* 23 U=1 offset+, U=0 offset- */ |
| 16805 | UInt bL = (insn >> 20) & 1; /* 20 L=1 load, L=0 store */ |
| 16806 | UInt bH = (insn >> 5) & 1; /* H=1 halfword, H=0 byte */ |
| 16807 | UInt bS = (insn >> 6) & 1; /* S=1 signed, S=0 unsigned */ |
| 16808 | UInt imm8 = ((insn >> 4) & 0xF0) | (insn & 0xF); /* 11:8, 3:0 */ |
| 16809 | |
| 16810 | /* Skip combinations that are either meaningless or already |
| 16811 | handled by main word-or-unsigned-byte load-store |
| 16812 | instructions. */ |
| 16813 | if (bS == 0 && bH == 0) /* "unsigned byte" */ |
| 16814 | goto after_load_store_sbyte_or_hword; |
| 16815 | if (bS == 1 && bL == 0) /* "signed store" */ |
| 16816 | goto after_load_store_sbyte_or_hword; |
| 16817 | |
| 16818 | /* Require 11:8 == 0 for Rn +/- Rm cases */ |
| 16819 | if ((summary & 32) != 0 && (imm8 & 0xF0) != 0) |
| 16820 | goto after_load_store_sbyte_or_hword; |
| 16821 | |
| 16822 | /* Skip some invalid cases, which would lead to two competing |
| 16823 | updates to the same register, or which are otherwise |
| 16824 | disallowed by the spec. */ |
| 16825 | switch (summary) { |
| 16826 | case 1 | 16: |
| 16827 | break; |
| 16828 | case 1 | 32: |
| 16829 | if (rM == 15) goto after_load_store_sbyte_or_hword; |
| 16830 | break; |
| 16831 | case 2 | 16: case 3 | 16: |
| 16832 | if (rN == 15) goto after_load_store_sbyte_or_hword; |
| 16833 | if (bL == 1 && rN == rD) goto after_load_store_sbyte_or_hword; |
| 16834 | break; |
| 16835 | case 2 | 32: case 3 | 32: |
| 16836 | if (rM == 15) goto after_load_store_sbyte_or_hword; |
| 16837 | if (rN == 15) goto after_load_store_sbyte_or_hword; |
| 16838 | if (rN == rM) goto after_load_store_sbyte_or_hword; |
| 16839 | if (bL == 1 && rN == rD) goto after_load_store_sbyte_or_hword; |
| 16840 | break; |
| 16841 | default: |
| 16842 | vassert(0); |
| 16843 | } |
| 16844 | |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16845 | /* If this is a branch, make it unconditional at this point. |
| 16846 | Doing conditional branches in-line is too complex (for now). |
| 16847 | Note that you'd have to be insane to use any of these loads to |
| 16848 | do a branch, since they only load 16 bits at most, but we |
| 16849 | handle it just in case. */ |
| 16850 | if (bL == 1 && rD == 15 && condT != IRTemp_INVALID) { |
| 16851 | // go uncond |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16852 | mk_skip_over_A32_if_cond_is_false( condT ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16853 | condT = IRTemp_INVALID; |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16854 | // now uncond |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16855 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16856 | |
| 16857 | /* compute the effective address. Bind it to a tmp since we |
| 16858 | may need to use it twice. */ |
| 16859 | IRExpr* eaE = NULL; |
| 16860 | switch (summary & 0xF0) { |
| 16861 | case 16: |
| 16862 | eaE = mk_EA_reg_plusminus_imm8( rN, bU, imm8, dis_buf ); |
| 16863 | break; |
| 16864 | case 32: |
| 16865 | eaE = mk_EA_reg_plusminus_reg( rN, bU, rM, dis_buf ); |
| 16866 | break; |
| 16867 | } |
| 16868 | vassert(eaE); |
| 16869 | IRTemp eaT = newTemp(Ity_I32); |
| 16870 | assign(eaT, eaE); |
| 16871 | |
| 16872 | /* get the old Rn value */ |
| 16873 | IRTemp rnT = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 16874 | assign(rnT, getIRegA(rN)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16875 | |
| 16876 | /* decide on the transfer address */ |
| 16877 | IRTemp taT = IRTemp_INVALID; |
| 16878 | switch (summary & 0x0F) { |
| 16879 | case 1: case 2: taT = eaT; break; |
| 16880 | case 3: taT = rnT; break; |
| 16881 | } |
| 16882 | vassert(taT != IRTemp_INVALID); |
| 16883 | |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16884 | /* ll previous value of rD, for dealing with conditional loads */ |
| 16885 | IRTemp llOldRd = newTemp(Ity_I32); |
| 16886 | assign(llOldRd, llGetIReg(rD)); |
| 16887 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16888 | /* halfword store H 1 L 0 S 0 |
| 16889 | uhalf load H 1 L 1 S 0 |
| 16890 | shalf load H 1 L 1 S 1 |
| 16891 | sbyte load H 0 L 1 S 1 |
| 16892 | */ |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 16893 | const HChar* name = NULL; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16894 | /* generate the transfer */ |
| 16895 | /**/ if (bH == 1 && bL == 0 && bS == 0) { // halfword store |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16896 | storeGuardedLE( mkexpr(taT), |
| 16897 | unop(Iop_32to16, getIRegA(rD)), condT ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16898 | name = "strh"; |
| 16899 | } |
| 16900 | else if (bH == 1 && bL == 1 && bS == 0) { // uhalf load |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16901 | IRTemp newRd = newTemp(Ity_I32); |
| 16902 | loadGuardedLE( newRd, ILGop_16Uto32, |
| 16903 | mkexpr(taT), mkexpr(llOldRd), condT ); |
| 16904 | putIRegA( rD, mkexpr(newRd), IRTemp_INVALID, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16905 | name = "ldrh"; |
| 16906 | } |
| 16907 | else if (bH == 1 && bL == 1 && bS == 1) { // shalf load |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16908 | IRTemp newRd = newTemp(Ity_I32); |
| 16909 | loadGuardedLE( newRd, ILGop_16Sto32, |
| 16910 | mkexpr(taT), mkexpr(llOldRd), condT ); |
| 16911 | putIRegA( rD, mkexpr(newRd), IRTemp_INVALID, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16912 | name = "ldrsh"; |
| 16913 | } |
| 16914 | else if (bH == 0 && bL == 1 && bS == 1) { // sbyte load |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16915 | IRTemp newRd = newTemp(Ity_I32); |
| 16916 | loadGuardedLE( newRd, ILGop_8Sto32, |
| 16917 | mkexpr(taT), mkexpr(llOldRd), condT ); |
| 16918 | putIRegA( rD, mkexpr(newRd), IRTemp_INVALID, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16919 | name = "ldrsb"; |
| 16920 | } |
| 16921 | else |
| 16922 | vassert(0); // should be assured by logic above |
| 16923 | |
| 16924 | /* Update Rn if necessary. */ |
| 16925 | switch (summary & 0x0F) { |
| 16926 | case 2: case 3: |
| 16927 | // should be assured by logic above: |
| 16928 | if (bL == 1) |
| 16929 | vassert(rD != rN); /* since we just wrote rD */ |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 16930 | putIRegA( rN, mkexpr(eaT), condT, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16931 | break; |
| 16932 | } |
| 16933 | |
| 16934 | switch (summary & 0x0F) { |
| 16935 | case 1: DIP("%s%s r%u, %s\n", name, nCC(INSN_COND), rD, dis_buf); |
| 16936 | break; |
| 16937 | case 2: DIP("%s%s r%u, %s! (at-EA-then-Rn=EA)\n", |
| 16938 | name, nCC(INSN_COND), rD, dis_buf); |
| 16939 | break; |
| 16940 | case 3: DIP("%s%s r%u, %s! (at-Rn-then-Rn=EA)\n", |
| 16941 | name, nCC(INSN_COND), rD, dis_buf); |
| 16942 | break; |
| 16943 | default: vassert(0); |
| 16944 | } |
| 16945 | |
| 16946 | /* XXX deal with alignment constraints */ |
| 16947 | |
| 16948 | goto decode_success; |
| 16949 | |
| 16950 | /* Complications: |
| 16951 | |
| 16952 | For all loads: if the Amode specifies base register |
| 16953 | writeback, and the same register is specified for Rd and Rn, |
| 16954 | the results are UNPREDICTABLE. |
| 16955 | |
| 16956 | For all loads and stores: if R15 is written, branch to |
| 16957 | that address afterwards. |
| 16958 | |
| 16959 | Misaligned halfword stores => Unpredictable |
| 16960 | Misaligned halfword loads => Unpredictable |
| 16961 | */ |
| 16962 | } |
| 16963 | |
| 16964 | after_load_store_sbyte_or_hword: |
| 16965 | |
| 16966 | /* --------------------- Load/store multiple -------------- */ |
| 16967 | // LD/STMIA LD/STMIB LD/STMDA LD/STMDB |
| 16968 | // Remarkably complex and difficult to get right |
| 16969 | // match 27:20 as 100XX0WL |
| 16970 | if (BITS8(1,0,0,0,0,0,0,0) == (INSN(27,20) & BITS8(1,1,1,0,0,1,0,0))) { |
| 16971 | // A5-50 LD/STMIA cond 1000 10WL Rn RegList |
| 16972 | // A5-51 LD/STMIB cond 1001 10WL Rn RegList |
| 16973 | // A5-53 LD/STMDA cond 1000 00WL Rn RegList |
| 16974 | // A5-53 LD/STMDB cond 1001 00WL Rn RegList |
| 16975 | // 28 24 20 16 0 |
| 16976 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 16977 | UInt bINC = (insn >> 23) & 1; |
| 16978 | UInt bBEFORE = (insn >> 24) & 1; |
| 16979 | |
| 16980 | UInt bL = (insn >> 20) & 1; /* load=1, store=0 */ |
| 16981 | UInt bW = (insn >> 21) & 1; /* Rn wback=1, no wback=0 */ |
| 16982 | UInt rN = (insn >> 16) & 0xF; |
| 16983 | UInt regList = insn & 0xFFFF; |
| 16984 | /* Skip some invalid cases, which would lead to two competing |
| 16985 | updates to the same register, or which are otherwise |
| 16986 | disallowed by the spec. Note the test above has required |
| 16987 | that S == 0, since that looks like a kernel-mode only thing. |
| 16988 | Done by forcing the real pattern, viz 100XXSWL to actually be |
| 16989 | 100XX0WL. */ |
| 16990 | if (rN == 15) goto after_load_store_multiple; |
| 16991 | // reglist can't be empty |
| 16992 | if (regList == 0) goto after_load_store_multiple; |
| 16993 | // if requested to writeback Rn, and this is a load instruction, |
| 16994 | // then Rn can't appear in RegList, since we'd have two competing |
| 16995 | // new values for Rn. We do however accept this case for store |
| 16996 | // instructions. |
| 16997 | if (bW == 1 && bL == 1 && ((1 << rN) & regList) > 0) |
| 16998 | goto after_load_store_multiple; |
| 16999 | |
| 17000 | /* Now, we can't do a conditional load or store, since that very |
| 17001 | likely will generate an exception. So we have to take a side |
| 17002 | exit at this point if the condition is false. */ |
| 17003 | if (condT != IRTemp_INVALID) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17004 | mk_skip_over_A32_if_cond_is_false( condT ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17005 | condT = IRTemp_INVALID; |
| 17006 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17007 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17008 | /* Ok, now we're unconditional. Generate the IR. */ |
| 17009 | mk_ldm_stm( True/*arm*/, rN, bINC, bBEFORE, bW, bL, regList ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17010 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17011 | DIP("%sm%c%c%s r%u%s, {0x%04x}\n", |
| 17012 | bL == 1 ? "ld" : "st", bINC ? 'i' : 'd', bBEFORE ? 'b' : 'a', |
| 17013 | nCC(INSN_COND), |
| 17014 | rN, bW ? "!" : "", regList); |
| 17015 | |
| 17016 | goto decode_success; |
| 17017 | } |
| 17018 | |
| 17019 | after_load_store_multiple: |
| 17020 | |
| 17021 | /* --------------------- Control flow --------------------- */ |
| 17022 | // B, BL (Branch, or Branch-and-Link, to immediate offset) |
| 17023 | // |
| 17024 | if (BITS8(1,0,1,0,0,0,0,0) == (INSN(27,20) & BITS8(1,1,1,0,0,0,0,0))) { |
| 17025 | UInt link = (insn >> 24) & 1; |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 17026 | UInt uimm24 = insn & ((1<<24)-1); uimm24 <<= 8; |
| 17027 | Int simm24 = (Int)uimm24; simm24 >>= 8; |
| 17028 | UInt dst = guest_R15_curr_instr_notENC + 8 + (((UInt)simm24) << 2); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17029 | IRJumpKind jk = link ? Ijk_Call : Ijk_Boring; |
| 17030 | if (link) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17031 | putIRegA(14, mkU32(guest_R15_curr_instr_notENC + 4), |
| 17032 | condT, Ijk_Boring); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17033 | } |
| 17034 | if (condT == IRTemp_INVALID) { |
| 17035 | /* unconditional transfer to 'dst'. See if we can simply |
| 17036 | continue tracing at the destination. */ |
florian | beac530 | 2014-12-31 12:09:38 +0000 | [diff] [blame] | 17037 | if (resteerOkFn( callback_opaque, dst )) { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17038 | /* yes */ |
sewardj | 984d9b1 | 2010-01-15 10:53:21 +0000 | [diff] [blame] | 17039 | dres.whatNext = Dis_ResteerU; |
florian | 0eaa35f | 2015-01-02 13:34:15 +0000 | [diff] [blame] | 17040 | dres.continueAt = dst; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17041 | } else { |
| 17042 | /* no; terminate the SB at this point. */ |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 17043 | llPutIReg(15, mkU32(dst)); |
| 17044 | dres.jk_StopHere = jk; |
| 17045 | dres.whatNext = Dis_StopHere; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17046 | } |
| 17047 | DIP("b%s 0x%x\n", link ? "l" : "", dst); |
| 17048 | } else { |
| 17049 | /* conditional transfer to 'dst' */ |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 17050 | const HChar* comment = ""; |
sewardj | 82f5688 | 2010-01-17 09:36:11 +0000 | [diff] [blame] | 17051 | |
| 17052 | /* First see if we can do some speculative chasing into one |
| 17053 | arm or the other. Be conservative and only chase if |
| 17054 | !link, that is, this is a normal conditional branch to a |
| 17055 | known destination. */ |
| 17056 | if (!link |
| 17057 | && resteerCisOk |
| 17058 | && vex_control.guest_chase_cond |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17059 | && dst < guest_R15_curr_instr_notENC |
florian | beac530 | 2014-12-31 12:09:38 +0000 | [diff] [blame] | 17060 | && resteerOkFn( callback_opaque, dst) ) { |
sewardj | 82f5688 | 2010-01-17 09:36:11 +0000 | [diff] [blame] | 17061 | /* Speculation: assume this backward branch is taken. So |
| 17062 | we need to emit a side-exit to the insn following this |
| 17063 | one, on the negation of the condition, and continue at |
| 17064 | the branch target address (dst). */ |
| 17065 | stmt( IRStmt_Exit( unop(Iop_Not1, |
| 17066 | unop(Iop_32to1, mkexpr(condT))), |
| 17067 | Ijk_Boring, |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 17068 | IRConst_U32(guest_R15_curr_instr_notENC+4), |
| 17069 | OFFB_R15T )); |
sewardj | 82f5688 | 2010-01-17 09:36:11 +0000 | [diff] [blame] | 17070 | dres.whatNext = Dis_ResteerC; |
florian | 0eaa35f | 2015-01-02 13:34:15 +0000 | [diff] [blame] | 17071 | dres.continueAt = (Addr32)dst; |
sewardj | 82f5688 | 2010-01-17 09:36:11 +0000 | [diff] [blame] | 17072 | comment = "(assumed taken)"; |
| 17073 | } |
| 17074 | else |
| 17075 | if (!link |
| 17076 | && resteerCisOk |
| 17077 | && vex_control.guest_chase_cond |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17078 | && dst >= guest_R15_curr_instr_notENC |
sewardj | 82f5688 | 2010-01-17 09:36:11 +0000 | [diff] [blame] | 17079 | && resteerOkFn( callback_opaque, |
florian | beac530 | 2014-12-31 12:09:38 +0000 | [diff] [blame] | 17080 | guest_R15_curr_instr_notENC+4) ) { |
sewardj | 82f5688 | 2010-01-17 09:36:11 +0000 | [diff] [blame] | 17081 | /* Speculation: assume this forward branch is not taken. |
| 17082 | So we need to emit a side-exit to dst (the dest) and |
| 17083 | continue disassembling at the insn immediately |
| 17084 | following this one. */ |
| 17085 | stmt( IRStmt_Exit( unop(Iop_32to1, mkexpr(condT)), |
| 17086 | Ijk_Boring, |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 17087 | IRConst_U32(dst), |
| 17088 | OFFB_R15T )); |
sewardj | 82f5688 | 2010-01-17 09:36:11 +0000 | [diff] [blame] | 17089 | dres.whatNext = Dis_ResteerC; |
florian | 0eaa35f | 2015-01-02 13:34:15 +0000 | [diff] [blame] | 17090 | dres.continueAt = guest_R15_curr_instr_notENC+4; |
sewardj | 82f5688 | 2010-01-17 09:36:11 +0000 | [diff] [blame] | 17091 | comment = "(assumed not taken)"; |
| 17092 | } |
| 17093 | else { |
| 17094 | /* Conservative default translation - end the block at |
| 17095 | this point. */ |
| 17096 | stmt( IRStmt_Exit( unop(Iop_32to1, mkexpr(condT)), |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 17097 | jk, IRConst_U32(dst), OFFB_R15T )); |
| 17098 | llPutIReg(15, mkU32(guest_R15_curr_instr_notENC + 4)); |
| 17099 | dres.jk_StopHere = Ijk_Boring; |
| 17100 | dres.whatNext = Dis_StopHere; |
sewardj | 82f5688 | 2010-01-17 09:36:11 +0000 | [diff] [blame] | 17101 | } |
| 17102 | DIP("b%s%s 0x%x %s\n", link ? "l" : "", nCC(INSN_COND), |
| 17103 | dst, comment); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17104 | } |
| 17105 | goto decode_success; |
| 17106 | } |
| 17107 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17108 | // B, BL (Branch, or Branch-and-Link, to a register) |
| 17109 | // NB: interworking branch |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17110 | if (INSN(27,20) == BITS8(0,0,0,1,0,0,1,0) |
| 17111 | && INSN(19,12) == BITS8(1,1,1,1,1,1,1,1) |
| 17112 | && (INSN(11,4) == BITS8(1,1,1,1,0,0,1,1) |
| 17113 | || INSN(11,4) == BITS8(1,1,1,1,0,0,0,1))) { |
sewardj | b706a0f | 2011-07-19 08:20:24 +0000 | [diff] [blame] | 17114 | IRTemp dst = newTemp(Ity_I32); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17115 | UInt link = (INSN(11,4) >> 1) & 1; |
| 17116 | UInt rM = INSN(3,0); |
| 17117 | // we don't decode the case (link && rM == 15), as that's |
| 17118 | // Unpredictable. |
| 17119 | if (!(link && rM == 15)) { |
| 17120 | if (condT != IRTemp_INVALID) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17121 | mk_skip_over_A32_if_cond_is_false( condT ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17122 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17123 | // rM contains an interworking address exactly as we require |
| 17124 | // (with continuation CPSR.T in bit 0), so we can use it |
| 17125 | // as-is, with no masking. |
sewardj | b706a0f | 2011-07-19 08:20:24 +0000 | [diff] [blame] | 17126 | assign( dst, getIRegA(rM) ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17127 | if (link) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17128 | putIRegA( 14, mkU32(guest_R15_curr_instr_notENC + 4), |
| 17129 | IRTemp_INVALID/*because AL*/, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17130 | } |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 17131 | llPutIReg(15, mkexpr(dst)); |
| 17132 | dres.jk_StopHere = link ? Ijk_Call |
| 17133 | : (rM == 14 ? Ijk_Ret : Ijk_Boring); |
| 17134 | dres.whatNext = Dis_StopHere; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17135 | if (condT == IRTemp_INVALID) { |
| 17136 | DIP("b%sx r%u\n", link ? "l" : "", rM); |
| 17137 | } else { |
| 17138 | DIP("b%sx%s r%u\n", link ? "l" : "", nCC(INSN_COND), rM); |
| 17139 | } |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 17140 | goto decode_success; |
| 17141 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17142 | /* else: (link && rM == 15): just fall through */ |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 17143 | } |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 17144 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17145 | /* --- NB: ARM interworking branches are in NV space, hence |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 17146 | are handled elsewhere by decode_NV_instruction_ARMv7_and_below. |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17147 | --- |
| 17148 | */ |
| 17149 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17150 | /* --------------------- Clz --------------------- */ |
| 17151 | // CLZ |
| 17152 | if (INSN(27,20) == BITS8(0,0,0,1,0,1,1,0) |
| 17153 | && INSN(19,16) == BITS4(1,1,1,1) |
| 17154 | && INSN(11,4) == BITS8(1,1,1,1,0,0,0,1)) { |
| 17155 | UInt rD = INSN(15,12); |
| 17156 | UInt rM = INSN(3,0); |
| 17157 | IRTemp arg = newTemp(Ity_I32); |
| 17158 | IRTemp res = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17159 | assign(arg, getIRegA(rM)); |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 17160 | assign(res, IRExpr_ITE( |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 17161 | binop(Iop_CmpEQ32, mkexpr(arg), mkU32(0)), |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 17162 | mkU32(32), |
| 17163 | unop(Iop_Clz32, mkexpr(arg)) |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17164 | )); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17165 | putIRegA(rD, mkexpr(res), condT, Ijk_Boring); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17166 | DIP("clz%s r%u, r%u\n", nCC(INSN_COND), rD, rM); |
| 17167 | goto decode_success; |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 17168 | } |
cerion | c60c01e | 2004-12-02 20:19:22 +0000 | [diff] [blame] | 17169 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17170 | /* --------------------- Mul etc --------------------- */ |
| 17171 | // MUL |
| 17172 | if (BITS8(0,0,0,0,0,0,0,0) == (INSN(27,20) & BITS8(1,1,1,1,1,1,1,0)) |
| 17173 | && INSN(15,12) == BITS4(0,0,0,0) |
| 17174 | && INSN(7,4) == BITS4(1,0,0,1)) { |
| 17175 | UInt bitS = (insn >> 20) & 1; /* 20:20 */ |
| 17176 | UInt rD = INSN(19,16); |
| 17177 | UInt rS = INSN(11,8); |
| 17178 | UInt rM = INSN(3,0); |
| 17179 | if (rD == 15 || rM == 15 || rS == 15) { |
| 17180 | /* Unpredictable; don't decode; fall through */ |
| 17181 | } else { |
| 17182 | IRTemp argL = newTemp(Ity_I32); |
| 17183 | IRTemp argR = newTemp(Ity_I32); |
| 17184 | IRTemp res = newTemp(Ity_I32); |
| 17185 | IRTemp oldC = IRTemp_INVALID; |
| 17186 | IRTemp oldV = IRTemp_INVALID; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17187 | assign( argL, getIRegA(rM)); |
| 17188 | assign( argR, getIRegA(rS)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17189 | assign( res, binop(Iop_Mul32, mkexpr(argL), mkexpr(argR)) ); |
| 17190 | if (bitS) { |
| 17191 | oldC = newTemp(Ity_I32); |
| 17192 | assign(oldC, mk_armg_calculate_flag_c()); |
| 17193 | oldV = newTemp(Ity_I32); |
| 17194 | assign(oldV, mk_armg_calculate_flag_v()); |
cerion | b85e8bb | 2005-02-16 08:54:33 +0000 | [diff] [blame] | 17195 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17196 | // now update guest state |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17197 | putIRegA( rD, mkexpr(res), condT, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17198 | if (bitS) { |
| 17199 | IRTemp pair = newTemp(Ity_I32); |
| 17200 | assign( pair, binop(Iop_Or32, |
| 17201 | binop(Iop_Shl32, mkexpr(oldC), mkU8(1)), |
| 17202 | mkexpr(oldV)) ); |
| 17203 | setFlags_D1_ND( ARMG_CC_OP_MUL, res, pair, condT ); |
cerion | b85e8bb | 2005-02-16 08:54:33 +0000 | [diff] [blame] | 17204 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17205 | DIP("mul%c%s r%u, r%u, r%u\n", |
| 17206 | bitS ? 's' : ' ', nCC(INSN_COND), rD, rM, rS); |
| 17207 | goto decode_success; |
cerion | b85e8bb | 2005-02-16 08:54:33 +0000 | [diff] [blame] | 17208 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17209 | /* fall through */ |
| 17210 | } |
cerion | f7da63d | 2004-12-09 19:04:57 +0000 | [diff] [blame] | 17211 | |
sewardj | 8bde7f1 | 2013-04-11 13:57:43 +0000 | [diff] [blame] | 17212 | /* --------------------- Integer Divides --------------------- */ |
| 17213 | // SDIV |
| 17214 | if (BITS8(0,1,1,1,0,0,0,1) == INSN(27,20) |
| 17215 | && INSN(15,12) == BITS4(1,1,1,1) |
| 17216 | && INSN(7,4) == BITS4(0,0,0,1)) { |
| 17217 | UInt rD = INSN(19,16); |
| 17218 | UInt rM = INSN(11,8); |
| 17219 | UInt rN = INSN(3,0); |
| 17220 | if (rD == 15 || rM == 15 || rN == 15) { |
| 17221 | /* Unpredictable; don't decode; fall through */ |
| 17222 | } else { |
| 17223 | IRTemp res = newTemp(Ity_I32); |
| 17224 | IRTemp argL = newTemp(Ity_I32); |
| 17225 | IRTemp argR = newTemp(Ity_I32); |
| 17226 | assign(argL, getIRegA(rN)); |
| 17227 | assign(argR, getIRegA(rM)); |
| 17228 | assign(res, binop(Iop_DivS32, mkexpr(argL), mkexpr(argR))); |
| 17229 | putIRegA(rD, mkexpr(res), condT, Ijk_Boring); |
| 17230 | DIP("sdiv r%u, r%u, r%u\n", rD, rN, rM); |
| 17231 | goto decode_success; |
| 17232 | } |
| 17233 | } |
| 17234 | |
| 17235 | // UDIV |
| 17236 | if (BITS8(0,1,1,1,0,0,1,1) == INSN(27,20) |
| 17237 | && INSN(15,12) == BITS4(1,1,1,1) |
| 17238 | && INSN(7,4) == BITS4(0,0,0,1)) { |
| 17239 | UInt rD = INSN(19,16); |
| 17240 | UInt rM = INSN(11,8); |
| 17241 | UInt rN = INSN(3,0); |
| 17242 | if (rD == 15 || rM == 15 || rN == 15) { |
| 17243 | /* Unpredictable; don't decode; fall through */ |
| 17244 | } else { |
| 17245 | IRTemp res = newTemp(Ity_I32); |
| 17246 | IRTemp argL = newTemp(Ity_I32); |
| 17247 | IRTemp argR = newTemp(Ity_I32); |
| 17248 | assign(argL, getIRegA(rN)); |
| 17249 | assign(argR, getIRegA(rM)); |
| 17250 | assign(res, binop(Iop_DivU32, mkexpr(argL), mkexpr(argR))); |
| 17251 | putIRegA(rD, mkexpr(res), condT, Ijk_Boring); |
| 17252 | DIP("udiv r%u, r%u, r%u\n", rD, rN, rM); |
| 17253 | goto decode_success; |
| 17254 | } |
| 17255 | } |
| 17256 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17257 | // MLA, MLS |
| 17258 | if (BITS8(0,0,0,0,0,0,1,0) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,0)) |
| 17259 | && INSN(7,4) == BITS4(1,0,0,1)) { |
| 17260 | UInt bitS = (insn >> 20) & 1; /* 20:20 */ |
| 17261 | UInt isMLS = (insn >> 22) & 1; /* 22:22 */ |
| 17262 | UInt rD = INSN(19,16); |
| 17263 | UInt rN = INSN(15,12); |
| 17264 | UInt rS = INSN(11,8); |
| 17265 | UInt rM = INSN(3,0); |
| 17266 | if (bitS == 1 && isMLS == 1) { |
| 17267 | /* This isn't allowed (MLS that sets flags). don't decode; |
| 17268 | fall through */ |
cerion | b85e8bb | 2005-02-16 08:54:33 +0000 | [diff] [blame] | 17269 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17270 | else |
| 17271 | if (rD == 15 || rM == 15 || rS == 15 || rN == 15) { |
| 17272 | /* Unpredictable; don't decode; fall through */ |
| 17273 | } else { |
| 17274 | IRTemp argL = newTemp(Ity_I32); |
| 17275 | IRTemp argR = newTemp(Ity_I32); |
| 17276 | IRTemp argP = newTemp(Ity_I32); |
| 17277 | IRTemp res = newTemp(Ity_I32); |
| 17278 | IRTemp oldC = IRTemp_INVALID; |
| 17279 | IRTemp oldV = IRTemp_INVALID; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17280 | assign( argL, getIRegA(rM)); |
| 17281 | assign( argR, getIRegA(rS)); |
| 17282 | assign( argP, getIRegA(rN)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17283 | assign( res, binop(isMLS ? Iop_Sub32 : Iop_Add32, |
| 17284 | mkexpr(argP), |
| 17285 | binop(Iop_Mul32, mkexpr(argL), mkexpr(argR)) )); |
| 17286 | if (bitS) { |
| 17287 | vassert(!isMLS); // guaranteed above |
| 17288 | oldC = newTemp(Ity_I32); |
| 17289 | assign(oldC, mk_armg_calculate_flag_c()); |
| 17290 | oldV = newTemp(Ity_I32); |
| 17291 | assign(oldV, mk_armg_calculate_flag_v()); |
| 17292 | } |
| 17293 | // now update guest state |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17294 | putIRegA( rD, mkexpr(res), condT, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17295 | if (bitS) { |
| 17296 | IRTemp pair = newTemp(Ity_I32); |
| 17297 | assign( pair, binop(Iop_Or32, |
| 17298 | binop(Iop_Shl32, mkexpr(oldC), mkU8(1)), |
| 17299 | mkexpr(oldV)) ); |
| 17300 | setFlags_D1_ND( ARMG_CC_OP_MUL, res, pair, condT ); |
| 17301 | } |
| 17302 | DIP("ml%c%c%s r%u, r%u, r%u, r%u\n", |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17303 | isMLS ? 's' : 'a', bitS ? 's' : ' ', |
| 17304 | nCC(INSN_COND), rD, rM, rS, rN); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17305 | goto decode_success; |
| 17306 | } |
| 17307 | /* fall through */ |
| 17308 | } |
| 17309 | |
| 17310 | // SMULL, UMULL |
| 17311 | if (BITS8(0,0,0,0,1,0,0,0) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,0)) |
| 17312 | && INSN(7,4) == BITS4(1,0,0,1)) { |
| 17313 | UInt bitS = (insn >> 20) & 1; /* 20:20 */ |
| 17314 | UInt rDhi = INSN(19,16); |
| 17315 | UInt rDlo = INSN(15,12); |
| 17316 | UInt rS = INSN(11,8); |
| 17317 | UInt rM = INSN(3,0); |
| 17318 | UInt isS = (INSN(27,20) >> 2) & 1; /* 22:22 */ |
| 17319 | if (rDhi == 15 || rDlo == 15 || rM == 15 || rS == 15 || rDhi == rDlo) { |
| 17320 | /* Unpredictable; don't decode; fall through */ |
| 17321 | } else { |
| 17322 | IRTemp argL = newTemp(Ity_I32); |
| 17323 | IRTemp argR = newTemp(Ity_I32); |
| 17324 | IRTemp res = newTemp(Ity_I64); |
| 17325 | IRTemp resHi = newTemp(Ity_I32); |
| 17326 | IRTemp resLo = newTemp(Ity_I32); |
| 17327 | IRTemp oldC = IRTemp_INVALID; |
| 17328 | IRTemp oldV = IRTemp_INVALID; |
| 17329 | IROp mulOp = isS ? Iop_MullS32 : Iop_MullU32; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17330 | assign( argL, getIRegA(rM)); |
| 17331 | assign( argR, getIRegA(rS)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17332 | assign( res, binop(mulOp, mkexpr(argL), mkexpr(argR)) ); |
| 17333 | assign( resHi, unop(Iop_64HIto32, mkexpr(res)) ); |
| 17334 | assign( resLo, unop(Iop_64to32, mkexpr(res)) ); |
| 17335 | if (bitS) { |
| 17336 | oldC = newTemp(Ity_I32); |
| 17337 | assign(oldC, mk_armg_calculate_flag_c()); |
| 17338 | oldV = newTemp(Ity_I32); |
| 17339 | assign(oldV, mk_armg_calculate_flag_v()); |
| 17340 | } |
| 17341 | // now update guest state |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17342 | putIRegA( rDhi, mkexpr(resHi), condT, Ijk_Boring ); |
| 17343 | putIRegA( rDlo, mkexpr(resLo), condT, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17344 | if (bitS) { |
| 17345 | IRTemp pair = newTemp(Ity_I32); |
| 17346 | assign( pair, binop(Iop_Or32, |
| 17347 | binop(Iop_Shl32, mkexpr(oldC), mkU8(1)), |
| 17348 | mkexpr(oldV)) ); |
| 17349 | setFlags_D1_D2_ND( ARMG_CC_OP_MULL, resLo, resHi, pair, condT ); |
| 17350 | } |
| 17351 | DIP("%cmull%c%s r%u, r%u, r%u, r%u\n", |
| 17352 | isS ? 's' : 'u', bitS ? 's' : ' ', |
| 17353 | nCC(INSN_COND), rDlo, rDhi, rM, rS); |
| 17354 | goto decode_success; |
| 17355 | } |
| 17356 | /* fall through */ |
| 17357 | } |
| 17358 | |
| 17359 | // SMLAL, UMLAL |
| 17360 | if (BITS8(0,0,0,0,1,0,1,0) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,0)) |
| 17361 | && INSN(7,4) == BITS4(1,0,0,1)) { |
| 17362 | UInt bitS = (insn >> 20) & 1; /* 20:20 */ |
| 17363 | UInt rDhi = INSN(19,16); |
| 17364 | UInt rDlo = INSN(15,12); |
| 17365 | UInt rS = INSN(11,8); |
| 17366 | UInt rM = INSN(3,0); |
| 17367 | UInt isS = (INSN(27,20) >> 2) & 1; /* 22:22 */ |
| 17368 | if (rDhi == 15 || rDlo == 15 || rM == 15 || rS == 15 || rDhi == rDlo) { |
| 17369 | /* Unpredictable; don't decode; fall through */ |
| 17370 | } else { |
| 17371 | IRTemp argL = newTemp(Ity_I32); |
| 17372 | IRTemp argR = newTemp(Ity_I32); |
| 17373 | IRTemp old = newTemp(Ity_I64); |
| 17374 | IRTemp res = newTemp(Ity_I64); |
| 17375 | IRTemp resHi = newTemp(Ity_I32); |
| 17376 | IRTemp resLo = newTemp(Ity_I32); |
| 17377 | IRTemp oldC = IRTemp_INVALID; |
| 17378 | IRTemp oldV = IRTemp_INVALID; |
| 17379 | IROp mulOp = isS ? Iop_MullS32 : Iop_MullU32; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17380 | assign( argL, getIRegA(rM)); |
| 17381 | assign( argR, getIRegA(rS)); |
| 17382 | assign( old, binop(Iop_32HLto64, getIRegA(rDhi), getIRegA(rDlo)) ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17383 | assign( res, binop(Iop_Add64, |
| 17384 | mkexpr(old), |
| 17385 | binop(mulOp, mkexpr(argL), mkexpr(argR))) ); |
| 17386 | assign( resHi, unop(Iop_64HIto32, mkexpr(res)) ); |
| 17387 | assign( resLo, unop(Iop_64to32, mkexpr(res)) ); |
| 17388 | if (bitS) { |
| 17389 | oldC = newTemp(Ity_I32); |
| 17390 | assign(oldC, mk_armg_calculate_flag_c()); |
| 17391 | oldV = newTemp(Ity_I32); |
| 17392 | assign(oldV, mk_armg_calculate_flag_v()); |
| 17393 | } |
| 17394 | // now update guest state |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17395 | putIRegA( rDhi, mkexpr(resHi), condT, Ijk_Boring ); |
| 17396 | putIRegA( rDlo, mkexpr(resLo), condT, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17397 | if (bitS) { |
| 17398 | IRTemp pair = newTemp(Ity_I32); |
| 17399 | assign( pair, binop(Iop_Or32, |
| 17400 | binop(Iop_Shl32, mkexpr(oldC), mkU8(1)), |
| 17401 | mkexpr(oldV)) ); |
| 17402 | setFlags_D1_D2_ND( ARMG_CC_OP_MULL, resLo, resHi, pair, condT ); |
| 17403 | } |
| 17404 | DIP("%cmlal%c%s r%u, r%u, r%u, r%u\n", |
| 17405 | isS ? 's' : 'u', bitS ? 's' : ' ', nCC(INSN_COND), |
| 17406 | rDlo, rDhi, rM, rS); |
| 17407 | goto decode_success; |
| 17408 | } |
| 17409 | /* fall through */ |
| 17410 | } |
| 17411 | |
sewardj | abf3945 | 2012-12-12 00:16:41 +0000 | [diff] [blame] | 17412 | // UMAAL |
| 17413 | if (BITS8(0,0,0,0,0,1,0,0) == INSN(27,20) && INSN(7,4) == BITS4(1,0,0,1)) { |
| 17414 | UInt rDhi = INSN(19,16); |
| 17415 | UInt rDlo = INSN(15,12); |
| 17416 | UInt rM = INSN(11,8); |
| 17417 | UInt rN = INSN(3,0); |
| 17418 | if (rDlo == 15 || rDhi == 15 || rN == 15 || rM == 15 || rDhi == rDlo) { |
| 17419 | /* Unpredictable; don't decode; fall through */ |
| 17420 | } else { |
| 17421 | IRTemp argN = newTemp(Ity_I32); |
| 17422 | IRTemp argM = newTemp(Ity_I32); |
| 17423 | IRTemp argDhi = newTemp(Ity_I32); |
| 17424 | IRTemp argDlo = newTemp(Ity_I32); |
| 17425 | IRTemp res = newTemp(Ity_I64); |
| 17426 | IRTemp resHi = newTemp(Ity_I32); |
| 17427 | IRTemp resLo = newTemp(Ity_I32); |
| 17428 | assign( argN, getIRegA(rN) ); |
| 17429 | assign( argM, getIRegA(rM) ); |
| 17430 | assign( argDhi, getIRegA(rDhi) ); |
| 17431 | assign( argDlo, getIRegA(rDlo) ); |
| 17432 | assign( res, |
| 17433 | binop(Iop_Add64, |
| 17434 | binop(Iop_Add64, |
| 17435 | binop(Iop_MullU32, mkexpr(argN), mkexpr(argM)), |
| 17436 | unop(Iop_32Uto64, mkexpr(argDhi))), |
| 17437 | unop(Iop_32Uto64, mkexpr(argDlo))) ); |
| 17438 | assign( resHi, unop(Iop_64HIto32, mkexpr(res)) ); |
| 17439 | assign( resLo, unop(Iop_64to32, mkexpr(res)) ); |
| 17440 | // now update guest state |
| 17441 | putIRegA( rDhi, mkexpr(resHi), condT, Ijk_Boring ); |
| 17442 | putIRegA( rDlo, mkexpr(resLo), condT, Ijk_Boring ); |
| 17443 | DIP("umaal %s r%u, r%u, r%u, r%u\n", |
| 17444 | nCC(INSN_COND), rDlo, rDhi, rN, rM); |
| 17445 | goto decode_success; |
| 17446 | } |
| 17447 | /* fall through */ |
| 17448 | } |
| 17449 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17450 | /* --------------------- Msr etc --------------------- */ |
| 17451 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 17452 | // MSR apsr, #imm |
| 17453 | if (INSN(27,20) == BITS8(0,0,1,1,0,0,1,0) |
| 17454 | && INSN(17,12) == BITS6(0,0,1,1,1,1)) { |
| 17455 | UInt write_ge = INSN(18,18); |
| 17456 | UInt write_nzcvq = INSN(19,19); |
| 17457 | if (write_nzcvq || write_ge) { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17458 | UInt imm = (INSN(11,0) >> 0) & 0xFF; |
| 17459 | UInt rot = 2 * ((INSN(11,0) >> 8) & 0xF); |
| 17460 | IRTemp immT = newTemp(Ity_I32); |
| 17461 | vassert(rot <= 30); |
| 17462 | imm = ROR32(imm, rot); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 17463 | assign(immT, mkU32(imm)); |
| 17464 | desynthesise_APSR( write_nzcvq, write_ge, immT, condT ); |
| 17465 | DIP("msr%s cpsr%s%sf, #0x%08x\n", nCC(INSN_COND), |
| 17466 | write_nzcvq ? "f" : "", write_ge ? "g" : "", imm); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17467 | goto decode_success; |
| 17468 | } |
| 17469 | /* fall through */ |
| 17470 | } |
| 17471 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 17472 | // MSR apsr, reg |
| 17473 | if (INSN(27,20) == BITS8(0,0,0,1,0,0,1,0) |
| 17474 | && INSN(17,12) == BITS6(0,0,1,1,1,1) |
| 17475 | && INSN(11,4) == BITS8(0,0,0,0,0,0,0,0)) { |
| 17476 | UInt rN = INSN(3,0); |
| 17477 | UInt write_ge = INSN(18,18); |
| 17478 | UInt write_nzcvq = INSN(19,19); |
| 17479 | if (rN != 15 && (write_nzcvq || write_ge)) { |
| 17480 | IRTemp rNt = newTemp(Ity_I32); |
| 17481 | assign(rNt, getIRegA(rN)); |
| 17482 | desynthesise_APSR( write_nzcvq, write_ge, rNt, condT ); |
| 17483 | DIP("msr%s cpsr_%s%s, r%u\n", nCC(INSN_COND), |
| 17484 | write_nzcvq ? "f" : "", write_ge ? "g" : "", rN); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17485 | goto decode_success; |
| 17486 | } |
| 17487 | /* fall through */ |
| 17488 | } |
| 17489 | |
| 17490 | // MRS rD, cpsr |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 17491 | if ((insn & 0x0FFF0FFF) == 0x010F0000) { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17492 | UInt rD = INSN(15,12); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 17493 | if (rD != 15) { |
| 17494 | IRTemp apsr = synthesise_APSR(); |
| 17495 | putIRegA( rD, mkexpr(apsr), condT, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17496 | DIP("mrs%s r%u, cpsr\n", nCC(INSN_COND), rD); |
| 17497 | goto decode_success; |
| 17498 | } |
| 17499 | /* fall through */ |
| 17500 | } |
| 17501 | |
| 17502 | /* --------------------- Svc --------------------- */ |
| 17503 | if (BITS8(1,1,1,1,0,0,0,0) == (INSN(27,20) & BITS8(1,1,1,1,0,0,0,0))) { |
| 17504 | UInt imm24 = (insn >> 0) & 0xFFFFFF; |
| 17505 | if (imm24 == 0) { |
| 17506 | /* A syscall. We can't do this conditionally, hence: */ |
| 17507 | if (condT != IRTemp_INVALID) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17508 | mk_skip_over_A32_if_cond_is_false( condT ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17509 | } |
| 17510 | // AL after here |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 17511 | llPutIReg(15, mkU32( guest_R15_curr_instr_notENC + 4 )); |
| 17512 | dres.jk_StopHere = Ijk_Sys_syscall; |
| 17513 | dres.whatNext = Dis_StopHere; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17514 | DIP("svc%s #0x%08x\n", nCC(INSN_COND), imm24); |
| 17515 | goto decode_success; |
| 17516 | } |
| 17517 | /* fall through */ |
| 17518 | } |
| 17519 | |
| 17520 | /* ------------------------ swp ------------------------ */ |
| 17521 | |
| 17522 | // SWP, SWPB |
| 17523 | if (BITS8(0,0,0,1,0,0,0,0) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
| 17524 | && BITS4(0,0,0,0) == INSN(11,8) |
| 17525 | && BITS4(1,0,0,1) == INSN(7,4)) { |
| 17526 | UInt rN = INSN(19,16); |
| 17527 | UInt rD = INSN(15,12); |
| 17528 | UInt rM = INSN(3,0); |
| 17529 | IRTemp tRn = newTemp(Ity_I32); |
| 17530 | IRTemp tNew = newTemp(Ity_I32); |
| 17531 | IRTemp tOld = IRTemp_INVALID; |
| 17532 | IRTemp tSC1 = newTemp(Ity_I1); |
| 17533 | UInt isB = (insn >> 22) & 1; |
| 17534 | |
| 17535 | if (rD == 15 || rN == 15 || rM == 15 || rN == rM || rN == rD) { |
| 17536 | /* undecodable; fall through */ |
| 17537 | } else { |
| 17538 | /* make unconditional */ |
| 17539 | if (condT != IRTemp_INVALID) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17540 | mk_skip_over_A32_if_cond_is_false( condT ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17541 | condT = IRTemp_INVALID; |
| 17542 | } |
| 17543 | /* Ok, now we're unconditional. Generate a LL-SC loop. */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17544 | assign(tRn, getIRegA(rN)); |
| 17545 | assign(tNew, getIRegA(rM)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17546 | if (isB) { |
| 17547 | /* swpb */ |
| 17548 | tOld = newTemp(Ity_I8); |
| 17549 | stmt( IRStmt_LLSC(Iend_LE, tOld, mkexpr(tRn), |
| 17550 | NULL/*=>isLL*/) ); |
| 17551 | stmt( IRStmt_LLSC(Iend_LE, tSC1, mkexpr(tRn), |
| 17552 | unop(Iop_32to8, mkexpr(tNew))) ); |
| 17553 | } else { |
| 17554 | /* swp */ |
| 17555 | tOld = newTemp(Ity_I32); |
| 17556 | stmt( IRStmt_LLSC(Iend_LE, tOld, mkexpr(tRn), |
| 17557 | NULL/*=>isLL*/) ); |
| 17558 | stmt( IRStmt_LLSC(Iend_LE, tSC1, mkexpr(tRn), |
| 17559 | mkexpr(tNew)) ); |
| 17560 | } |
| 17561 | stmt( IRStmt_Exit(unop(Iop_Not1, mkexpr(tSC1)), |
| 17562 | /*Ijk_NoRedir*/Ijk_Boring, |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 17563 | IRConst_U32(guest_R15_curr_instr_notENC), |
| 17564 | OFFB_R15T )); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17565 | putIRegA(rD, isB ? unop(Iop_8Uto32, mkexpr(tOld)) : mkexpr(tOld), |
| 17566 | IRTemp_INVALID, Ijk_Boring); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17567 | DIP("swp%s%s r%u, r%u, [r%u]\n", |
| 17568 | isB ? "b" : "", nCC(INSN_COND), rD, rM, rN); |
| 17569 | goto decode_success; |
| 17570 | } |
| 17571 | /* fall through */ |
| 17572 | } |
| 17573 | |
| 17574 | /* ----------------------------------------------------------- */ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17575 | /* -- ARMv6 instructions -- */ |
| 17576 | /* ----------------------------------------------------------- */ |
| 17577 | |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17578 | /* ------------------- {ldr,str}ex{,b,h,d} ------------------- */ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17579 | |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17580 | // LDREXD, LDREX, LDREXH, LDREXB |
| 17581 | if (0x01900F9F == (insn & 0x0F900FFF)) { |
| 17582 | UInt rT = INSN(15,12); |
| 17583 | UInt rN = INSN(19,16); |
| 17584 | IRType ty = Ity_INVALID; |
| 17585 | IROp widen = Iop_INVALID; |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 17586 | const HChar* nm = NULL; |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17587 | Bool valid = True; |
| 17588 | switch (INSN(22,21)) { |
| 17589 | case 0: nm = ""; ty = Ity_I32; break; |
| 17590 | case 1: nm = "d"; ty = Ity_I64; break; |
| 17591 | case 2: nm = "b"; ty = Ity_I8; widen = Iop_8Uto32; break; |
| 17592 | case 3: nm = "h"; ty = Ity_I16; widen = Iop_16Uto32; break; |
| 17593 | default: vassert(0); |
| 17594 | } |
| 17595 | if (ty == Ity_I32 || ty == Ity_I16 || ty == Ity_I8) { |
| 17596 | if (rT == 15 || rN == 15) |
| 17597 | valid = False; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17598 | } else { |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17599 | vassert(ty == Ity_I64); |
| 17600 | if ((rT & 1) == 1 || rT == 14 || rN == 15) |
| 17601 | valid = False; |
| 17602 | } |
| 17603 | if (valid) { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17604 | IRTemp res; |
| 17605 | /* make unconditional */ |
| 17606 | if (condT != IRTemp_INVALID) { |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17607 | mk_skip_over_A32_if_cond_is_false( condT ); |
| 17608 | condT = IRTemp_INVALID; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17609 | } |
| 17610 | /* Ok, now we're unconditional. Do the load. */ |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17611 | res = newTemp(ty); |
| 17612 | // FIXME: assumes little-endian guest |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17613 | stmt( IRStmt_LLSC(Iend_LE, res, getIRegA(rN), |
| 17614 | NULL/*this is a load*/) ); |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17615 | if (ty == Ity_I64) { |
| 17616 | // FIXME: assumes little-endian guest |
| 17617 | putIRegA(rT+0, unop(Iop_64to32, mkexpr(res)), |
| 17618 | IRTemp_INVALID, Ijk_Boring); |
| 17619 | putIRegA(rT+1, unop(Iop_64HIto32, mkexpr(res)), |
| 17620 | IRTemp_INVALID, Ijk_Boring); |
| 17621 | DIP("ldrex%s%s r%u, r%u, [r%u]\n", |
| 17622 | nm, nCC(INSN_COND), rT+0, rT+1, rN); |
| 17623 | } else { |
| 17624 | putIRegA(rT, widen == Iop_INVALID |
| 17625 | ? mkexpr(res) : unop(widen, mkexpr(res)), |
| 17626 | IRTemp_INVALID, Ijk_Boring); |
| 17627 | DIP("ldrex%s%s r%u, [r%u]\n", nm, nCC(INSN_COND), rT, rN); |
| 17628 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17629 | goto decode_success; |
| 17630 | } |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17631 | /* undecodable; fall through */ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17632 | } |
| 17633 | |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17634 | // STREXD, STREX, STREXH, STREXB |
| 17635 | if (0x01800F90 == (insn & 0x0F900FF0)) { |
| 17636 | UInt rT = INSN(3,0); |
| 17637 | UInt rN = INSN(19,16); |
| 17638 | UInt rD = INSN(15,12); |
| 17639 | IRType ty = Ity_INVALID; |
| 17640 | IROp narrow = Iop_INVALID; |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 17641 | const HChar* nm = NULL; |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17642 | Bool valid = True; |
| 17643 | switch (INSN(22,21)) { |
| 17644 | case 0: nm = ""; ty = Ity_I32; break; |
| 17645 | case 1: nm = "d"; ty = Ity_I64; break; |
| 17646 | case 2: nm = "b"; ty = Ity_I8; narrow = Iop_32to8; break; |
| 17647 | case 3: nm = "h"; ty = Ity_I16; narrow = Iop_32to16; break; |
| 17648 | default: vassert(0); |
| 17649 | } |
| 17650 | if (ty == Ity_I32 || ty == Ity_I16 || ty == Ity_I8) { |
| 17651 | if (rD == 15 || rN == 15 || rT == 15 |
| 17652 | || rD == rN || rD == rT) |
| 17653 | valid = False; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17654 | } else { |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17655 | vassert(ty == Ity_I64); |
| 17656 | if (rD == 15 || (rT & 1) == 1 || rT == 14 || rN == 15 |
| 17657 | || rD == rN || rD == rT || rD == rT+1) |
| 17658 | valid = False; |
| 17659 | } |
| 17660 | if (valid) { |
| 17661 | IRTemp resSC1, resSC32, data; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17662 | /* make unconditional */ |
| 17663 | if (condT != IRTemp_INVALID) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17664 | mk_skip_over_A32_if_cond_is_false( condT ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17665 | condT = IRTemp_INVALID; |
| 17666 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17667 | /* Ok, now we're unconditional. Do the store. */ |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17668 | data = newTemp(ty); |
| 17669 | assign(data, |
| 17670 | ty == Ity_I64 |
| 17671 | // FIXME: assumes little-endian guest |
| 17672 | ? binop(Iop_32HLto64, getIRegA(rT+1), getIRegA(rT+0)) |
| 17673 | : narrow == Iop_INVALID |
| 17674 | ? getIRegA(rT) |
| 17675 | : unop(narrow, getIRegA(rT))); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17676 | resSC1 = newTemp(Ity_I1); |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17677 | // FIXME: assumes little-endian guest |
| 17678 | stmt( IRStmt_LLSC(Iend_LE, resSC1, getIRegA(rN), mkexpr(data)) ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17679 | |
| 17680 | /* Set rD to 1 on failure, 0 on success. Currently we have |
| 17681 | resSC1 == 0 on failure, 1 on success. */ |
| 17682 | resSC32 = newTemp(Ity_I32); |
| 17683 | assign(resSC32, |
| 17684 | unop(Iop_1Uto32, unop(Iop_Not1, mkexpr(resSC1)))); |
| 17685 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17686 | putIRegA(rD, mkexpr(resSC32), |
| 17687 | IRTemp_INVALID, Ijk_Boring); |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 17688 | if (ty == Ity_I64) { |
| 17689 | DIP("strex%s%s r%u, r%u, r%u, [r%u]\n", |
| 17690 | nm, nCC(INSN_COND), rD, rT, rT+1, rN); |
| 17691 | } else { |
| 17692 | DIP("strex%s%s r%u, r%u, [r%u]\n", |
| 17693 | nm, nCC(INSN_COND), rD, rT, rN); |
| 17694 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17695 | goto decode_success; |
| 17696 | } |
| 17697 | /* fall through */ |
| 17698 | } |
| 17699 | |
| 17700 | /* --------------------- movw, movt --------------------- */ |
| 17701 | if (0x03000000 == (insn & 0x0FF00000) |
| 17702 | || 0x03400000 == (insn & 0x0FF00000)) /* pray for CSE */ { |
| 17703 | UInt rD = INSN(15,12); |
| 17704 | UInt imm16 = (insn & 0xFFF) | ((insn >> 4) & 0x0000F000); |
| 17705 | UInt isT = (insn >> 22) & 1; |
| 17706 | if (rD == 15) { |
| 17707 | /* forget it */ |
| 17708 | } else { |
| 17709 | if (isT) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17710 | putIRegA(rD, |
| 17711 | binop(Iop_Or32, |
| 17712 | binop(Iop_And32, getIRegA(rD), mkU32(0xFFFF)), |
| 17713 | mkU32(imm16 << 16)), |
| 17714 | condT, Ijk_Boring); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17715 | DIP("movt%s r%u, #0x%04x\n", nCC(INSN_COND), rD, imm16); |
| 17716 | goto decode_success; |
| 17717 | } else { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17718 | putIRegA(rD, mkU32(imm16), condT, Ijk_Boring); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17719 | DIP("movw%s r%u, #0x%04x\n", nCC(INSN_COND), rD, imm16); |
| 17720 | goto decode_success; |
| 17721 | } |
| 17722 | } |
| 17723 | /* fall through */ |
| 17724 | } |
| 17725 | |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 17726 | /* ----------- uxtb, sxtb, uxth, sxth, uxtb16, sxtb16 ----------- */ |
| 17727 | /* FIXME: this is an exact duplicate of the Thumb version. They |
| 17728 | should be commoned up. */ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17729 | if (BITS8(0,1,1,0,1, 0,0,0) == (INSN(27,20) & BITS8(1,1,1,1,1,0,0,0)) |
| 17730 | && BITS4(1,1,1,1) == INSN(19,16) |
| 17731 | && BITS4(0,1,1,1) == INSN(7,4) |
| 17732 | && BITS4(0,0, 0,0) == (INSN(11,8) & BITS4(0,0,1,1))) { |
| 17733 | UInt subopc = INSN(27,20) & BITS8(0,0,0,0,0, 1,1,1); |
| 17734 | if (subopc != BITS4(0,0,0,1) && subopc != BITS4(0,1,0,1)) { |
| 17735 | Int rot = (INSN(11,8) >> 2) & 3; |
| 17736 | UInt rM = INSN(3,0); |
| 17737 | UInt rD = INSN(15,12); |
| 17738 | IRTemp srcT = newTemp(Ity_I32); |
| 17739 | IRTemp rotT = newTemp(Ity_I32); |
| 17740 | IRTemp dstT = newTemp(Ity_I32); |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 17741 | const HChar* nm = "???"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17742 | assign(srcT, getIRegA(rM)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17743 | assign(rotT, genROR32(srcT, 8 * rot)); /* 0, 8, 16 or 24 only */ |
| 17744 | switch (subopc) { |
| 17745 | case BITS4(0,1,1,0): // UXTB |
| 17746 | assign(dstT, unop(Iop_8Uto32, unop(Iop_32to8, mkexpr(rotT)))); |
| 17747 | nm = "uxtb"; |
| 17748 | break; |
| 17749 | case BITS4(0,0,1,0): // SXTB |
| 17750 | assign(dstT, unop(Iop_8Sto32, unop(Iop_32to8, mkexpr(rotT)))); |
| 17751 | nm = "sxtb"; |
| 17752 | break; |
| 17753 | case BITS4(0,1,1,1): // UXTH |
| 17754 | assign(dstT, unop(Iop_16Uto32, unop(Iop_32to16, mkexpr(rotT)))); |
| 17755 | nm = "uxth"; |
| 17756 | break; |
| 17757 | case BITS4(0,0,1,1): // SXTH |
| 17758 | assign(dstT, unop(Iop_16Sto32, unop(Iop_32to16, mkexpr(rotT)))); |
| 17759 | nm = "sxth"; |
| 17760 | break; |
| 17761 | case BITS4(0,1,0,0): // UXTB16 |
| 17762 | assign(dstT, binop(Iop_And32, mkexpr(rotT), mkU32(0x00FF00FF))); |
| 17763 | nm = "uxtb16"; |
| 17764 | break; |
| 17765 | case BITS4(0,0,0,0): { // SXTB16 |
| 17766 | IRTemp lo32 = newTemp(Ity_I32); |
| 17767 | IRTemp hi32 = newTemp(Ity_I32); |
| 17768 | assign(lo32, binop(Iop_And32, mkexpr(rotT), mkU32(0xFF))); |
| 17769 | assign(hi32, binop(Iop_Shr32, mkexpr(rotT), mkU8(16))); |
| 17770 | assign( |
| 17771 | dstT, |
| 17772 | binop(Iop_Or32, |
| 17773 | binop(Iop_And32, |
| 17774 | unop(Iop_8Sto32, |
| 17775 | unop(Iop_32to8, mkexpr(lo32))), |
| 17776 | mkU32(0xFFFF)), |
| 17777 | binop(Iop_Shl32, |
| 17778 | unop(Iop_8Sto32, |
| 17779 | unop(Iop_32to8, mkexpr(hi32))), |
| 17780 | mkU8(16)) |
| 17781 | )); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 17782 | nm = "sxtb16"; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17783 | break; |
| 17784 | } |
| 17785 | default: |
| 17786 | vassert(0); // guarded by "if" above |
| 17787 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17788 | putIRegA(rD, mkexpr(dstT), condT, Ijk_Boring); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 17789 | DIP("%s%s r%u, r%u, ROR #%d\n", nm, nCC(INSN_COND), rD, rM, rot); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17790 | goto decode_success; |
| 17791 | } |
| 17792 | /* fall through */ |
| 17793 | } |
| 17794 | |
| 17795 | /* ------------------- bfi, bfc ------------------- */ |
| 17796 | if (BITS8(0,1,1,1,1,1,0, 0) == (INSN(27,20) & BITS8(1,1,1,1,1,1,1,0)) |
| 17797 | && BITS4(0, 0,0,1) == (INSN(7,4) & BITS4(0,1,1,1))) { |
| 17798 | UInt rD = INSN(15,12); |
| 17799 | UInt rN = INSN(3,0); |
| 17800 | UInt msb = (insn >> 16) & 0x1F; /* 20:16 */ |
| 17801 | UInt lsb = (insn >> 7) & 0x1F; /* 11:7 */ |
| 17802 | if (rD == 15 || msb < lsb) { |
| 17803 | /* undecodable; fall through */ |
| 17804 | } else { |
| 17805 | IRTemp src = newTemp(Ity_I32); |
| 17806 | IRTemp olddst = newTemp(Ity_I32); |
| 17807 | IRTemp newdst = newTemp(Ity_I32); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 17808 | UInt mask = ((UInt)1) << (msb - lsb); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17809 | mask = (mask - 1) + mask; |
| 17810 | vassert(mask != 0); // guaranteed by "msb < lsb" check above |
| 17811 | mask <<= lsb; |
| 17812 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17813 | assign(src, rN == 15 ? mkU32(0) : getIRegA(rN)); |
| 17814 | assign(olddst, getIRegA(rD)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17815 | assign(newdst, |
| 17816 | binop(Iop_Or32, |
| 17817 | binop(Iop_And32, |
| 17818 | binop(Iop_Shl32, mkexpr(src), mkU8(lsb)), |
| 17819 | mkU32(mask)), |
| 17820 | binop(Iop_And32, |
| 17821 | mkexpr(olddst), |
| 17822 | mkU32(~mask))) |
| 17823 | ); |
| 17824 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17825 | putIRegA(rD, mkexpr(newdst), condT, Ijk_Boring); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17826 | |
| 17827 | if (rN == 15) { |
| 17828 | DIP("bfc%s r%u, #%u, #%u\n", |
| 17829 | nCC(INSN_COND), rD, lsb, msb-lsb+1); |
| 17830 | } else { |
| 17831 | DIP("bfi%s r%u, r%u, #%u, #%u\n", |
| 17832 | nCC(INSN_COND), rD, rN, lsb, msb-lsb+1); |
| 17833 | } |
| 17834 | goto decode_success; |
| 17835 | } |
| 17836 | /* fall through */ |
| 17837 | } |
| 17838 | |
| 17839 | /* ------------------- {u,s}bfx ------------------- */ |
| 17840 | if (BITS8(0,1,1,1,1,0,1,0) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,0)) |
| 17841 | && BITS4(0,1,0,1) == (INSN(7,4) & BITS4(0,1,1,1))) { |
| 17842 | UInt rD = INSN(15,12); |
| 17843 | UInt rN = INSN(3,0); |
| 17844 | UInt wm1 = (insn >> 16) & 0x1F; /* 20:16 */ |
| 17845 | UInt lsb = (insn >> 7) & 0x1F; /* 11:7 */ |
| 17846 | UInt msb = lsb + wm1; |
| 17847 | UInt isU = (insn >> 22) & 1; /* 22:22 */ |
| 17848 | if (rD == 15 || rN == 15 || msb >= 32) { |
| 17849 | /* undecodable; fall through */ |
| 17850 | } else { |
| 17851 | IRTemp src = newTemp(Ity_I32); |
| 17852 | IRTemp tmp = newTemp(Ity_I32); |
| 17853 | IRTemp res = newTemp(Ity_I32); |
| 17854 | UInt mask = ((1 << wm1) - 1) + (1 << wm1); |
| 17855 | vassert(msb >= 0 && msb <= 31); |
| 17856 | vassert(mask != 0); // guaranteed by msb being in 0 .. 31 inclusive |
| 17857 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17858 | assign(src, getIRegA(rN)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17859 | assign(tmp, binop(Iop_And32, |
| 17860 | binop(Iop_Shr32, mkexpr(src), mkU8(lsb)), |
| 17861 | mkU32(mask))); |
| 17862 | assign(res, binop(isU ? Iop_Shr32 : Iop_Sar32, |
| 17863 | binop(Iop_Shl32, mkexpr(tmp), mkU8(31-wm1)), |
| 17864 | mkU8(31-wm1))); |
| 17865 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17866 | putIRegA(rD, mkexpr(res), condT, Ijk_Boring); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17867 | |
| 17868 | DIP("%s%s r%u, r%u, #%u, #%u\n", |
| 17869 | isU ? "ubfx" : "sbfx", |
| 17870 | nCC(INSN_COND), rD, rN, lsb, wm1 + 1); |
| 17871 | goto decode_success; |
| 17872 | } |
| 17873 | /* fall through */ |
| 17874 | } |
| 17875 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17876 | /* --------------------- Load/store doubleword ------------- */ |
| 17877 | // LDRD STRD |
| 17878 | /* 31 27 23 19 15 11 7 3 # highest bit |
| 17879 | 28 24 20 16 12 8 4 0 |
| 17880 | A5-36 1 | 16 cond 0001 U100 Rn Rd im4h 11S1 im4l |
| 17881 | A5-38 1 | 32 cond 0001 U000 Rn Rd 0000 11S1 Rm |
| 17882 | A5-40 2 | 16 cond 0001 U110 Rn Rd im4h 11S1 im4l |
| 17883 | A5-42 2 | 32 cond 0001 U010 Rn Rd 0000 11S1 Rm |
| 17884 | A5-44 3 | 16 cond 0000 U100 Rn Rd im4h 11S1 im4l |
| 17885 | A5-46 3 | 32 cond 0000 U000 Rn Rd 0000 11S1 Rm |
| 17886 | */ |
| 17887 | /* case coding: |
| 17888 | 1 at-ea (access at ea) |
| 17889 | 2 at-ea-then-upd (access at ea, then Rn = ea) |
| 17890 | 3 at-Rn-then-upd (access at Rn, then Rn = ea) |
| 17891 | ea coding |
| 17892 | 16 Rn +/- imm8 |
| 17893 | 32 Rn +/- Rm |
| 17894 | */ |
| 17895 | /* Quickly skip over all of this for hopefully most instructions */ |
| 17896 | if ((INSN(27,24) & BITS4(1,1,1,0)) != BITS4(0,0,0,0)) |
| 17897 | goto after_load_store_doubleword; |
| 17898 | |
| 17899 | /* Check the "11S1" thing. */ |
| 17900 | if ((INSN(7,4) & BITS4(1,1,0,1)) != BITS4(1,1,0,1)) |
| 17901 | goto after_load_store_doubleword; |
| 17902 | |
| 17903 | summary = 0; |
| 17904 | |
| 17905 | /**/ if (INSN(27,24) == BITS4(0,0,0,1) && INSN(22,20) == BITS3(1,0,0)) { |
| 17906 | summary = 1 | 16; |
| 17907 | } |
| 17908 | else if (INSN(27,24) == BITS4(0,0,0,1) && INSN(22,20) == BITS3(0,0,0)) { |
| 17909 | summary = 1 | 32; |
| 17910 | } |
| 17911 | else if (INSN(27,24) == BITS4(0,0,0,1) && INSN(22,20) == BITS3(1,1,0)) { |
| 17912 | summary = 2 | 16; |
| 17913 | } |
| 17914 | else if (INSN(27,24) == BITS4(0,0,0,1) && INSN(22,20) == BITS3(0,1,0)) { |
| 17915 | summary = 2 | 32; |
| 17916 | } |
| 17917 | else if (INSN(27,24) == BITS4(0,0,0,0) && INSN(22,20) == BITS3(1,0,0)) { |
| 17918 | summary = 3 | 16; |
| 17919 | } |
| 17920 | else if (INSN(27,24) == BITS4(0,0,0,0) && INSN(22,20) == BITS3(0,0,0)) { |
| 17921 | summary = 3 | 32; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17922 | } |
| 17923 | else goto after_load_store_doubleword; |
| 17924 | |
| 17925 | { UInt rN = (insn >> 16) & 0xF; /* 19:16 */ |
| 17926 | UInt rD = (insn >> 12) & 0xF; /* 15:12 */ |
| 17927 | UInt rM = (insn >> 0) & 0xF; /* 3:0 */ |
| 17928 | UInt bU = (insn >> 23) & 1; /* 23 U=1 offset+, U=0 offset- */ |
| 17929 | UInt bS = (insn >> 5) & 1; /* S=1 store, S=0 load */ |
| 17930 | UInt imm8 = ((insn >> 4) & 0xF0) | (insn & 0xF); /* 11:8, 3:0 */ |
| 17931 | |
| 17932 | /* Require rD to be an even numbered register */ |
| 17933 | if ((rD & 1) != 0) |
| 17934 | goto after_load_store_doubleword; |
| 17935 | |
| 17936 | /* Require 11:8 == 0 for Rn +/- Rm cases */ |
| 17937 | if ((summary & 32) != 0 && (imm8 & 0xF0) != 0) |
| 17938 | goto after_load_store_doubleword; |
| 17939 | |
| 17940 | /* Skip some invalid cases, which would lead to two competing |
| 17941 | updates to the same register, or which are otherwise |
| 17942 | disallowed by the spec. */ |
| 17943 | switch (summary) { |
| 17944 | case 1 | 16: |
| 17945 | break; |
| 17946 | case 1 | 32: |
| 17947 | if (rM == 15) goto after_load_store_doubleword; |
| 17948 | break; |
| 17949 | case 2 | 16: case 3 | 16: |
| 17950 | if (rN == 15) goto after_load_store_doubleword; |
| 17951 | if (bS == 0 && (rN == rD || rN == rD+1)) |
| 17952 | goto after_load_store_doubleword; |
| 17953 | break; |
| 17954 | case 2 | 32: case 3 | 32: |
| 17955 | if (rM == 15) goto after_load_store_doubleword; |
| 17956 | if (rN == 15) goto after_load_store_doubleword; |
| 17957 | if (rN == rM) goto after_load_store_doubleword; |
| 17958 | if (bS == 0 && (rN == rD || rN == rD+1)) |
| 17959 | goto after_load_store_doubleword; |
| 17960 | break; |
| 17961 | default: |
| 17962 | vassert(0); |
| 17963 | } |
| 17964 | |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 17965 | /* If this is a branch, make it unconditional at this point. |
| 17966 | Doing conditional branches in-line is too complex (for |
| 17967 | now). */ |
| 17968 | vassert((rD & 1) == 0); /* from tests above */ |
| 17969 | if (bS == 0 && rD+1 == 15 && condT != IRTemp_INVALID) { |
| 17970 | // go uncond |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17971 | mk_skip_over_A32_if_cond_is_false( condT ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17972 | condT = IRTemp_INVALID; |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 17973 | // now uncond |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17974 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17975 | |
| 17976 | /* compute the effective address. Bind it to a tmp since we |
| 17977 | may need to use it twice. */ |
| 17978 | IRExpr* eaE = NULL; |
| 17979 | switch (summary & 0xF0) { |
| 17980 | case 16: |
| 17981 | eaE = mk_EA_reg_plusminus_imm8( rN, bU, imm8, dis_buf ); |
| 17982 | break; |
| 17983 | case 32: |
| 17984 | eaE = mk_EA_reg_plusminus_reg( rN, bU, rM, dis_buf ); |
| 17985 | break; |
| 17986 | } |
| 17987 | vassert(eaE); |
| 17988 | IRTemp eaT = newTemp(Ity_I32); |
| 17989 | assign(eaT, eaE); |
| 17990 | |
| 17991 | /* get the old Rn value */ |
| 17992 | IRTemp rnT = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 17993 | assign(rnT, getIRegA(rN)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 17994 | |
| 17995 | /* decide on the transfer address */ |
| 17996 | IRTemp taT = IRTemp_INVALID; |
| 17997 | switch (summary & 0x0F) { |
| 17998 | case 1: case 2: taT = eaT; break; |
| 17999 | case 3: taT = rnT; break; |
| 18000 | } |
| 18001 | vassert(taT != IRTemp_INVALID); |
| 18002 | |
| 18003 | /* XXX deal with alignment constraints */ |
| 18004 | /* XXX: but the A8 doesn't seem to trap for misaligned loads, so, |
| 18005 | ignore alignment issues for the time being. */ |
| 18006 | |
sewardj | 92001d4 | 2013-04-26 10:24:31 +0000 | [diff] [blame] | 18007 | /* For almost all cases, we do the writeback after the transfers. |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 18008 | However, that leaves the stack "uncovered" in cases like: |
sewardj | 92001d4 | 2013-04-26 10:24:31 +0000 | [diff] [blame] | 18009 | strd rD, [sp, #-8] |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 18010 | strd rD, [sp, #-16] |
sewardj | 92001d4 | 2013-04-26 10:24:31 +0000 | [diff] [blame] | 18011 | In which case, do the writeback to SP now, instead of later. |
| 18012 | This is bad in that it makes the insn non-restartable if the |
| 18013 | accesses fault, but at least keeps Memcheck happy. */ |
| 18014 | Bool writeback_already_done = False; |
| 18015 | if (bS == 1 /*store*/ && summary == (2 | 16) |
| 18016 | && rN == 13 && rN != rD && rN != rD+1 |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 18017 | && bU == 0/*minus*/ |
| 18018 | && (imm8 == 8 || imm8 == 16)) { |
sewardj | 92001d4 | 2013-04-26 10:24:31 +0000 | [diff] [blame] | 18019 | putIRegA( rN, mkexpr(eaT), condT, Ijk_Boring ); |
| 18020 | writeback_already_done = True; |
| 18021 | } |
| 18022 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18023 | /* doubleword store S 1 |
| 18024 | doubleword load S 0 |
| 18025 | */ |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 18026 | const HChar* name = NULL; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18027 | /* generate the transfers */ |
| 18028 | if (bS == 1) { // doubleword store |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 18029 | storeGuardedLE( binop(Iop_Add32, mkexpr(taT), mkU32(0)), |
| 18030 | getIRegA(rD+0), condT ); |
| 18031 | storeGuardedLE( binop(Iop_Add32, mkexpr(taT), mkU32(4)), |
| 18032 | getIRegA(rD+1), condT ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18033 | name = "strd"; |
| 18034 | } else { // doubleword load |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 18035 | IRTemp oldRd0 = newTemp(Ity_I32); |
| 18036 | IRTemp oldRd1 = newTemp(Ity_I32); |
| 18037 | assign(oldRd0, llGetIReg(rD+0)); |
| 18038 | assign(oldRd1, llGetIReg(rD+1)); |
| 18039 | IRTemp newRd0 = newTemp(Ity_I32); |
| 18040 | IRTemp newRd1 = newTemp(Ity_I32); |
| 18041 | loadGuardedLE( newRd0, ILGop_Ident32, |
| 18042 | binop(Iop_Add32, mkexpr(taT), mkU32(0)), |
| 18043 | mkexpr(oldRd0), condT ); |
| 18044 | putIRegA( rD+0, mkexpr(newRd0), IRTemp_INVALID, Ijk_Boring ); |
| 18045 | loadGuardedLE( newRd1, ILGop_Ident32, |
| 18046 | binop(Iop_Add32, mkexpr(taT), mkU32(4)), |
| 18047 | mkexpr(oldRd1), condT ); |
| 18048 | putIRegA( rD+1, mkexpr(newRd1), IRTemp_INVALID, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18049 | name = "ldrd"; |
| 18050 | } |
| 18051 | |
| 18052 | /* Update Rn if necessary. */ |
| 18053 | switch (summary & 0x0F) { |
| 18054 | case 2: case 3: |
| 18055 | // should be assured by logic above: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 18056 | vassert(rN != 15); /* from checks above */ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18057 | if (bS == 0) { |
| 18058 | vassert(rD+0 != rN); /* since we just wrote rD+0 */ |
| 18059 | vassert(rD+1 != rN); /* since we just wrote rD+1 */ |
| 18060 | } |
sewardj | 92001d4 | 2013-04-26 10:24:31 +0000 | [diff] [blame] | 18061 | if (!writeback_already_done) |
| 18062 | putIRegA( rN, mkexpr(eaT), condT, Ijk_Boring ); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18063 | break; |
| 18064 | } |
| 18065 | |
| 18066 | switch (summary & 0x0F) { |
| 18067 | case 1: DIP("%s%s r%u, %s\n", name, nCC(INSN_COND), rD, dis_buf); |
| 18068 | break; |
| 18069 | case 2: DIP("%s%s r%u, %s! (at-EA-then-Rn=EA)\n", |
| 18070 | name, nCC(INSN_COND), rD, dis_buf); |
| 18071 | break; |
| 18072 | case 3: DIP("%s%s r%u, %s! (at-Rn-then-Rn=EA)\n", |
| 18073 | name, nCC(INSN_COND), rD, dis_buf); |
| 18074 | break; |
| 18075 | default: vassert(0); |
| 18076 | } |
| 18077 | |
| 18078 | goto decode_success; |
| 18079 | } |
| 18080 | |
| 18081 | after_load_store_doubleword: |
| 18082 | |
sewardj | 04ac5de | 2010-03-08 14:49:03 +0000 | [diff] [blame] | 18083 | /* ------------------- {s,u}xtab ------------- */ |
| 18084 | if (BITS8(0,1,1,0,1,0,1,0) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18085 | && BITS4(0,0,0,0) == (INSN(11,8) & BITS4(0,0,1,1)) |
| 18086 | && BITS4(0,1,1,1) == INSN(7,4)) { |
| 18087 | UInt rN = INSN(19,16); |
| 18088 | UInt rD = INSN(15,12); |
| 18089 | UInt rM = INSN(3,0); |
| 18090 | UInt rot = (insn >> 10) & 3; |
sewardj | 04ac5de | 2010-03-08 14:49:03 +0000 | [diff] [blame] | 18091 | UInt isU = INSN(22,22); |
| 18092 | if (rN == 15/*it's {S,U}XTB*/ || rD == 15 || rM == 15) { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18093 | /* undecodable; fall through */ |
| 18094 | } else { |
| 18095 | IRTemp srcL = newTemp(Ity_I32); |
| 18096 | IRTemp srcR = newTemp(Ity_I32); |
| 18097 | IRTemp res = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18098 | assign(srcR, getIRegA(rM)); |
| 18099 | assign(srcL, getIRegA(rN)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18100 | assign(res, binop(Iop_Add32, |
| 18101 | mkexpr(srcL), |
sewardj | 04ac5de | 2010-03-08 14:49:03 +0000 | [diff] [blame] | 18102 | unop(isU ? Iop_8Uto32 : Iop_8Sto32, |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18103 | unop(Iop_32to8, |
| 18104 | genROR32(srcR, 8 * rot))))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18105 | putIRegA(rD, mkexpr(res), condT, Ijk_Boring); |
sewardj | 04ac5de | 2010-03-08 14:49:03 +0000 | [diff] [blame] | 18106 | DIP("%cxtab%s r%u, r%u, r%u, ror #%u\n", |
| 18107 | isU ? 'u' : 's', nCC(INSN_COND), rD, rN, rM, rot); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18108 | goto decode_success; |
| 18109 | } |
| 18110 | /* fall through */ |
| 18111 | } |
| 18112 | |
sewardj | 04ac5de | 2010-03-08 14:49:03 +0000 | [diff] [blame] | 18113 | /* ------------------- {s,u}xtah ------------- */ |
| 18114 | if (BITS8(0,1,1,0,1,0,1,1) == (INSN(27,20) & BITS8(1,1,1,1,1,0,1,1)) |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18115 | && BITS4(0,0,0,0) == (INSN(11,8) & BITS4(0,0,1,1)) |
| 18116 | && BITS4(0,1,1,1) == INSN(7,4)) { |
| 18117 | UInt rN = INSN(19,16); |
| 18118 | UInt rD = INSN(15,12); |
| 18119 | UInt rM = INSN(3,0); |
| 18120 | UInt rot = (insn >> 10) & 3; |
sewardj | 04ac5de | 2010-03-08 14:49:03 +0000 | [diff] [blame] | 18121 | UInt isU = INSN(22,22); |
| 18122 | if (rN == 15/*it's {S,U}XTH*/ || rD == 15 || rM == 15) { |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18123 | /* undecodable; fall through */ |
| 18124 | } else { |
| 18125 | IRTemp srcL = newTemp(Ity_I32); |
| 18126 | IRTemp srcR = newTemp(Ity_I32); |
| 18127 | IRTemp res = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18128 | assign(srcR, getIRegA(rM)); |
| 18129 | assign(srcL, getIRegA(rN)); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18130 | assign(res, binop(Iop_Add32, |
| 18131 | mkexpr(srcL), |
sewardj | 04ac5de | 2010-03-08 14:49:03 +0000 | [diff] [blame] | 18132 | unop(isU ? Iop_16Uto32 : Iop_16Sto32, |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18133 | unop(Iop_32to16, |
| 18134 | genROR32(srcR, 8 * rot))))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18135 | putIRegA(rD, mkexpr(res), condT, Ijk_Boring); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18136 | |
sewardj | 04ac5de | 2010-03-08 14:49:03 +0000 | [diff] [blame] | 18137 | DIP("%cxtah%s r%u, r%u, r%u, ror #%u\n", |
| 18138 | isU ? 'u' : 's', nCC(INSN_COND), rD, rN, rM, rot); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18139 | goto decode_success; |
| 18140 | } |
| 18141 | /* fall through */ |
| 18142 | } |
| 18143 | |
sewardj | db1c20a | 2010-09-22 22:26:40 +0000 | [diff] [blame] | 18144 | /* ------------------- rev16, rev ------------------ */ |
| 18145 | if (INSN(27,16) == 0x6BF |
| 18146 | && (INSN(11,4) == 0xFB/*rev16*/ || INSN(11,4) == 0xF3/*rev*/)) { |
| 18147 | Bool isREV = INSN(11,4) == 0xF3; |
| 18148 | UInt rM = INSN(3,0); |
| 18149 | UInt rD = INSN(15,12); |
| 18150 | if (rM != 15 && rD != 15) { |
| 18151 | IRTemp rMt = newTemp(Ity_I32); |
| 18152 | assign(rMt, getIRegA(rM)); |
sewardj | 27312d3 | 2010-09-26 00:48:41 +0000 | [diff] [blame] | 18153 | IRTemp res = isREV ? gen_REV(rMt) : gen_REV16(rMt); |
| 18154 | putIRegA(rD, mkexpr(res), condT, Ijk_Boring); |
sewardj | db1c20a | 2010-09-22 22:26:40 +0000 | [diff] [blame] | 18155 | DIP("rev%s%s r%u, r%u\n", isREV ? "" : "16", |
| 18156 | nCC(INSN_COND), rD, rM); |
| 18157 | goto decode_success; |
| 18158 | } |
| 18159 | } |
| 18160 | |
sewardj | a325ee1 | 2013-07-04 20:35:42 +0000 | [diff] [blame] | 18161 | /* ------------------- revsh ----------------------- */ |
| 18162 | if (INSN(27,16) == 0x6FF && INSN(11,4) == 0xFB) { |
| 18163 | UInt rM = INSN(3,0); |
| 18164 | UInt rD = INSN(15,12); |
| 18165 | if (rM != 15 && rD != 15) { |
| 18166 | IRTemp irt_rM = newTemp(Ity_I32); |
| 18167 | IRTemp irt_hi = newTemp(Ity_I32); |
| 18168 | IRTemp irt_low = newTemp(Ity_I32); |
| 18169 | IRTemp irt_res = newTemp(Ity_I32); |
| 18170 | assign(irt_rM, getIRegA(rM)); |
| 18171 | assign(irt_hi, |
| 18172 | binop(Iop_Sar32, |
| 18173 | binop(Iop_Shl32, mkexpr(irt_rM), mkU8(24)), |
| 18174 | mkU8(16) |
| 18175 | ) |
| 18176 | ); |
| 18177 | assign(irt_low, |
| 18178 | binop(Iop_And32, |
| 18179 | binop(Iop_Shr32, mkexpr(irt_rM), mkU8(8)), |
| 18180 | mkU32(0xFF) |
| 18181 | ) |
| 18182 | ); |
| 18183 | assign(irt_res, |
| 18184 | binop(Iop_Or32, mkexpr(irt_hi), mkexpr(irt_low)) |
| 18185 | ); |
| 18186 | putIRegA(rD, mkexpr(irt_res), condT, Ijk_Boring); |
| 18187 | DIP("revsh%s r%u, r%u\n", nCC(INSN_COND), rD, rM); |
| 18188 | goto decode_success; |
| 18189 | } |
| 18190 | } |
| 18191 | |
sewardj | 389239a | 2010-09-24 21:59:55 +0000 | [diff] [blame] | 18192 | /* ------------------- rbit ------------------ */ |
| 18193 | if (INSN(27,16) == 0x6FF && INSN(11,4) == 0xF3) { |
| 18194 | UInt rD = INSN(15,12); |
| 18195 | UInt rM = INSN(3,0); |
| 18196 | if (rD != 15 && rM != 15) { |
| 18197 | IRTemp arg = newTemp(Ity_I32); |
| 18198 | assign(arg, getIRegA(rM)); |
| 18199 | IRTemp res = gen_BITREV(arg); |
| 18200 | putIRegA(rD, mkexpr(res), condT, Ijk_Boring); |
| 18201 | DIP("rbit r%u, r%u\n", rD, rM); |
| 18202 | goto decode_success; |
| 18203 | } |
| 18204 | } |
| 18205 | |
sewardj | db1c20a | 2010-09-22 22:26:40 +0000 | [diff] [blame] | 18206 | /* ------------------- smmul ------------------ */ |
| 18207 | if (INSN(27,20) == BITS8(0,1,1,1,0,1,0,1) |
| 18208 | && INSN(15,12) == BITS4(1,1,1,1) |
| 18209 | && (INSN(7,4) & BITS4(1,1,0,1)) == BITS4(0,0,0,1)) { |
| 18210 | UInt bitR = INSN(5,5); |
| 18211 | UInt rD = INSN(19,16); |
| 18212 | UInt rM = INSN(11,8); |
| 18213 | UInt rN = INSN(3,0); |
| 18214 | if (rD != 15 && rM != 15 && rN != 15) { |
| 18215 | IRExpr* res |
| 18216 | = unop(Iop_64HIto32, |
| 18217 | binop(Iop_Add64, |
| 18218 | binop(Iop_MullS32, getIRegA(rN), getIRegA(rM)), |
| 18219 | mkU64(bitR ? 0x80000000ULL : 0ULL))); |
| 18220 | putIRegA(rD, res, condT, Ijk_Boring); |
| 18221 | DIP("smmul%s%s r%u, r%u, r%u\n", |
| 18222 | nCC(INSN_COND), bitR ? "r" : "", rD, rN, rM); |
| 18223 | goto decode_success; |
| 18224 | } |
| 18225 | } |
| 18226 | |
sewardj | ccff5f4 | 2013-04-23 08:56:43 +0000 | [diff] [blame] | 18227 | /* ------------------- smmla ------------------ */ |
| 18228 | if (INSN(27,20) == BITS8(0,1,1,1,0,1,0,1) |
| 18229 | && INSN(15,12) != BITS4(1,1,1,1) |
| 18230 | && (INSN(7,4) & BITS4(1,1,0,1)) == BITS4(0,0,0,1)) { |
| 18231 | UInt bitR = INSN(5,5); |
| 18232 | UInt rD = INSN(19,16); |
| 18233 | UInt rA = INSN(15,12); |
| 18234 | UInt rM = INSN(11,8); |
| 18235 | UInt rN = INSN(3,0); |
| 18236 | if (rD != 15 && rM != 15 && rN != 15) { |
| 18237 | IRExpr* res |
| 18238 | = unop(Iop_64HIto32, |
| 18239 | binop(Iop_Add64, |
| 18240 | binop(Iop_Add64, |
| 18241 | binop(Iop_32HLto64, getIRegA(rA), mkU32(0)), |
| 18242 | binop(Iop_MullS32, getIRegA(rN), getIRegA(rM))), |
| 18243 | mkU64(bitR ? 0x80000000ULL : 0ULL))); |
| 18244 | putIRegA(rD, res, condT, Ijk_Boring); |
| 18245 | DIP("smmla%s%s r%u, r%u, r%u, r%u\n", |
| 18246 | nCC(INSN_COND), bitR ? "r" : "", rD, rN, rM, rA); |
| 18247 | goto decode_success; |
| 18248 | } |
| 18249 | } |
| 18250 | |
sewardj | 3b22fc8 | 2013-09-29 18:24:15 +0000 | [diff] [blame] | 18251 | /* -------------- (A1) LDRT reg+/-#imm12 -------------- */ |
| 18252 | /* Load Register Unprivileged: |
| 18253 | ldrt<c> Rt, [Rn] {, #+/-imm12} |
| 18254 | */ |
| 18255 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,1,0,0,0,0,1,1) ) { |
| 18256 | UInt rT = INSN(15,12); |
| 18257 | UInt rN = INSN(19,16); |
| 18258 | UInt imm12 = INSN(11,0); |
| 18259 | UInt bU = INSN(23,23); |
| 18260 | Bool valid = True; |
| 18261 | if (rT == 15 || rN == 15 || rN == rT) valid = False; |
| 18262 | if (valid) { |
| 18263 | IRTemp newRt = newTemp(Ity_I32); |
| 18264 | loadGuardedLE( newRt, |
| 18265 | ILGop_Ident32, getIRegA(rN), getIRegA(rT), condT ); |
| 18266 | putIRegA(rT, mkexpr(newRt), IRTemp_INVALID, Ijk_Boring); |
| 18267 | IRExpr* erN = binop(bU ? Iop_Add32 : Iop_Sub32, |
| 18268 | getIRegA(rN), mkU32(imm12)); |
| 18269 | putIRegA(rN, erN, condT, Ijk_Boring); |
| 18270 | DIP("ldrt%s r%u, [r%u], #%c%u\n", |
| 18271 | nCC(INSN_COND), rT, rN, bU ? '+' : '-', imm12); |
| 18272 | goto decode_success; |
| 18273 | } |
| 18274 | } |
| 18275 | |
| 18276 | /* -------------- (A2) LDRT reg+/-reg with shift -------------- */ |
| 18277 | /* Load Register Unprivileged: |
| 18278 | ldrt<c> Rt, [Rn], +/-Rm{, shift} |
| 18279 | */ |
| 18280 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,1,1,0,0,0,1,1) |
| 18281 | && INSN(4,4) == 0 ) { |
| 18282 | UInt rT = INSN(15,12); |
| 18283 | UInt rN = INSN(19,16); |
| 18284 | UInt rM = INSN(3,0); |
| 18285 | UInt imm5 = INSN(11,7); |
| 18286 | UInt bU = INSN(23,23); |
| 18287 | UInt type = INSN(6,5); |
| 18288 | Bool valid = True; |
| 18289 | if (rT == 15 || rN == 15 || rN == rT || rM == 15 |
| 18290 | /* || (ArchVersion() < 6 && rM == rN) */) |
| 18291 | valid = False; |
| 18292 | if (valid) { |
| 18293 | IRTemp newRt = newTemp(Ity_I32); |
| 18294 | loadGuardedLE( newRt, |
| 18295 | ILGop_Ident32, getIRegA(rN), getIRegA(rT), condT ); |
| 18296 | putIRegA(rT, mkexpr(newRt), IRTemp_INVALID, Ijk_Boring); |
| 18297 | // dis_buf generated is slightly bogus, in fact. |
| 18298 | IRExpr* erN = mk_EA_reg_plusminus_shifted_reg(rN, bU, rM, |
| 18299 | type, imm5, dis_buf); |
| 18300 | putIRegA(rN, erN, condT, Ijk_Boring); |
| 18301 | DIP("ldrt%s r%u, %s\n", nCC(INSN_COND), rT, dis_buf); |
| 18302 | goto decode_success; |
| 18303 | } |
| 18304 | } |
| 18305 | |
| 18306 | /* -------------- (A1) LDRBT reg+/-#imm12 -------------- */ |
| 18307 | /* Load Register Byte Unprivileged: |
| 18308 | ldrbt<c> Rt, [Rn], #+/-imm12 |
| 18309 | */ |
| 18310 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,1,0,0,0,1,1,1) ) { |
| 18311 | UInt rT = INSN(15,12); |
| 18312 | UInt rN = INSN(19,16); |
| 18313 | UInt imm12 = INSN(11,0); |
| 18314 | UInt bU = INSN(23,23); |
| 18315 | Bool valid = True; |
| 18316 | if (rT == 15 || rN == 15 || rN == rT) valid = False; |
| 18317 | if (valid) { |
| 18318 | IRTemp newRt = newTemp(Ity_I32); |
| 18319 | loadGuardedLE( newRt, |
| 18320 | ILGop_8Uto32, getIRegA(rN), getIRegA(rT), condT ); |
| 18321 | putIRegA(rT, mkexpr(newRt), IRTemp_INVALID, Ijk_Boring); |
| 18322 | IRExpr* erN = binop(bU ? Iop_Add32 : Iop_Sub32, |
| 18323 | getIRegA(rN), mkU32(imm12)); |
| 18324 | putIRegA(rN, erN, condT, Ijk_Boring); |
| 18325 | DIP("ldrbt%s r%u, [r%u], #%c%u\n", |
| 18326 | nCC(INSN_COND), rT, rN, bU ? '+' : '-', imm12); |
| 18327 | goto decode_success; |
| 18328 | } |
| 18329 | } |
| 18330 | |
| 18331 | /* -------------- (A2) LDRBT reg+/-reg with shift -------------- */ |
| 18332 | /* Load Register Byte Unprivileged: |
| 18333 | ldrbt<c> Rt, [Rn], +/-Rm{, shift} |
| 18334 | */ |
| 18335 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,1,1,0,0,1,1,1) |
| 18336 | && INSN(4,4) == 0 ) { |
| 18337 | UInt rT = INSN(15,12); |
| 18338 | UInt rN = INSN(19,16); |
| 18339 | UInt rM = INSN(3,0); |
| 18340 | UInt imm5 = INSN(11,7); |
| 18341 | UInt bU = INSN(23,23); |
| 18342 | UInt type = INSN(6,5); |
| 18343 | Bool valid = True; |
| 18344 | if (rT == 15 || rN == 15 || rN == rT || rM == 15 |
| 18345 | /* || (ArchVersion() < 6 && rM == rN) */) |
| 18346 | valid = False; |
| 18347 | if (valid) { |
| 18348 | IRTemp newRt = newTemp(Ity_I32); |
| 18349 | loadGuardedLE( newRt, |
| 18350 | ILGop_8Uto32, getIRegA(rN), getIRegA(rT), condT ); |
| 18351 | putIRegA(rT, mkexpr(newRt), IRTemp_INVALID, Ijk_Boring); |
| 18352 | // dis_buf generated is slightly bogus, in fact. |
| 18353 | IRExpr* erN = mk_EA_reg_plusminus_shifted_reg(rN, bU, rM, |
| 18354 | type, imm5, dis_buf); |
| 18355 | putIRegA(rN, erN, condT, Ijk_Boring); |
| 18356 | DIP("ldrbt%s r%u, %s\n", nCC(INSN_COND), rT, dis_buf); |
| 18357 | goto decode_success; |
| 18358 | } |
| 18359 | } |
| 18360 | |
| 18361 | /* -------------- (A1) LDRHT reg+#imm8 -------------- */ |
| 18362 | /* Load Register Halfword Unprivileged: |
| 18363 | ldrht<c> Rt, [Rn] {, #+/-imm8} |
| 18364 | */ |
| 18365 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,0,0,0,0,1,1,1) |
| 18366 | && INSN(7,4) == BITS4(1,0,1,1) ) { |
| 18367 | UInt rT = INSN(15,12); |
| 18368 | UInt rN = INSN(19,16); |
| 18369 | UInt bU = INSN(23,23); |
| 18370 | UInt imm4H = INSN(11,8); |
| 18371 | UInt imm4L = INSN(3,0); |
| 18372 | UInt imm8 = (imm4H << 4) | imm4L; |
| 18373 | Bool valid = True; |
| 18374 | if (rT == 15 || rN == 15 || rN == rT) |
| 18375 | valid = False; |
| 18376 | if (valid) { |
| 18377 | IRTemp newRt = newTemp(Ity_I32); |
| 18378 | loadGuardedLE( newRt, |
| 18379 | ILGop_16Uto32, getIRegA(rN), getIRegA(rT), condT ); |
| 18380 | putIRegA(rT, mkexpr(newRt), IRTemp_INVALID, Ijk_Boring); |
| 18381 | IRExpr* erN = binop(bU ? Iop_Add32 : Iop_Sub32, |
| 18382 | getIRegA(rN), mkU32(imm8)); |
| 18383 | putIRegA(rN, erN, condT, Ijk_Boring); |
| 18384 | DIP("ldrht%s r%u, [r%u], #%c%u\n", |
| 18385 | nCC(INSN_COND), rT, rN, bU ? '+' : '-', imm8); |
| 18386 | goto decode_success; |
| 18387 | } |
| 18388 | } |
| 18389 | |
| 18390 | /* -------------- (A2) LDRHT reg+/-reg -------------- */ |
| 18391 | /* Load Register Halfword Unprivileged: |
| 18392 | ldrht<c> Rt, [Rn], +/-Rm |
| 18393 | */ |
| 18394 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,0,0,0,0,0,1,1) |
| 18395 | && INSN(11,4) == BITS8(0,0,0,0,1,0,1,1) ) { |
| 18396 | UInt rT = INSN(15,12); |
| 18397 | UInt rN = INSN(19,16); |
| 18398 | UInt rM = INSN(3,0); |
| 18399 | UInt bU = INSN(23,23); |
| 18400 | Bool valid = True; |
| 18401 | if (rT == 15 || rN == 15 || rN == rT || rM == 15) |
| 18402 | valid = False; |
| 18403 | if (valid) { |
| 18404 | IRTemp newRt = newTemp(Ity_I32); |
| 18405 | loadGuardedLE( newRt, |
| 18406 | ILGop_16Uto32, getIRegA(rN), getIRegA(rT), condT ); |
| 18407 | putIRegA(rT, mkexpr(newRt), IRTemp_INVALID, Ijk_Boring); |
| 18408 | IRExpr* erN = binop(bU ? Iop_Add32 : Iop_Sub32, |
| 18409 | getIRegA(rN), getIRegA(rM)); |
| 18410 | putIRegA(rN, erN, condT, Ijk_Boring); |
| 18411 | DIP("ldrht%s r%u, [r%u], %cr%u\n", |
| 18412 | nCC(INSN_COND), rT, rN, bU ? '+' : '-', rM); |
| 18413 | goto decode_success; |
| 18414 | } |
| 18415 | } |
| 18416 | |
| 18417 | /* -------------- (A1) LDRSHT reg+#imm8 -------------- */ |
| 18418 | /* Load Register Signed Halfword Unprivileged: |
| 18419 | ldrsht<c> Rt, [Rn] {, #+/-imm8} |
| 18420 | */ |
| 18421 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,0,0,0,0,1,1,1) |
| 18422 | && INSN(7,4) == BITS4(1,1,1,1)) { |
| 18423 | UInt rT = INSN(15,12); |
| 18424 | UInt rN = INSN(19,16); |
| 18425 | UInt bU = INSN(23,23); |
| 18426 | UInt imm4H = INSN(11,8); |
| 18427 | UInt imm4L = INSN(3,0); |
| 18428 | UInt imm8 = (imm4H << 4) | imm4L; |
| 18429 | Bool valid = True; |
| 18430 | if (rN == 15 || rT == 15 || rN == rT) |
| 18431 | valid = False; |
| 18432 | if (valid) { |
| 18433 | IRTemp newRt = newTemp(Ity_I32); |
| 18434 | loadGuardedLE( newRt, |
| 18435 | ILGop_16Sto32, getIRegA(rN), getIRegA(rT), condT ); |
| 18436 | putIRegA(rT, mkexpr(newRt), IRTemp_INVALID, Ijk_Boring); |
| 18437 | IRExpr* erN = binop(bU ? Iop_Add32 : Iop_Sub32, |
| 18438 | getIRegA(rN), mkU32(imm8)); |
| 18439 | putIRegA(rN, erN, condT, Ijk_Boring); |
| 18440 | DIP("ldrsht%s r%u, [r%u], #%c%u\n", |
| 18441 | nCC(INSN_COND), rT, rN, bU ? '+' : '-', imm8); |
| 18442 | goto decode_success; |
| 18443 | } |
| 18444 | } |
| 18445 | |
| 18446 | /* -------------- (A2) LDRSHT reg+/-reg -------------- */ |
| 18447 | /* Load Register Signed Halfword Unprivileged: |
| 18448 | ldrsht<c> Rt, [Rn], +/-Rm |
| 18449 | */ |
| 18450 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,0,0,0,0,0,1,1) |
| 18451 | && INSN(11,4) == BITS8(0,0,0,0,1,1,1,1)) { |
| 18452 | UInt rT = INSN(15,12); |
| 18453 | UInt rN = INSN(19,16); |
| 18454 | UInt rM = INSN(3,0); |
| 18455 | UInt bU = INSN(23,23); |
| 18456 | Bool valid = True; |
| 18457 | if (rN == 15 || rT == 15 || rN == rT || rM == 15) |
| 18458 | valid = False; |
| 18459 | if (valid) { |
| 18460 | IRTemp newRt = newTemp(Ity_I32); |
| 18461 | loadGuardedLE( newRt, |
| 18462 | ILGop_16Sto32, getIRegA(rN), getIRegA(rT), condT ); |
| 18463 | putIRegA(rT, mkexpr(newRt), IRTemp_INVALID, Ijk_Boring); |
| 18464 | IRExpr* erN = binop(bU ? Iop_Add32 : Iop_Sub32, |
| 18465 | getIRegA(rN), getIRegA(rM)); |
| 18466 | putIRegA(rN, erN, condT, Ijk_Boring); |
| 18467 | DIP("ldrsht%s r%u, [r%u], %cr%u\n", |
| 18468 | nCC(INSN_COND), rT, rN, bU ? '+' : '-', rM); |
| 18469 | goto decode_success; |
| 18470 | } |
| 18471 | } |
| 18472 | |
| 18473 | /* -------------- (A1) LDRSBT reg+#imm8 -------------- */ |
| 18474 | /* Load Register Signed Byte Unprivileged: |
| 18475 | ldrsbt<c> Rt, [Rn] {, #+/-imm8} |
| 18476 | */ |
| 18477 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,0,0,0,0,1,1,1) |
| 18478 | && INSN(7,4) == BITS4(1,1,0,1)) { |
| 18479 | UInt rT = INSN(15,12); |
| 18480 | UInt rN = INSN(19,16); |
| 18481 | UInt bU = INSN(23,23); |
| 18482 | UInt imm4H = INSN(11,8); |
| 18483 | UInt imm4L = INSN(3,0); |
| 18484 | UInt imm8 = (imm4H << 4) | imm4L; |
| 18485 | Bool valid = True; |
| 18486 | if (rT == 15 || rN == 15 || rN == rT) |
| 18487 | valid = False; |
| 18488 | if (valid) { |
| 18489 | IRTemp newRt = newTemp(Ity_I32); |
| 18490 | loadGuardedLE( newRt, |
| 18491 | ILGop_8Sto32, getIRegA(rN), getIRegA(rT), condT ); |
| 18492 | putIRegA(rT, mkexpr(newRt), IRTemp_INVALID, Ijk_Boring); |
| 18493 | IRExpr* erN = binop(bU ? Iop_Add32 : Iop_Sub32, |
| 18494 | getIRegA(rN), mkU32(imm8)); |
| 18495 | putIRegA(rN, erN, condT, Ijk_Boring); |
| 18496 | DIP("ldrsbt%s r%u, [r%u], #%c%u\n", |
| 18497 | nCC(INSN_COND), rT, rN, bU ? '+' : '-', imm8); |
| 18498 | goto decode_success; |
| 18499 | } |
| 18500 | } |
| 18501 | |
| 18502 | /* -------------- (A2) LDRSBT reg+/-reg -------------- */ |
| 18503 | /* Load Register Signed Byte Unprivileged: |
| 18504 | ldrsbt<c> Rt, [Rn], +/-Rm |
| 18505 | */ |
| 18506 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,0,0,0,0,0,1,1) |
| 18507 | && INSN(11,4) == BITS8(0,0,0,0,1,1,0,1)) { |
| 18508 | UInt rT = INSN(15,12); |
| 18509 | UInt rN = INSN(19,16); |
| 18510 | UInt bU = INSN(23,23); |
| 18511 | UInt rM = INSN(3,0); |
| 18512 | Bool valid = True; |
| 18513 | if (rT == 15 || rN == 15 || rN == rT || rM == 15) |
| 18514 | valid = False; |
| 18515 | if (valid) { |
| 18516 | IRTemp newRt = newTemp(Ity_I32); |
| 18517 | loadGuardedLE( newRt, |
| 18518 | ILGop_8Sto32, getIRegA(rN), getIRegA(rT), condT ); |
| 18519 | putIRegA(rT, mkexpr(newRt), IRTemp_INVALID, Ijk_Boring); |
| 18520 | IRExpr* erN = binop(bU ? Iop_Add32 : Iop_Sub32, |
| 18521 | getIRegA(rN), getIRegA(rM)); |
| 18522 | putIRegA(rN, erN, condT, Ijk_Boring); |
| 18523 | DIP("ldrsbt%s r%u, [r%u], %cr%u\n", |
| 18524 | nCC(INSN_COND), rT, rN, bU ? '+' : '-', rM); |
| 18525 | goto decode_success; |
| 18526 | } |
| 18527 | } |
| 18528 | |
| 18529 | /* -------------- (A1) STRBT reg+#imm12 -------------- */ |
| 18530 | /* Store Register Byte Unprivileged: |
| 18531 | strbt<c> Rt, [Rn], #+/-imm12 |
| 18532 | */ |
| 18533 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,1,0,0,0,1,1,0) ) { |
| 18534 | UInt rT = INSN(15,12); |
| 18535 | UInt rN = INSN(19,16); |
| 18536 | UInt imm12 = INSN(11,0); |
| 18537 | UInt bU = INSN(23,23); |
| 18538 | Bool valid = True; |
| 18539 | if (rT == 15 || rN == 15 || rN == rT) valid = False; |
| 18540 | if (valid) { |
| 18541 | IRExpr* address = getIRegA(rN); |
| 18542 | IRExpr* data = unop(Iop_32to8, getIRegA(rT)); |
| 18543 | storeGuardedLE( address, data, condT); |
| 18544 | IRExpr* newRn = binop(bU ? Iop_Add32 : Iop_Sub32, |
| 18545 | getIRegA(rN), mkU32(imm12)); |
| 18546 | putIRegA(rN, newRn, condT, Ijk_Boring); |
| 18547 | DIP("strbt%s r%u, [r%u], #%c%u\n", |
| 18548 | nCC(INSN_COND), rT, rN, bU ? '+' : '-', imm12); |
| 18549 | goto decode_success; |
| 18550 | } |
| 18551 | } |
| 18552 | |
| 18553 | /* -------------- (A2) STRBT reg+/-reg -------------- */ |
| 18554 | /* Store Register Byte Unprivileged: |
| 18555 | strbt<c> Rt, [Rn], +/-Rm{, shift} |
| 18556 | */ |
| 18557 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,1,1,0,0,1,1,0) |
| 18558 | && INSN(4,4) == 0) { |
| 18559 | UInt rT = INSN(15,12); |
| 18560 | UInt rN = INSN(19,16); |
| 18561 | UInt imm5 = INSN(11,7); |
| 18562 | UInt type = INSN(6,5); |
| 18563 | UInt rM = INSN(3,0); |
| 18564 | UInt bU = INSN(23,23); |
| 18565 | Bool valid = True; |
| 18566 | if (rT == 15 || rN == 15 || rN == rT || rM == 15) valid = False; |
| 18567 | if (valid) { |
| 18568 | IRExpr* address = getIRegA(rN); |
| 18569 | IRExpr* data = unop(Iop_32to8, getIRegA(rT)); |
| 18570 | storeGuardedLE( address, data, condT); |
| 18571 | // dis_buf generated is slightly bogus, in fact. |
| 18572 | IRExpr* erN = mk_EA_reg_plusminus_shifted_reg(rN, bU, rM, |
| 18573 | type, imm5, dis_buf); |
| 18574 | putIRegA(rN, erN, condT, Ijk_Boring); |
| 18575 | DIP("strbt%s r%u, %s\n", nCC(INSN_COND), rT, dis_buf); |
| 18576 | goto decode_success; |
| 18577 | } |
| 18578 | } |
| 18579 | |
| 18580 | /* -------------- (A1) STRHT reg+#imm8 -------------- */ |
| 18581 | /* Store Register Halfword Unprivileged: |
| 18582 | strht<c> Rt, [Rn], #+/-imm8 |
| 18583 | */ |
| 18584 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,0,0,0,0,1,1,0) |
| 18585 | && INSN(7,4) == BITS4(1,0,1,1) ) { |
| 18586 | UInt rT = INSN(15,12); |
| 18587 | UInt rN = INSN(19,16); |
| 18588 | UInt imm4H = INSN(11,8); |
| 18589 | UInt imm4L = INSN(3,0); |
| 18590 | UInt imm8 = (imm4H << 4) | imm4L; |
| 18591 | UInt bU = INSN(23,23); |
| 18592 | Bool valid = True; |
| 18593 | if (rT == 15 || rN == 15 || rN == rT) valid = False; |
| 18594 | if (valid) { |
| 18595 | IRExpr* address = getIRegA(rN); |
| 18596 | IRExpr* data = unop(Iop_32to16, getIRegA(rT)); |
| 18597 | storeGuardedLE( address, data, condT); |
| 18598 | IRExpr* newRn = binop(bU ? Iop_Add32 : Iop_Sub32, |
| 18599 | getIRegA(rN), mkU32(imm8)); |
| 18600 | putIRegA(rN, newRn, condT, Ijk_Boring); |
| 18601 | DIP("strht%s r%u, [r%u], #%c%u\n", |
| 18602 | nCC(INSN_COND), rT, rN, bU ? '+' : '-', imm8); |
| 18603 | goto decode_success; |
| 18604 | } |
| 18605 | } |
| 18606 | |
| 18607 | /* -------------- (A2) STRHT reg+reg -------------- */ |
| 18608 | /* Store Register Halfword Unprivileged: |
| 18609 | strht<c> Rt, [Rn], +/-Rm |
| 18610 | */ |
| 18611 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,0,0,0,0,0,1,0) |
| 18612 | && INSN(11,4) == BITS8(0,0,0,0,1,0,1,1) ) { |
| 18613 | UInt rT = INSN(15,12); |
| 18614 | UInt rN = INSN(19,16); |
| 18615 | UInt rM = INSN(3,0); |
| 18616 | UInt bU = INSN(23,23); |
| 18617 | Bool valid = True; |
| 18618 | if (rT == 15 || rN == 15 || rN == rT || rM == 15) valid = False; |
| 18619 | if (valid) { |
| 18620 | IRExpr* address = getIRegA(rN); |
| 18621 | IRExpr* data = unop(Iop_32to16, getIRegA(rT)); |
| 18622 | storeGuardedLE( address, data, condT); |
| 18623 | IRExpr* newRn = binop(bU ? Iop_Add32 : Iop_Sub32, |
| 18624 | getIRegA(rN), getIRegA(rM)); |
| 18625 | putIRegA(rN, newRn, condT, Ijk_Boring); |
| 18626 | DIP("strht%s r%u, [r%u], %cr%u\n", |
| 18627 | nCC(INSN_COND), rT, rN, bU ? '+' : '-', rM); |
| 18628 | goto decode_success; |
| 18629 | } |
| 18630 | } |
| 18631 | |
| 18632 | /* -------------- (A1) STRT reg+imm12 -------------- */ |
| 18633 | /* Store Register Unprivileged: |
| 18634 | strt<c> Rt, [Rn], #+/-imm12 |
| 18635 | */ |
| 18636 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,1,0,0,0,0,1,0) ) { |
| 18637 | UInt rT = INSN(15,12); |
| 18638 | UInt rN = INSN(19,16); |
| 18639 | UInt imm12 = INSN(11,0); |
| 18640 | UInt bU = INSN(23,23); |
| 18641 | Bool valid = True; |
| 18642 | if (rN == 15 || rN == rT) valid = False; |
| 18643 | if (valid) { |
| 18644 | IRExpr* address = getIRegA(rN); |
| 18645 | storeGuardedLE( address, getIRegA(rT), condT); |
| 18646 | IRExpr* newRn = binop(bU ? Iop_Add32 : Iop_Sub32, |
| 18647 | getIRegA(rN), mkU32(imm12)); |
| 18648 | putIRegA(rN, newRn, condT, Ijk_Boring); |
| 18649 | DIP("strt%s r%u, [r%u], %c%u\n", |
| 18650 | nCC(INSN_COND), rT, rN, bU ? '+' : '-', imm12); |
| 18651 | goto decode_success; |
| 18652 | } |
| 18653 | } |
| 18654 | |
| 18655 | /* -------------- (A2) STRT reg+reg -------------- */ |
| 18656 | /* Store Register Unprivileged: |
| 18657 | strt<c> Rt, [Rn], +/-Rm{, shift} |
| 18658 | */ |
| 18659 | if ( (INSN(27,20) & BITS8(1,1,1,1,0,1,1,1)) == BITS8(0,1,1,0,0,0,1,0) |
| 18660 | && INSN(4,4) == 0 ) { |
| 18661 | UInt rT = INSN(15,12); |
| 18662 | UInt rN = INSN(19,16); |
| 18663 | UInt rM = INSN(3,0); |
| 18664 | UInt type = INSN(6,5); |
| 18665 | UInt imm5 = INSN(11,7); |
| 18666 | UInt bU = INSN(23,23); |
| 18667 | Bool valid = True; |
| 18668 | if (rN == 15 || rN == rT || rM == 15) valid = False; |
| 18669 | /* FIXME We didn't do: |
| 18670 | if ArchVersion() < 6 && rM == rN then UNPREDICTABLE */ |
| 18671 | if (valid) { |
| 18672 | storeGuardedLE( getIRegA(rN), getIRegA(rT), condT); |
| 18673 | // dis_buf generated is slightly bogus, in fact. |
| 18674 | IRExpr* erN = mk_EA_reg_plusminus_shifted_reg(rN, bU, rM, |
| 18675 | type, imm5, dis_buf); |
| 18676 | putIRegA(rN, erN, condT, Ijk_Boring); |
| 18677 | DIP("strt%s r%u, %s\n", nCC(INSN_COND), rT, dis_buf); |
| 18678 | goto decode_success; |
| 18679 | } |
| 18680 | } |
| 18681 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18682 | /* ----------------------------------------------------------- */ |
| 18683 | /* -- ARMv7 instructions -- */ |
| 18684 | /* ----------------------------------------------------------- */ |
| 18685 | |
| 18686 | /* -------------- read CP15 TPIDRURO register ------------- */ |
sewardj | 61d7d98 | 2015-08-20 16:17:11 +0000 | [diff] [blame] | 18687 | /* mrc p15, 0, r0, c13, c0, 3 up to |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18688 | mrc p15, 0, r14, c13, c0, 3 |
| 18689 | */ |
| 18690 | /* I don't know whether this is really v7-only. But anyway, we |
| 18691 | have to support it since arm-linux uses TPIDRURO as a thread |
| 18692 | state register. */ |
| 18693 | if (0x0E1D0F70 == (insn & 0x0FFF0FFF)) { |
| 18694 | UInt rD = INSN(15,12); |
| 18695 | if (rD <= 14) { |
| 18696 | /* skip r15, that's too stupid to handle */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18697 | putIRegA(rD, IRExpr_Get(OFFB_TPIDRURO, Ity_I32), |
| 18698 | condT, Ijk_Boring); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18699 | DIP("mrc%s p15,0, r%u, c13, c0, 3\n", nCC(INSN_COND), rD); |
| 18700 | goto decode_success; |
| 18701 | } |
| 18702 | /* fall through */ |
| 18703 | } |
| 18704 | |
sewardj | 61d7d98 | 2015-08-20 16:17:11 +0000 | [diff] [blame] | 18705 | /* -------------- read CP15 PMUSRENR register ------------- */ |
| 18706 | /* mrc p15, 0, r0, c9, c14, 0 up to |
| 18707 | mrc p15, 0, r14, c9, c14, 0 |
| 18708 | */ |
| 18709 | /* A program reading this register is really asking "which |
| 18710 | performance monitoring registes are available in user space? |
| 18711 | The simple answer here is to return zero, meaning "none". See |
| 18712 | #345984. */ |
| 18713 | if (0x0E190F1E == (insn & 0x0FFF0FFF)) { |
| 18714 | UInt rD = INSN(15,12); |
| 18715 | if (rD <= 14) { |
| 18716 | /* skip r15, that's too stupid to handle */ |
| 18717 | putIRegA(rD, mkU32(0), condT, Ijk_Boring); |
| 18718 | DIP("mrc%s p15,0, r%u, c9, c14, 0\n", nCC(INSN_COND), rD); |
| 18719 | goto decode_success; |
| 18720 | } |
| 18721 | /* fall through */ |
| 18722 | } |
| 18723 | |
sewardj | 412098c | 2010-05-04 08:48:43 +0000 | [diff] [blame] | 18724 | /* Handle various kinds of barriers. This is rather indiscriminate |
| 18725 | in the sense that they are all turned into an IR Fence, which |
| 18726 | means we don't know which they are, so the back end has to |
| 18727 | re-emit them all when it comes acrosss an IR Fence. |
| 18728 | */ |
sewardj | 1331f4d | 2012-02-17 15:07:09 +0000 | [diff] [blame] | 18729 | /* v6 */ /* mcr 15, 0, rT, c7, c10, 5 */ |
| 18730 | if (0xEE070FBA == (insn & 0xFFFF0FFF)) { |
| 18731 | UInt rT = INSN(15,12); |
| 18732 | if (rT <= 14) { |
| 18733 | /* mcr 15, 0, rT, c7, c10, 5 (v6) equiv to DMB (v7). Data |
| 18734 | Memory Barrier -- ensures ordering of memory accesses. */ |
| 18735 | stmt( IRStmt_MBE(Imbe_Fence) ); |
| 18736 | DIP("mcr 15, 0, r%u, c7, c10, 5 (data memory barrier)\n", rT); |
| 18737 | goto decode_success; |
| 18738 | } |
| 18739 | /* fall through */ |
| 18740 | } |
| 18741 | /* other flavours of barrier */ |
sewardj | 412098c | 2010-05-04 08:48:43 +0000 | [diff] [blame] | 18742 | switch (insn) { |
| 18743 | case 0xEE070F9A: /* v6 */ |
| 18744 | /* mcr 15, 0, r0, c7, c10, 4 (v6) equiv to DSB (v7). Data |
| 18745 | Synch Barrier -- ensures completion of memory accesses. */ |
| 18746 | stmt( IRStmt_MBE(Imbe_Fence) ); |
| 18747 | DIP("mcr 15, 0, r0, c7, c10, 4 (data synch barrier)\n"); |
| 18748 | goto decode_success; |
sewardj | 412098c | 2010-05-04 08:48:43 +0000 | [diff] [blame] | 18749 | case 0xEE070F95: /* v6 */ |
| 18750 | /* mcr 15, 0, r0, c7, c5, 4 (v6) equiv to ISB (v7). |
| 18751 | Instruction Synchronisation Barrier (or Flush Prefetch |
| 18752 | Buffer) -- a pipe flush, I think. I suspect we could |
| 18753 | ignore those, but to be on the safe side emit a fence |
| 18754 | anyway. */ |
| 18755 | stmt( IRStmt_MBE(Imbe_Fence) ); |
| 18756 | DIP("mcr 15, 0, r0, c7, c5, 4 (insn synch barrier)\n"); |
| 18757 | goto decode_success; |
| 18758 | default: |
| 18759 | break; |
| 18760 | } |
| 18761 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18762 | /* ----------------------------------------------------------- */ |
sewardj | 5e65812 | 2015-08-17 13:55:41 +0000 | [diff] [blame] | 18763 | /* -- Hints -- */ |
| 18764 | /* ----------------------------------------------------------- */ |
| 18765 | |
| 18766 | switch (insn & 0x0FFFFFFF) { |
| 18767 | /* ------------------- NOP ------------------ */ |
| 18768 | case 0x0320F000: |
| 18769 | DIP("nop%s\n", nCC(INSN_COND)); |
| 18770 | goto decode_success; |
| 18771 | /* ------------------- YIELD ------------------ */ |
| 18772 | case 0x0320F001: |
| 18773 | /* Continue after conditionally yielding. */ |
| 18774 | DIP("yield%s\n", nCC(INSN_COND)); |
| 18775 | stmt( IRStmt_Exit( unop(Iop_32to1, |
| 18776 | condT == IRTemp_INVALID |
| 18777 | ? mkU32(1) : mkexpr(condT)), |
| 18778 | Ijk_Yield, |
| 18779 | IRConst_U32(guest_R15_curr_instr_notENC + 4), |
| 18780 | OFFB_R15T )); |
| 18781 | goto decode_success; |
| 18782 | default: |
| 18783 | break; |
| 18784 | } |
| 18785 | |
| 18786 | /* ----------------------------------------------------------- */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18787 | /* -- VFP (CP 10, CP 11) instructions (in ARM mode) -- */ |
| 18788 | /* ----------------------------------------------------------- */ |
| 18789 | |
| 18790 | if (INSN_COND != ARMCondNV) { |
| 18791 | Bool ok_vfp = decode_CP10_CP11_instruction ( |
| 18792 | &dres, INSN(27,0), condT, INSN_COND, |
| 18793 | False/*!isT*/ |
| 18794 | ); |
| 18795 | if (ok_vfp) |
| 18796 | goto decode_success; |
| 18797 | } |
| 18798 | |
| 18799 | /* ----------------------------------------------------------- */ |
| 18800 | /* -- NEON instructions (in ARM mode) -- */ |
| 18801 | /* ----------------------------------------------------------- */ |
| 18802 | |
| 18803 | /* These are all in NV space, and so are taken care of (far) above, |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 18804 | by a call from this function to |
| 18805 | decode_NV_instruction_ARMv7_and_below(). */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18806 | |
| 18807 | /* ----------------------------------------------------------- */ |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 18808 | /* -- v6 media instructions (in ARM mode) -- */ |
| 18809 | /* ----------------------------------------------------------- */ |
| 18810 | |
| 18811 | { Bool ok_v6m = decode_V6MEDIA_instruction( |
| 18812 | &dres, INSN(27,0), condT, INSN_COND, |
| 18813 | False/*!isT*/ |
| 18814 | ); |
| 18815 | if (ok_v6m) |
| 18816 | goto decode_success; |
| 18817 | } |
| 18818 | |
| 18819 | /* ----------------------------------------------------------- */ |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 18820 | /* -- v8 instructions (in ARM mode) -- */ |
| 18821 | /* ----------------------------------------------------------- */ |
| 18822 | |
| 18823 | after_v7_decoder: |
| 18824 | |
| 18825 | /* If we get here, it means that all attempts to decode the |
| 18826 | instruction as ARMv7 or earlier have failed. So, if we're doing |
| 18827 | ARMv8 or later, here is the point to try for it. */ |
| 18828 | |
| 18829 | if (VEX_ARM_ARCHLEVEL(archinfo->hwcaps) >= 8) { |
| 18830 | Bool ok_v8 |
| 18831 | = decode_V8_instruction( &dres, insn, condT, False/*!isT*/, |
| 18832 | IRTemp_INVALID, IRTemp_INVALID ); |
| 18833 | if (ok_v8) |
| 18834 | goto decode_success; |
| 18835 | } |
| 18836 | |
| 18837 | /* ----------------------------------------------------------- */ |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18838 | /* -- Undecodable -- */ |
| 18839 | /* ----------------------------------------------------------- */ |
| 18840 | |
| 18841 | goto decode_failure; |
| 18842 | /*NOTREACHED*/ |
| 18843 | |
| 18844 | decode_failure: |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 18845 | /* All decode failures end up here. */ |
sewardj | 442e51a | 2012-12-06 18:08:04 +0000 | [diff] [blame] | 18846 | if (sigill_diag) { |
| 18847 | vex_printf("disInstr(arm): unhandled instruction: " |
| 18848 | "0x%x\n", insn); |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 18849 | vex_printf(" cond=%d(0x%x) 27:20=%d(0x%02x) " |
sewardj | 442e51a | 2012-12-06 18:08:04 +0000 | [diff] [blame] | 18850 | "4:4=%d " |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 18851 | "3:0=%d(0x%x)\n", |
sewardj | 442e51a | 2012-12-06 18:08:04 +0000 | [diff] [blame] | 18852 | (Int)INSN_COND, (UInt)INSN_COND, |
| 18853 | (Int)INSN(27,20), (UInt)INSN(27,20), |
| 18854 | (Int)INSN(4,4), |
| 18855 | (Int)INSN(3,0), (UInt)INSN(3,0) ); |
| 18856 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18857 | |
| 18858 | /* Tell the dispatcher that this insn cannot be decoded, and so has |
| 18859 | not been executed, and (is currently) the next to be executed. |
| 18860 | R15 should be up-to-date since it made so at the start of each |
| 18861 | insn, but nevertheless be paranoid and update it again right |
| 18862 | now. */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18863 | vassert(0 == (guest_R15_curr_instr_notENC & 3)); |
| 18864 | llPutIReg( 15, mkU32(guest_R15_curr_instr_notENC) ); |
philippe | 2faf591 | 2014-08-11 22:45:47 +0000 | [diff] [blame] | 18865 | dres.len = 0; |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 18866 | dres.whatNext = Dis_StopHere; |
| 18867 | dres.jk_StopHere = Ijk_NoDecode; |
philippe | 2faf591 | 2014-08-11 22:45:47 +0000 | [diff] [blame] | 18868 | dres.continueAt = 0; |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18869 | return dres; |
| 18870 | |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 18871 | decode_success: |
| 18872 | /* All decode successes end up here. */ |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 18873 | DIP("\n"); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18874 | |
| 18875 | vassert(dres.len == 4 || dres.len == 20); |
| 18876 | |
| 18877 | /* Now then. Do we have an implicit jump to r15 to deal with? */ |
| 18878 | if (r15written) { |
| 18879 | /* If we get jump to deal with, we assume that there's been no |
| 18880 | other competing branch stuff previously generated for this |
| 18881 | insn. That's reasonable, in the sense that the ARM insn set |
| 18882 | appears to declare as "Unpredictable" any instruction which |
| 18883 | generates more than one possible new value for r15. Hence |
| 18884 | just assert. The decoders themselves should check against |
| 18885 | all such instructions which are thusly Unpredictable, and |
| 18886 | decline to decode them. Hence we should never get here if we |
| 18887 | have competing new values for r15, and hence it is safe to |
| 18888 | assert here. */ |
| 18889 | vassert(dres.whatNext == Dis_Continue); |
| 18890 | vassert(irsb->next == NULL); |
sewardj | f6d2cf9 | 2011-05-03 11:08:39 +0000 | [diff] [blame] | 18891 | vassert(irsb->jumpkind == Ijk_Boring); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18892 | /* If r15 is unconditionally written, terminate the block by |
| 18893 | jumping to it. If it's conditionally written, still |
| 18894 | terminate the block (a shame, but we can't do side exits to |
| 18895 | arbitrary destinations), but first jump to the next |
| 18896 | instruction if the condition doesn't hold. */ |
| 18897 | /* We can't use getIReg(15) to get the destination, since that |
| 18898 | will produce r15+8, which isn't what we want. Must use |
| 18899 | llGetIReg(15) instead. */ |
| 18900 | if (r15guard == IRTemp_INVALID) { |
| 18901 | /* unconditional */ |
| 18902 | } else { |
| 18903 | /* conditional */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18904 | stmt( IRStmt_Exit( |
| 18905 | unop(Iop_32to1, |
| 18906 | binop(Iop_Xor32, |
| 18907 | mkexpr(r15guard), mkU32(1))), |
| 18908 | r15kind, |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 18909 | IRConst_U32(guest_R15_curr_instr_notENC + 4), |
| 18910 | OFFB_R15T |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18911 | )); |
| 18912 | } |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 18913 | /* This seems crazy, but we're required to finish the insn with |
| 18914 | a write to the guest PC. As usual we rely on ir_opt to tidy |
| 18915 | up later. */ |
| 18916 | llPutIReg(15, llGetIReg(15)); |
| 18917 | dres.whatNext = Dis_StopHere; |
| 18918 | dres.jk_StopHere = r15kind; |
| 18919 | } else { |
| 18920 | /* Set up the end-state in the normal way. */ |
| 18921 | switch (dres.whatNext) { |
| 18922 | case Dis_Continue: |
| 18923 | llPutIReg(15, mkU32(dres.len + guest_R15_curr_instr_notENC)); |
| 18924 | break; |
| 18925 | case Dis_ResteerU: |
| 18926 | case Dis_ResteerC: |
| 18927 | llPutIReg(15, mkU32(dres.continueAt)); |
| 18928 | break; |
| 18929 | case Dis_StopHere: |
| 18930 | break; |
| 18931 | default: |
| 18932 | vassert(0); |
| 18933 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18934 | } |
| 18935 | |
| 18936 | return dres; |
| 18937 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 18938 | # undef INSN_COND |
sewardj | 80bea7b | 2010-01-09 11:43:21 +0000 | [diff] [blame] | 18939 | # undef INSN |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 18940 | } |
| 18941 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18942 | |
| 18943 | /*------------------------------------------------------------*/ |
| 18944 | /*--- Disassemble a single Thumb2 instruction ---*/ |
| 18945 | /*------------------------------------------------------------*/ |
| 18946 | |
sewardj | 66c8c9b | 2011-07-04 16:58:40 +0000 | [diff] [blame] | 18947 | static const UChar it_length_table[256]; /* fwds */ |
| 18948 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18949 | /* NB: in Thumb mode we do fetches of regs with getIRegT, which |
| 18950 | automagically adds 4 to fetches of r15. However, writes to regs |
| 18951 | are done with putIRegT, which disallows writes to r15. Hence any |
| 18952 | r15 writes and associated jumps have to be done "by hand". */ |
| 18953 | |
| 18954 | /* Disassemble a single Thumb instruction into IR. The instruction is |
| 18955 | located in host memory at guest_instr, and has (decoded) guest IP |
| 18956 | of guest_R15_curr_instr_notENC, which will have been set before the |
| 18957 | call here. */ |
| 18958 | |
| 18959 | static |
| 18960 | DisResult disInstr_THUMB_WRK ( |
florian | beac530 | 2014-12-31 12:09:38 +0000 | [diff] [blame] | 18961 | Bool (*resteerOkFn) ( /*opaque*/void*, Addr ), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18962 | Bool resteerCisOk, |
| 18963 | void* callback_opaque, |
florian | 8462d11 | 2014-09-24 15:18:09 +0000 | [diff] [blame] | 18964 | const UChar* guest_instr, |
florian | cacba8e | 2014-12-15 18:58:07 +0000 | [diff] [blame] | 18965 | const VexArchInfo* archinfo, |
| 18966 | const VexAbiInfo* abiinfo, |
sewardj | 442e51a | 2012-12-06 18:08:04 +0000 | [diff] [blame] | 18967 | Bool sigill_diag |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18968 | ) |
| 18969 | { |
| 18970 | /* A macro to fish bits out of insn0. There's also INSN1, to fish |
| 18971 | bits out of insn1, but that's defined only after the end of the |
| 18972 | 16-bit insn decoder, so as to stop it mistakenly being used |
| 18973 | therein. */ |
| 18974 | # define INSN0(_bMax,_bMin) SLICE_UInt(((UInt)insn0), (_bMax), (_bMin)) |
| 18975 | |
| 18976 | DisResult dres; |
sewardj | 5276ff5 | 2012-07-14 14:21:56 +0000 | [diff] [blame] | 18977 | UShort insn0; /* first 16 bits of the insn */ |
| 18978 | UShort insn1; /* second 16 bits of the insn */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18979 | HChar dis_buf[128]; // big enough to hold LDMIA etc text |
| 18980 | |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 18981 | /* Summary result of the ITxxx backwards analysis: False == safe |
| 18982 | but suboptimal. */ |
| 18983 | Bool guaranteedUnconditional = False; |
| 18984 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18985 | /* Set result defaults. */ |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 18986 | dres.whatNext = Dis_Continue; |
| 18987 | dres.len = 2; |
| 18988 | dres.continueAt = 0; |
| 18989 | dres.jk_StopHere = Ijk_INVALID; |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 18990 | dres.hint = Dis_HintNone; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 18991 | |
| 18992 | /* Set default actions for post-insn handling of writes to r15, if |
| 18993 | required. */ |
| 18994 | r15written = False; |
| 18995 | r15guard = IRTemp_INVALID; /* unconditional */ |
| 18996 | r15kind = Ijk_Boring; |
| 18997 | |
| 18998 | /* Insns could be 2 or 4 bytes long. Just get the first 16 bits at |
| 18999 | this point. If we need the second 16, get them later. We can't |
| 19000 | get them both out immediately because it risks a fault (very |
| 19001 | unlikely, but ..) if the second 16 bits aren't actually |
| 19002 | necessary. */ |
| 19003 | insn0 = getUShortLittleEndianly( guest_instr ); |
sewardj | 5276ff5 | 2012-07-14 14:21:56 +0000 | [diff] [blame] | 19004 | insn1 = 0; /* We'll get it later, once we know we need it. */ |
| 19005 | |
| 19006 | /* Similarly, will set this later. */ |
| 19007 | IRTemp old_itstate = IRTemp_INVALID; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19008 | |
| 19009 | if (0) vex_printf("insn: 0x%x\n", insn0); |
| 19010 | |
| 19011 | DIP("\t(thumb) 0x%x: ", (UInt)guest_R15_curr_instr_notENC); |
| 19012 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19013 | vassert(0 == (guest_R15_curr_instr_notENC & 1)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19014 | |
| 19015 | /* ----------------------------------------------------------- */ |
| 19016 | /* Spot "Special" instructions (see comment at top of file). */ |
| 19017 | { |
florian | 8462d11 | 2014-09-24 15:18:09 +0000 | [diff] [blame] | 19018 | const UChar* code = guest_instr; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19019 | /* Spot the 16-byte preamble: |
| 19020 | |
| 19021 | ea4f 0cfc mov.w ip, ip, ror #3 |
| 19022 | ea4f 3c7c mov.w ip, ip, ror #13 |
| 19023 | ea4f 7c7c mov.w ip, ip, ror #29 |
| 19024 | ea4f 4cfc mov.w ip, ip, ror #19 |
| 19025 | */ |
| 19026 | UInt word1 = 0x0CFCEA4F; |
| 19027 | UInt word2 = 0x3C7CEA4F; |
| 19028 | UInt word3 = 0x7C7CEA4F; |
| 19029 | UInt word4 = 0x4CFCEA4F; |
| 19030 | if (getUIntLittleEndianly(code+ 0) == word1 && |
| 19031 | getUIntLittleEndianly(code+ 4) == word2 && |
| 19032 | getUIntLittleEndianly(code+ 8) == word3 && |
| 19033 | getUIntLittleEndianly(code+12) == word4) { |
| 19034 | /* Got a "Special" instruction preamble. Which one is it? */ |
| 19035 | // 0x 0A 0A EA 4A |
| 19036 | if (getUIntLittleEndianly(code+16) == 0x0A0AEA4A |
| 19037 | /* orr.w r10,r10,r10 */) { |
| 19038 | /* R3 = client_request ( R4 ) */ |
| 19039 | DIP("r3 = client_request ( %%r4 )\n"); |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 19040 | llPutIReg(15, mkU32( (guest_R15_curr_instr_notENC + 20) | 1 )); |
| 19041 | dres.jk_StopHere = Ijk_ClientReq; |
| 19042 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19043 | goto decode_success; |
| 19044 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19045 | else |
| 19046 | // 0x 0B 0B EA 4B |
sewardj | c070405 | 2010-08-22 22:21:19 +0000 | [diff] [blame] | 19047 | if (getUIntLittleEndianly(code+16) == 0x0B0BEA4B |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19048 | /* orr r11,r11,r11 */) { |
| 19049 | /* R3 = guest_NRADDR */ |
| 19050 | DIP("r3 = guest_NRADDR\n"); |
| 19051 | dres.len = 20; |
| 19052 | llPutIReg(3, IRExpr_Get( OFFB_NRADDR, Ity_I32 )); |
| 19053 | goto decode_success; |
| 19054 | } |
| 19055 | else |
| 19056 | // 0x 0C 0C EA 4C |
sewardj | c070405 | 2010-08-22 22:21:19 +0000 | [diff] [blame] | 19057 | if (getUIntLittleEndianly(code+16) == 0x0C0CEA4C |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19058 | /* orr r12,r12,r12 */) { |
| 19059 | /* branch-and-link-to-noredir R4 */ |
| 19060 | DIP("branch-and-link-to-noredir r4\n"); |
sewardj | c070405 | 2010-08-22 22:21:19 +0000 | [diff] [blame] | 19061 | llPutIReg(14, mkU32( (guest_R15_curr_instr_notENC + 20) | 1 )); |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 19062 | llPutIReg(15, getIRegT(4)); |
| 19063 | dres.jk_StopHere = Ijk_NoRedir; |
| 19064 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19065 | goto decode_success; |
| 19066 | } |
sewardj | a0d8eb8 | 2012-09-03 21:48:42 +0000 | [diff] [blame] | 19067 | else |
| 19068 | // 0x 09 09 EA 49 |
| 19069 | if (getUIntLittleEndianly(code+16) == 0x0909EA49 |
| 19070 | /* orr r9,r9,r9 */) { |
| 19071 | /* IR injection */ |
| 19072 | DIP("IR injection\n"); |
| 19073 | vex_inject_ir(irsb, Iend_LE); |
| 19074 | // Invalidate the current insn. The reason is that the IRop we're |
| 19075 | // injecting here can change. In which case the translation has to |
| 19076 | // be redone. For ease of handling, we simply invalidate all the |
| 19077 | // time. |
sewardj | 05f5e01 | 2014-05-04 10:52:11 +0000 | [diff] [blame] | 19078 | stmt(IRStmt_Put(OFFB_CMSTART, mkU32(guest_R15_curr_instr_notENC))); |
| 19079 | stmt(IRStmt_Put(OFFB_CMLEN, mkU32(20))); |
sewardj | a0d8eb8 | 2012-09-03 21:48:42 +0000 | [diff] [blame] | 19080 | llPutIReg(15, mkU32( (guest_R15_curr_instr_notENC + 20) | 1 )); |
| 19081 | dres.whatNext = Dis_StopHere; |
sewardj | 05f5e01 | 2014-05-04 10:52:11 +0000 | [diff] [blame] | 19082 | dres.jk_StopHere = Ijk_InvalICache; |
sewardj | a0d8eb8 | 2012-09-03 21:48:42 +0000 | [diff] [blame] | 19083 | goto decode_success; |
| 19084 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19085 | /* We don't know what it is. Set insn0 so decode_failure |
| 19086 | can print the insn following the Special-insn preamble. */ |
| 19087 | insn0 = getUShortLittleEndianly(code+16); |
| 19088 | goto decode_failure; |
| 19089 | /*NOTREACHED*/ |
| 19090 | } |
| 19091 | |
| 19092 | } |
| 19093 | |
| 19094 | /* ----------------------------------------------------------- */ |
| 19095 | |
| 19096 | /* Main Thumb instruction decoder starts here. It's a series of |
| 19097 | switches which examine ever longer bit sequences at the MSB of |
| 19098 | the instruction word, first for 16-bit insns, then for 32-bit |
| 19099 | insns. */ |
| 19100 | |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19101 | /* --- BEGIN ITxxx optimisation analysis --- */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19102 | /* This is a crucial optimisation for the ITState boilerplate that |
| 19103 | follows. Examine the 9 halfwords preceding this instruction, |
| 19104 | and if we are absolutely sure that none of them constitute an |
| 19105 | 'it' instruction, then we can be sure that this instruction is |
| 19106 | not under the control of any 'it' instruction, and so |
| 19107 | guest_ITSTATE must be zero. So write zero into ITSTATE right |
| 19108 | now, so that iropt can fold out almost all of the resulting |
| 19109 | junk. |
| 19110 | |
| 19111 | If we aren't sure, we can always safely skip this step. So be a |
| 19112 | bit conservative about it: only poke around in the same page as |
| 19113 | this instruction, lest we get a fault from the previous page |
| 19114 | that would not otherwise have happened. The saving grace is |
| 19115 | that such skipping is pretty rare -- it only happens, |
| 19116 | statistically, 18/4096ths of the time, so is judged unlikely to |
| 19117 | be a performance problems. |
| 19118 | |
| 19119 | FIXME: do better. Take into account the number of insns covered |
| 19120 | by any IT insns we find, to rule out cases where an IT clearly |
| 19121 | cannot cover this instruction. This would improve behaviour for |
| 19122 | branch targets immediately following an IT-guarded group that is |
| 19123 | not of full length. Eg, (and completely ignoring issues of 16- |
| 19124 | vs 32-bit insn length): |
| 19125 | |
| 19126 | ite cond |
| 19127 | insn1 |
| 19128 | insn2 |
| 19129 | label: insn3 |
| 19130 | insn4 |
| 19131 | |
| 19132 | The 'it' only conditionalises insn1 and insn2. However, the |
| 19133 | current analysis is conservative and considers insn3 and insn4 |
| 19134 | also possibly guarded. Hence if 'label:' is the start of a hot |
| 19135 | loop we will get a big performance hit. |
| 19136 | */ |
| 19137 | { |
| 19138 | /* Summary result of this analysis: False == safe but |
| 19139 | suboptimal. */ |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19140 | vassert(guaranteedUnconditional == False); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19141 | |
| 19142 | UInt pc = guest_R15_curr_instr_notENC; |
| 19143 | vassert(0 == (pc & 1)); |
| 19144 | |
| 19145 | UInt pageoff = pc & 0xFFF; |
| 19146 | if (pageoff >= 18) { |
| 19147 | /* It's safe to poke about in the 9 halfwords preceding this |
| 19148 | insn. So, have a look at them. */ |
sewardj | 66c8c9b | 2011-07-04 16:58:40 +0000 | [diff] [blame] | 19149 | guaranteedUnconditional = True; /* assume no 'it' insn found, |
| 19150 | till we do */ |
sewardj | df86f16 | 2010-08-22 18:47:30 +0000 | [diff] [blame] | 19151 | UShort* hwp = (UShort*)(HWord)pc; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19152 | Int i; |
| 19153 | for (i = -1; i >= -9; i--) { |
| 19154 | /* We're in the same page. (True, but commented out due |
| 19155 | to expense.) */ |
| 19156 | /* |
| 19157 | vassert( ( ((UInt)(&hwp[i])) & 0xFFFFF000 ) |
| 19158 | == ( pc & 0xFFFFF000 ) ); |
| 19159 | */ |
| 19160 | /* All valid IT instructions must have the form 0xBFxy, |
sewardj | 66c8c9b | 2011-07-04 16:58:40 +0000 | [diff] [blame] | 19161 | where x can be anything, but y must be nonzero. Find |
| 19162 | the number of insns covered by it (1 .. 4) and check to |
| 19163 | see if it can possibly reach up to the instruction in |
| 19164 | question. Some (x,y) combinations mean UNPREDICTABLE, |
| 19165 | and the table is constructed to be conservative by |
| 19166 | returning 4 for those cases, so the analysis is safe |
| 19167 | even if the code uses unpredictable IT instructions (in |
| 19168 | which case its authors are nuts, but hey.) */ |
| 19169 | UShort hwp_i = hwp[i]; |
| 19170 | if (UNLIKELY((hwp_i & 0xFF00) == 0xBF00 && (hwp_i & 0xF) != 0)) { |
| 19171 | /* might be an 'it' insn. */ |
| 19172 | /* # guarded insns */ |
| 19173 | Int n_guarded = (Int)it_length_table[hwp_i & 0xFF]; |
| 19174 | vassert(n_guarded >= 1 && n_guarded <= 4); |
| 19175 | if (n_guarded * 2 /* # guarded HWs, worst case */ |
| 19176 | > (-(i+1))) /* -(i+1): # remaining HWs after the IT */ |
| 19177 | /* -(i+0) also seems to work, even though I think |
| 19178 | it's wrong. I don't understand that. */ |
| 19179 | guaranteedUnconditional = False; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19180 | break; |
| 19181 | } |
| 19182 | } |
| 19183 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19184 | } |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19185 | /* --- END ITxxx optimisation analysis --- */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19186 | |
| 19187 | /* Generate the guarding condition for this insn, by examining |
| 19188 | ITSTATE. Assign it to condT. Also, generate new |
| 19189 | values for ITSTATE ready for stuffing back into the |
| 19190 | guest state, but don't actually do the Put yet, since it will |
| 19191 | need to stuffed back in only after the instruction gets to a |
| 19192 | point where it is sure to complete. Mostly we let the code at |
| 19193 | decode_success handle this, but in cases where the insn contains |
| 19194 | a side exit, we have to update them before the exit. */ |
| 19195 | |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19196 | /* If the ITxxx optimisation analysis above could not prove that |
| 19197 | this instruction is guaranteed unconditional, we insert a |
| 19198 | lengthy IR preamble to compute the guarding condition at |
| 19199 | runtime. If it can prove it (which obviously we hope is the |
| 19200 | normal case) then we insert a minimal preamble, which is |
| 19201 | equivalent to setting guest_ITSTATE to zero and then folding |
| 19202 | that through the full preamble (which completely disappears). */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19203 | |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19204 | IRTemp condT = IRTemp_INVALID; |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19205 | IRTemp cond_AND_notInIT_T = IRTemp_INVALID; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19206 | |
sewardj | 5276ff5 | 2012-07-14 14:21:56 +0000 | [diff] [blame] | 19207 | IRTemp new_itstate = IRTemp_INVALID; |
| 19208 | vassert(old_itstate == IRTemp_INVALID); |
| 19209 | |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19210 | if (guaranteedUnconditional) { |
| 19211 | /* BEGIN "partial eval { ITSTATE = 0; STANDARD_PREAMBLE; }" */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19212 | |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19213 | // ITSTATE = 0 :: I32 |
| 19214 | IRTemp z32 = newTemp(Ity_I32); |
| 19215 | assign(z32, mkU32(0)); |
| 19216 | put_ITSTATE(z32); |
| 19217 | |
| 19218 | // old_itstate = 0 :: I32 |
| 19219 | // |
| 19220 | // old_itstate = get_ITSTATE(); |
| 19221 | old_itstate = z32; /* 0 :: I32 */ |
| 19222 | |
| 19223 | // new_itstate = old_itstate >> 8 |
| 19224 | // = 0 >> 8 |
| 19225 | // = 0 :: I32 |
| 19226 | // |
| 19227 | // new_itstate = newTemp(Ity_I32); |
| 19228 | // assign(new_itstate, |
| 19229 | // binop(Iop_Shr32, mkexpr(old_itstate), mkU8(8))); |
| 19230 | new_itstate = z32; |
| 19231 | |
| 19232 | // ITSTATE = 0 :: I32(again) |
| 19233 | // |
| 19234 | // put_ITSTATE(new_itstate); |
| 19235 | |
| 19236 | // condT1 = calc_cond_dyn( xor(and(old_istate,0xF0), 0xE0) ) |
| 19237 | // = calc_cond_dyn( xor(0,0xE0) ) |
| 19238 | // = calc_cond_dyn ( 0xE0 ) |
| 19239 | // = 1 :: I32 |
| 19240 | // Not that this matters, since the computed value is not used: |
| 19241 | // see condT folding below |
| 19242 | // |
| 19243 | // IRTemp condT1 = newTemp(Ity_I32); |
| 19244 | // assign(condT1, |
| 19245 | // mk_armg_calculate_condition_dyn( |
| 19246 | // binop(Iop_Xor32, |
| 19247 | // binop(Iop_And32, mkexpr(old_itstate), mkU32(0xF0)), |
| 19248 | // mkU32(0xE0)) |
| 19249 | // ) |
| 19250 | // ); |
| 19251 | |
| 19252 | // condT = 32to8(and32(old_itstate,0xF0)) == 0 ? 1 : condT1 |
| 19253 | // = 32to8(and32(0,0xF0)) == 0 ? 1 : condT1 |
| 19254 | // = 32to8(0) == 0 ? 1 : condT1 |
| 19255 | // = 0 == 0 ? 1 : condT1 |
| 19256 | // = 1 |
| 19257 | // |
| 19258 | // condT = newTemp(Ity_I32); |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 19259 | // assign(condT, IRExpr_ITE( |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19260 | // unop(Iop_32to8, binop(Iop_And32, |
| 19261 | // mkexpr(old_itstate), |
| 19262 | // mkU32(0xF0))), |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 19263 | // mkexpr(condT1), |
| 19264 | // mkU32(1)) |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19265 | // )); |
| 19266 | condT = newTemp(Ity_I32); |
| 19267 | assign(condT, mkU32(1)); |
| 19268 | |
| 19269 | // notInITt = xor32(and32(old_itstate, 1), 1) |
| 19270 | // = xor32(and32(0, 1), 1) |
| 19271 | // = xor32(0, 1) |
| 19272 | // = 1 :: I32 |
| 19273 | // |
| 19274 | // IRTemp notInITt = newTemp(Ity_I32); |
| 19275 | // assign(notInITt, |
| 19276 | // binop(Iop_Xor32, |
| 19277 | // binop(Iop_And32, mkexpr(old_itstate), mkU32(1)), |
| 19278 | // mkU32(1))); |
| 19279 | |
| 19280 | // cond_AND_notInIT_T = and32(notInITt, condT) |
| 19281 | // = and32(1, 1) |
| 19282 | // = 1 |
| 19283 | // |
| 19284 | // cond_AND_notInIT_T = newTemp(Ity_I32); |
| 19285 | // assign(cond_AND_notInIT_T, |
| 19286 | // binop(Iop_And32, mkexpr(notInITt), mkexpr(condT))); |
| 19287 | cond_AND_notInIT_T = condT; /* 1 :: I32 */ |
| 19288 | |
| 19289 | /* END "partial eval { ITSTATE = 0; STANDARD_PREAMBLE; }" */ |
| 19290 | } else { |
| 19291 | /* BEGIN { STANDARD PREAMBLE; } */ |
| 19292 | |
| 19293 | old_itstate = get_ITSTATE(); |
| 19294 | |
| 19295 | new_itstate = newTemp(Ity_I32); |
| 19296 | assign(new_itstate, |
| 19297 | binop(Iop_Shr32, mkexpr(old_itstate), mkU8(8))); |
| 19298 | |
| 19299 | put_ITSTATE(new_itstate); |
| 19300 | |
| 19301 | /* Same strategy as for ARM insns: generate a condition |
| 19302 | temporary at this point (or IRTemp_INVALID, meaning |
| 19303 | unconditional). We leave it to lower-level instruction |
| 19304 | decoders to decide whether they can generate straight-line |
| 19305 | code, or whether they must generate a side exit before the |
| 19306 | instruction. condT :: Ity_I32 and is always either zero or |
| 19307 | one. */ |
| 19308 | IRTemp condT1 = newTemp(Ity_I32); |
| 19309 | assign(condT1, |
| 19310 | mk_armg_calculate_condition_dyn( |
| 19311 | binop(Iop_Xor32, |
| 19312 | binop(Iop_And32, mkexpr(old_itstate), mkU32(0xF0)), |
| 19313 | mkU32(0xE0)) |
| 19314 | ) |
| 19315 | ); |
| 19316 | |
| 19317 | /* This is a bit complex, but needed to make Memcheck understand |
| 19318 | that, if the condition in old_itstate[7:4] denotes AL (that |
| 19319 | is, if this instruction is to be executed unconditionally), |
| 19320 | then condT does not depend on the results of calling the |
| 19321 | helper. |
| 19322 | |
| 19323 | We test explicitly for old_itstate[7:4] == AL ^ 0xE, and in |
| 19324 | that case set condT directly to 1. Else we use the results |
| 19325 | of the helper. Since old_itstate is always defined and |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 19326 | because Memcheck does lazy V-bit propagation through ITE, |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19327 | this will cause condT to always be a defined 1 if the |
| 19328 | condition is 'AL'. From an execution semantics point of view |
| 19329 | this is irrelevant since we're merely duplicating part of the |
| 19330 | behaviour of the helper. But it makes it clear to Memcheck, |
| 19331 | in this case, that condT does not in fact depend on the |
| 19332 | contents of the condition code thunk. Without it, we get |
| 19333 | quite a lot of false errors. |
| 19334 | |
| 19335 | So, just to clarify: from a straight semantics point of view, |
| 19336 | we can simply do "assign(condT, mkexpr(condT1))", and the |
| 19337 | simulator still runs fine. It's just that we get loads of |
| 19338 | false errors from Memcheck. */ |
| 19339 | condT = newTemp(Ity_I32); |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 19340 | assign(condT, IRExpr_ITE( |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 19341 | binop(Iop_CmpNE32, binop(Iop_And32, |
| 19342 | mkexpr(old_itstate), |
| 19343 | mkU32(0xF0)), |
| 19344 | mkU32(0)), |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 19345 | mkexpr(condT1), |
| 19346 | mkU32(1) |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19347 | )); |
| 19348 | |
| 19349 | /* Something we don't have in ARM: generate a 0 or 1 value |
| 19350 | indicating whether or not we are in an IT block (NB: 0 = in |
| 19351 | IT block, 1 = not in IT block). This is used to gate |
| 19352 | condition code updates in 16-bit Thumb instructions. */ |
| 19353 | IRTemp notInITt = newTemp(Ity_I32); |
| 19354 | assign(notInITt, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19355 | binop(Iop_Xor32, |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19356 | binop(Iop_And32, mkexpr(old_itstate), mkU32(1)), |
| 19357 | mkU32(1))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19358 | |
sewardj | a002def | 2010-10-05 22:29:49 +0000 | [diff] [blame] | 19359 | /* Compute 'condT && notInITt' -- that is, the instruction is |
| 19360 | going to execute, and we're not in an IT block. This is the |
| 19361 | gating condition for updating condition codes in 16-bit Thumb |
| 19362 | instructions, except for CMP, CMN and TST. */ |
| 19363 | cond_AND_notInIT_T = newTemp(Ity_I32); |
| 19364 | assign(cond_AND_notInIT_T, |
| 19365 | binop(Iop_And32, mkexpr(notInITt), mkexpr(condT))); |
| 19366 | /* END { STANDARD PREAMBLE; } */ |
| 19367 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19368 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19369 | |
| 19370 | /* At this point: |
| 19371 | * ITSTATE has been updated |
| 19372 | * condT holds the guarding condition for this instruction (0 or 1), |
| 19373 | * notInITt is 1 if we're in "normal" code, 0 if in an IT block |
| 19374 | * cond_AND_notInIT_T is the AND of the above two. |
| 19375 | |
| 19376 | If the instruction proper can't trap, then there's nothing else |
| 19377 | to do w.r.t. ITSTATE -- just go and and generate IR for the |
| 19378 | insn, taking into account the guarding condition. |
| 19379 | |
| 19380 | If, however, the instruction might trap, then we must back up |
| 19381 | ITSTATE to the old value, and re-update it after the potentially |
| 19382 | trapping IR section. A trap can happen either via a memory |
| 19383 | reference or because we need to throw SIGILL. |
| 19384 | |
| 19385 | If an instruction has a side exit, we need to be sure that any |
| 19386 | ITSTATE backup is re-updated before the side exit. |
| 19387 | */ |
| 19388 | |
| 19389 | /* ----------------------------------------------------------- */ |
| 19390 | /* -- -- */ |
| 19391 | /* -- Thumb 16-bit integer instructions -- */ |
| 19392 | /* -- -- */ |
| 19393 | /* -- IMPORTANT: references to insn1 or INSN1 are -- */ |
| 19394 | /* -- not allowed in this section -- */ |
| 19395 | /* -- -- */ |
| 19396 | /* ----------------------------------------------------------- */ |
| 19397 | |
| 19398 | /* 16-bit instructions inside an IT block, apart from CMP, CMN and |
| 19399 | TST, do not set the condition codes. Hence we must dynamically |
| 19400 | test for this case for every condition code update. */ |
| 19401 | |
| 19402 | IROp anOp = Iop_INVALID; |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 19403 | const HChar* anOpNm = NULL; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19404 | |
| 19405 | /* ================ 16-bit 15:6 cases ================ */ |
| 19406 | |
| 19407 | switch (INSN0(15,6)) { |
| 19408 | |
| 19409 | case 0x10a: // CMP |
| 19410 | case 0x10b: { // CMN |
| 19411 | /* ---------------- CMP Rn, Rm ---------------- */ |
| 19412 | Bool isCMN = INSN0(15,6) == 0x10b; |
| 19413 | UInt rN = INSN0(2,0); |
| 19414 | UInt rM = INSN0(5,3); |
| 19415 | IRTemp argL = newTemp(Ity_I32); |
| 19416 | IRTemp argR = newTemp(Ity_I32); |
| 19417 | assign( argL, getIRegT(rN) ); |
| 19418 | assign( argR, getIRegT(rM) ); |
| 19419 | /* Update flags regardless of whether in an IT block or not. */ |
| 19420 | setFlags_D1_D2( isCMN ? ARMG_CC_OP_ADD : ARMG_CC_OP_SUB, |
| 19421 | argL, argR, condT ); |
| 19422 | DIP("%s r%u, r%u\n", isCMN ? "cmn" : "cmp", rN, rM); |
| 19423 | goto decode_success; |
| 19424 | } |
| 19425 | |
| 19426 | case 0x108: { |
| 19427 | /* ---------------- TST Rn, Rm ---------------- */ |
| 19428 | UInt rN = INSN0(2,0); |
| 19429 | UInt rM = INSN0(5,3); |
| 19430 | IRTemp oldC = newTemp(Ity_I32); |
| 19431 | IRTemp oldV = newTemp(Ity_I32); |
| 19432 | IRTemp res = newTemp(Ity_I32); |
| 19433 | assign( oldC, mk_armg_calculate_flag_c() ); |
| 19434 | assign( oldV, mk_armg_calculate_flag_v() ); |
| 19435 | assign( res, binop(Iop_And32, getIRegT(rN), getIRegT(rM)) ); |
| 19436 | /* Update flags regardless of whether in an IT block or not. */ |
| 19437 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, oldC, oldV, condT ); |
| 19438 | DIP("tst r%u, r%u\n", rN, rM); |
| 19439 | goto decode_success; |
| 19440 | } |
| 19441 | |
| 19442 | case 0x109: { |
| 19443 | /* ---------------- NEGS Rd, Rm ---------------- */ |
| 19444 | /* Rd = -Rm */ |
| 19445 | UInt rM = INSN0(5,3); |
| 19446 | UInt rD = INSN0(2,0); |
| 19447 | IRTemp arg = newTemp(Ity_I32); |
| 19448 | IRTemp zero = newTemp(Ity_I32); |
| 19449 | assign(arg, getIRegT(rM)); |
| 19450 | assign(zero, mkU32(0)); |
| 19451 | // rD can never be r15 |
| 19452 | putIRegT(rD, binop(Iop_Sub32, mkexpr(zero), mkexpr(arg)), condT); |
| 19453 | setFlags_D1_D2( ARMG_CC_OP_SUB, zero, arg, cond_AND_notInIT_T); |
| 19454 | DIP("negs r%u, r%u\n", rD, rM); |
| 19455 | goto decode_success; |
| 19456 | } |
| 19457 | |
| 19458 | case 0x10F: { |
| 19459 | /* ---------------- MVNS Rd, Rm ---------------- */ |
| 19460 | /* Rd = ~Rm */ |
| 19461 | UInt rM = INSN0(5,3); |
| 19462 | UInt rD = INSN0(2,0); |
| 19463 | IRTemp oldV = newTemp(Ity_I32); |
| 19464 | IRTemp oldC = newTemp(Ity_I32); |
| 19465 | IRTemp res = newTemp(Ity_I32); |
| 19466 | assign( oldV, mk_armg_calculate_flag_v() ); |
| 19467 | assign( oldC, mk_armg_calculate_flag_c() ); |
| 19468 | assign(res, unop(Iop_Not32, getIRegT(rM))); |
| 19469 | // rD can never be r15 |
| 19470 | putIRegT(rD, mkexpr(res), condT); |
| 19471 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, oldC, oldV, |
| 19472 | cond_AND_notInIT_T ); |
| 19473 | DIP("mvns r%u, r%u\n", rD, rM); |
| 19474 | goto decode_success; |
| 19475 | } |
| 19476 | |
| 19477 | case 0x10C: |
| 19478 | /* ---------------- ORRS Rd, Rm ---------------- */ |
| 19479 | anOp = Iop_Or32; anOpNm = "orr"; goto and_orr_eor_mul; |
| 19480 | case 0x100: |
| 19481 | /* ---------------- ANDS Rd, Rm ---------------- */ |
| 19482 | anOp = Iop_And32; anOpNm = "and"; goto and_orr_eor_mul; |
| 19483 | case 0x101: |
| 19484 | /* ---------------- EORS Rd, Rm ---------------- */ |
| 19485 | anOp = Iop_Xor32; anOpNm = "eor"; goto and_orr_eor_mul; |
| 19486 | case 0x10d: |
| 19487 | /* ---------------- MULS Rd, Rm ---------------- */ |
| 19488 | anOp = Iop_Mul32; anOpNm = "mul"; goto and_orr_eor_mul; |
| 19489 | and_orr_eor_mul: { |
| 19490 | /* Rd = Rd `op` Rm */ |
| 19491 | UInt rM = INSN0(5,3); |
| 19492 | UInt rD = INSN0(2,0); |
| 19493 | IRTemp res = newTemp(Ity_I32); |
| 19494 | IRTemp oldV = newTemp(Ity_I32); |
| 19495 | IRTemp oldC = newTemp(Ity_I32); |
| 19496 | assign( oldV, mk_armg_calculate_flag_v() ); |
| 19497 | assign( oldC, mk_armg_calculate_flag_c() ); |
| 19498 | assign( res, binop(anOp, getIRegT(rD), getIRegT(rM) )); |
| 19499 | // not safe to read guest state after here |
| 19500 | // rD can never be r15 |
| 19501 | putIRegT(rD, mkexpr(res), condT); |
| 19502 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, oldC, oldV, |
| 19503 | cond_AND_notInIT_T ); |
| 19504 | DIP("%s r%u, r%u\n", anOpNm, rD, rM); |
| 19505 | goto decode_success; |
| 19506 | } |
| 19507 | |
| 19508 | case 0x10E: { |
| 19509 | /* ---------------- BICS Rd, Rm ---------------- */ |
| 19510 | /* Rd = Rd & ~Rm */ |
| 19511 | UInt rM = INSN0(5,3); |
| 19512 | UInt rD = INSN0(2,0); |
| 19513 | IRTemp res = newTemp(Ity_I32); |
| 19514 | IRTemp oldV = newTemp(Ity_I32); |
| 19515 | IRTemp oldC = newTemp(Ity_I32); |
| 19516 | assign( oldV, mk_armg_calculate_flag_v() ); |
| 19517 | assign( oldC, mk_armg_calculate_flag_c() ); |
| 19518 | assign( res, binop(Iop_And32, getIRegT(rD), |
| 19519 | unop(Iop_Not32, getIRegT(rM) ))); |
| 19520 | // not safe to read guest state after here |
| 19521 | // rD can never be r15 |
| 19522 | putIRegT(rD, mkexpr(res), condT); |
| 19523 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, oldC, oldV, |
| 19524 | cond_AND_notInIT_T ); |
| 19525 | DIP("bics r%u, r%u\n", rD, rM); |
| 19526 | goto decode_success; |
| 19527 | } |
| 19528 | |
| 19529 | case 0x105: { |
| 19530 | /* ---------------- ADCS Rd, Rm ---------------- */ |
| 19531 | /* Rd = Rd + Rm + oldC */ |
| 19532 | UInt rM = INSN0(5,3); |
| 19533 | UInt rD = INSN0(2,0); |
| 19534 | IRTemp argL = newTemp(Ity_I32); |
| 19535 | IRTemp argR = newTemp(Ity_I32); |
| 19536 | IRTemp oldC = newTemp(Ity_I32); |
| 19537 | IRTemp res = newTemp(Ity_I32); |
| 19538 | assign(argL, getIRegT(rD)); |
| 19539 | assign(argR, getIRegT(rM)); |
| 19540 | assign(oldC, mk_armg_calculate_flag_c()); |
| 19541 | assign(res, binop(Iop_Add32, |
| 19542 | binop(Iop_Add32, mkexpr(argL), mkexpr(argR)), |
| 19543 | mkexpr(oldC))); |
| 19544 | // rD can never be r15 |
| 19545 | putIRegT(rD, mkexpr(res), condT); |
| 19546 | setFlags_D1_D2_ND( ARMG_CC_OP_ADC, argL, argR, oldC, |
| 19547 | cond_AND_notInIT_T ); |
| 19548 | DIP("adcs r%u, r%u\n", rD, rM); |
| 19549 | goto decode_success; |
| 19550 | } |
| 19551 | |
| 19552 | case 0x106: { |
| 19553 | /* ---------------- SBCS Rd, Rm ---------------- */ |
| 19554 | /* Rd = Rd - Rm - (oldC ^ 1) */ |
| 19555 | UInt rM = INSN0(5,3); |
| 19556 | UInt rD = INSN0(2,0); |
| 19557 | IRTemp argL = newTemp(Ity_I32); |
| 19558 | IRTemp argR = newTemp(Ity_I32); |
| 19559 | IRTemp oldC = newTemp(Ity_I32); |
| 19560 | IRTemp res = newTemp(Ity_I32); |
| 19561 | assign(argL, getIRegT(rD)); |
| 19562 | assign(argR, getIRegT(rM)); |
| 19563 | assign(oldC, mk_armg_calculate_flag_c()); |
| 19564 | assign(res, binop(Iop_Sub32, |
| 19565 | binop(Iop_Sub32, mkexpr(argL), mkexpr(argR)), |
| 19566 | binop(Iop_Xor32, mkexpr(oldC), mkU32(1)))); |
| 19567 | // rD can never be r15 |
| 19568 | putIRegT(rD, mkexpr(res), condT); |
| 19569 | setFlags_D1_D2_ND( ARMG_CC_OP_SBB, argL, argR, oldC, |
| 19570 | cond_AND_notInIT_T ); |
| 19571 | DIP("sbcs r%u, r%u\n", rD, rM); |
| 19572 | goto decode_success; |
| 19573 | } |
| 19574 | |
| 19575 | case 0x2CB: { |
| 19576 | /* ---------------- UXTB Rd, Rm ---------------- */ |
| 19577 | /* Rd = 8Uto32(Rm) */ |
| 19578 | UInt rM = INSN0(5,3); |
| 19579 | UInt rD = INSN0(2,0); |
| 19580 | putIRegT(rD, binop(Iop_And32, getIRegT(rM), mkU32(0xFF)), |
| 19581 | condT); |
| 19582 | DIP("uxtb r%u, r%u\n", rD, rM); |
| 19583 | goto decode_success; |
| 19584 | } |
| 19585 | |
| 19586 | case 0x2C9: { |
| 19587 | /* ---------------- SXTB Rd, Rm ---------------- */ |
| 19588 | /* Rd = 8Sto32(Rm) */ |
| 19589 | UInt rM = INSN0(5,3); |
| 19590 | UInt rD = INSN0(2,0); |
| 19591 | putIRegT(rD, binop(Iop_Sar32, |
| 19592 | binop(Iop_Shl32, getIRegT(rM), mkU8(24)), |
| 19593 | mkU8(24)), |
| 19594 | condT); |
| 19595 | DIP("sxtb r%u, r%u\n", rD, rM); |
| 19596 | goto decode_success; |
| 19597 | } |
| 19598 | |
| 19599 | case 0x2CA: { |
| 19600 | /* ---------------- UXTH Rd, Rm ---------------- */ |
| 19601 | /* Rd = 16Uto32(Rm) */ |
| 19602 | UInt rM = INSN0(5,3); |
| 19603 | UInt rD = INSN0(2,0); |
| 19604 | putIRegT(rD, binop(Iop_And32, getIRegT(rM), mkU32(0xFFFF)), |
| 19605 | condT); |
| 19606 | DIP("uxth r%u, r%u\n", rD, rM); |
| 19607 | goto decode_success; |
| 19608 | } |
| 19609 | |
| 19610 | case 0x2C8: { |
| 19611 | /* ---------------- SXTH Rd, Rm ---------------- */ |
| 19612 | /* Rd = 16Sto32(Rm) */ |
| 19613 | UInt rM = INSN0(5,3); |
| 19614 | UInt rD = INSN0(2,0); |
| 19615 | putIRegT(rD, binop(Iop_Sar32, |
| 19616 | binop(Iop_Shl32, getIRegT(rM), mkU8(16)), |
| 19617 | mkU8(16)), |
| 19618 | condT); |
| 19619 | DIP("sxth r%u, r%u\n", rD, rM); |
| 19620 | goto decode_success; |
| 19621 | } |
| 19622 | |
| 19623 | case 0x102: // LSLS |
| 19624 | case 0x103: // LSRS |
| 19625 | case 0x104: // ASRS |
| 19626 | case 0x107: { // RORS |
| 19627 | /* ---------------- LSLS Rs, Rd ---------------- */ |
| 19628 | /* ---------------- LSRS Rs, Rd ---------------- */ |
| 19629 | /* ---------------- ASRS Rs, Rd ---------------- */ |
| 19630 | /* ---------------- RORS Rs, Rd ---------------- */ |
| 19631 | /* Rd = Rd `op` Rs, and set flags */ |
| 19632 | UInt rS = INSN0(5,3); |
| 19633 | UInt rD = INSN0(2,0); |
| 19634 | IRTemp oldV = newTemp(Ity_I32); |
| 19635 | IRTemp rDt = newTemp(Ity_I32); |
| 19636 | IRTemp rSt = newTemp(Ity_I32); |
| 19637 | IRTemp res = newTemp(Ity_I32); |
| 19638 | IRTemp resC = newTemp(Ity_I32); |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 19639 | const HChar* wot = "???"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19640 | assign(rSt, getIRegT(rS)); |
| 19641 | assign(rDt, getIRegT(rD)); |
| 19642 | assign(oldV, mk_armg_calculate_flag_v()); |
| 19643 | /* Does not appear to be the standard 'how' encoding. */ |
| 19644 | switch (INSN0(15,6)) { |
| 19645 | case 0x102: |
| 19646 | compute_result_and_C_after_LSL_by_reg( |
| 19647 | dis_buf, &res, &resC, rDt, rSt, rD, rS |
| 19648 | ); |
| 19649 | wot = "lsl"; |
| 19650 | break; |
| 19651 | case 0x103: |
| 19652 | compute_result_and_C_after_LSR_by_reg( |
| 19653 | dis_buf, &res, &resC, rDt, rSt, rD, rS |
| 19654 | ); |
| 19655 | wot = "lsr"; |
| 19656 | break; |
| 19657 | case 0x104: |
| 19658 | compute_result_and_C_after_ASR_by_reg( |
| 19659 | dis_buf, &res, &resC, rDt, rSt, rD, rS |
| 19660 | ); |
| 19661 | wot = "asr"; |
| 19662 | break; |
| 19663 | case 0x107: |
| 19664 | compute_result_and_C_after_ROR_by_reg( |
| 19665 | dis_buf, &res, &resC, rDt, rSt, rD, rS |
| 19666 | ); |
| 19667 | wot = "ror"; |
| 19668 | break; |
| 19669 | default: |
| 19670 | /*NOTREACHED*/vassert(0); |
| 19671 | } |
| 19672 | // not safe to read guest state after this point |
| 19673 | putIRegT(rD, mkexpr(res), condT); |
| 19674 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, resC, oldV, |
| 19675 | cond_AND_notInIT_T ); |
| 19676 | DIP("%ss r%u, r%u\n", wot, rS, rD); |
| 19677 | goto decode_success; |
| 19678 | } |
| 19679 | |
sewardj | 27312d3 | 2010-09-26 00:48:41 +0000 | [diff] [blame] | 19680 | case 0x2E8: // REV |
| 19681 | case 0x2E9: { // REV16 |
| 19682 | /* ---------------- REV Rd, Rm ---------------- */ |
| 19683 | /* ---------------- REV16 Rd, Rm ---------------- */ |
| 19684 | UInt rM = INSN0(5,3); |
| 19685 | UInt rD = INSN0(2,0); |
| 19686 | Bool isREV = INSN0(15,6) == 0x2E8; |
| 19687 | IRTemp arg = newTemp(Ity_I32); |
| 19688 | assign(arg, getIRegT(rM)); |
| 19689 | IRTemp res = isREV ? gen_REV(arg) : gen_REV16(arg); |
| 19690 | putIRegT(rD, mkexpr(res), condT); |
| 19691 | DIP("rev%s r%u, r%u\n", isREV ? "" : "16", rD, rM); |
| 19692 | goto decode_success; |
| 19693 | } |
| 19694 | |
sewardj | a325ee1 | 2013-07-04 20:35:42 +0000 | [diff] [blame] | 19695 | case 0x2EB: { // REVSH |
| 19696 | /* ---------------- REVSH Rd, Rn ---------------- */ |
| 19697 | UInt rM = INSN0(5,3); |
| 19698 | UInt rD = INSN0(2,0); |
| 19699 | IRTemp irt_rM = newTemp(Ity_I32); |
| 19700 | IRTemp irt_hi = newTemp(Ity_I32); |
| 19701 | IRTemp irt_low = newTemp(Ity_I32); |
| 19702 | IRTemp irt_res = newTemp(Ity_I32); |
| 19703 | assign(irt_rM, getIRegT(rM)); |
| 19704 | assign(irt_hi, |
| 19705 | binop(Iop_Sar32, |
| 19706 | binop(Iop_Shl32, mkexpr(irt_rM), mkU8(24)), |
| 19707 | mkU8(16) |
| 19708 | ) |
| 19709 | ); |
| 19710 | assign(irt_low, |
| 19711 | binop(Iop_And32, |
| 19712 | binop(Iop_Shr32, mkexpr(irt_rM), mkU8(8)), |
| 19713 | mkU32(0xFF) |
| 19714 | ) |
| 19715 | ); |
| 19716 | assign(irt_res, |
| 19717 | binop(Iop_Or32, mkexpr(irt_hi), mkexpr(irt_low)) |
| 19718 | ); |
| 19719 | putIRegT(rD, mkexpr(irt_res), condT); |
| 19720 | DIP("revsh r%u, r%u\n", rD, rM); |
| 19721 | goto decode_success; |
| 19722 | } |
| 19723 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19724 | default: |
| 19725 | break; /* examine the next shortest prefix */ |
| 19726 | |
| 19727 | } |
| 19728 | |
| 19729 | |
| 19730 | /* ================ 16-bit 15:7 cases ================ */ |
| 19731 | |
| 19732 | switch (INSN0(15,7)) { |
| 19733 | |
| 19734 | case BITS9(1,0,1,1,0,0,0,0,0): { |
| 19735 | /* ------------ ADD SP, #imm7 * 4 ------------ */ |
| 19736 | UInt uimm7 = INSN0(6,0); |
| 19737 | putIRegT(13, binop(Iop_Add32, getIRegT(13), mkU32(uimm7 * 4)), |
| 19738 | condT); |
| 19739 | DIP("add sp, #%u\n", uimm7 * 4); |
| 19740 | goto decode_success; |
| 19741 | } |
| 19742 | |
| 19743 | case BITS9(1,0,1,1,0,0,0,0,1): { |
| 19744 | /* ------------ SUB SP, #imm7 * 4 ------------ */ |
| 19745 | UInt uimm7 = INSN0(6,0); |
| 19746 | putIRegT(13, binop(Iop_Sub32, getIRegT(13), mkU32(uimm7 * 4)), |
| 19747 | condT); |
| 19748 | DIP("sub sp, #%u\n", uimm7 * 4); |
| 19749 | goto decode_success; |
| 19750 | } |
| 19751 | |
| 19752 | case BITS9(0,1,0,0,0,1,1,1,0): { |
| 19753 | /* ---------------- BX rM ---------------- */ |
| 19754 | /* Branch to reg, and optionally switch modes. Reg contains a |
| 19755 | suitably encoded address therefore (w CPSR.T at the bottom). |
| 19756 | Have to special-case r15, as usual. */ |
| 19757 | UInt rM = (INSN0(6,6) << 3) | INSN0(5,3); |
sewardj | 8e074cc | 2010-09-02 21:14:10 +0000 | [diff] [blame] | 19758 | if (BITS3(0,0,0) == INSN0(2,0)) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19759 | IRTemp dst = newTemp(Ity_I32); |
| 19760 | gen_SIGILL_T_if_in_but_NLI_ITBlock(old_itstate, new_itstate); |
| 19761 | mk_skip_over_T16_if_cond_is_false(condT); |
| 19762 | condT = IRTemp_INVALID; |
| 19763 | // now uncond |
| 19764 | if (rM <= 14) { |
| 19765 | assign( dst, getIRegT(rM) ); |
| 19766 | } else { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19767 | vassert(rM == 15); |
| 19768 | assign( dst, mkU32(guest_R15_curr_instr_notENC + 4) ); |
| 19769 | } |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 19770 | llPutIReg(15, mkexpr(dst)); |
| 19771 | dres.jk_StopHere = rM == 14 ? Ijk_Ret : Ijk_Boring; |
| 19772 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19773 | DIP("bx r%u (possibly switch to ARM mode)\n", rM); |
| 19774 | goto decode_success; |
| 19775 | } |
| 19776 | break; |
| 19777 | } |
| 19778 | |
| 19779 | /* ---------------- BLX rM ---------------- */ |
| 19780 | /* Branch and link to interworking address in rM. */ |
| 19781 | case BITS9(0,1,0,0,0,1,1,1,1): { |
| 19782 | if (BITS3(0,0,0) == INSN0(2,0)) { |
| 19783 | UInt rM = (INSN0(6,6) << 3) | INSN0(5,3); |
| 19784 | IRTemp dst = newTemp(Ity_I32); |
| 19785 | if (rM <= 14) { |
| 19786 | gen_SIGILL_T_if_in_but_NLI_ITBlock(old_itstate, new_itstate); |
| 19787 | mk_skip_over_T16_if_cond_is_false(condT); |
| 19788 | condT = IRTemp_INVALID; |
| 19789 | // now uncond |
| 19790 | /* We're returning to Thumb code, hence "| 1" */ |
| 19791 | assign( dst, getIRegT(rM) ); |
| 19792 | putIRegT( 14, mkU32( (guest_R15_curr_instr_notENC + 2) | 1 ), |
| 19793 | IRTemp_INVALID ); |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 19794 | llPutIReg(15, mkexpr(dst)); |
| 19795 | dres.jk_StopHere = Ijk_Call; |
| 19796 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19797 | DIP("blx r%u (possibly switch to ARM mode)\n", rM); |
| 19798 | goto decode_success; |
| 19799 | } |
| 19800 | /* else unpredictable, fall through */ |
| 19801 | } |
| 19802 | break; |
| 19803 | } |
| 19804 | |
| 19805 | default: |
| 19806 | break; /* examine the next shortest prefix */ |
| 19807 | |
| 19808 | } |
| 19809 | |
| 19810 | |
| 19811 | /* ================ 16-bit 15:8 cases ================ */ |
| 19812 | |
| 19813 | switch (INSN0(15,8)) { |
| 19814 | |
| 19815 | case BITS8(1,1,0,1,1,1,1,1): { |
| 19816 | /* ---------------- SVC ---------------- */ |
| 19817 | UInt imm8 = INSN0(7,0); |
| 19818 | if (imm8 == 0) { |
| 19819 | /* A syscall. We can't do this conditionally, hence: */ |
| 19820 | mk_skip_over_T16_if_cond_is_false( condT ); |
| 19821 | // FIXME: what if we have to back up and restart this insn? |
| 19822 | // then ITSTATE will be wrong (we'll have it as "used") |
| 19823 | // when it isn't. Correct is to save ITSTATE in a |
| 19824 | // stash pseudo-reg, and back up from that if we have to |
| 19825 | // restart. |
| 19826 | // uncond after here |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 19827 | llPutIReg(15, mkU32( (guest_R15_curr_instr_notENC + 2) | 1 )); |
| 19828 | dres.jk_StopHere = Ijk_Sys_syscall; |
| 19829 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19830 | DIP("svc #0x%08x\n", imm8); |
| 19831 | goto decode_success; |
| 19832 | } |
| 19833 | /* else fall through */ |
| 19834 | break; |
| 19835 | } |
| 19836 | |
| 19837 | case BITS8(0,1,0,0,0,1,0,0): { |
| 19838 | /* ---------------- ADD(HI) Rd, Rm ---------------- */ |
| 19839 | UInt h1 = INSN0(7,7); |
| 19840 | UInt h2 = INSN0(6,6); |
| 19841 | UInt rM = (h2 << 3) | INSN0(5,3); |
| 19842 | UInt rD = (h1 << 3) | INSN0(2,0); |
| 19843 | //if (h1 == 0 && h2 == 0) { // Original T1 was more restrictive |
| 19844 | if (rD == 15 && rM == 15) { |
| 19845 | // then it's invalid |
| 19846 | } else { |
| 19847 | IRTemp res = newTemp(Ity_I32); |
| 19848 | assign( res, binop(Iop_Add32, getIRegT(rD), getIRegT(rM) )); |
| 19849 | if (rD != 15) { |
| 19850 | putIRegT( rD, mkexpr(res), condT ); |
| 19851 | } else { |
| 19852 | /* Only allowed outside or last-in IT block; SIGILL if not so. */ |
| 19853 | gen_SIGILL_T_if_in_but_NLI_ITBlock(old_itstate, new_itstate); |
| 19854 | /* jump over insn if not selected */ |
| 19855 | mk_skip_over_T16_if_cond_is_false(condT); |
| 19856 | condT = IRTemp_INVALID; |
| 19857 | // now uncond |
| 19858 | /* non-interworking branch */ |
sewardj | 0a4f4bb | 2014-03-15 08:14:06 +0000 | [diff] [blame] | 19859 | llPutIReg(15, binop(Iop_Or32, mkexpr(res), mkU32(1))); |
| 19860 | dres.jk_StopHere = Ijk_Boring; |
| 19861 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19862 | } |
| 19863 | DIP("add(hi) r%u, r%u\n", rD, rM); |
| 19864 | goto decode_success; |
| 19865 | } |
| 19866 | break; |
| 19867 | } |
| 19868 | |
| 19869 | case BITS8(0,1,0,0,0,1,0,1): { |
| 19870 | /* ---------------- CMP(HI) Rd, Rm ---------------- */ |
| 19871 | UInt h1 = INSN0(7,7); |
| 19872 | UInt h2 = INSN0(6,6); |
| 19873 | UInt rM = (h2 << 3) | INSN0(5,3); |
| 19874 | UInt rN = (h1 << 3) | INSN0(2,0); |
| 19875 | if (h1 != 0 || h2 != 0) { |
| 19876 | IRTemp argL = newTemp(Ity_I32); |
| 19877 | IRTemp argR = newTemp(Ity_I32); |
| 19878 | assign( argL, getIRegT(rN) ); |
| 19879 | assign( argR, getIRegT(rM) ); |
| 19880 | /* Update flags regardless of whether in an IT block or not. */ |
| 19881 | setFlags_D1_D2( ARMG_CC_OP_SUB, argL, argR, condT ); |
| 19882 | DIP("cmphi r%u, r%u\n", rN, rM); |
| 19883 | goto decode_success; |
| 19884 | } |
| 19885 | break; |
| 19886 | } |
| 19887 | |
| 19888 | case BITS8(0,1,0,0,0,1,1,0): { |
| 19889 | /* ---------------- MOV(HI) Rd, Rm ---------------- */ |
| 19890 | UInt h1 = INSN0(7,7); |
| 19891 | UInt h2 = INSN0(6,6); |
| 19892 | UInt rM = (h2 << 3) | INSN0(5,3); |
| 19893 | UInt rD = (h1 << 3) | INSN0(2,0); |
| 19894 | /* The old ARM ARM seems to disallow the case where both Rd and |
| 19895 | Rm are "low" registers, but newer versions allow it. */ |
| 19896 | if (1 /*h1 != 0 || h2 != 0*/) { |
| 19897 | IRTemp val = newTemp(Ity_I32); |
| 19898 | assign( val, getIRegT(rM) ); |
| 19899 | if (rD != 15) { |
| 19900 | putIRegT( rD, mkexpr(val), condT ); |
| 19901 | } else { |
| 19902 | /* Only allowed outside or last-in IT block; SIGILL if not so. */ |
| 19903 | gen_SIGILL_T_if_in_but_NLI_ITBlock(old_itstate, new_itstate); |
| 19904 | /* jump over insn if not selected */ |
| 19905 | mk_skip_over_T16_if_cond_is_false(condT); |
| 19906 | condT = IRTemp_INVALID; |
| 19907 | // now uncond |
| 19908 | /* non-interworking branch */ |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 19909 | llPutIReg(15, binop(Iop_Or32, mkexpr(val), mkU32(1))); |
| 19910 | dres.jk_StopHere = rM == 14 ? Ijk_Ret : Ijk_Boring; |
| 19911 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19912 | } |
| 19913 | DIP("mov r%u, r%u\n", rD, rM); |
| 19914 | goto decode_success; |
| 19915 | } |
| 19916 | break; |
| 19917 | } |
| 19918 | |
| 19919 | case BITS8(1,0,1,1,1,1,1,1): { |
| 19920 | /* ---------------- IT (if-then) ---------------- */ |
| 19921 | UInt firstcond = INSN0(7,4); |
| 19922 | UInt mask = INSN0(3,0); |
| 19923 | UInt newITSTATE = 0; |
| 19924 | /* This is the ITSTATE represented as described in |
| 19925 | libvex_guest_arm.h. It is not the ARM ARM representation. */ |
florian | 5df8ab0 | 2012-10-13 19:34:19 +0000 | [diff] [blame] | 19926 | HChar c1 = '.'; |
| 19927 | HChar c2 = '.'; |
| 19928 | HChar c3 = '.'; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19929 | Bool valid = compute_ITSTATE( &newITSTATE, &c1, &c2, &c3, |
| 19930 | firstcond, mask ); |
| 19931 | if (valid && firstcond != 0xF/*NV*/) { |
| 19932 | /* Not allowed in an IT block; SIGILL if so. */ |
| 19933 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 19934 | |
| 19935 | IRTemp t = newTemp(Ity_I32); |
| 19936 | assign(t, mkU32(newITSTATE)); |
| 19937 | put_ITSTATE(t); |
| 19938 | |
| 19939 | DIP("it%c%c%c %s\n", c1, c2, c3, nCC(firstcond)); |
| 19940 | goto decode_success; |
| 19941 | } |
| 19942 | break; |
| 19943 | } |
| 19944 | |
| 19945 | case BITS8(1,0,1,1,0,0,0,1): |
| 19946 | case BITS8(1,0,1,1,0,0,1,1): |
| 19947 | case BITS8(1,0,1,1,1,0,0,1): |
| 19948 | case BITS8(1,0,1,1,1,0,1,1): { |
| 19949 | /* ---------------- CB{N}Z ---------------- */ |
| 19950 | UInt rN = INSN0(2,0); |
| 19951 | UInt bOP = INSN0(11,11); |
| 19952 | UInt imm32 = (INSN0(9,9) << 6) | (INSN0(7,3) << 1); |
| 19953 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 19954 | /* It's a conditional branch forward. */ |
| 19955 | IRTemp kond = newTemp(Ity_I1); |
| 19956 | assign( kond, binop(bOP ? Iop_CmpNE32 : Iop_CmpEQ32, |
| 19957 | getIRegT(rN), mkU32(0)) ); |
| 19958 | |
| 19959 | vassert(0 == (guest_R15_curr_instr_notENC & 1)); |
| 19960 | /* Looks like the nearest insn we can branch to is the one after |
| 19961 | next. That makes sense, as there's no point in being able to |
| 19962 | encode a conditional branch to the next instruction. */ |
| 19963 | UInt dst = (guest_R15_curr_instr_notENC + 4 + imm32) | 1; |
| 19964 | stmt(IRStmt_Exit( mkexpr(kond), |
| 19965 | Ijk_Boring, |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 19966 | IRConst_U32(toUInt(dst)), |
| 19967 | OFFB_R15T )); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19968 | DIP("cb%s r%u, 0x%x\n", bOP ? "nz" : "z", rN, dst - 1); |
| 19969 | goto decode_success; |
| 19970 | } |
| 19971 | |
| 19972 | default: |
| 19973 | break; /* examine the next shortest prefix */ |
| 19974 | |
| 19975 | } |
| 19976 | |
| 19977 | |
| 19978 | /* ================ 16-bit 15:9 cases ================ */ |
| 19979 | |
| 19980 | switch (INSN0(15,9)) { |
| 19981 | |
| 19982 | case BITS7(1,0,1,1,0,1,0): { |
| 19983 | /* ---------------- PUSH ---------------- */ |
| 19984 | /* This is a bit like STMxx, but way simpler. Complications we |
| 19985 | don't have to deal with: |
| 19986 | * SP being one of the transferred registers |
| 19987 | * direction (increment vs decrement) |
| 19988 | * before-vs-after-ness |
| 19989 | */ |
| 19990 | Int i, nRegs; |
| 19991 | UInt bitR = INSN0(8,8); |
| 19992 | UInt regList = INSN0(7,0); |
| 19993 | if (bitR) regList |= (1 << 14); |
| 19994 | |
sewardj | b0f1df2 | 2012-06-11 21:54:58 +0000 | [diff] [blame] | 19995 | /* At least one register must be transferred, else result is |
| 19996 | UNPREDICTABLE. */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 19997 | if (regList != 0) { |
| 19998 | /* Since we can't generate a guaranteed non-trapping IR |
| 19999 | sequence, (1) jump over the insn if it is gated false, and |
| 20000 | (2) back out the ITSTATE update. */ |
| 20001 | mk_skip_over_T16_if_cond_is_false(condT); |
| 20002 | condT = IRTemp_INVALID; |
| 20003 | put_ITSTATE(old_itstate); |
| 20004 | // now uncond |
| 20005 | |
| 20006 | nRegs = 0; |
| 20007 | for (i = 0; i < 16; i++) { |
| 20008 | if ((regList & (1 << i)) != 0) |
| 20009 | nRegs++; |
| 20010 | } |
sewardj | b0f1df2 | 2012-06-11 21:54:58 +0000 | [diff] [blame] | 20011 | vassert(nRegs >= 1 && nRegs <= 9); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20012 | |
| 20013 | /* Move SP down first of all, so we're "covered". And don't |
| 20014 | mess with its alignment. */ |
| 20015 | IRTemp newSP = newTemp(Ity_I32); |
| 20016 | assign(newSP, binop(Iop_Sub32, getIRegT(13), mkU32(4 * nRegs))); |
| 20017 | putIRegT(13, mkexpr(newSP), IRTemp_INVALID); |
| 20018 | |
| 20019 | /* Generate a transfer base address as a forced-aligned |
| 20020 | version of the final SP value. */ |
| 20021 | IRTemp base = newTemp(Ity_I32); |
| 20022 | assign(base, binop(Iop_And32, mkexpr(newSP), mkU32(~3))); |
| 20023 | |
| 20024 | /* Now the transfers */ |
| 20025 | nRegs = 0; |
| 20026 | for (i = 0; i < 16; i++) { |
| 20027 | if ((regList & (1 << i)) != 0) { |
| 20028 | storeLE( binop(Iop_Add32, mkexpr(base), mkU32(4 * nRegs)), |
| 20029 | getIRegT(i) ); |
| 20030 | nRegs++; |
| 20031 | } |
| 20032 | } |
| 20033 | |
| 20034 | /* Reinstate the ITSTATE update. */ |
| 20035 | put_ITSTATE(new_itstate); |
| 20036 | |
| 20037 | DIP("push {%s0x%04x}\n", bitR ? "lr," : "", regList & 0xFF); |
| 20038 | goto decode_success; |
| 20039 | } |
| 20040 | break; |
| 20041 | } |
| 20042 | |
| 20043 | case BITS7(1,0,1,1,1,1,0): { |
| 20044 | /* ---------------- POP ---------------- */ |
| 20045 | Int i, nRegs; |
| 20046 | UInt bitR = INSN0(8,8); |
| 20047 | UInt regList = INSN0(7,0); |
| 20048 | |
sewardj | b0f1df2 | 2012-06-11 21:54:58 +0000 | [diff] [blame] | 20049 | /* At least one register must be transferred, else result is |
| 20050 | UNPREDICTABLE. */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20051 | if (regList != 0 || bitR) { |
| 20052 | /* Since we can't generate a guaranteed non-trapping IR |
| 20053 | sequence, (1) jump over the insn if it is gated false, and |
| 20054 | (2) back out the ITSTATE update. */ |
| 20055 | mk_skip_over_T16_if_cond_is_false(condT); |
| 20056 | condT = IRTemp_INVALID; |
| 20057 | put_ITSTATE(old_itstate); |
| 20058 | // now uncond |
| 20059 | |
| 20060 | nRegs = 0; |
| 20061 | for (i = 0; i < 8; i++) { |
| 20062 | if ((regList & (1 << i)) != 0) |
| 20063 | nRegs++; |
| 20064 | } |
sewardj | b0f1df2 | 2012-06-11 21:54:58 +0000 | [diff] [blame] | 20065 | vassert(nRegs >= 0 && nRegs <= 8); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20066 | vassert(bitR == 0 || bitR == 1); |
| 20067 | |
| 20068 | IRTemp oldSP = newTemp(Ity_I32); |
| 20069 | assign(oldSP, getIRegT(13)); |
| 20070 | |
| 20071 | /* Generate a transfer base address as a forced-aligned |
| 20072 | version of the original SP value. */ |
| 20073 | IRTemp base = newTemp(Ity_I32); |
| 20074 | assign(base, binop(Iop_And32, mkexpr(oldSP), mkU32(~3))); |
| 20075 | |
| 20076 | /* Compute a new value for SP, but don't install it yet, so |
| 20077 | that we're "covered" until all the transfers are done. |
| 20078 | And don't mess with its alignment. */ |
| 20079 | IRTemp newSP = newTemp(Ity_I32); |
| 20080 | assign(newSP, binop(Iop_Add32, mkexpr(oldSP), |
| 20081 | mkU32(4 * (nRegs + bitR)))); |
| 20082 | |
| 20083 | /* Now the transfers, not including PC */ |
| 20084 | nRegs = 0; |
| 20085 | for (i = 0; i < 8; i++) { |
| 20086 | if ((regList & (1 << i)) != 0) { |
| 20087 | putIRegT(i, loadLE( Ity_I32, |
| 20088 | binop(Iop_Add32, mkexpr(base), |
| 20089 | mkU32(4 * nRegs))), |
| 20090 | IRTemp_INVALID ); |
| 20091 | nRegs++; |
| 20092 | } |
| 20093 | } |
| 20094 | |
| 20095 | IRTemp newPC = IRTemp_INVALID; |
| 20096 | if (bitR) { |
| 20097 | newPC = newTemp(Ity_I32); |
| 20098 | assign( newPC, loadLE( Ity_I32, |
| 20099 | binop(Iop_Add32, mkexpr(base), |
| 20100 | mkU32(4 * nRegs)))); |
| 20101 | } |
| 20102 | |
| 20103 | /* Now we can safely install the new SP value */ |
| 20104 | putIRegT(13, mkexpr(newSP), IRTemp_INVALID); |
| 20105 | |
| 20106 | /* Reinstate the ITSTATE update. */ |
| 20107 | put_ITSTATE(new_itstate); |
| 20108 | |
| 20109 | /* now, do we also have to do a branch? If so, it turns out |
| 20110 | that the new PC value is encoded exactly as we need it to |
| 20111 | be -- with CPSR.T in the bottom bit. So we can simply use |
| 20112 | it as is, no need to mess with it. Note, therefore, this |
| 20113 | is an interworking return. */ |
| 20114 | if (bitR) { |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 20115 | llPutIReg(15, mkexpr(newPC)); |
| 20116 | dres.jk_StopHere = Ijk_Ret; |
| 20117 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20118 | } |
| 20119 | |
| 20120 | DIP("pop {%s0x%04x}\n", bitR ? "pc," : "", regList & 0xFF); |
| 20121 | goto decode_success; |
| 20122 | } |
| 20123 | break; |
| 20124 | } |
| 20125 | |
| 20126 | case BITS7(0,0,0,1,1,1,0): /* ADDS */ |
| 20127 | case BITS7(0,0,0,1,1,1,1): { /* SUBS */ |
| 20128 | /* ---------------- ADDS Rd, Rn, #uimm3 ---------------- */ |
| 20129 | /* ---------------- SUBS Rd, Rn, #uimm3 ---------------- */ |
| 20130 | UInt uimm3 = INSN0(8,6); |
| 20131 | UInt rN = INSN0(5,3); |
| 20132 | UInt rD = INSN0(2,0); |
| 20133 | UInt isSub = INSN0(9,9); |
| 20134 | IRTemp argL = newTemp(Ity_I32); |
| 20135 | IRTemp argR = newTemp(Ity_I32); |
| 20136 | assign( argL, getIRegT(rN) ); |
| 20137 | assign( argR, mkU32(uimm3) ); |
| 20138 | putIRegT(rD, binop(isSub ? Iop_Sub32 : Iop_Add32, |
| 20139 | mkexpr(argL), mkexpr(argR)), |
| 20140 | condT); |
| 20141 | setFlags_D1_D2( isSub ? ARMG_CC_OP_SUB : ARMG_CC_OP_ADD, |
| 20142 | argL, argR, cond_AND_notInIT_T ); |
| 20143 | DIP("%s r%u, r%u, #%u\n", isSub ? "subs" : "adds", rD, rN, uimm3); |
| 20144 | goto decode_success; |
| 20145 | } |
| 20146 | |
| 20147 | case BITS7(0,0,0,1,1,0,0): /* ADDS */ |
| 20148 | case BITS7(0,0,0,1,1,0,1): { /* SUBS */ |
| 20149 | /* ---------------- ADDS Rd, Rn, Rm ---------------- */ |
| 20150 | /* ---------------- SUBS Rd, Rn, Rm ---------------- */ |
| 20151 | UInt rM = INSN0(8,6); |
| 20152 | UInt rN = INSN0(5,3); |
| 20153 | UInt rD = INSN0(2,0); |
| 20154 | UInt isSub = INSN0(9,9); |
| 20155 | IRTemp argL = newTemp(Ity_I32); |
| 20156 | IRTemp argR = newTemp(Ity_I32); |
| 20157 | assign( argL, getIRegT(rN) ); |
| 20158 | assign( argR, getIRegT(rM) ); |
| 20159 | putIRegT( rD, binop(isSub ? Iop_Sub32 : Iop_Add32, |
| 20160 | mkexpr(argL), mkexpr(argR)), |
| 20161 | condT ); |
| 20162 | setFlags_D1_D2( isSub ? ARMG_CC_OP_SUB : ARMG_CC_OP_ADD, |
| 20163 | argL, argR, cond_AND_notInIT_T ); |
| 20164 | DIP("%s r%u, r%u, r%u\n", isSub ? "subs" : "adds", rD, rN, rM); |
| 20165 | goto decode_success; |
| 20166 | } |
| 20167 | |
| 20168 | case BITS7(0,1,0,1,0,0,0): /* STR */ |
| 20169 | case BITS7(0,1,0,1,1,0,0): { /* LDR */ |
| 20170 | /* ------------- LDR Rd, [Rn, Rm] ------------- */ |
| 20171 | /* ------------- STR Rd, [Rn, Rm] ------------- */ |
| 20172 | /* LDR/STR Rd, [Rn + Rm] */ |
| 20173 | UInt rD = INSN0(2,0); |
| 20174 | UInt rN = INSN0(5,3); |
| 20175 | UInt rM = INSN0(8,6); |
| 20176 | UInt isLD = INSN0(11,11); |
| 20177 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20178 | IRExpr* ea = binop(Iop_Add32, getIRegT(rN), getIRegT(rM)); |
| 20179 | put_ITSTATE(old_itstate); // backout |
| 20180 | if (isLD) { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20181 | IRTemp tD = newTemp(Ity_I32); |
| 20182 | loadGuardedLE( tD, ILGop_Ident32, ea, llGetIReg(rD), condT ); |
| 20183 | putIRegT(rD, mkexpr(tD), IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20184 | } else { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20185 | storeGuardedLE(ea, getIRegT(rD), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20186 | } |
| 20187 | put_ITSTATE(new_itstate); // restore |
| 20188 | |
| 20189 | DIP("%s r%u, [r%u, r%u]\n", isLD ? "ldr" : "str", rD, rN, rM); |
| 20190 | goto decode_success; |
| 20191 | } |
| 20192 | |
| 20193 | case BITS7(0,1,0,1,0,0,1): |
| 20194 | case BITS7(0,1,0,1,1,0,1): { |
| 20195 | /* ------------- LDRH Rd, [Rn, Rm] ------------- */ |
| 20196 | /* ------------- STRH Rd, [Rn, Rm] ------------- */ |
| 20197 | /* LDRH/STRH Rd, [Rn + Rm] */ |
| 20198 | UInt rD = INSN0(2,0); |
| 20199 | UInt rN = INSN0(5,3); |
| 20200 | UInt rM = INSN0(8,6); |
| 20201 | UInt isLD = INSN0(11,11); |
| 20202 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20203 | IRExpr* ea = binop(Iop_Add32, getIRegT(rN), getIRegT(rM)); |
| 20204 | put_ITSTATE(old_itstate); // backout |
| 20205 | if (isLD) { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20206 | IRTemp tD = newTemp(Ity_I32); |
| 20207 | loadGuardedLE(tD, ILGop_16Uto32, ea, llGetIReg(rD), condT); |
| 20208 | putIRegT(rD, mkexpr(tD), IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20209 | } else { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20210 | storeGuardedLE( ea, unop(Iop_32to16, getIRegT(rD)), condT ); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20211 | } |
| 20212 | put_ITSTATE(new_itstate); // restore |
| 20213 | |
| 20214 | DIP("%sh r%u, [r%u, r%u]\n", isLD ? "ldr" : "str", rD, rN, rM); |
| 20215 | goto decode_success; |
| 20216 | } |
| 20217 | |
| 20218 | case BITS7(0,1,0,1,1,1,1): { |
| 20219 | /* ------------- LDRSH Rd, [Rn, Rm] ------------- */ |
| 20220 | /* LDRSH Rd, [Rn + Rm] */ |
| 20221 | UInt rD = INSN0(2,0); |
| 20222 | UInt rN = INSN0(5,3); |
| 20223 | UInt rM = INSN0(8,6); |
| 20224 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20225 | IRExpr* ea = binop(Iop_Add32, getIRegT(rN), getIRegT(rM)); |
| 20226 | put_ITSTATE(old_itstate); // backout |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20227 | IRTemp tD = newTemp(Ity_I32); |
| 20228 | loadGuardedLE(tD, ILGop_16Sto32, ea, llGetIReg(rD), condT); |
| 20229 | putIRegT(rD, mkexpr(tD), IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20230 | put_ITSTATE(new_itstate); // restore |
| 20231 | |
| 20232 | DIP("ldrsh r%u, [r%u, r%u]\n", rD, rN, rM); |
| 20233 | goto decode_success; |
| 20234 | } |
| 20235 | |
| 20236 | case BITS7(0,1,0,1,0,1,1): { |
| 20237 | /* ------------- LDRSB Rd, [Rn, Rm] ------------- */ |
| 20238 | /* LDRSB Rd, [Rn + Rm] */ |
| 20239 | UInt rD = INSN0(2,0); |
| 20240 | UInt rN = INSN0(5,3); |
| 20241 | UInt rM = INSN0(8,6); |
| 20242 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20243 | IRExpr* ea = binop(Iop_Add32, getIRegT(rN), getIRegT(rM)); |
| 20244 | put_ITSTATE(old_itstate); // backout |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20245 | IRTemp tD = newTemp(Ity_I32); |
| 20246 | loadGuardedLE(tD, ILGop_8Sto32, ea, llGetIReg(rD), condT); |
| 20247 | putIRegT(rD, mkexpr(tD), IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20248 | put_ITSTATE(new_itstate); // restore |
| 20249 | |
| 20250 | DIP("ldrsb r%u, [r%u, r%u]\n", rD, rN, rM); |
| 20251 | goto decode_success; |
| 20252 | } |
| 20253 | |
| 20254 | case BITS7(0,1,0,1,0,1,0): |
| 20255 | case BITS7(0,1,0,1,1,1,0): { |
| 20256 | /* ------------- LDRB Rd, [Rn, Rm] ------------- */ |
| 20257 | /* ------------- STRB Rd, [Rn, Rm] ------------- */ |
| 20258 | /* LDRB/STRB Rd, [Rn + Rm] */ |
| 20259 | UInt rD = INSN0(2,0); |
| 20260 | UInt rN = INSN0(5,3); |
| 20261 | UInt rM = INSN0(8,6); |
| 20262 | UInt isLD = INSN0(11,11); |
| 20263 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20264 | IRExpr* ea = binop(Iop_Add32, getIRegT(rN), getIRegT(rM)); |
| 20265 | put_ITSTATE(old_itstate); // backout |
| 20266 | if (isLD) { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20267 | IRTemp tD = newTemp(Ity_I32); |
| 20268 | loadGuardedLE(tD, ILGop_8Uto32, ea, llGetIReg(rD), condT); |
| 20269 | putIRegT(rD, mkexpr(tD), IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20270 | } else { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20271 | storeGuardedLE( ea, unop(Iop_32to8, getIRegT(rD)), condT ); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20272 | } |
| 20273 | put_ITSTATE(new_itstate); // restore |
| 20274 | |
| 20275 | DIP("%sb r%u, [r%u, r%u]\n", isLD ? "ldr" : "str", rD, rN, rM); |
| 20276 | goto decode_success; |
| 20277 | } |
| 20278 | |
| 20279 | default: |
| 20280 | break; /* examine the next shortest prefix */ |
| 20281 | |
| 20282 | } |
| 20283 | |
| 20284 | |
| 20285 | /* ================ 16-bit 15:11 cases ================ */ |
| 20286 | |
| 20287 | switch (INSN0(15,11)) { |
| 20288 | |
| 20289 | case BITS5(0,0,1,1,0): |
| 20290 | case BITS5(0,0,1,1,1): { |
| 20291 | /* ---------------- ADDS Rn, #uimm8 ---------------- */ |
| 20292 | /* ---------------- SUBS Rn, #uimm8 ---------------- */ |
| 20293 | UInt isSub = INSN0(11,11); |
| 20294 | UInt rN = INSN0(10,8); |
| 20295 | UInt uimm8 = INSN0(7,0); |
| 20296 | IRTemp argL = newTemp(Ity_I32); |
| 20297 | IRTemp argR = newTemp(Ity_I32); |
| 20298 | assign( argL, getIRegT(rN) ); |
| 20299 | assign( argR, mkU32(uimm8) ); |
| 20300 | putIRegT( rN, binop(isSub ? Iop_Sub32 : Iop_Add32, |
| 20301 | mkexpr(argL), mkexpr(argR)), condT ); |
| 20302 | setFlags_D1_D2( isSub ? ARMG_CC_OP_SUB : ARMG_CC_OP_ADD, |
| 20303 | argL, argR, cond_AND_notInIT_T ); |
| 20304 | DIP("%s r%u, #%u\n", isSub ? "subs" : "adds", rN, uimm8); |
| 20305 | goto decode_success; |
| 20306 | } |
| 20307 | |
| 20308 | case BITS5(1,0,1,0,0): { |
| 20309 | /* ---------------- ADD rD, PC, #imm8 * 4 ---------------- */ |
| 20310 | /* a.k.a. ADR */ |
| 20311 | /* rD = align4(PC) + imm8 * 4 */ |
| 20312 | UInt rD = INSN0(10,8); |
| 20313 | UInt imm8 = INSN0(7,0); |
| 20314 | putIRegT(rD, binop(Iop_Add32, |
sewardj | df86f16 | 2010-08-22 18:47:30 +0000 | [diff] [blame] | 20315 | binop(Iop_And32, getIRegT(15), mkU32(~3U)), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20316 | mkU32(imm8 * 4)), |
| 20317 | condT); |
| 20318 | DIP("add r%u, pc, #%u\n", rD, imm8 * 4); |
| 20319 | goto decode_success; |
| 20320 | } |
| 20321 | |
| 20322 | case BITS5(1,0,1,0,1): { |
| 20323 | /* ---------------- ADD rD, SP, #imm8 * 4 ---------------- */ |
| 20324 | UInt rD = INSN0(10,8); |
| 20325 | UInt imm8 = INSN0(7,0); |
| 20326 | putIRegT(rD, binop(Iop_Add32, getIRegT(13), mkU32(imm8 * 4)), |
| 20327 | condT); |
| 20328 | DIP("add r%u, r13, #%u\n", rD, imm8 * 4); |
| 20329 | goto decode_success; |
| 20330 | } |
| 20331 | |
| 20332 | case BITS5(0,0,1,0,1): { |
| 20333 | /* ---------------- CMP Rn, #uimm8 ---------------- */ |
| 20334 | UInt rN = INSN0(10,8); |
| 20335 | UInt uimm8 = INSN0(7,0); |
| 20336 | IRTemp argL = newTemp(Ity_I32); |
| 20337 | IRTemp argR = newTemp(Ity_I32); |
| 20338 | assign( argL, getIRegT(rN) ); |
| 20339 | assign( argR, mkU32(uimm8) ); |
| 20340 | /* Update flags regardless of whether in an IT block or not. */ |
| 20341 | setFlags_D1_D2( ARMG_CC_OP_SUB, argL, argR, condT ); |
| 20342 | DIP("cmp r%u, #%u\n", rN, uimm8); |
| 20343 | goto decode_success; |
| 20344 | } |
| 20345 | |
| 20346 | case BITS5(0,0,1,0,0): { |
| 20347 | /* -------------- (T1) MOVS Rn, #uimm8 -------------- */ |
| 20348 | UInt rD = INSN0(10,8); |
| 20349 | UInt uimm8 = INSN0(7,0); |
| 20350 | IRTemp oldV = newTemp(Ity_I32); |
| 20351 | IRTemp oldC = newTemp(Ity_I32); |
| 20352 | IRTemp res = newTemp(Ity_I32); |
| 20353 | assign( oldV, mk_armg_calculate_flag_v() ); |
| 20354 | assign( oldC, mk_armg_calculate_flag_c() ); |
| 20355 | assign( res, mkU32(uimm8) ); |
| 20356 | putIRegT(rD, mkexpr(res), condT); |
| 20357 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, oldC, oldV, |
| 20358 | cond_AND_notInIT_T ); |
| 20359 | DIP("movs r%u, #%u\n", rD, uimm8); |
| 20360 | goto decode_success; |
| 20361 | } |
| 20362 | |
| 20363 | case BITS5(0,1,0,0,1): { |
| 20364 | /* ------------- LDR Rd, [PC, #imm8 * 4] ------------- */ |
| 20365 | /* LDR Rd, [align4(PC) + imm8 * 4] */ |
| 20366 | UInt rD = INSN0(10,8); |
| 20367 | UInt imm8 = INSN0(7,0); |
| 20368 | IRTemp ea = newTemp(Ity_I32); |
| 20369 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20370 | assign(ea, binop(Iop_Add32, |
sewardj | df86f16 | 2010-08-22 18:47:30 +0000 | [diff] [blame] | 20371 | binop(Iop_And32, getIRegT(15), mkU32(~3U)), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20372 | mkU32(imm8 * 4))); |
| 20373 | put_ITSTATE(old_itstate); // backout |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20374 | IRTemp tD = newTemp(Ity_I32); |
| 20375 | loadGuardedLE( tD, ILGop_Ident32, mkexpr(ea), llGetIReg(rD), condT ); |
| 20376 | putIRegT(rD, mkexpr(tD), IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20377 | put_ITSTATE(new_itstate); // restore |
| 20378 | |
| 20379 | DIP("ldr r%u, [pc, #%u]\n", rD, imm8 * 4); |
| 20380 | goto decode_success; |
| 20381 | } |
| 20382 | |
| 20383 | case BITS5(0,1,1,0,0): /* STR */ |
| 20384 | case BITS5(0,1,1,0,1): { /* LDR */ |
| 20385 | /* ------------- LDR Rd, [Rn, #imm5 * 4] ------------- */ |
| 20386 | /* ------------- STR Rd, [Rn, #imm5 * 4] ------------- */ |
| 20387 | /* LDR/STR Rd, [Rn + imm5 * 4] */ |
| 20388 | UInt rD = INSN0(2,0); |
| 20389 | UInt rN = INSN0(5,3); |
| 20390 | UInt imm5 = INSN0(10,6); |
| 20391 | UInt isLD = INSN0(11,11); |
| 20392 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20393 | IRExpr* ea = binop(Iop_Add32, getIRegT(rN), mkU32(imm5 * 4)); |
| 20394 | put_ITSTATE(old_itstate); // backout |
| 20395 | if (isLD) { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20396 | IRTemp tD = newTemp(Ity_I32); |
| 20397 | loadGuardedLE( tD, ILGop_Ident32, ea, llGetIReg(rD), condT ); |
| 20398 | putIRegT(rD, mkexpr(tD), IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20399 | } else { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20400 | storeGuardedLE( ea, getIRegT(rD), condT ); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20401 | } |
| 20402 | put_ITSTATE(new_itstate); // restore |
| 20403 | |
| 20404 | DIP("%s r%u, [r%u, #%u]\n", isLD ? "ldr" : "str", rD, rN, imm5 * 4); |
| 20405 | goto decode_success; |
| 20406 | } |
| 20407 | |
| 20408 | case BITS5(1,0,0,0,0): /* STRH */ |
| 20409 | case BITS5(1,0,0,0,1): { /* LDRH */ |
| 20410 | /* ------------- LDRH Rd, [Rn, #imm5 * 2] ------------- */ |
| 20411 | /* ------------- STRH Rd, [Rn, #imm5 * 2] ------------- */ |
| 20412 | /* LDRH/STRH Rd, [Rn + imm5 * 2] */ |
| 20413 | UInt rD = INSN0(2,0); |
| 20414 | UInt rN = INSN0(5,3); |
| 20415 | UInt imm5 = INSN0(10,6); |
| 20416 | UInt isLD = INSN0(11,11); |
| 20417 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20418 | IRExpr* ea = binop(Iop_Add32, getIRegT(rN), mkU32(imm5 * 2)); |
| 20419 | put_ITSTATE(old_itstate); // backout |
| 20420 | if (isLD) { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20421 | IRTemp tD = newTemp(Ity_I32); |
| 20422 | loadGuardedLE( tD, ILGop_16Uto32, ea, llGetIReg(rD), condT ); |
| 20423 | putIRegT(rD, mkexpr(tD), IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20424 | } else { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20425 | storeGuardedLE( ea, unop(Iop_32to16, getIRegT(rD)), condT ); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20426 | } |
| 20427 | put_ITSTATE(new_itstate); // restore |
| 20428 | |
| 20429 | DIP("%sh r%u, [r%u, #%u]\n", isLD ? "ldr" : "str", rD, rN, imm5 * 2); |
| 20430 | goto decode_success; |
| 20431 | } |
| 20432 | |
| 20433 | case BITS5(0,1,1,1,0): /* STRB */ |
| 20434 | case BITS5(0,1,1,1,1): { /* LDRB */ |
| 20435 | /* ------------- LDRB Rd, [Rn, #imm5] ------------- */ |
| 20436 | /* ------------- STRB Rd, [Rn, #imm5] ------------- */ |
| 20437 | /* LDRB/STRB Rd, [Rn + imm5] */ |
| 20438 | UInt rD = INSN0(2,0); |
| 20439 | UInt rN = INSN0(5,3); |
| 20440 | UInt imm5 = INSN0(10,6); |
| 20441 | UInt isLD = INSN0(11,11); |
| 20442 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20443 | IRExpr* ea = binop(Iop_Add32, getIRegT(rN), mkU32(imm5)); |
| 20444 | put_ITSTATE(old_itstate); // backout |
| 20445 | if (isLD) { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20446 | IRTemp tD = newTemp(Ity_I32); |
| 20447 | loadGuardedLE( tD, ILGop_8Uto32, ea, llGetIReg(rD), condT ); |
| 20448 | putIRegT(rD, mkexpr(tD), IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20449 | } else { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20450 | storeGuardedLE( ea, unop(Iop_32to8, getIRegT(rD)), condT ); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20451 | } |
| 20452 | put_ITSTATE(new_itstate); // restore |
| 20453 | |
| 20454 | DIP("%sb r%u, [r%u, #%u]\n", isLD ? "ldr" : "str", rD, rN, imm5); |
| 20455 | goto decode_success; |
| 20456 | } |
| 20457 | |
| 20458 | case BITS5(1,0,0,1,0): /* STR */ |
| 20459 | case BITS5(1,0,0,1,1): { /* LDR */ |
| 20460 | /* ------------- LDR Rd, [SP, #imm8 * 4] ------------- */ |
| 20461 | /* ------------- STR Rd, [SP, #imm8 * 4] ------------- */ |
| 20462 | /* LDR/STR Rd, [SP + imm8 * 4] */ |
| 20463 | UInt rD = INSN0(10,8); |
| 20464 | UInt imm8 = INSN0(7,0); |
| 20465 | UInt isLD = INSN0(11,11); |
| 20466 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20467 | IRExpr* ea = binop(Iop_Add32, getIRegT(13), mkU32(imm8 * 4)); |
| 20468 | put_ITSTATE(old_itstate); // backout |
| 20469 | if (isLD) { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20470 | IRTemp tD = newTemp(Ity_I32); |
| 20471 | loadGuardedLE( tD, ILGop_Ident32, ea, llGetIReg(rD), condT ); |
| 20472 | putIRegT(rD, mkexpr(tD), IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20473 | } else { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 20474 | storeGuardedLE(ea, getIRegT(rD), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20475 | } |
| 20476 | put_ITSTATE(new_itstate); // restore |
| 20477 | |
| 20478 | DIP("%s r%u, [sp, #%u]\n", isLD ? "ldr" : "str", rD, imm8 * 4); |
| 20479 | goto decode_success; |
| 20480 | } |
| 20481 | |
| 20482 | case BITS5(1,1,0,0,1): { |
| 20483 | /* ------------- LDMIA Rn!, {reglist} ------------- */ |
| 20484 | Int i, nRegs = 0; |
| 20485 | UInt rN = INSN0(10,8); |
| 20486 | UInt list = INSN0(7,0); |
| 20487 | /* Empty lists aren't allowed. */ |
| 20488 | if (list != 0) { |
| 20489 | mk_skip_over_T16_if_cond_is_false(condT); |
| 20490 | condT = IRTemp_INVALID; |
| 20491 | put_ITSTATE(old_itstate); |
| 20492 | // now uncond |
| 20493 | |
| 20494 | IRTemp oldRn = newTemp(Ity_I32); |
| 20495 | IRTemp base = newTemp(Ity_I32); |
| 20496 | assign(oldRn, getIRegT(rN)); |
sewardj | df86f16 | 2010-08-22 18:47:30 +0000 | [diff] [blame] | 20497 | assign(base, binop(Iop_And32, mkexpr(oldRn), mkU32(~3U))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20498 | for (i = 0; i < 8; i++) { |
| 20499 | if (0 == (list & (1 << i))) |
| 20500 | continue; |
| 20501 | nRegs++; |
| 20502 | putIRegT( |
| 20503 | i, loadLE(Ity_I32, |
| 20504 | binop(Iop_Add32, mkexpr(base), |
| 20505 | mkU32(nRegs * 4 - 4))), |
| 20506 | IRTemp_INVALID |
| 20507 | ); |
| 20508 | } |
| 20509 | /* Only do the writeback for rN if it isn't in the list of |
| 20510 | registers to be transferred. */ |
| 20511 | if (0 == (list & (1 << rN))) { |
| 20512 | putIRegT(rN, |
| 20513 | binop(Iop_Add32, mkexpr(oldRn), |
| 20514 | mkU32(nRegs * 4)), |
| 20515 | IRTemp_INVALID |
| 20516 | ); |
| 20517 | } |
| 20518 | |
| 20519 | /* Reinstate the ITSTATE update. */ |
| 20520 | put_ITSTATE(new_itstate); |
| 20521 | |
| 20522 | DIP("ldmia r%u!, {0x%04x}\n", rN, list); |
| 20523 | goto decode_success; |
| 20524 | } |
| 20525 | break; |
| 20526 | } |
| 20527 | |
| 20528 | case BITS5(1,1,0,0,0): { |
| 20529 | /* ------------- STMIA Rn!, {reglist} ------------- */ |
| 20530 | Int i, nRegs = 0; |
| 20531 | UInt rN = INSN0(10,8); |
| 20532 | UInt list = INSN0(7,0); |
| 20533 | /* Empty lists aren't allowed. Also, if rN is in the list then |
| 20534 | it must be the lowest numbered register in the list. */ |
| 20535 | Bool valid = list != 0; |
| 20536 | if (valid && 0 != (list & (1 << rN))) { |
| 20537 | for (i = 0; i < rN; i++) { |
| 20538 | if (0 != (list & (1 << i))) |
| 20539 | valid = False; |
| 20540 | } |
| 20541 | } |
| 20542 | if (valid) { |
| 20543 | mk_skip_over_T16_if_cond_is_false(condT); |
| 20544 | condT = IRTemp_INVALID; |
| 20545 | put_ITSTATE(old_itstate); |
| 20546 | // now uncond |
| 20547 | |
| 20548 | IRTemp oldRn = newTemp(Ity_I32); |
| 20549 | IRTemp base = newTemp(Ity_I32); |
| 20550 | assign(oldRn, getIRegT(rN)); |
sewardj | df86f16 | 2010-08-22 18:47:30 +0000 | [diff] [blame] | 20551 | assign(base, binop(Iop_And32, mkexpr(oldRn), mkU32(~3U))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20552 | for (i = 0; i < 8; i++) { |
| 20553 | if (0 == (list & (1 << i))) |
| 20554 | continue; |
| 20555 | nRegs++; |
| 20556 | storeLE( binop(Iop_Add32, mkexpr(base), mkU32(nRegs * 4 - 4)), |
| 20557 | getIRegT(i) ); |
| 20558 | } |
| 20559 | /* Always do the writeback. */ |
| 20560 | putIRegT(rN, |
| 20561 | binop(Iop_Add32, mkexpr(oldRn), |
| 20562 | mkU32(nRegs * 4)), |
| 20563 | IRTemp_INVALID); |
| 20564 | |
| 20565 | /* Reinstate the ITSTATE update. */ |
| 20566 | put_ITSTATE(new_itstate); |
| 20567 | |
| 20568 | DIP("stmia r%u!, {0x%04x}\n", rN, list); |
| 20569 | goto decode_success; |
| 20570 | } |
| 20571 | break; |
| 20572 | } |
| 20573 | |
| 20574 | case BITS5(0,0,0,0,0): /* LSLS */ |
| 20575 | case BITS5(0,0,0,0,1): /* LSRS */ |
| 20576 | case BITS5(0,0,0,1,0): { /* ASRS */ |
| 20577 | /* ---------------- LSLS Rd, Rm, #imm5 ---------------- */ |
| 20578 | /* ---------------- LSRS Rd, Rm, #imm5 ---------------- */ |
| 20579 | /* ---------------- ASRS Rd, Rm, #imm5 ---------------- */ |
| 20580 | UInt rD = INSN0(2,0); |
| 20581 | UInt rM = INSN0(5,3); |
| 20582 | UInt imm5 = INSN0(10,6); |
| 20583 | IRTemp res = newTemp(Ity_I32); |
| 20584 | IRTemp resC = newTemp(Ity_I32); |
| 20585 | IRTemp rMt = newTemp(Ity_I32); |
| 20586 | IRTemp oldV = newTemp(Ity_I32); |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 20587 | const HChar* wot = "???"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20588 | assign(rMt, getIRegT(rM)); |
| 20589 | assign(oldV, mk_armg_calculate_flag_v()); |
| 20590 | /* Looks like INSN0(12,11) are the standard 'how' encoding. |
| 20591 | Could compactify if the ROR case later appears. */ |
| 20592 | switch (INSN0(15,11)) { |
| 20593 | case BITS5(0,0,0,0,0): |
| 20594 | compute_result_and_C_after_LSL_by_imm5( |
| 20595 | dis_buf, &res, &resC, rMt, imm5, rM |
| 20596 | ); |
| 20597 | wot = "lsl"; |
| 20598 | break; |
| 20599 | case BITS5(0,0,0,0,1): |
| 20600 | compute_result_and_C_after_LSR_by_imm5( |
| 20601 | dis_buf, &res, &resC, rMt, imm5, rM |
| 20602 | ); |
| 20603 | wot = "lsr"; |
| 20604 | break; |
| 20605 | case BITS5(0,0,0,1,0): |
| 20606 | compute_result_and_C_after_ASR_by_imm5( |
| 20607 | dis_buf, &res, &resC, rMt, imm5, rM |
| 20608 | ); |
| 20609 | wot = "asr"; |
| 20610 | break; |
| 20611 | default: |
| 20612 | /*NOTREACHED*/vassert(0); |
| 20613 | } |
| 20614 | // not safe to read guest state after this point |
| 20615 | putIRegT(rD, mkexpr(res), condT); |
| 20616 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, resC, oldV, |
| 20617 | cond_AND_notInIT_T ); |
| 20618 | /* ignore buf and roll our own output */ |
| 20619 | DIP("%ss r%u, r%u, #%u\n", wot, rD, rM, imm5); |
| 20620 | goto decode_success; |
| 20621 | } |
| 20622 | |
| 20623 | case BITS5(1,1,1,0,0): { |
| 20624 | /* ---------------- B #simm11 ---------------- */ |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 20625 | UInt uimm11 = INSN0(10,0); uimm11 <<= 21; |
| 20626 | Int simm11 = (Int)uimm11; simm11 >>= 20; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20627 | UInt dst = simm11 + guest_R15_curr_instr_notENC + 4; |
| 20628 | /* Only allowed outside or last-in IT block; SIGILL if not so. */ |
| 20629 | gen_SIGILL_T_if_in_but_NLI_ITBlock(old_itstate, new_itstate); |
| 20630 | // and skip this insn if not selected; being cleverer is too |
| 20631 | // difficult |
| 20632 | mk_skip_over_T16_if_cond_is_false(condT); |
| 20633 | condT = IRTemp_INVALID; |
| 20634 | // now uncond |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 20635 | llPutIReg(15, mkU32( dst | 1 /*CPSR.T*/ )); |
| 20636 | dres.jk_StopHere = Ijk_Boring; |
| 20637 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20638 | DIP("b 0x%x\n", dst); |
| 20639 | goto decode_success; |
| 20640 | } |
| 20641 | |
| 20642 | default: |
| 20643 | break; /* examine the next shortest prefix */ |
| 20644 | |
| 20645 | } |
| 20646 | |
| 20647 | |
| 20648 | /* ================ 16-bit 15:12 cases ================ */ |
| 20649 | |
| 20650 | switch (INSN0(15,12)) { |
| 20651 | |
| 20652 | case BITS4(1,1,0,1): { |
| 20653 | /* ---------------- Bcond #simm8 ---------------- */ |
| 20654 | UInt cond = INSN0(11,8); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 20655 | UInt uimm8 = INSN0(7,0); uimm8 <<= 24; |
| 20656 | Int simm8 = (Int)uimm8; simm8 >>= 23; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20657 | UInt dst = simm8 + guest_R15_curr_instr_notENC + 4; |
| 20658 | if (cond != ARMCondAL && cond != ARMCondNV) { |
| 20659 | /* Not allowed in an IT block; SIGILL if so. */ |
| 20660 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 20661 | |
| 20662 | IRTemp kondT = newTemp(Ity_I32); |
| 20663 | assign( kondT, mk_armg_calculate_condition(cond) ); |
| 20664 | stmt( IRStmt_Exit( unop(Iop_32to1, mkexpr(kondT)), |
| 20665 | Ijk_Boring, |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 20666 | IRConst_U32(dst | 1/*CPSR.T*/), |
| 20667 | OFFB_R15T )); |
| 20668 | llPutIReg(15, mkU32( (guest_R15_curr_instr_notENC + 2) |
| 20669 | | 1 /*CPSR.T*/ )); |
| 20670 | dres.jk_StopHere = Ijk_Boring; |
| 20671 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20672 | DIP("b%s 0x%x\n", nCC(cond), dst); |
| 20673 | goto decode_success; |
| 20674 | } |
| 20675 | break; |
| 20676 | } |
| 20677 | |
| 20678 | default: |
| 20679 | break; /* hmm, nothing matched */ |
| 20680 | |
| 20681 | } |
| 20682 | |
| 20683 | /* ================ 16-bit misc cases ================ */ |
| 20684 | |
sewardj | 0ef8d9e | 2013-03-01 21:13:24 +0000 | [diff] [blame] | 20685 | switch (INSN0(15,0)) { |
| 20686 | case 0xBF00: |
| 20687 | /* ------ NOP ------ */ |
| 20688 | DIP("nop\n"); |
| 20689 | goto decode_success; |
sewardj | 5e65812 | 2015-08-17 13:55:41 +0000 | [diff] [blame] | 20690 | case 0xBF10: // YIELD |
| 20691 | case 0xBF20: // WFE |
| 20692 | /* ------ WFE, YIELD ------ */ |
| 20693 | /* Both appear to get used as a spin-loop hints. Do the usual thing, |
sewardj | 0ef8d9e | 2013-03-01 21:13:24 +0000 | [diff] [blame] | 20694 | which is to continue after yielding. */ |
| 20695 | stmt( IRStmt_Exit( unop(Iop_32to1, mkexpr(condT)), |
| 20696 | Ijk_Yield, |
| 20697 | IRConst_U32((guest_R15_curr_instr_notENC + 2) |
| 20698 | | 1 /*CPSR.T*/), |
| 20699 | OFFB_R15T )); |
sewardj | 5e65812 | 2015-08-17 13:55:41 +0000 | [diff] [blame] | 20700 | Bool isWFE = INSN0(15,0) == 0xBF20; |
| 20701 | DIP(isWFE ? "wfe\n" : "yield\n"); |
sewardj | 0ef8d9e | 2013-03-01 21:13:24 +0000 | [diff] [blame] | 20702 | goto decode_success; |
| 20703 | case 0xBF40: |
| 20704 | /* ------ SEV ------ */ |
| 20705 | /* Treat this as a no-op. Any matching WFEs won't really |
| 20706 | cause the host CPU to snooze; they just cause V to try to |
| 20707 | run some other thread for a while. So there's no point in |
| 20708 | really doing anything for SEV. */ |
| 20709 | DIP("sev\n"); |
| 20710 | goto decode_success; |
| 20711 | default: |
| 20712 | break; /* fall through */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20713 | } |
| 20714 | |
| 20715 | /* ----------------------------------------------------------- */ |
| 20716 | /* -- -- */ |
| 20717 | /* -- Thumb 32-bit integer instructions -- */ |
| 20718 | /* -- -- */ |
| 20719 | /* ----------------------------------------------------------- */ |
| 20720 | |
| 20721 | # define INSN1(_bMax,_bMin) SLICE_UInt(((UInt)insn1), (_bMax), (_bMin)) |
| 20722 | |
| 20723 | /* second 16 bits of the instruction, if any */ |
sewardj | 5276ff5 | 2012-07-14 14:21:56 +0000 | [diff] [blame] | 20724 | vassert(insn1 == 0); |
| 20725 | insn1 = getUShortLittleEndianly( guest_instr+2 ); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20726 | |
| 20727 | anOp = Iop_INVALID; /* paranoia */ |
| 20728 | anOpNm = NULL; /* paranoia */ |
| 20729 | |
| 20730 | /* Change result defaults to suit 32-bit insns. */ |
| 20731 | vassert(dres.whatNext == Dis_Continue); |
| 20732 | vassert(dres.len == 2); |
| 20733 | vassert(dres.continueAt == 0); |
| 20734 | dres.len = 4; |
| 20735 | |
| 20736 | /* ---------------- BL/BLX simm26 ---------------- */ |
| 20737 | if (BITS5(1,1,1,1,0) == INSN0(15,11) && BITS2(1,1) == INSN1(15,14)) { |
| 20738 | UInt isBL = INSN1(12,12); |
| 20739 | UInt bS = INSN0(10,10); |
| 20740 | UInt bJ1 = INSN1(13,13); |
| 20741 | UInt bJ2 = INSN1(11,11); |
| 20742 | UInt bI1 = 1 ^ (bJ1 ^ bS); |
| 20743 | UInt bI2 = 1 ^ (bJ2 ^ bS); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 20744 | UInt uimm25 |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20745 | = (bS << (1 + 1 + 10 + 11 + 1)) |
| 20746 | | (bI1 << (1 + 10 + 11 + 1)) |
| 20747 | | (bI2 << (10 + 11 + 1)) |
| 20748 | | (INSN0(9,0) << (11 + 1)) |
| 20749 | | (INSN1(10,0) << 1); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 20750 | uimm25 <<= 7; |
| 20751 | Int simm25 = (Int)uimm25; |
| 20752 | simm25 >>= 7; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20753 | |
| 20754 | vassert(0 == (guest_R15_curr_instr_notENC & 1)); |
| 20755 | UInt dst = simm25 + guest_R15_curr_instr_notENC + 4; |
| 20756 | |
| 20757 | /* One further validity case to check: in the case of BLX |
| 20758 | (not-BL), that insn1[0] must be zero. */ |
| 20759 | Bool valid = True; |
| 20760 | if (isBL == 0 && INSN1(0,0) == 1) valid = False; |
| 20761 | if (valid) { |
| 20762 | /* Only allowed outside or last-in IT block; SIGILL if not so. */ |
| 20763 | gen_SIGILL_T_if_in_but_NLI_ITBlock(old_itstate, new_itstate); |
| 20764 | // and skip this insn if not selected; being cleverer is too |
| 20765 | // difficult |
| 20766 | mk_skip_over_T32_if_cond_is_false(condT); |
| 20767 | condT = IRTemp_INVALID; |
| 20768 | // now uncond |
| 20769 | |
| 20770 | /* We're returning to Thumb code, hence "| 1" */ |
| 20771 | putIRegT( 14, mkU32( (guest_R15_curr_instr_notENC + 4) | 1 ), |
| 20772 | IRTemp_INVALID); |
| 20773 | if (isBL) { |
| 20774 | /* BL: unconditional T -> T call */ |
| 20775 | /* we're calling Thumb code, hence "| 1" */ |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 20776 | llPutIReg(15, mkU32( dst | 1 )); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20777 | DIP("bl 0x%x (stay in Thumb mode)\n", dst); |
| 20778 | } else { |
| 20779 | /* BLX: unconditional T -> A call */ |
| 20780 | /* we're calling ARM code, hence "& 3" to align to a |
| 20781 | valid ARM insn address */ |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 20782 | llPutIReg(15, mkU32( dst & ~3 )); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20783 | DIP("blx 0x%x (switch to ARM mode)\n", dst & ~3); |
| 20784 | } |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 20785 | dres.whatNext = Dis_StopHere; |
| 20786 | dres.jk_StopHere = Ijk_Call; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20787 | goto decode_success; |
| 20788 | } |
| 20789 | } |
| 20790 | |
| 20791 | /* ---------------- {LD,ST}M{IA,DB} ---------------- */ |
| 20792 | if (0x3a2 == INSN0(15,6) // {LD,ST}MIA |
| 20793 | || 0x3a4 == INSN0(15,6)) { // {LD,ST}MDB |
| 20794 | UInt bW = INSN0(5,5); /* writeback Rn ? */ |
| 20795 | UInt bL = INSN0(4,4); |
| 20796 | UInt rN = INSN0(3,0); |
| 20797 | UInt bP = INSN1(15,15); /* reglist entry for r15 */ |
| 20798 | UInt bM = INSN1(14,14); /* reglist entry for r14 */ |
| 20799 | UInt rLmost = INSN1(12,0); /* reglist entry for r0 .. 12 */ |
| 20800 | UInt rL13 = INSN1(13,13); /* must be zero */ |
| 20801 | UInt regList = 0; |
| 20802 | Bool valid = True; |
| 20803 | |
| 20804 | UInt bINC = 1; |
| 20805 | UInt bBEFORE = 0; |
| 20806 | if (INSN0(15,6) == 0x3a4) { |
| 20807 | bINC = 0; |
| 20808 | bBEFORE = 1; |
| 20809 | } |
| 20810 | |
| 20811 | /* detect statically invalid cases, and construct the final |
| 20812 | reglist */ |
| 20813 | if (rL13 == 1) |
| 20814 | valid = False; |
| 20815 | |
| 20816 | if (bL == 1) { |
| 20817 | regList = (bP << 15) | (bM << 14) | rLmost; |
| 20818 | if (rN == 15) valid = False; |
| 20819 | if (popcount32(regList) < 2) valid = False; |
| 20820 | if (bP == 1 && bM == 1) valid = False; |
| 20821 | if (bW == 1 && (regList & (1<<rN))) valid = False; |
| 20822 | } else { |
| 20823 | regList = (bM << 14) | rLmost; |
| 20824 | if (bP == 1) valid = False; |
| 20825 | if (rN == 15) valid = False; |
| 20826 | if (popcount32(regList) < 2) valid = False; |
| 20827 | if (bW == 1 && (regList & (1<<rN))) valid = False; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20828 | } |
| 20829 | |
| 20830 | if (valid) { |
| 20831 | if (bL == 1 && bP == 1) { |
| 20832 | // We'll be writing the PC. Hence: |
| 20833 | /* Only allowed outside or last-in IT block; SIGILL if not so. */ |
| 20834 | gen_SIGILL_T_if_in_but_NLI_ITBlock(old_itstate, new_itstate); |
| 20835 | } |
| 20836 | |
| 20837 | /* Go uncond: */ |
| 20838 | mk_skip_over_T32_if_cond_is_false(condT); |
| 20839 | condT = IRTemp_INVALID; |
| 20840 | // now uncond |
| 20841 | |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 20842 | /* Generate the IR. This might generate a write to R15. */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20843 | mk_ldm_stm(False/*!arm*/, rN, bINC, bBEFORE, bW, bL, regList); |
| 20844 | |
| 20845 | if (bL == 1 && (regList & (1<<15))) { |
| 20846 | // If we wrote to R15, we have an interworking return to |
| 20847 | // deal with. |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 20848 | llPutIReg(15, llGetIReg(15)); |
| 20849 | dres.jk_StopHere = Ijk_Ret; |
| 20850 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20851 | } |
| 20852 | |
| 20853 | DIP("%sm%c%c r%u%s, {0x%04x}\n", |
| 20854 | bL == 1 ? "ld" : "st", bINC ? 'i' : 'd', bBEFORE ? 'b' : 'a', |
| 20855 | rN, bW ? "!" : "", regList); |
| 20856 | |
| 20857 | goto decode_success; |
| 20858 | } |
| 20859 | } |
| 20860 | |
| 20861 | /* -------------- (T3) ADD{S}.W Rd, Rn, #constT -------------- */ |
| 20862 | if (INSN0(15,11) == BITS5(1,1,1,1,0) |
| 20863 | && INSN0(9,5) == BITS5(0,1,0,0,0) |
| 20864 | && INSN1(15,15) == 0) { |
| 20865 | UInt bS = INSN0(4,4); |
| 20866 | UInt rN = INSN0(3,0); |
| 20867 | UInt rD = INSN1(11,8); |
| 20868 | Bool valid = !isBadRegT(rN) && !isBadRegT(rD); |
sewardj | 51016d1 | 2011-10-26 15:06:25 +0000 | [diff] [blame] | 20869 | /* but allow "add.w reg, sp, #constT" for reg != PC */ |
| 20870 | if (!valid && rD <= 14 && rN == 13) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20871 | valid = True; |
| 20872 | if (valid) { |
| 20873 | IRTemp argL = newTemp(Ity_I32); |
| 20874 | IRTemp argR = newTemp(Ity_I32); |
| 20875 | IRTemp res = newTemp(Ity_I32); |
| 20876 | UInt imm32 = thumbExpandImm_from_I0_I1(NULL, insn0, insn1); |
| 20877 | assign(argL, getIRegT(rN)); |
| 20878 | assign(argR, mkU32(imm32)); |
| 20879 | assign(res, binop(Iop_Add32, mkexpr(argL), mkexpr(argR))); |
| 20880 | putIRegT(rD, mkexpr(res), condT); |
| 20881 | if (bS == 1) |
| 20882 | setFlags_D1_D2( ARMG_CC_OP_ADD, argL, argR, condT ); |
| 20883 | DIP("add%s.w r%u, r%u, #%u\n", |
| 20884 | bS == 1 ? "s" : "", rD, rN, imm32); |
| 20885 | goto decode_success; |
| 20886 | } |
| 20887 | } |
| 20888 | |
sewardj | dbf3d59 | 2011-07-08 15:36:59 +0000 | [diff] [blame] | 20889 | /* ---------------- (T4) ADDW Rd, Rn, #uimm12 -------------- */ |
| 20890 | if (INSN0(15,11) == BITS5(1,1,1,1,0) |
| 20891 | && INSN0(9,4) == BITS6(1,0,0,0,0,0) |
| 20892 | && INSN1(15,15) == 0) { |
| 20893 | UInt rN = INSN0(3,0); |
| 20894 | UInt rD = INSN1(11,8); |
| 20895 | Bool valid = !isBadRegT(rN) && !isBadRegT(rD); |
sewardj | 51016d1 | 2011-10-26 15:06:25 +0000 | [diff] [blame] | 20896 | /* but allow "addw reg, sp, #uimm12" for reg != PC */ |
| 20897 | if (!valid && rD <= 14 && rN == 13) |
sewardj | dbf3d59 | 2011-07-08 15:36:59 +0000 | [diff] [blame] | 20898 | valid = True; |
| 20899 | if (valid) { |
| 20900 | IRTemp argL = newTemp(Ity_I32); |
| 20901 | IRTemp argR = newTemp(Ity_I32); |
| 20902 | IRTemp res = newTemp(Ity_I32); |
| 20903 | UInt imm12 = (INSN0(10,10) << 11) | (INSN1(14,12) << 8) | INSN1(7,0); |
| 20904 | assign(argL, getIRegT(rN)); |
| 20905 | assign(argR, mkU32(imm12)); |
| 20906 | assign(res, binop(Iop_Add32, mkexpr(argL), mkexpr(argR))); |
| 20907 | putIRegT(rD, mkexpr(res), condT); |
| 20908 | DIP("addw r%u, r%u, #%u\n", rD, rN, imm12); |
| 20909 | goto decode_success; |
| 20910 | } |
| 20911 | } |
| 20912 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20913 | /* ---------------- (T2) CMP.W Rn, #constT ---------------- */ |
| 20914 | /* ---------------- (T2) CMN.W Rn, #constT ---------------- */ |
| 20915 | if (INSN0(15,11) == BITS5(1,1,1,1,0) |
| 20916 | && ( INSN0(9,4) == BITS6(0,1,1,0,1,1) // CMP |
| 20917 | || INSN0(9,4) == BITS6(0,1,0,0,0,1)) // CMN |
| 20918 | && INSN1(15,15) == 0 |
| 20919 | && INSN1(11,8) == BITS4(1,1,1,1)) { |
| 20920 | UInt rN = INSN0(3,0); |
| 20921 | if (rN != 15) { |
| 20922 | IRTemp argL = newTemp(Ity_I32); |
| 20923 | IRTemp argR = newTemp(Ity_I32); |
| 20924 | Bool isCMN = INSN0(9,4) == BITS6(0,1,0,0,0,1); |
| 20925 | UInt imm32 = thumbExpandImm_from_I0_I1(NULL, insn0, insn1); |
| 20926 | assign(argL, getIRegT(rN)); |
| 20927 | assign(argR, mkU32(imm32)); |
| 20928 | setFlags_D1_D2( isCMN ? ARMG_CC_OP_ADD : ARMG_CC_OP_SUB, |
| 20929 | argL, argR, condT ); |
| 20930 | DIP("%s.w r%u, #%u\n", isCMN ? "cmn" : "cmp", rN, imm32); |
| 20931 | goto decode_success; |
| 20932 | } |
| 20933 | } |
| 20934 | |
| 20935 | /* -------------- (T1) TST.W Rn, #constT -------------- */ |
| 20936 | /* -------------- (T1) TEQ.W Rn, #constT -------------- */ |
| 20937 | if (INSN0(15,11) == BITS5(1,1,1,1,0) |
| 20938 | && ( INSN0(9,4) == BITS6(0,0,0,0,0,1) // TST |
| 20939 | || INSN0(9,4) == BITS6(0,0,1,0,0,1)) // TEQ |
| 20940 | && INSN1(15,15) == 0 |
| 20941 | && INSN1(11,8) == BITS4(1,1,1,1)) { |
| 20942 | UInt rN = INSN0(3,0); |
| 20943 | if (!isBadRegT(rN)) { // yes, really, it's inconsistent with CMP.W |
| 20944 | Bool isTST = INSN0(9,4) == BITS6(0,0,0,0,0,1); |
| 20945 | IRTemp argL = newTemp(Ity_I32); |
| 20946 | IRTemp argR = newTemp(Ity_I32); |
| 20947 | IRTemp res = newTemp(Ity_I32); |
| 20948 | IRTemp oldV = newTemp(Ity_I32); |
| 20949 | IRTemp oldC = newTemp(Ity_I32); |
| 20950 | Bool updC = False; |
| 20951 | UInt imm32 = thumbExpandImm_from_I0_I1(&updC, insn0, insn1); |
| 20952 | assign(argL, getIRegT(rN)); |
| 20953 | assign(argR, mkU32(imm32)); |
| 20954 | assign(res, binop(isTST ? Iop_And32 : Iop_Xor32, |
| 20955 | mkexpr(argL), mkexpr(argR))); |
| 20956 | assign( oldV, mk_armg_calculate_flag_v() ); |
| 20957 | assign( oldC, updC |
| 20958 | ? mkU32((imm32 >> 31) & 1) |
| 20959 | : mk_armg_calculate_flag_c() ); |
| 20960 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, oldC, oldV, condT ); |
| 20961 | DIP("%s.w r%u, #%u\n", isTST ? "tst" : "teq", rN, imm32); |
| 20962 | goto decode_success; |
| 20963 | } |
| 20964 | } |
| 20965 | |
| 20966 | /* -------------- (T3) SUB{S}.W Rd, Rn, #constT -------------- */ |
| 20967 | /* -------------- (T3) RSB{S}.W Rd, Rn, #constT -------------- */ |
| 20968 | if (INSN0(15,11) == BITS5(1,1,1,1,0) |
| 20969 | && (INSN0(9,5) == BITS5(0,1,1,0,1) // SUB |
| 20970 | || INSN0(9,5) == BITS5(0,1,1,1,0)) // RSB |
| 20971 | && INSN1(15,15) == 0) { |
| 20972 | Bool isRSB = INSN0(9,5) == BITS5(0,1,1,1,0); |
| 20973 | UInt bS = INSN0(4,4); |
| 20974 | UInt rN = INSN0(3,0); |
| 20975 | UInt rD = INSN1(11,8); |
| 20976 | Bool valid = !isBadRegT(rN) && !isBadRegT(rD); |
sewardj | 15c0104 | 2011-03-24 11:14:02 +0000 | [diff] [blame] | 20977 | /* but allow "sub{s}.w reg, sp, #constT |
| 20978 | this is (T2) of "SUB (SP minus immediate)" */ |
| 20979 | if (!valid && !isRSB && rN == 13 && rD != 15) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 20980 | valid = True; |
| 20981 | if (valid) { |
| 20982 | IRTemp argL = newTemp(Ity_I32); |
| 20983 | IRTemp argR = newTemp(Ity_I32); |
| 20984 | IRTemp res = newTemp(Ity_I32); |
| 20985 | UInt imm32 = thumbExpandImm_from_I0_I1(NULL, insn0, insn1); |
| 20986 | assign(argL, getIRegT(rN)); |
| 20987 | assign(argR, mkU32(imm32)); |
| 20988 | assign(res, isRSB |
| 20989 | ? binop(Iop_Sub32, mkexpr(argR), mkexpr(argL)) |
| 20990 | : binop(Iop_Sub32, mkexpr(argL), mkexpr(argR))); |
| 20991 | putIRegT(rD, mkexpr(res), condT); |
| 20992 | if (bS == 1) { |
| 20993 | if (isRSB) |
| 20994 | setFlags_D1_D2( ARMG_CC_OP_SUB, argR, argL, condT ); |
| 20995 | else |
| 20996 | setFlags_D1_D2( ARMG_CC_OP_SUB, argL, argR, condT ); |
| 20997 | } |
| 20998 | DIP("%s%s.w r%u, r%u, #%u\n", |
| 20999 | isRSB ? "rsb" : "sub", bS == 1 ? "s" : "", rD, rN, imm32); |
| 21000 | goto decode_success; |
| 21001 | } |
| 21002 | } |
| 21003 | |
sewardj | dbf3d59 | 2011-07-08 15:36:59 +0000 | [diff] [blame] | 21004 | /* -------------- (T4) SUBW Rd, Rn, #uimm12 ------------------- */ |
| 21005 | if (INSN0(15,11) == BITS5(1,1,1,1,0) |
| 21006 | && INSN0(9,4) == BITS6(1,0,1,0,1,0) |
| 21007 | && INSN1(15,15) == 0) { |
| 21008 | UInt rN = INSN0(3,0); |
| 21009 | UInt rD = INSN1(11,8); |
| 21010 | Bool valid = !isBadRegT(rN) && !isBadRegT(rD); |
| 21011 | /* but allow "subw sp, sp, #uimm12" */ |
| 21012 | if (!valid && rD == 13 && rN == 13) |
| 21013 | valid = True; |
| 21014 | if (valid) { |
| 21015 | IRTemp argL = newTemp(Ity_I32); |
| 21016 | IRTemp argR = newTemp(Ity_I32); |
| 21017 | IRTemp res = newTemp(Ity_I32); |
| 21018 | UInt imm12 = (INSN0(10,10) << 11) | (INSN1(14,12) << 8) | INSN1(7,0); |
| 21019 | assign(argL, getIRegT(rN)); |
| 21020 | assign(argR, mkU32(imm12)); |
| 21021 | assign(res, binop(Iop_Sub32, mkexpr(argL), mkexpr(argR))); |
| 21022 | putIRegT(rD, mkexpr(res), condT); |
| 21023 | DIP("subw r%u, r%u, #%u\n", rD, rN, imm12); |
| 21024 | goto decode_success; |
| 21025 | } |
| 21026 | } |
| 21027 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21028 | /* -------------- (T1) ADC{S}.W Rd, Rn, #constT -------------- */ |
| 21029 | /* -------------- (T1) SBC{S}.W Rd, Rn, #constT -------------- */ |
| 21030 | if (INSN0(15,11) == BITS5(1,1,1,1,0) |
| 21031 | && ( INSN0(9,5) == BITS5(0,1,0,1,0) // ADC |
| 21032 | || INSN0(9,5) == BITS5(0,1,0,1,1)) // SBC |
| 21033 | && INSN1(15,15) == 0) { |
| 21034 | /* ADC: Rd = Rn + constT + oldC */ |
| 21035 | /* SBC: Rd = Rn - constT - (oldC ^ 1) */ |
| 21036 | UInt bS = INSN0(4,4); |
| 21037 | UInt rN = INSN0(3,0); |
| 21038 | UInt rD = INSN1(11,8); |
| 21039 | if (!isBadRegT(rN) && !isBadRegT(rD)) { |
| 21040 | IRTemp argL = newTemp(Ity_I32); |
| 21041 | IRTemp argR = newTemp(Ity_I32); |
| 21042 | IRTemp res = newTemp(Ity_I32); |
| 21043 | IRTemp oldC = newTemp(Ity_I32); |
| 21044 | UInt imm32 = thumbExpandImm_from_I0_I1(NULL, insn0, insn1); |
| 21045 | assign(argL, getIRegT(rN)); |
| 21046 | assign(argR, mkU32(imm32)); |
| 21047 | assign(oldC, mk_armg_calculate_flag_c() ); |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 21048 | const HChar* nm = "???"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21049 | switch (INSN0(9,5)) { |
| 21050 | case BITS5(0,1,0,1,0): // ADC |
| 21051 | nm = "adc"; |
| 21052 | assign(res, |
| 21053 | binop(Iop_Add32, |
| 21054 | binop(Iop_Add32, mkexpr(argL), mkexpr(argR)), |
| 21055 | mkexpr(oldC) )); |
| 21056 | putIRegT(rD, mkexpr(res), condT); |
| 21057 | if (bS) |
| 21058 | setFlags_D1_D2_ND( ARMG_CC_OP_ADC, |
| 21059 | argL, argR, oldC, condT ); |
| 21060 | break; |
| 21061 | case BITS5(0,1,0,1,1): // SBC |
| 21062 | nm = "sbc"; |
| 21063 | assign(res, |
| 21064 | binop(Iop_Sub32, |
| 21065 | binop(Iop_Sub32, mkexpr(argL), mkexpr(argR)), |
| 21066 | binop(Iop_Xor32, mkexpr(oldC), mkU32(1)) )); |
| 21067 | putIRegT(rD, mkexpr(res), condT); |
| 21068 | if (bS) |
| 21069 | setFlags_D1_D2_ND( ARMG_CC_OP_SBB, |
| 21070 | argL, argR, oldC, condT ); |
| 21071 | break; |
| 21072 | default: |
| 21073 | vassert(0); |
| 21074 | } |
| 21075 | DIP("%s%s.w r%u, r%u, #%u\n", |
| 21076 | nm, bS == 1 ? "s" : "", rD, rN, imm32); |
| 21077 | goto decode_success; |
| 21078 | } |
| 21079 | } |
| 21080 | |
| 21081 | /* -------------- (T1) ORR{S}.W Rd, Rn, #constT -------------- */ |
| 21082 | /* -------------- (T1) AND{S}.W Rd, Rn, #constT -------------- */ |
| 21083 | /* -------------- (T1) BIC{S}.W Rd, Rn, #constT -------------- */ |
| 21084 | /* -------------- (T1) EOR{S}.W Rd, Rn, #constT -------------- */ |
| 21085 | if (INSN0(15,11) == BITS5(1,1,1,1,0) |
| 21086 | && ( INSN0(9,5) == BITS5(0,0,0,1,0) // ORR |
| 21087 | || INSN0(9,5) == BITS5(0,0,0,0,0) // AND |
| 21088 | || INSN0(9,5) == BITS5(0,0,0,0,1) // BIC |
sewardj | 04d6da3 | 2010-09-25 22:06:12 +0000 | [diff] [blame] | 21089 | || INSN0(9,5) == BITS5(0,0,1,0,0) // EOR |
| 21090 | || INSN0(9,5) == BITS5(0,0,0,1,1)) // ORN |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21091 | && INSN1(15,15) == 0) { |
| 21092 | UInt bS = INSN0(4,4); |
| 21093 | UInt rN = INSN0(3,0); |
| 21094 | UInt rD = INSN1(11,8); |
| 21095 | if (!isBadRegT(rN) && !isBadRegT(rD)) { |
sewardj | 04d6da3 | 2010-09-25 22:06:12 +0000 | [diff] [blame] | 21096 | Bool notArgR = False; |
| 21097 | IROp op = Iop_INVALID; |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 21098 | const HChar* nm = "???"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21099 | switch (INSN0(9,5)) { |
| 21100 | case BITS5(0,0,0,1,0): op = Iop_Or32; nm = "orr"; break; |
| 21101 | case BITS5(0,0,0,0,0): op = Iop_And32; nm = "and"; break; |
| 21102 | case BITS5(0,0,0,0,1): op = Iop_And32; nm = "bic"; |
sewardj | 04d6da3 | 2010-09-25 22:06:12 +0000 | [diff] [blame] | 21103 | notArgR = True; break; |
| 21104 | case BITS5(0,0,1,0,0): op = Iop_Xor32; nm = "eor"; break; |
| 21105 | case BITS5(0,0,0,1,1): op = Iop_Or32; nm = "orn"; |
| 21106 | notArgR = True; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21107 | default: vassert(0); |
| 21108 | } |
| 21109 | IRTemp argL = newTemp(Ity_I32); |
| 21110 | IRTemp argR = newTemp(Ity_I32); |
| 21111 | IRTemp res = newTemp(Ity_I32); |
| 21112 | Bool updC = False; |
| 21113 | UInt imm32 = thumbExpandImm_from_I0_I1(&updC, insn0, insn1); |
| 21114 | assign(argL, getIRegT(rN)); |
sewardj | 04d6da3 | 2010-09-25 22:06:12 +0000 | [diff] [blame] | 21115 | assign(argR, mkU32(notArgR ? ~imm32 : imm32)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21116 | assign(res, binop(op, mkexpr(argL), mkexpr(argR))); |
| 21117 | putIRegT(rD, mkexpr(res), condT); |
| 21118 | if (bS) { |
| 21119 | IRTemp oldV = newTemp(Ity_I32); |
| 21120 | IRTemp oldC = newTemp(Ity_I32); |
| 21121 | assign( oldV, mk_armg_calculate_flag_v() ); |
| 21122 | assign( oldC, updC |
| 21123 | ? mkU32((imm32 >> 31) & 1) |
| 21124 | : mk_armg_calculate_flag_c() ); |
| 21125 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, oldC, oldV, |
| 21126 | condT ); |
| 21127 | } |
| 21128 | DIP("%s%s.w r%u, r%u, #%u\n", |
| 21129 | nm, bS == 1 ? "s" : "", rD, rN, imm32); |
| 21130 | goto decode_success; |
| 21131 | } |
| 21132 | } |
| 21133 | |
| 21134 | /* ---------- (T3) ADD{S}.W Rd, Rn, Rm, {shift} ---------- */ |
| 21135 | /* ---------- (T3) SUB{S}.W Rd, Rn, Rm, {shift} ---------- */ |
| 21136 | /* ---------- (T3) RSB{S}.W Rd, Rn, Rm, {shift} ---------- */ |
| 21137 | if (INSN0(15,9) == BITS7(1,1,1,0,1,0,1) |
| 21138 | && ( INSN0(8,5) == BITS4(1,0,0,0) // add subopc |
| 21139 | || INSN0(8,5) == BITS4(1,1,0,1) // sub subopc |
| 21140 | || INSN0(8,5) == BITS4(1,1,1,0)) // rsb subopc |
| 21141 | && INSN1(15,15) == 0) { |
| 21142 | UInt rN = INSN0(3,0); |
| 21143 | UInt rD = INSN1(11,8); |
| 21144 | UInt rM = INSN1(3,0); |
| 21145 | UInt bS = INSN0(4,4); |
| 21146 | UInt imm5 = (INSN1(14,12) << 2) | INSN1(7,6); |
| 21147 | UInt how = INSN1(5,4); |
| 21148 | |
| 21149 | Bool valid = !isBadRegT(rD) && !isBadRegT(rN) && !isBadRegT(rM); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 21150 | /* but allow "add.w reg, sp, reg, lsl #N for N=0..31 |
sewardj | 15c0104 | 2011-03-24 11:14:02 +0000 | [diff] [blame] | 21151 | (T3) "ADD (SP plus register) */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21152 | if (!valid && INSN0(8,5) == BITS4(1,0,0,0) // add |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 21153 | && rD != 15 && rN == 13 && imm5 <= 31 && how == 0) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21154 | valid = True; |
| 21155 | } |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 21156 | /* also allow "sub.w reg, sp, reg lsl #N for N=0 .. 5 |
sewardj | 15c0104 | 2011-03-24 11:14:02 +0000 | [diff] [blame] | 21157 | (T1) "SUB (SP minus register) */ |
| 21158 | if (!valid && INSN0(8,5) == BITS4(1,1,0,1) // sub |
Elliott Hughes | ed39800 | 2017-06-21 14:41:24 -0700 | [diff] [blame] | 21159 | && rD != 15 && rN == 13 && imm5 <= 5 && how == 0) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21160 | valid = True; |
| 21161 | } |
| 21162 | if (valid) { |
| 21163 | Bool swap = False; |
| 21164 | IROp op = Iop_INVALID; |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 21165 | const HChar* nm = "???"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21166 | switch (INSN0(8,5)) { |
| 21167 | case BITS4(1,0,0,0): op = Iop_Add32; nm = "add"; break; |
| 21168 | case BITS4(1,1,0,1): op = Iop_Sub32; nm = "sub"; break; |
| 21169 | case BITS4(1,1,1,0): op = Iop_Sub32; nm = "rsb"; |
| 21170 | swap = True; break; |
| 21171 | default: vassert(0); |
| 21172 | } |
| 21173 | |
| 21174 | IRTemp argL = newTemp(Ity_I32); |
| 21175 | assign(argL, getIRegT(rN)); |
| 21176 | |
| 21177 | IRTemp rMt = newTemp(Ity_I32); |
| 21178 | assign(rMt, getIRegT(rM)); |
| 21179 | |
| 21180 | IRTemp argR = newTemp(Ity_I32); |
| 21181 | compute_result_and_C_after_shift_by_imm5( |
| 21182 | dis_buf, &argR, NULL, rMt, how, imm5, rM |
| 21183 | ); |
| 21184 | |
| 21185 | IRTemp res = newTemp(Ity_I32); |
| 21186 | assign(res, swap |
| 21187 | ? binop(op, mkexpr(argR), mkexpr(argL)) |
| 21188 | : binop(op, mkexpr(argL), mkexpr(argR))); |
| 21189 | |
| 21190 | putIRegT(rD, mkexpr(res), condT); |
| 21191 | if (bS) { |
| 21192 | switch (op) { |
| 21193 | case Iop_Add32: |
| 21194 | setFlags_D1_D2( ARMG_CC_OP_ADD, argL, argR, condT ); |
| 21195 | break; |
| 21196 | case Iop_Sub32: |
| 21197 | if (swap) |
| 21198 | setFlags_D1_D2( ARMG_CC_OP_SUB, argR, argL, condT ); |
| 21199 | else |
| 21200 | setFlags_D1_D2( ARMG_CC_OP_SUB, argL, argR, condT ); |
| 21201 | break; |
| 21202 | default: |
| 21203 | vassert(0); |
| 21204 | } |
| 21205 | } |
| 21206 | |
| 21207 | DIP("%s%s.w r%u, r%u, %s\n", |
| 21208 | nm, bS ? "s" : "", rD, rN, dis_buf); |
| 21209 | goto decode_success; |
| 21210 | } |
| 21211 | } |
| 21212 | |
| 21213 | /* ---------- (T3) ADC{S}.W Rd, Rn, Rm, {shift} ---------- */ |
| 21214 | /* ---------- (T2) SBC{S}.W Rd, Rn, Rm, {shift} ---------- */ |
| 21215 | if (INSN0(15,9) == BITS7(1,1,1,0,1,0,1) |
| 21216 | && ( INSN0(8,5) == BITS4(1,0,1,0) // adc subopc |
| 21217 | || INSN0(8,5) == BITS4(1,0,1,1)) // sbc subopc |
| 21218 | && INSN1(15,15) == 0) { |
| 21219 | /* ADC: Rd = Rn + shifter_operand + oldC */ |
| 21220 | /* SBC: Rd = Rn - shifter_operand - (oldC ^ 1) */ |
| 21221 | UInt rN = INSN0(3,0); |
| 21222 | UInt rD = INSN1(11,8); |
| 21223 | UInt rM = INSN1(3,0); |
| 21224 | if (!isBadRegT(rD) && !isBadRegT(rN) && !isBadRegT(rM)) { |
| 21225 | UInt bS = INSN0(4,4); |
| 21226 | UInt imm5 = (INSN1(14,12) << 2) | INSN1(7,6); |
| 21227 | UInt how = INSN1(5,4); |
| 21228 | |
| 21229 | IRTemp argL = newTemp(Ity_I32); |
| 21230 | assign(argL, getIRegT(rN)); |
| 21231 | |
| 21232 | IRTemp rMt = newTemp(Ity_I32); |
| 21233 | assign(rMt, getIRegT(rM)); |
| 21234 | |
| 21235 | IRTemp oldC = newTemp(Ity_I32); |
| 21236 | assign(oldC, mk_armg_calculate_flag_c()); |
| 21237 | |
| 21238 | IRTemp argR = newTemp(Ity_I32); |
| 21239 | compute_result_and_C_after_shift_by_imm5( |
| 21240 | dis_buf, &argR, NULL, rMt, how, imm5, rM |
| 21241 | ); |
| 21242 | |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 21243 | const HChar* nm = "???"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21244 | IRTemp res = newTemp(Ity_I32); |
| 21245 | switch (INSN0(8,5)) { |
| 21246 | case BITS4(1,0,1,0): // ADC |
| 21247 | nm = "adc"; |
| 21248 | assign(res, |
| 21249 | binop(Iop_Add32, |
| 21250 | binop(Iop_Add32, mkexpr(argL), mkexpr(argR)), |
| 21251 | mkexpr(oldC) )); |
| 21252 | putIRegT(rD, mkexpr(res), condT); |
| 21253 | if (bS) |
| 21254 | setFlags_D1_D2_ND( ARMG_CC_OP_ADC, |
| 21255 | argL, argR, oldC, condT ); |
| 21256 | break; |
| 21257 | case BITS4(1,0,1,1): // SBC |
| 21258 | nm = "sbc"; |
| 21259 | assign(res, |
| 21260 | binop(Iop_Sub32, |
| 21261 | binop(Iop_Sub32, mkexpr(argL), mkexpr(argR)), |
| 21262 | binop(Iop_Xor32, mkexpr(oldC), mkU32(1)) )); |
| 21263 | putIRegT(rD, mkexpr(res), condT); |
| 21264 | if (bS) |
| 21265 | setFlags_D1_D2_ND( ARMG_CC_OP_SBB, |
| 21266 | argL, argR, oldC, condT ); |
| 21267 | break; |
| 21268 | default: |
| 21269 | vassert(0); |
| 21270 | } |
| 21271 | |
| 21272 | DIP("%s%s.w r%u, r%u, %s\n", |
| 21273 | nm, bS ? "s" : "", rD, rN, dis_buf); |
| 21274 | goto decode_success; |
| 21275 | } |
| 21276 | } |
| 21277 | |
| 21278 | /* ---------- (T3) AND{S}.W Rd, Rn, Rm, {shift} ---------- */ |
| 21279 | /* ---------- (T3) ORR{S}.W Rd, Rn, Rm, {shift} ---------- */ |
| 21280 | /* ---------- (T3) EOR{S}.W Rd, Rn, Rm, {shift} ---------- */ |
| 21281 | /* ---------- (T3) BIC{S}.W Rd, Rn, Rm, {shift} ---------- */ |
sewardj | 04d6da3 | 2010-09-25 22:06:12 +0000 | [diff] [blame] | 21282 | /* ---------- (T1) ORN{S}.W Rd, Rn, Rm, {shift} ---------- */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21283 | if (INSN0(15,9) == BITS7(1,1,1,0,1,0,1) |
| 21284 | && ( INSN0(8,5) == BITS4(0,0,0,0) // and subopc |
| 21285 | || INSN0(8,5) == BITS4(0,0,1,0) // orr subopc |
| 21286 | || INSN0(8,5) == BITS4(0,1,0,0) // eor subopc |
sewardj | 04d6da3 | 2010-09-25 22:06:12 +0000 | [diff] [blame] | 21287 | || INSN0(8,5) == BITS4(0,0,0,1) // bic subopc |
| 21288 | || INSN0(8,5) == BITS4(0,0,1,1)) // orn subopc |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21289 | && INSN1(15,15) == 0) { |
| 21290 | UInt rN = INSN0(3,0); |
| 21291 | UInt rD = INSN1(11,8); |
| 21292 | UInt rM = INSN1(3,0); |
| 21293 | if (!isBadRegT(rD) && !isBadRegT(rN) && !isBadRegT(rM)) { |
sewardj | 04d6da3 | 2010-09-25 22:06:12 +0000 | [diff] [blame] | 21294 | Bool notArgR = False; |
| 21295 | IROp op = Iop_INVALID; |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 21296 | const HChar* nm = "???"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21297 | switch (INSN0(8,5)) { |
| 21298 | case BITS4(0,0,0,0): op = Iop_And32; nm = "and"; break; |
| 21299 | case BITS4(0,0,1,0): op = Iop_Or32; nm = "orr"; break; |
| 21300 | case BITS4(0,1,0,0): op = Iop_Xor32; nm = "eor"; break; |
| 21301 | case BITS4(0,0,0,1): op = Iop_And32; nm = "bic"; |
sewardj | 04d6da3 | 2010-09-25 22:06:12 +0000 | [diff] [blame] | 21302 | notArgR = True; break; |
| 21303 | case BITS4(0,0,1,1): op = Iop_Or32; nm = "orn"; |
| 21304 | notArgR = True; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21305 | default: vassert(0); |
| 21306 | } |
| 21307 | UInt bS = INSN0(4,4); |
| 21308 | UInt imm5 = (INSN1(14,12) << 2) | INSN1(7,6); |
| 21309 | UInt how = INSN1(5,4); |
| 21310 | |
| 21311 | IRTemp rNt = newTemp(Ity_I32); |
| 21312 | assign(rNt, getIRegT(rN)); |
| 21313 | |
| 21314 | IRTemp rMt = newTemp(Ity_I32); |
| 21315 | assign(rMt, getIRegT(rM)); |
| 21316 | |
| 21317 | IRTemp argR = newTemp(Ity_I32); |
| 21318 | IRTemp oldC = bS ? newTemp(Ity_I32) : IRTemp_INVALID; |
| 21319 | |
| 21320 | compute_result_and_C_after_shift_by_imm5( |
| 21321 | dis_buf, &argR, bS ? &oldC : NULL, rMt, how, imm5, rM |
| 21322 | ); |
| 21323 | |
| 21324 | IRTemp res = newTemp(Ity_I32); |
sewardj | 04d6da3 | 2010-09-25 22:06:12 +0000 | [diff] [blame] | 21325 | if (notArgR) { |
| 21326 | vassert(op == Iop_And32 || op == Iop_Or32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21327 | assign(res, binop(op, mkexpr(rNt), |
| 21328 | unop(Iop_Not32, mkexpr(argR)))); |
| 21329 | } else { |
| 21330 | assign(res, binop(op, mkexpr(rNt), mkexpr(argR))); |
| 21331 | } |
| 21332 | |
| 21333 | putIRegT(rD, mkexpr(res), condT); |
| 21334 | if (bS) { |
| 21335 | IRTemp oldV = newTemp(Ity_I32); |
| 21336 | assign( oldV, mk_armg_calculate_flag_v() ); |
| 21337 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, oldC, oldV, |
| 21338 | condT ); |
| 21339 | } |
| 21340 | |
| 21341 | DIP("%s%s.w r%u, r%u, %s\n", |
| 21342 | nm, bS ? "s" : "", rD, rN, dis_buf); |
| 21343 | goto decode_success; |
| 21344 | } |
| 21345 | } |
| 21346 | |
| 21347 | /* -------------- (T?) LSL{S}.W Rd, Rn, Rm -------------- */ |
| 21348 | /* -------------- (T?) LSR{S}.W Rd, Rn, Rm -------------- */ |
| 21349 | /* -------------- (T?) ASR{S}.W Rd, Rn, Rm -------------- */ |
| 21350 | /* -------------- (T?) ROR{S}.W Rd, Rn, Rm -------------- */ |
| 21351 | if (INSN0(15,7) == BITS9(1,1,1,1,1,0,1,0,0) |
| 21352 | && INSN1(15,12) == BITS4(1,1,1,1) |
| 21353 | && INSN1(7,4) == BITS4(0,0,0,0)) { |
| 21354 | UInt how = INSN0(6,5); // standard encoding |
| 21355 | UInt rN = INSN0(3,0); |
| 21356 | UInt rD = INSN1(11,8); |
| 21357 | UInt rM = INSN1(3,0); |
| 21358 | UInt bS = INSN0(4,4); |
| 21359 | Bool valid = !isBadRegT(rN) && !isBadRegT(rM) && !isBadRegT(rD); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21360 | if (valid) { |
| 21361 | IRTemp rNt = newTemp(Ity_I32); |
| 21362 | IRTemp rMt = newTemp(Ity_I32); |
| 21363 | IRTemp res = newTemp(Ity_I32); |
| 21364 | IRTemp oldC = bS ? newTemp(Ity_I32) : IRTemp_INVALID; |
| 21365 | IRTemp oldV = bS ? newTemp(Ity_I32) : IRTemp_INVALID; |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 21366 | const HChar* nms[4] = { "lsl", "lsr", "asr", "ror" }; |
| 21367 | const HChar* nm = nms[how]; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21368 | assign(rNt, getIRegT(rN)); |
| 21369 | assign(rMt, getIRegT(rM)); |
| 21370 | compute_result_and_C_after_shift_by_reg( |
| 21371 | dis_buf, &res, bS ? &oldC : NULL, |
| 21372 | rNt, how, rMt, rN, rM |
| 21373 | ); |
| 21374 | if (bS) |
| 21375 | assign(oldV, mk_armg_calculate_flag_v()); |
| 21376 | putIRegT(rD, mkexpr(res), condT); |
| 21377 | if (bS) { |
| 21378 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, oldC, oldV, |
| 21379 | condT ); |
| 21380 | } |
| 21381 | DIP("%s%s.w r%u, r%u, r%u\n", |
| 21382 | nm, bS ? "s" : "", rD, rN, rM); |
| 21383 | goto decode_success; |
| 21384 | } |
| 21385 | } |
| 21386 | |
| 21387 | /* ------------ (T?) MOV{S}.W Rd, Rn, {shift} ------------ */ |
| 21388 | /* ------------ (T?) MVN{S}.W Rd, Rn, {shift} ------------ */ |
| 21389 | if ((INSN0(15,0) & 0xFFCF) == 0xEA4F |
| 21390 | && INSN1(15,15) == 0) { |
sewardj | f88e39d | 2015-08-17 07:29:07 +0000 | [diff] [blame] | 21391 | UInt rD = INSN1(11,8); |
| 21392 | UInt rN = INSN1(3,0); |
| 21393 | UInt bS = INSN0(4,4); |
| 21394 | UInt isMVN = INSN0(5,5); |
| 21395 | Bool regsOK = (bS || isMVN) |
| 21396 | ? (!isBadRegT(rD) && !isBadRegT(rN)) |
| 21397 | : (rD != 15 && rN != 15 && (rD != 13 || rN != 13)); |
| 21398 | if (regsOK) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21399 | UInt imm5 = (INSN1(14,12) << 2) | INSN1(7,6); |
| 21400 | UInt how = INSN1(5,4); |
| 21401 | |
| 21402 | IRTemp rNt = newTemp(Ity_I32); |
| 21403 | assign(rNt, getIRegT(rN)); |
| 21404 | |
| 21405 | IRTemp oldRn = newTemp(Ity_I32); |
| 21406 | IRTemp oldC = bS ? newTemp(Ity_I32) : IRTemp_INVALID; |
| 21407 | compute_result_and_C_after_shift_by_imm5( |
| 21408 | dis_buf, &oldRn, bS ? &oldC : NULL, rNt, how, imm5, rN |
| 21409 | ); |
| 21410 | |
| 21411 | IRTemp res = newTemp(Ity_I32); |
| 21412 | assign(res, isMVN ? unop(Iop_Not32, mkexpr(oldRn)) |
| 21413 | : mkexpr(oldRn)); |
| 21414 | |
| 21415 | putIRegT(rD, mkexpr(res), condT); |
| 21416 | if (bS) { |
| 21417 | IRTemp oldV = newTemp(Ity_I32); |
| 21418 | assign( oldV, mk_armg_calculate_flag_v() ); |
| 21419 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, oldC, oldV, condT); |
| 21420 | } |
| 21421 | DIP("%s%s.w r%u, %s\n", |
| 21422 | isMVN ? "mvn" : "mov", bS ? "s" : "", rD, dis_buf); |
| 21423 | goto decode_success; |
| 21424 | } |
| 21425 | } |
| 21426 | |
| 21427 | /* -------------- (T?) TST.W Rn, Rm, {shift} -------------- */ |
| 21428 | /* -------------- (T?) TEQ.W Rn, Rm, {shift} -------------- */ |
| 21429 | if (INSN0(15,9) == BITS7(1,1,1,0,1,0,1) |
| 21430 | && ( INSN0(8,4) == BITS5(0,0,0,0,1) // TST |
| 21431 | || INSN0(8,4) == BITS5(0,1,0,0,1)) // TEQ |
| 21432 | && INSN1(15,15) == 0 |
| 21433 | && INSN1(11,8) == BITS4(1,1,1,1)) { |
| 21434 | UInt rN = INSN0(3,0); |
| 21435 | UInt rM = INSN1(3,0); |
| 21436 | if (!isBadRegT(rN) && !isBadRegT(rM)) { |
| 21437 | Bool isTST = INSN0(8,4) == BITS5(0,0,0,0,1); |
| 21438 | |
| 21439 | UInt how = INSN1(5,4); |
| 21440 | UInt imm5 = (INSN1(14,12) << 2) | INSN1(7,6); |
| 21441 | |
| 21442 | IRTemp argL = newTemp(Ity_I32); |
| 21443 | assign(argL, getIRegT(rN)); |
| 21444 | |
| 21445 | IRTemp rMt = newTemp(Ity_I32); |
| 21446 | assign(rMt, getIRegT(rM)); |
| 21447 | |
| 21448 | IRTemp argR = newTemp(Ity_I32); |
| 21449 | IRTemp oldC = newTemp(Ity_I32); |
| 21450 | compute_result_and_C_after_shift_by_imm5( |
| 21451 | dis_buf, &argR, &oldC, rMt, how, imm5, rM |
| 21452 | ); |
| 21453 | |
| 21454 | IRTemp oldV = newTemp(Ity_I32); |
| 21455 | assign( oldV, mk_armg_calculate_flag_v() ); |
| 21456 | |
| 21457 | IRTemp res = newTemp(Ity_I32); |
| 21458 | assign(res, binop(isTST ? Iop_And32 : Iop_Xor32, |
| 21459 | mkexpr(argL), mkexpr(argR))); |
| 21460 | |
| 21461 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, oldC, oldV, |
| 21462 | condT ); |
| 21463 | DIP("%s.w r%u, %s\n", isTST ? "tst" : "teq", rN, dis_buf); |
| 21464 | goto decode_success; |
| 21465 | } |
| 21466 | } |
| 21467 | |
| 21468 | /* -------------- (T3) CMP.W Rn, Rm, {shift} -------------- */ |
| 21469 | /* -------------- (T2) CMN.W Rn, Rm, {shift} -------------- */ |
| 21470 | if (INSN0(15,9) == BITS7(1,1,1,0,1,0,1) |
| 21471 | && ( INSN0(8,4) == BITS5(1,1,0,1,1) // CMP |
| 21472 | || INSN0(8,4) == BITS5(1,0,0,0,1)) // CMN |
| 21473 | && INSN1(15,15) == 0 |
| 21474 | && INSN1(11,8) == BITS4(1,1,1,1)) { |
| 21475 | UInt rN = INSN0(3,0); |
| 21476 | UInt rM = INSN1(3,0); |
| 21477 | if (!isBadRegT(rN) && !isBadRegT(rM)) { |
| 21478 | Bool isCMN = INSN0(8,4) == BITS5(1,0,0,0,1); |
| 21479 | UInt how = INSN1(5,4); |
| 21480 | UInt imm5 = (INSN1(14,12) << 2) | INSN1(7,6); |
| 21481 | |
| 21482 | IRTemp argL = newTemp(Ity_I32); |
| 21483 | assign(argL, getIRegT(rN)); |
| 21484 | |
| 21485 | IRTemp rMt = newTemp(Ity_I32); |
| 21486 | assign(rMt, getIRegT(rM)); |
| 21487 | |
| 21488 | IRTemp argR = newTemp(Ity_I32); |
| 21489 | compute_result_and_C_after_shift_by_imm5( |
| 21490 | dis_buf, &argR, NULL, rMt, how, imm5, rM |
| 21491 | ); |
| 21492 | |
| 21493 | setFlags_D1_D2( isCMN ? ARMG_CC_OP_ADD : ARMG_CC_OP_SUB, |
| 21494 | argL, argR, condT ); |
| 21495 | |
| 21496 | DIP("%s.w r%u, %s\n", isCMN ? "cmn" : "cmp", rN, dis_buf); |
| 21497 | goto decode_success; |
| 21498 | } |
| 21499 | } |
| 21500 | |
| 21501 | /* -------------- (T2) MOV{S}.W Rd, #constT -------------- */ |
| 21502 | /* -------------- (T2) MVN{S}.W Rd, #constT -------------- */ |
| 21503 | if (INSN0(15,11) == BITS5(1,1,1,1,0) |
| 21504 | && ( INSN0(9,5) == BITS5(0,0,0,1,0) // MOV |
| 21505 | || INSN0(9,5) == BITS5(0,0,0,1,1)) // MVN |
| 21506 | && INSN0(3,0) == BITS4(1,1,1,1) |
| 21507 | && INSN1(15,15) == 0) { |
| 21508 | UInt rD = INSN1(11,8); |
| 21509 | if (!isBadRegT(rD)) { |
| 21510 | Bool updC = False; |
| 21511 | UInt bS = INSN0(4,4); |
| 21512 | Bool isMVN = INSN0(5,5) == 1; |
| 21513 | UInt imm32 = thumbExpandImm_from_I0_I1(&updC, insn0, insn1); |
| 21514 | IRTemp res = newTemp(Ity_I32); |
| 21515 | assign(res, mkU32(isMVN ? ~imm32 : imm32)); |
| 21516 | putIRegT(rD, mkexpr(res), condT); |
| 21517 | if (bS) { |
| 21518 | IRTemp oldV = newTemp(Ity_I32); |
| 21519 | IRTemp oldC = newTemp(Ity_I32); |
| 21520 | assign( oldV, mk_armg_calculate_flag_v() ); |
| 21521 | assign( oldC, updC |
| 21522 | ? mkU32((imm32 >> 31) & 1) |
| 21523 | : mk_armg_calculate_flag_c() ); |
| 21524 | setFlags_D1_D2_ND( ARMG_CC_OP_LOGIC, res, oldC, oldV, |
| 21525 | condT ); |
| 21526 | } |
| 21527 | DIP("%s%s.w r%u, #%u\n", |
| 21528 | isMVN ? "mvn" : "mov", bS ? "s" : "", rD, imm32); |
| 21529 | goto decode_success; |
| 21530 | } |
| 21531 | } |
| 21532 | |
| 21533 | /* -------------- (T3) MOVW Rd, #imm16 -------------- */ |
| 21534 | if (INSN0(15,11) == BITS5(1,1,1,1,0) |
| 21535 | && INSN0(9,4) == BITS6(1,0,0,1,0,0) |
| 21536 | && INSN1(15,15) == 0) { |
| 21537 | UInt rD = INSN1(11,8); |
| 21538 | if (!isBadRegT(rD)) { |
| 21539 | UInt imm16 = (INSN0(3,0) << 12) | (INSN0(10,10) << 11) |
| 21540 | | (INSN1(14,12) << 8) | INSN1(7,0); |
| 21541 | putIRegT(rD, mkU32(imm16), condT); |
| 21542 | DIP("movw r%u, #%u\n", rD, imm16); |
| 21543 | goto decode_success; |
| 21544 | } |
| 21545 | } |
| 21546 | |
| 21547 | /* ---------------- MOVT Rd, #imm16 ---------------- */ |
| 21548 | if (INSN0(15,11) == BITS5(1,1,1,1,0) |
| 21549 | && INSN0(9,4) == BITS6(1,0,1,1,0,0) |
| 21550 | && INSN1(15,15) == 0) { |
| 21551 | UInt rD = INSN1(11,8); |
| 21552 | if (!isBadRegT(rD)) { |
| 21553 | UInt imm16 = (INSN0(3,0) << 12) | (INSN0(10,10) << 11) |
| 21554 | | (INSN1(14,12) << 8) | INSN1(7,0); |
| 21555 | IRTemp res = newTemp(Ity_I32); |
| 21556 | assign(res, |
| 21557 | binop(Iop_Or32, |
| 21558 | binop(Iop_And32, getIRegT(rD), mkU32(0xFFFF)), |
| 21559 | mkU32(imm16 << 16))); |
| 21560 | putIRegT(rD, mkexpr(res), condT); |
| 21561 | DIP("movt r%u, #%u\n", rD, imm16); |
| 21562 | goto decode_success; |
| 21563 | } |
| 21564 | } |
| 21565 | |
| 21566 | /* ---------------- LD/ST reg+/-#imm8 ---------------- */ |
| 21567 | /* Loads and stores of the form: |
| 21568 | op Rt, [Rn, #-imm8] or |
| 21569 | op Rt, [Rn], #+/-imm8 or |
| 21570 | op Rt, [Rn, #+/-imm8]! |
| 21571 | where op is one of |
| 21572 | ldrb ldrh ldr ldrsb ldrsh |
| 21573 | strb strh str |
| 21574 | */ |
| 21575 | if (INSN0(15,9) == BITS7(1,1,1,1,1,0,0) && INSN1(11,11) == 1) { |
| 21576 | Bool valid = True; |
| 21577 | Bool syned = False; |
| 21578 | Bool isST = False; |
| 21579 | IRType ty = Ity_I8; |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 21580 | const HChar* nm = "???"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21581 | |
| 21582 | switch (INSN0(8,4)) { |
| 21583 | case BITS5(0,0,0,0,0): // strb |
| 21584 | nm = "strb"; isST = True; break; |
| 21585 | case BITS5(0,0,0,0,1): // ldrb |
| 21586 | nm = "ldrb"; break; |
| 21587 | case BITS5(1,0,0,0,1): // ldrsb |
| 21588 | nm = "ldrsb"; syned = True; break; |
| 21589 | case BITS5(0,0,0,1,0): // strh |
| 21590 | nm = "strh"; ty = Ity_I16; isST = True; break; |
| 21591 | case BITS5(0,0,0,1,1): // ldrh |
| 21592 | nm = "ldrh"; ty = Ity_I16; break; |
| 21593 | case BITS5(1,0,0,1,1): // ldrsh |
| 21594 | nm = "ldrsh"; ty = Ity_I16; syned = True; break; |
| 21595 | case BITS5(0,0,1,0,0): // str |
| 21596 | nm = "str"; ty = Ity_I32; isST = True; break; |
| 21597 | case BITS5(0,0,1,0,1): |
| 21598 | nm = "ldr"; ty = Ity_I32; break; // ldr |
| 21599 | default: |
| 21600 | valid = False; break; |
| 21601 | } |
| 21602 | |
| 21603 | UInt rN = INSN0(3,0); |
| 21604 | UInt rT = INSN1(15,12); |
| 21605 | UInt bP = INSN1(10,10); |
| 21606 | UInt bU = INSN1(9,9); |
| 21607 | UInt bW = INSN1(8,8); |
| 21608 | UInt imm8 = INSN1(7,0); |
| 21609 | Bool loadsPC = False; |
| 21610 | |
| 21611 | if (valid) { |
| 21612 | if (bP == 1 && bU == 1 && bW == 0) |
| 21613 | valid = False; |
| 21614 | if (bP == 0 && bW == 0) |
| 21615 | valid = False; |
| 21616 | if (rN == 15) |
| 21617 | valid = False; |
| 21618 | if (bW == 1 && rN == rT) |
| 21619 | valid = False; |
| 21620 | if (ty == Ity_I8 || ty == Ity_I16) { |
| 21621 | if (isBadRegT(rT)) |
| 21622 | valid = False; |
| 21623 | } else { |
| 21624 | /* ty == Ity_I32 */ |
| 21625 | if (isST && rT == 15) |
| 21626 | valid = False; |
| 21627 | if (!isST && rT == 15) |
| 21628 | loadsPC = True; |
| 21629 | } |
| 21630 | } |
| 21631 | |
| 21632 | if (valid) { |
| 21633 | // if it's a branch, it can't happen in the middle of an IT block |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21634 | // Also, if it is a branch, make it unconditional at this point. |
| 21635 | // Doing conditional branches in-line is too complex (for now) |
| 21636 | if (loadsPC) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21637 | gen_SIGILL_T_if_in_but_NLI_ITBlock(old_itstate, new_itstate); |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21638 | // go uncond |
| 21639 | mk_skip_over_T32_if_cond_is_false(condT); |
| 21640 | condT = IRTemp_INVALID; |
| 21641 | // now uncond |
| 21642 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21643 | |
| 21644 | IRTemp preAddr = newTemp(Ity_I32); |
| 21645 | assign(preAddr, getIRegT(rN)); |
| 21646 | |
| 21647 | IRTemp postAddr = newTemp(Ity_I32); |
| 21648 | assign(postAddr, binop(bU == 1 ? Iop_Add32 : Iop_Sub32, |
| 21649 | mkexpr(preAddr), mkU32(imm8))); |
| 21650 | |
| 21651 | IRTemp transAddr = bP == 1 ? postAddr : preAddr; |
| 21652 | |
| 21653 | if (isST) { |
| 21654 | |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21655 | /* Store. If necessary, update the base register before |
| 21656 | the store itself, so that the common idiom of "str rX, |
| 21657 | [sp, #-4]!" (store rX at sp-4, then do new sp = sp-4, |
| 21658 | a.k.a "push rX") doesn't cause Memcheck to complain |
| 21659 | that the access is below the stack pointer. Also, not |
| 21660 | updating sp before the store confuses Valgrind's |
| 21661 | dynamic stack-extending logic. So do it before the |
| 21662 | store. Hence we need to snarf the store data before |
| 21663 | doing the basereg update. */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21664 | |
| 21665 | /* get hold of the data to be stored */ |
| 21666 | IRTemp oldRt = newTemp(Ity_I32); |
| 21667 | assign(oldRt, getIRegT(rT)); |
| 21668 | |
| 21669 | /* Update Rn if necessary. */ |
| 21670 | if (bW == 1) { |
| 21671 | vassert(rN != rT); // assured by validity check above |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21672 | putIRegT(rN, mkexpr(postAddr), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21673 | } |
| 21674 | |
| 21675 | /* generate the transfer */ |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21676 | IRExpr* data = NULL; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21677 | switch (ty) { |
| 21678 | case Ity_I8: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21679 | data = unop(Iop_32to8, mkexpr(oldRt)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21680 | break; |
| 21681 | case Ity_I16: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21682 | data = unop(Iop_32to16, mkexpr(oldRt)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21683 | break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21684 | case Ity_I32: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21685 | data = mkexpr(oldRt); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21686 | break; |
| 21687 | default: |
| 21688 | vassert(0); |
| 21689 | } |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21690 | storeGuardedLE(mkexpr(transAddr), data, condT); |
| 21691 | |
| 21692 | } else { |
| 21693 | |
| 21694 | /* Load. */ |
| 21695 | IRTemp llOldRt = newTemp(Ity_I32); |
| 21696 | assign(llOldRt, llGetIReg(rT)); |
| 21697 | |
| 21698 | /* generate the transfer */ |
| 21699 | IRTemp newRt = newTemp(Ity_I32); |
| 21700 | IRLoadGOp widen = ILGop_INVALID; |
| 21701 | switch (ty) { |
| 21702 | case Ity_I8: |
| 21703 | widen = syned ? ILGop_8Sto32 : ILGop_8Uto32; break; |
| 21704 | case Ity_I16: |
| 21705 | widen = syned ? ILGop_16Sto32 : ILGop_16Uto32; break; |
| 21706 | case Ity_I32: |
| 21707 | widen = ILGop_Ident32; break; |
| 21708 | default: |
| 21709 | vassert(0); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21710 | } |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21711 | loadGuardedLE(newRt, widen, |
| 21712 | mkexpr(transAddr), mkexpr(llOldRt), condT); |
| 21713 | if (rT == 15) { |
| 21714 | vassert(loadsPC); |
| 21715 | /* We'll do the write to the PC just below */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21716 | } else { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21717 | vassert(!loadsPC); |
| 21718 | /* IRTemp_INVALID is OK here because in the case where |
| 21719 | condT is false at run time, we're just putting the |
| 21720 | old rT value back. */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21721 | putIRegT(rT, mkexpr(newRt), IRTemp_INVALID); |
| 21722 | } |
| 21723 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21724 | /* Update Rn if necessary. */ |
| 21725 | if (bW == 1) { |
| 21726 | vassert(rN != rT); // assured by validity check above |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21727 | putIRegT(rN, mkexpr(postAddr), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21728 | } |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 21729 | |
| 21730 | if (loadsPC) { |
| 21731 | /* Presumably this is an interworking branch. */ |
| 21732 | vassert(rN != 15); // assured by validity check above |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21733 | vassert(rT == 15); |
| 21734 | vassert(condT == IRTemp_INVALID); /* due to check above */ |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 21735 | llPutIReg(15, mkexpr(newRt)); |
| 21736 | dres.jk_StopHere = Ijk_Boring; /* or _Ret ? */ |
| 21737 | dres.whatNext = Dis_StopHere; |
| 21738 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21739 | } |
| 21740 | |
| 21741 | if (bP == 1 && bW == 0) { |
| 21742 | DIP("%s.w r%u, [r%u, #%c%u]\n", |
| 21743 | nm, rT, rN, bU ? '+' : '-', imm8); |
| 21744 | } |
| 21745 | else if (bP == 1 && bW == 1) { |
| 21746 | DIP("%s.w r%u, [r%u, #%c%u]!\n", |
| 21747 | nm, rT, rN, bU ? '+' : '-', imm8); |
| 21748 | } |
| 21749 | else { |
| 21750 | vassert(bP == 0 && bW == 1); |
| 21751 | DIP("%s.w r%u, [r%u], #%c%u\n", |
| 21752 | nm, rT, rN, bU ? '+' : '-', imm8); |
| 21753 | } |
| 21754 | |
| 21755 | goto decode_success; |
| 21756 | } |
| 21757 | } |
| 21758 | |
| 21759 | /* ------------- LD/ST reg+(reg<<imm2) ------------- */ |
| 21760 | /* Loads and stores of the form: |
| 21761 | op Rt, [Rn, Rm, LSL #imm8] |
| 21762 | where op is one of |
| 21763 | ldrb ldrh ldr ldrsb ldrsh |
| 21764 | strb strh str |
| 21765 | */ |
| 21766 | if (INSN0(15,9) == BITS7(1,1,1,1,1,0,0) |
| 21767 | && INSN1(11,6) == BITS6(0,0,0,0,0,0)) { |
| 21768 | Bool valid = True; |
| 21769 | Bool syned = False; |
| 21770 | Bool isST = False; |
| 21771 | IRType ty = Ity_I8; |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 21772 | const HChar* nm = "???"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21773 | |
| 21774 | switch (INSN0(8,4)) { |
| 21775 | case BITS5(0,0,0,0,0): // strb |
| 21776 | nm = "strb"; isST = True; break; |
| 21777 | case BITS5(0,0,0,0,1): // ldrb |
| 21778 | nm = "ldrb"; break; |
| 21779 | case BITS5(1,0,0,0,1): // ldrsb |
| 21780 | nm = "ldrsb"; syned = True; break; |
| 21781 | case BITS5(0,0,0,1,0): // strh |
| 21782 | nm = "strh"; ty = Ity_I16; isST = True; break; |
| 21783 | case BITS5(0,0,0,1,1): // ldrh |
| 21784 | nm = "ldrh"; ty = Ity_I16; break; |
| 21785 | case BITS5(1,0,0,1,1): // ldrsh |
| 21786 | nm = "ldrsh"; ty = Ity_I16; syned = True; break; |
| 21787 | case BITS5(0,0,1,0,0): // str |
| 21788 | nm = "str"; ty = Ity_I32; isST = True; break; |
| 21789 | case BITS5(0,0,1,0,1): |
| 21790 | nm = "ldr"; ty = Ity_I32; break; // ldr |
| 21791 | default: |
| 21792 | valid = False; break; |
| 21793 | } |
| 21794 | |
| 21795 | UInt rN = INSN0(3,0); |
| 21796 | UInt rM = INSN1(3,0); |
| 21797 | UInt rT = INSN1(15,12); |
| 21798 | UInt imm2 = INSN1(5,4); |
| 21799 | Bool loadsPC = False; |
| 21800 | |
| 21801 | if (ty == Ity_I8 || ty == Ity_I16) { |
| 21802 | /* all 8- and 16-bit load and store cases have the |
| 21803 | same exclusion set. */ |
| 21804 | if (rN == 15 || isBadRegT(rT) || isBadRegT(rM)) |
| 21805 | valid = False; |
| 21806 | } else { |
| 21807 | vassert(ty == Ity_I32); |
| 21808 | if (rN == 15 || isBadRegT(rM)) |
| 21809 | valid = False; |
| 21810 | if (isST && rT == 15) |
| 21811 | valid = False; |
| 21812 | /* If it is a load and rT is 15, that's only allowable if we |
| 21813 | not in an IT block, or are the last in it. Need to insert |
| 21814 | a dynamic check for that. */ |
| 21815 | if (!isST && rT == 15) |
| 21816 | loadsPC = True; |
| 21817 | } |
| 21818 | |
| 21819 | if (valid) { |
| 21820 | // if it's a branch, it can't happen in the middle of an IT block |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21821 | // Also, if it is a branch, make it unconditional at this point. |
| 21822 | // Doing conditional branches in-line is too complex (for now) |
| 21823 | if (loadsPC) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21824 | gen_SIGILL_T_if_in_but_NLI_ITBlock(old_itstate, new_itstate); |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21825 | // go uncond |
| 21826 | mk_skip_over_T32_if_cond_is_false(condT); |
| 21827 | condT = IRTemp_INVALID; |
| 21828 | // now uncond |
| 21829 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21830 | |
| 21831 | IRTemp transAddr = newTemp(Ity_I32); |
| 21832 | assign(transAddr, |
| 21833 | binop( Iop_Add32, |
| 21834 | getIRegT(rN), |
| 21835 | binop(Iop_Shl32, getIRegT(rM), mkU8(imm2)) )); |
| 21836 | |
| 21837 | if (isST) { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21838 | |
| 21839 | /* get hold of the data to be stored */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21840 | IRTemp oldRt = newTemp(Ity_I32); |
| 21841 | assign(oldRt, getIRegT(rT)); |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21842 | |
| 21843 | /* generate the transfer */ |
| 21844 | IRExpr* data = NULL; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21845 | switch (ty) { |
| 21846 | case Ity_I8: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21847 | data = unop(Iop_32to8, mkexpr(oldRt)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21848 | break; |
| 21849 | case Ity_I16: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21850 | data = unop(Iop_32to16, mkexpr(oldRt)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21851 | break; |
| 21852 | case Ity_I32: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21853 | data = mkexpr(oldRt); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21854 | break; |
| 21855 | default: |
| 21856 | vassert(0); |
| 21857 | } |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21858 | storeGuardedLE(mkexpr(transAddr), data, condT); |
| 21859 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21860 | } else { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21861 | |
| 21862 | /* Load. */ |
| 21863 | IRTemp llOldRt = newTemp(Ity_I32); |
| 21864 | assign(llOldRt, llGetIReg(rT)); |
| 21865 | |
| 21866 | /* generate the transfer */ |
| 21867 | IRTemp newRt = newTemp(Ity_I32); |
| 21868 | IRLoadGOp widen = ILGop_INVALID; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21869 | switch (ty) { |
| 21870 | case Ity_I8: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21871 | widen = syned ? ILGop_8Sto32 : ILGop_8Uto32; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21872 | case Ity_I16: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21873 | widen = syned ? ILGop_16Sto32 : ILGop_16Uto32; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21874 | case Ity_I32: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21875 | widen = ILGop_Ident32; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21876 | default: |
| 21877 | vassert(0); |
| 21878 | } |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21879 | loadGuardedLE(newRt, widen, |
| 21880 | mkexpr(transAddr), mkexpr(llOldRt), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21881 | |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21882 | if (rT == 15) { |
| 21883 | vassert(loadsPC); |
| 21884 | /* We'll do the write to the PC just below */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21885 | } else { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21886 | vassert(!loadsPC); |
| 21887 | /* IRTemp_INVALID is OK here because in the case where |
| 21888 | condT is false at run time, we're just putting the |
| 21889 | old rT value back. */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21890 | putIRegT(rT, mkexpr(newRt), IRTemp_INVALID); |
| 21891 | } |
| 21892 | |
| 21893 | if (loadsPC) { |
| 21894 | /* Presumably this is an interworking branch. */ |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21895 | vassert(rN != 15); // assured by validity check above |
| 21896 | vassert(rT == 15); |
| 21897 | vassert(condT == IRTemp_INVALID); /* due to check above */ |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 21898 | llPutIReg(15, mkexpr(newRt)); |
| 21899 | dres.jk_StopHere = Ijk_Boring; /* or _Ret ? */ |
| 21900 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21901 | } |
| 21902 | } |
| 21903 | |
| 21904 | DIP("%s.w r%u, [r%u, r%u, LSL #%u]\n", |
| 21905 | nm, rT, rN, rM, imm2); |
| 21906 | |
| 21907 | goto decode_success; |
| 21908 | } |
| 21909 | } |
| 21910 | |
| 21911 | /* --------------- LD/ST reg+imm12 --------------- */ |
| 21912 | /* Loads and stores of the form: |
sewardj | f36d9af | 2014-05-13 14:44:21 +0000 | [diff] [blame] | 21913 | op Rt, [Rn, #+-imm12] |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21914 | where op is one of |
| 21915 | ldrb ldrh ldr ldrsb ldrsh |
| 21916 | strb strh str |
| 21917 | */ |
| 21918 | if (INSN0(15,9) == BITS7(1,1,1,1,1,0,0)) { |
| 21919 | Bool valid = True; |
sewardj | f36d9af | 2014-05-13 14:44:21 +0000 | [diff] [blame] | 21920 | Bool syned = INSN0(8,8) == 1; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21921 | Bool isST = False; |
| 21922 | IRType ty = Ity_I8; |
sewardj | f36d9af | 2014-05-13 14:44:21 +0000 | [diff] [blame] | 21923 | UInt bU = INSN0(7,7); // 1: +imm 0: -imm |
| 21924 | // -imm is only supported by literal versions |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 21925 | const HChar* nm = "???"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21926 | |
sewardj | f36d9af | 2014-05-13 14:44:21 +0000 | [diff] [blame] | 21927 | switch (INSN0(6,4)) { |
| 21928 | case BITS3(0,0,0): // strb |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21929 | nm = "strb"; isST = True; break; |
sewardj | f36d9af | 2014-05-13 14:44:21 +0000 | [diff] [blame] | 21930 | case BITS3(0,0,1): // ldrb |
| 21931 | nm = syned ? "ldrsb" : "ldrb"; break; |
| 21932 | case BITS3(0,1,0): // strh |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21933 | nm = "strh"; ty = Ity_I16; isST = True; break; |
sewardj | f36d9af | 2014-05-13 14:44:21 +0000 | [diff] [blame] | 21934 | case BITS3(0,1,1): // ldrh |
| 21935 | nm = syned ? "ldrsh" : "ldrh"; ty = Ity_I16; break; |
| 21936 | case BITS3(1,0,0): // str |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21937 | nm = "str"; ty = Ity_I32; isST = True; break; |
sewardj | f36d9af | 2014-05-13 14:44:21 +0000 | [diff] [blame] | 21938 | case BITS3(1,0,1): |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21939 | nm = "ldr"; ty = Ity_I32; break; // ldr |
| 21940 | default: |
| 21941 | valid = False; break; |
| 21942 | } |
| 21943 | |
| 21944 | UInt rN = INSN0(3,0); |
| 21945 | UInt rT = INSN1(15,12); |
| 21946 | UInt imm12 = INSN1(11,0); |
| 21947 | Bool loadsPC = False; |
| 21948 | |
sewardj | f36d9af | 2014-05-13 14:44:21 +0000 | [diff] [blame] | 21949 | if (rN != 15 && bU == 0) { |
| 21950 | // only pc supports #-imm12 |
| 21951 | valid = False; |
| 21952 | } |
| 21953 | |
| 21954 | if (isST) { |
| 21955 | if (syned) valid = False; |
| 21956 | if (rN == 15 || rT == 15) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21957 | valid = False; |
| 21958 | } else { |
sewardj | f36d9af | 2014-05-13 14:44:21 +0000 | [diff] [blame] | 21959 | /* For a 32-bit load, rT == 15 is only allowable if we are not |
| 21960 | in an IT block, or are the last in it. Need to insert |
| 21961 | a dynamic check for that. Also, in this particular |
| 21962 | case, rN == 15 is allowable. In this case however, the |
| 21963 | value obtained for rN is (apparently) |
| 21964 | "word-align(address of current insn + 4)". */ |
| 21965 | if (rT == 15) { |
| 21966 | if (ty == Ity_I32) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21967 | loadsPC = True; |
sewardj | f36d9af | 2014-05-13 14:44:21 +0000 | [diff] [blame] | 21968 | else // Can't do it for B/H loads |
| 21969 | valid = False; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21970 | } |
| 21971 | } |
| 21972 | |
| 21973 | if (valid) { |
| 21974 | // if it's a branch, it can't happen in the middle of an IT block |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21975 | // Also, if it is a branch, make it unconditional at this point. |
| 21976 | // Doing conditional branches in-line is too complex (for now) |
| 21977 | if (loadsPC) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21978 | gen_SIGILL_T_if_in_but_NLI_ITBlock(old_itstate, new_itstate); |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21979 | // go uncond |
| 21980 | mk_skip_over_T32_if_cond_is_false(condT); |
| 21981 | condT = IRTemp_INVALID; |
| 21982 | // now uncond |
| 21983 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21984 | |
| 21985 | IRTemp rNt = newTemp(Ity_I32); |
| 21986 | if (rN == 15) { |
sewardj | f36d9af | 2014-05-13 14:44:21 +0000 | [diff] [blame] | 21987 | vassert(!isST); |
| 21988 | assign(rNt, binop(Iop_And32, getIRegT(15), mkU32(~3))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21989 | } else { |
| 21990 | assign(rNt, getIRegT(rN)); |
| 21991 | } |
| 21992 | |
| 21993 | IRTemp transAddr = newTemp(Ity_I32); |
| 21994 | assign(transAddr, |
sewardj | f36d9af | 2014-05-13 14:44:21 +0000 | [diff] [blame] | 21995 | binop(bU == 1 ? Iop_Add32 : Iop_Sub32, |
| 21996 | mkexpr(rNt), mkU32(imm12))); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 21997 | |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 21998 | IRTemp oldRt = newTemp(Ity_I32); |
| 21999 | assign(oldRt, getIRegT(rT)); |
| 22000 | |
| 22001 | IRTemp llOldRt = newTemp(Ity_I32); |
| 22002 | assign(llOldRt, llGetIReg(rT)); |
| 22003 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22004 | if (isST) { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22005 | IRExpr* data = NULL; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22006 | switch (ty) { |
| 22007 | case Ity_I8: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22008 | data = unop(Iop_32to8, mkexpr(oldRt)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22009 | break; |
| 22010 | case Ity_I16: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22011 | data = unop(Iop_32to16, mkexpr(oldRt)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22012 | break; |
| 22013 | case Ity_I32: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22014 | data = mkexpr(oldRt); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22015 | break; |
| 22016 | default: |
| 22017 | vassert(0); |
| 22018 | } |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22019 | storeGuardedLE(mkexpr(transAddr), data, condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22020 | } else { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22021 | IRTemp newRt = newTemp(Ity_I32); |
| 22022 | IRLoadGOp widen = ILGop_INVALID; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22023 | switch (ty) { |
| 22024 | case Ity_I8: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22025 | widen = syned ? ILGop_8Sto32 : ILGop_8Uto32; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22026 | case Ity_I16: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22027 | widen = syned ? ILGop_16Sto32 : ILGop_16Uto32; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22028 | case Ity_I32: |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22029 | widen = ILGop_Ident32; break; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22030 | default: |
| 22031 | vassert(0); |
| 22032 | } |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22033 | loadGuardedLE(newRt, widen, |
| 22034 | mkexpr(transAddr), mkexpr(llOldRt), condT); |
| 22035 | if (rT == 15) { |
| 22036 | vassert(loadsPC); |
| 22037 | /* We'll do the write to the PC just below */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22038 | } else { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22039 | vassert(!loadsPC); |
| 22040 | /* IRTemp_INVALID is OK here because in the case where |
| 22041 | condT is false at run time, we're just putting the |
| 22042 | old rT value back. */ |
| 22043 | putIRegT(rT, mkexpr(newRt), IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22044 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22045 | |
| 22046 | if (loadsPC) { |
| 22047 | /* Presumably this is an interworking branch. */ |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22048 | vassert(rT == 15); |
| 22049 | vassert(condT == IRTemp_INVALID); /* due to check above */ |
| 22050 | llPutIReg(15, mkexpr(newRt)); |
sewardj | bd11c71 | 2014-05-13 15:54:14 +0000 | [diff] [blame] | 22051 | dres.jk_StopHere = Ijk_Boring; |
| 22052 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22053 | } |
| 22054 | } |
| 22055 | |
| 22056 | DIP("%s.w r%u, [r%u, +#%u]\n", nm, rT, rN, imm12); |
| 22057 | |
| 22058 | goto decode_success; |
| 22059 | } |
| 22060 | } |
| 22061 | |
| 22062 | /* -------------- LDRD/STRD reg+/-#imm8 -------------- */ |
| 22063 | /* Doubleword loads and stores of the form: |
sewardj | 055e93a | 2014-03-15 08:33:06 +0000 | [diff] [blame] | 22064 | ldrd/strd Rt, Rt2, [Rn, #+/-imm8] or |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22065 | ldrd/strd Rt, Rt2, [Rn], #+/-imm8 or |
| 22066 | ldrd/strd Rt, Rt2, [Rn, #+/-imm8]! |
| 22067 | */ |
| 22068 | if (INSN0(15,9) == BITS7(1,1,1,0,1,0,0) && INSN0(6,6) == 1) { |
| 22069 | UInt bP = INSN0(8,8); |
| 22070 | UInt bU = INSN0(7,7); |
| 22071 | UInt bW = INSN0(5,5); |
| 22072 | UInt bL = INSN0(4,4); // 1: load 0: store |
| 22073 | UInt rN = INSN0(3,0); |
| 22074 | UInt rT = INSN1(15,12); |
| 22075 | UInt rT2 = INSN1(11,8); |
| 22076 | UInt imm8 = INSN1(7,0); |
| 22077 | |
| 22078 | Bool valid = True; |
| 22079 | if (bP == 0 && bW == 0) valid = False; |
| 22080 | if (bW == 1 && (rN == rT || rN == rT2)) valid = False; |
| 22081 | if (isBadRegT(rT) || isBadRegT(rT2)) valid = False; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22082 | if (bL == 1 && rT == rT2) valid = False; |
sewardj | 055e93a | 2014-03-15 08:33:06 +0000 | [diff] [blame] | 22083 | /* It's OK to use PC as the base register only in the |
| 22084 | following case: ldrd Rt, Rt2, [PC, #+/-imm8] */ |
| 22085 | if (rN == 15 && (bL == 0/*store*/ |
| 22086 | || bW == 1/*wb*/)) valid = False; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22087 | |
| 22088 | if (valid) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22089 | IRTemp preAddr = newTemp(Ity_I32); |
sewardj | 1cf79f7 | 2014-04-01 11:00:36 +0000 | [diff] [blame] | 22090 | assign(preAddr, 15 == rN |
| 22091 | ? binop(Iop_And32, getIRegT(15), mkU32(~3U)) |
| 22092 | : getIRegT(rN)); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22093 | |
| 22094 | IRTemp postAddr = newTemp(Ity_I32); |
| 22095 | assign(postAddr, binop(bU == 1 ? Iop_Add32 : Iop_Sub32, |
| 22096 | mkexpr(preAddr), mkU32(imm8 << 2))); |
| 22097 | |
| 22098 | IRTemp transAddr = bP == 1 ? postAddr : preAddr; |
| 22099 | |
sewardj | 92001d4 | 2013-04-26 10:24:31 +0000 | [diff] [blame] | 22100 | /* For almost all cases, we do the writeback after the transfers. |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 22101 | However, that leaves the stack "uncovered" in cases like: |
sewardj | 92001d4 | 2013-04-26 10:24:31 +0000 | [diff] [blame] | 22102 | strd rD, [sp, #-8] |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 22103 | strd rD, [sp, #-16] |
sewardj | 92001d4 | 2013-04-26 10:24:31 +0000 | [diff] [blame] | 22104 | In which case, do the writeback to SP now, instead of later. |
| 22105 | This is bad in that it makes the insn non-restartable if the |
| 22106 | accesses fault, but at least keeps Memcheck happy. */ |
| 22107 | Bool writeback_already_done = False; |
| 22108 | if (bL == 0/*store*/ && bW == 1/*wb*/ |
| 22109 | && rN == 13 && rN != rT && rN != rT2 |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 22110 | && bU == 0/*minus*/ |
| 22111 | && ((imm8 << 2) == 8 || (imm8 << 2) == 16)) { |
sewardj | 92001d4 | 2013-04-26 10:24:31 +0000 | [diff] [blame] | 22112 | putIRegT(rN, mkexpr(postAddr), condT); |
| 22113 | writeback_already_done = True; |
| 22114 | } |
| 22115 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22116 | if (bL == 0) { |
| 22117 | IRTemp oldRt = newTemp(Ity_I32); |
| 22118 | IRTemp oldRt2 = newTemp(Ity_I32); |
| 22119 | assign(oldRt, getIRegT(rT)); |
| 22120 | assign(oldRt2, getIRegT(rT2)); |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22121 | storeGuardedLE( mkexpr(transAddr), |
| 22122 | mkexpr(oldRt), condT ); |
| 22123 | storeGuardedLE( binop(Iop_Add32, mkexpr(transAddr), mkU32(4)), |
| 22124 | mkexpr(oldRt2), condT ); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22125 | } else { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22126 | IRTemp oldRt = newTemp(Ity_I32); |
| 22127 | IRTemp oldRt2 = newTemp(Ity_I32); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22128 | IRTemp newRt = newTemp(Ity_I32); |
| 22129 | IRTemp newRt2 = newTemp(Ity_I32); |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22130 | assign(oldRt, llGetIReg(rT)); |
| 22131 | assign(oldRt2, llGetIReg(rT2)); |
| 22132 | loadGuardedLE( newRt, ILGop_Ident32, |
| 22133 | mkexpr(transAddr), |
| 22134 | mkexpr(oldRt), condT ); |
| 22135 | loadGuardedLE( newRt2, ILGop_Ident32, |
| 22136 | binop(Iop_Add32, mkexpr(transAddr), mkU32(4)), |
| 22137 | mkexpr(oldRt2), condT ); |
| 22138 | /* Put unconditionally, since we already switched on the condT |
| 22139 | in the guarded loads. */ |
| 22140 | putIRegT(rT, mkexpr(newRt), IRTemp_INVALID); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22141 | putIRegT(rT2, mkexpr(newRt2), IRTemp_INVALID); |
| 22142 | } |
| 22143 | |
sewardj | 92001d4 | 2013-04-26 10:24:31 +0000 | [diff] [blame] | 22144 | if (bW == 1 && !writeback_already_done) { |
sewardj | cfe046e | 2013-01-17 14:23:53 +0000 | [diff] [blame] | 22145 | putIRegT(rN, mkexpr(postAddr), condT); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22146 | } |
| 22147 | |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 22148 | const HChar* nm = bL ? "ldrd" : "strd"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22149 | |
| 22150 | if (bP == 1 && bW == 0) { |
| 22151 | DIP("%s.w r%u, r%u, [r%u, #%c%u]\n", |
| 22152 | nm, rT, rT2, rN, bU ? '+' : '-', imm8 << 2); |
| 22153 | } |
| 22154 | else if (bP == 1 && bW == 1) { |
| 22155 | DIP("%s.w r%u, r%u, [r%u, #%c%u]!\n", |
| 22156 | nm, rT, rT2, rN, bU ? '+' : '-', imm8 << 2); |
| 22157 | } |
| 22158 | else { |
| 22159 | vassert(bP == 0 && bW == 1); |
| 22160 | DIP("%s.w r%u, r%u, [r%u], #%c%u\n", |
| 22161 | nm, rT, rT2, rN, bU ? '+' : '-', imm8 << 2); |
| 22162 | } |
| 22163 | |
| 22164 | goto decode_success; |
| 22165 | } |
| 22166 | } |
| 22167 | |
| 22168 | /* -------------- (T3) Bcond.W label -------------- */ |
| 22169 | /* This variant carries its own condition, so can't be part of an |
| 22170 | IT block ... */ |
| 22171 | if (INSN0(15,11) == BITS5(1,1,1,1,0) |
| 22172 | && INSN1(15,14) == BITS2(1,0) |
| 22173 | && INSN1(12,12) == 0) { |
| 22174 | UInt cond = INSN0(9,6); |
| 22175 | if (cond != ARMCondAL && cond != ARMCondNV) { |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 22176 | UInt uimm21 |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22177 | = (INSN0(10,10) << (1 + 1 + 6 + 11 + 1)) |
| 22178 | | (INSN1(11,11) << (1 + 6 + 11 + 1)) |
| 22179 | | (INSN1(13,13) << (6 + 11 + 1)) |
| 22180 | | (INSN0(5,0) << (11 + 1)) |
| 22181 | | (INSN1(10,0) << 1); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 22182 | uimm21 <<= 11; |
| 22183 | Int simm21 = (Int)uimm21; |
| 22184 | simm21 >>= 11; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22185 | |
| 22186 | vassert(0 == (guest_R15_curr_instr_notENC & 1)); |
| 22187 | UInt dst = simm21 + guest_R15_curr_instr_notENC + 4; |
| 22188 | |
| 22189 | /* Not allowed in an IT block; SIGILL if so. */ |
| 22190 | gen_SIGILL_T_if_in_ITBlock(old_itstate, new_itstate); |
| 22191 | |
| 22192 | IRTemp kondT = newTemp(Ity_I32); |
| 22193 | assign( kondT, mk_armg_calculate_condition(cond) ); |
| 22194 | stmt( IRStmt_Exit( unop(Iop_32to1, mkexpr(kondT)), |
| 22195 | Ijk_Boring, |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 22196 | IRConst_U32(dst | 1/*CPSR.T*/), |
| 22197 | OFFB_R15T )); |
| 22198 | llPutIReg(15, mkU32( (guest_R15_curr_instr_notENC + 4) |
| 22199 | | 1 /*CPSR.T*/ )); |
| 22200 | dres.jk_StopHere = Ijk_Boring; |
| 22201 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22202 | DIP("b%s.w 0x%x\n", nCC(cond), dst); |
| 22203 | goto decode_success; |
| 22204 | } |
| 22205 | } |
| 22206 | |
| 22207 | /* ---------------- (T4) B.W label ---------------- */ |
| 22208 | /* ... whereas this variant doesn't carry its own condition, so it |
| 22209 | has to be either unconditional or the conditional by virtue of |
| 22210 | being the last in an IT block. The upside is that there's 4 |
| 22211 | more bits available for the jump offset, so it has a 16-times |
| 22212 | greater branch range than the T3 variant. */ |
| 22213 | if (INSN0(15,11) == BITS5(1,1,1,1,0) |
| 22214 | && INSN1(15,14) == BITS2(1,0) |
| 22215 | && INSN1(12,12) == 1) { |
| 22216 | if (1) { |
| 22217 | UInt bS = INSN0(10,10); |
| 22218 | UInt bJ1 = INSN1(13,13); |
| 22219 | UInt bJ2 = INSN1(11,11); |
| 22220 | UInt bI1 = 1 ^ (bJ1 ^ bS); |
| 22221 | UInt bI2 = 1 ^ (bJ2 ^ bS); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 22222 | UInt uimm25 |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22223 | = (bS << (1 + 1 + 10 + 11 + 1)) |
| 22224 | | (bI1 << (1 + 10 + 11 + 1)) |
| 22225 | | (bI2 << (10 + 11 + 1)) |
| 22226 | | (INSN0(9,0) << (11 + 1)) |
| 22227 | | (INSN1(10,0) << 1); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 22228 | uimm25 <<= 7; |
| 22229 | Int simm25 = (Int)uimm25; |
| 22230 | simm25 >>= 7; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22231 | |
| 22232 | vassert(0 == (guest_R15_curr_instr_notENC & 1)); |
| 22233 | UInt dst = simm25 + guest_R15_curr_instr_notENC + 4; |
| 22234 | |
| 22235 | /* If in an IT block, must be the last insn. */ |
| 22236 | gen_SIGILL_T_if_in_but_NLI_ITBlock(old_itstate, new_itstate); |
| 22237 | |
| 22238 | // go uncond |
| 22239 | mk_skip_over_T32_if_cond_is_false(condT); |
| 22240 | condT = IRTemp_INVALID; |
| 22241 | // now uncond |
| 22242 | |
| 22243 | // branch to dst |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 22244 | llPutIReg(15, mkU32( dst | 1 /*CPSR.T*/ )); |
| 22245 | dres.jk_StopHere = Ijk_Boring; |
| 22246 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22247 | DIP("b.w 0x%x\n", dst); |
| 22248 | goto decode_success; |
| 22249 | } |
| 22250 | } |
| 22251 | |
| 22252 | /* ------------------ TBB, TBH ------------------ */ |
| 22253 | if (INSN0(15,4) == 0xE8D && INSN1(15,5) == 0x780) { |
| 22254 | UInt rN = INSN0(3,0); |
| 22255 | UInt rM = INSN1(3,0); |
| 22256 | UInt bH = INSN1(4,4); |
| 22257 | if (bH/*ATC*/ || (rN != 13 && !isBadRegT(rM))) { |
| 22258 | /* Must be last or not-in IT block */ |
| 22259 | gen_SIGILL_T_if_in_but_NLI_ITBlock(old_itstate, new_itstate); |
| 22260 | /* Go uncond */ |
| 22261 | mk_skip_over_T32_if_cond_is_false(condT); |
| 22262 | condT = IRTemp_INVALID; |
| 22263 | |
| 22264 | IRExpr* ea |
| 22265 | = binop(Iop_Add32, |
| 22266 | getIRegT(rN), |
| 22267 | bH ? binop(Iop_Shl32, getIRegT(rM), mkU8(1)) |
| 22268 | : getIRegT(rM)); |
| 22269 | |
| 22270 | IRTemp delta = newTemp(Ity_I32); |
| 22271 | if (bH) { |
| 22272 | assign(delta, unop(Iop_16Uto32, loadLE(Ity_I16, ea))); |
| 22273 | } else { |
| 22274 | assign(delta, unop(Iop_8Uto32, loadLE(Ity_I8, ea))); |
| 22275 | } |
| 22276 | |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 22277 | llPutIReg( |
| 22278 | 15, |
| 22279 | binop(Iop_Or32, |
| 22280 | binop(Iop_Add32, |
| 22281 | getIRegT(15), |
| 22282 | binop(Iop_Shl32, mkexpr(delta), mkU8(1)) |
| 22283 | ), |
| 22284 | mkU32(1) |
| 22285 | )); |
| 22286 | dres.jk_StopHere = Ijk_Boring; |
| 22287 | dres.whatNext = Dis_StopHere; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22288 | DIP("tb%c [r%u, r%u%s]\n", |
| 22289 | bH ? 'h' : 'b', rN, rM, bH ? ", LSL #1" : ""); |
| 22290 | goto decode_success; |
| 22291 | } |
| 22292 | } |
| 22293 | |
| 22294 | /* ------------------ UBFX ------------------ */ |
| 22295 | /* ------------------ SBFX ------------------ */ |
| 22296 | /* There's also ARM versions of same, but it doesn't seem worth the |
| 22297 | hassle to common up the handling (it's only a couple of C |
| 22298 | statements). */ |
| 22299 | if ((INSN0(15,4) == 0xF3C // UBFX |
| 22300 | || INSN0(15,4) == 0xF34) // SBFX |
| 22301 | && INSN1(15,15) == 0 && INSN1(5,5) == 0) { |
| 22302 | UInt rN = INSN0(3,0); |
| 22303 | UInt rD = INSN1(11,8); |
| 22304 | UInt lsb = (INSN1(14,12) << 2) | INSN1(7,6); |
| 22305 | UInt wm1 = INSN1(4,0); |
| 22306 | UInt msb = lsb + wm1; |
| 22307 | if (!isBadRegT(rD) && !isBadRegT(rN) && msb <= 31) { |
| 22308 | Bool isU = INSN0(15,4) == 0xF3C; |
| 22309 | IRTemp src = newTemp(Ity_I32); |
| 22310 | IRTemp tmp = newTemp(Ity_I32); |
| 22311 | IRTemp res = newTemp(Ity_I32); |
| 22312 | UInt mask = ((1 << wm1) - 1) + (1 << wm1); |
| 22313 | vassert(msb >= 0 && msb <= 31); |
| 22314 | vassert(mask != 0); // guaranteed by msb being in 0 .. 31 inclusive |
| 22315 | |
| 22316 | assign(src, getIRegT(rN)); |
| 22317 | assign(tmp, binop(Iop_And32, |
| 22318 | binop(Iop_Shr32, mkexpr(src), mkU8(lsb)), |
| 22319 | mkU32(mask))); |
| 22320 | assign(res, binop(isU ? Iop_Shr32 : Iop_Sar32, |
| 22321 | binop(Iop_Shl32, mkexpr(tmp), mkU8(31-wm1)), |
| 22322 | mkU8(31-wm1))); |
| 22323 | |
| 22324 | putIRegT(rD, mkexpr(res), condT); |
| 22325 | |
| 22326 | DIP("%s r%u, r%u, #%u, #%u\n", |
| 22327 | isU ? "ubfx" : "sbfx", rD, rN, lsb, wm1 + 1); |
| 22328 | goto decode_success; |
| 22329 | } |
| 22330 | } |
| 22331 | |
| 22332 | /* ------------------ UXTB ------------------ */ |
| 22333 | /* ------------------ UXTH ------------------ */ |
| 22334 | /* ------------------ SXTB ------------------ */ |
| 22335 | /* ------------------ SXTH ------------------ */ |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 22336 | /* ----------------- UXTB16 ----------------- */ |
| 22337 | /* ----------------- SXTB16 ----------------- */ |
| 22338 | /* FIXME: this is an exact duplicate of the ARM version. They |
| 22339 | should be commoned up. */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22340 | if ((INSN0(15,0) == 0xFA5F // UXTB |
| 22341 | || INSN0(15,0) == 0xFA1F // UXTH |
| 22342 | || INSN0(15,0) == 0xFA4F // SXTB |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 22343 | || INSN0(15,0) == 0xFA0F // SXTH |
| 22344 | || INSN0(15,0) == 0xFA3F // UXTB16 |
| 22345 | || INSN0(15,0) == 0xFA2F) // SXTB16 |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22346 | && INSN1(15,12) == BITS4(1,1,1,1) |
| 22347 | && INSN1(7,6) == BITS2(1,0)) { |
| 22348 | UInt rD = INSN1(11,8); |
| 22349 | UInt rM = INSN1(3,0); |
| 22350 | UInt rot = INSN1(5,4); |
| 22351 | if (!isBadRegT(rD) && !isBadRegT(rM)) { |
florian | 55085f8 | 2012-11-21 00:36:55 +0000 | [diff] [blame] | 22352 | const HChar* nm = "???"; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22353 | IRTemp srcT = newTemp(Ity_I32); |
| 22354 | IRTemp rotT = newTemp(Ity_I32); |
| 22355 | IRTemp dstT = newTemp(Ity_I32); |
| 22356 | assign(srcT, getIRegT(rM)); |
| 22357 | assign(rotT, genROR32(srcT, 8 * rot)); |
| 22358 | switch (INSN0(15,0)) { |
| 22359 | case 0xFA5F: // UXTB |
| 22360 | nm = "uxtb"; |
| 22361 | assign(dstT, unop(Iop_8Uto32, |
| 22362 | unop(Iop_32to8, mkexpr(rotT)))); |
| 22363 | break; |
| 22364 | case 0xFA1F: // UXTH |
| 22365 | nm = "uxth"; |
| 22366 | assign(dstT, unop(Iop_16Uto32, |
| 22367 | unop(Iop_32to16, mkexpr(rotT)))); |
| 22368 | break; |
| 22369 | case 0xFA4F: // SXTB |
| 22370 | nm = "sxtb"; |
| 22371 | assign(dstT, unop(Iop_8Sto32, |
| 22372 | unop(Iop_32to8, mkexpr(rotT)))); |
| 22373 | break; |
| 22374 | case 0xFA0F: // SXTH |
| 22375 | nm = "sxth"; |
| 22376 | assign(dstT, unop(Iop_16Sto32, |
| 22377 | unop(Iop_32to16, mkexpr(rotT)))); |
| 22378 | break; |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 22379 | case 0xFA3F: // UXTB16 |
| 22380 | nm = "uxtb16"; |
| 22381 | assign(dstT, binop(Iop_And32, mkexpr(rotT), |
| 22382 | mkU32(0x00FF00FF))); |
| 22383 | break; |
| 22384 | case 0xFA2F: { // SXTB16 |
| 22385 | nm = "sxtb16"; |
| 22386 | IRTemp lo32 = newTemp(Ity_I32); |
| 22387 | IRTemp hi32 = newTemp(Ity_I32); |
| 22388 | assign(lo32, binop(Iop_And32, mkexpr(rotT), mkU32(0xFF))); |
| 22389 | assign(hi32, binop(Iop_Shr32, mkexpr(rotT), mkU8(16))); |
| 22390 | assign( |
| 22391 | dstT, |
| 22392 | binop(Iop_Or32, |
| 22393 | binop(Iop_And32, |
| 22394 | unop(Iop_8Sto32, |
| 22395 | unop(Iop_32to8, mkexpr(lo32))), |
| 22396 | mkU32(0xFFFF)), |
| 22397 | binop(Iop_Shl32, |
| 22398 | unop(Iop_8Sto32, |
| 22399 | unop(Iop_32to8, mkexpr(hi32))), |
| 22400 | mkU8(16)) |
| 22401 | )); |
| 22402 | break; |
| 22403 | } |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22404 | default: |
| 22405 | vassert(0); |
| 22406 | } |
| 22407 | putIRegT(rD, mkexpr(dstT), condT); |
| 22408 | DIP("%s r%u, r%u, ror #%u\n", nm, rD, rM, 8 * rot); |
| 22409 | goto decode_success; |
| 22410 | } |
| 22411 | } |
| 22412 | |
| 22413 | /* -------------- MUL.W Rd, Rn, Rm -------------- */ |
| 22414 | if (INSN0(15,4) == 0xFB0 |
| 22415 | && (INSN1(15,0) & 0xF0F0) == 0xF000) { |
| 22416 | UInt rN = INSN0(3,0); |
| 22417 | UInt rD = INSN1(11,8); |
| 22418 | UInt rM = INSN1(3,0); |
| 22419 | if (!isBadRegT(rD) && !isBadRegT(rN) && !isBadRegT(rM)) { |
| 22420 | IRTemp res = newTemp(Ity_I32); |
| 22421 | assign(res, binop(Iop_Mul32, getIRegT(rN), getIRegT(rM))); |
| 22422 | putIRegT(rD, mkexpr(res), condT); |
| 22423 | DIP("mul.w r%u, r%u, r%u\n", rD, rN, rM); |
| 22424 | goto decode_success; |
| 22425 | } |
| 22426 | } |
| 22427 | |
sewardj | 8bde7f1 | 2013-04-11 13:57:43 +0000 | [diff] [blame] | 22428 | /* -------------- SDIV.W Rd, Rn, Rm -------------- */ |
| 22429 | if (INSN0(15,4) == 0xFB9 |
| 22430 | && (INSN1(15,0) & 0xF0F0) == 0xF0F0) { |
| 22431 | UInt rN = INSN0(3,0); |
| 22432 | UInt rD = INSN1(11,8); |
| 22433 | UInt rM = INSN1(3,0); |
| 22434 | if (!isBadRegT(rD) && !isBadRegT(rN) && !isBadRegT(rM)) { |
| 22435 | IRTemp res = newTemp(Ity_I32); |
| 22436 | IRTemp argL = newTemp(Ity_I32); |
| 22437 | IRTemp argR = newTemp(Ity_I32); |
| 22438 | assign(argL, getIRegT(rN)); |
| 22439 | assign(argR, getIRegT(rM)); |
| 22440 | assign(res, binop(Iop_DivS32, mkexpr(argL), mkexpr(argR))); |
| 22441 | putIRegT(rD, mkexpr(res), condT); |
| 22442 | DIP("sdiv.w r%u, r%u, r%u\n", rD, rN, rM); |
| 22443 | goto decode_success; |
| 22444 | } |
| 22445 | } |
| 22446 | |
| 22447 | /* -------------- UDIV.W Rd, Rn, Rm -------------- */ |
| 22448 | if (INSN0(15,4) == 0xFBB |
| 22449 | && (INSN1(15,0) & 0xF0F0) == 0xF0F0) { |
| 22450 | UInt rN = INSN0(3,0); |
| 22451 | UInt rD = INSN1(11,8); |
| 22452 | UInt rM = INSN1(3,0); |
| 22453 | if (!isBadRegT(rD) && !isBadRegT(rN) && !isBadRegT(rM)) { |
| 22454 | IRTemp res = newTemp(Ity_I32); |
| 22455 | IRTemp argL = newTemp(Ity_I32); |
| 22456 | IRTemp argR = newTemp(Ity_I32); |
| 22457 | assign(argL, getIRegT(rN)); |
| 22458 | assign(argR, getIRegT(rM)); |
| 22459 | assign(res, binop(Iop_DivU32, mkexpr(argL), mkexpr(argR))); |
| 22460 | putIRegT(rD, mkexpr(res), condT); |
| 22461 | DIP("udiv.w r%u, r%u, r%u\n", rD, rN, rM); |
| 22462 | goto decode_success; |
| 22463 | } |
| 22464 | } |
| 22465 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22466 | /* ------------------ {U,S}MULL ------------------ */ |
| 22467 | if ((INSN0(15,4) == 0xFB8 || INSN0(15,4) == 0xFBA) |
| 22468 | && INSN1(7,4) == BITS4(0,0,0,0)) { |
| 22469 | UInt isU = INSN0(5,5); |
| 22470 | UInt rN = INSN0(3,0); |
| 22471 | UInt rDlo = INSN1(15,12); |
| 22472 | UInt rDhi = INSN1(11,8); |
| 22473 | UInt rM = INSN1(3,0); |
| 22474 | if (!isBadRegT(rDhi) && !isBadRegT(rDlo) |
| 22475 | && !isBadRegT(rN) && !isBadRegT(rM) && rDlo != rDhi) { |
| 22476 | IRTemp res = newTemp(Ity_I64); |
| 22477 | assign(res, binop(isU ? Iop_MullU32 : Iop_MullS32, |
| 22478 | getIRegT(rN), getIRegT(rM))); |
| 22479 | putIRegT( rDhi, unop(Iop_64HIto32, mkexpr(res)), condT ); |
| 22480 | putIRegT( rDlo, unop(Iop_64to32, mkexpr(res)), condT ); |
| 22481 | DIP("%cmull r%u, r%u, r%u, r%u\n", |
| 22482 | isU ? 'u' : 's', rDlo, rDhi, rN, rM); |
| 22483 | goto decode_success; |
| 22484 | } |
| 22485 | } |
| 22486 | |
| 22487 | /* ------------------ ML{A,S} ------------------ */ |
| 22488 | if (INSN0(15,4) == 0xFB0 |
| 22489 | && ( INSN1(7,4) == BITS4(0,0,0,0) // MLA |
| 22490 | || INSN1(7,4) == BITS4(0,0,0,1))) { // MLS |
| 22491 | UInt rN = INSN0(3,0); |
| 22492 | UInt rA = INSN1(15,12); |
| 22493 | UInt rD = INSN1(11,8); |
| 22494 | UInt rM = INSN1(3,0); |
| 22495 | if (!isBadRegT(rD) && !isBadRegT(rN) |
| 22496 | && !isBadRegT(rM) && !isBadRegT(rA)) { |
| 22497 | Bool isMLA = INSN1(7,4) == BITS4(0,0,0,0); |
| 22498 | IRTemp res = newTemp(Ity_I32); |
| 22499 | assign(res, |
| 22500 | binop(isMLA ? Iop_Add32 : Iop_Sub32, |
| 22501 | getIRegT(rA), |
| 22502 | binop(Iop_Mul32, getIRegT(rN), getIRegT(rM)))); |
| 22503 | putIRegT(rD, mkexpr(res), condT); |
| 22504 | DIP("%s r%u, r%u, r%u, r%u\n", |
| 22505 | isMLA ? "mla" : "mls", rD, rN, rM, rA); |
| 22506 | goto decode_success; |
| 22507 | } |
| 22508 | } |
| 22509 | |
| 22510 | /* ------------------ (T3) ADR ------------------ */ |
| 22511 | if ((INSN0(15,0) == 0xF20F || INSN0(15,0) == 0xF60F) |
| 22512 | && INSN1(15,15) == 0) { |
| 22513 | /* rD = align4(PC) + imm32 */ |
| 22514 | UInt rD = INSN1(11,8); |
| 22515 | if (!isBadRegT(rD)) { |
| 22516 | UInt imm32 = (INSN0(10,10) << 11) |
| 22517 | | (INSN1(14,12) << 8) | INSN1(7,0); |
| 22518 | putIRegT(rD, binop(Iop_Add32, |
sewardj | df86f16 | 2010-08-22 18:47:30 +0000 | [diff] [blame] | 22519 | binop(Iop_And32, getIRegT(15), mkU32(~3U)), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22520 | mkU32(imm32)), |
| 22521 | condT); |
| 22522 | DIP("add r%u, pc, #%u\n", rD, imm32); |
| 22523 | goto decode_success; |
| 22524 | } |
| 22525 | } |
| 22526 | |
| 22527 | /* ----------------- (T1) UMLAL ----------------- */ |
| 22528 | /* ----------------- (T1) SMLAL ----------------- */ |
| 22529 | if ((INSN0(15,4) == 0xFBE // UMLAL |
| 22530 | || INSN0(15,4) == 0xFBC) // SMLAL |
| 22531 | && INSN1(7,4) == BITS4(0,0,0,0)) { |
| 22532 | UInt rN = INSN0(3,0); |
| 22533 | UInt rDlo = INSN1(15,12); |
| 22534 | UInt rDhi = INSN1(11,8); |
| 22535 | UInt rM = INSN1(3,0); |
| 22536 | if (!isBadRegT(rDlo) && !isBadRegT(rDhi) && !isBadRegT(rN) |
| 22537 | && !isBadRegT(rM) && rDhi != rDlo) { |
| 22538 | Bool isS = INSN0(15,4) == 0xFBC; |
| 22539 | IRTemp argL = newTemp(Ity_I32); |
| 22540 | IRTemp argR = newTemp(Ity_I32); |
| 22541 | IRTemp old = newTemp(Ity_I64); |
| 22542 | IRTemp res = newTemp(Ity_I64); |
| 22543 | IRTemp resHi = newTemp(Ity_I32); |
| 22544 | IRTemp resLo = newTemp(Ity_I32); |
| 22545 | IROp mulOp = isS ? Iop_MullS32 : Iop_MullU32; |
| 22546 | assign( argL, getIRegT(rM)); |
| 22547 | assign( argR, getIRegT(rN)); |
| 22548 | assign( old, binop(Iop_32HLto64, getIRegT(rDhi), getIRegT(rDlo)) ); |
| 22549 | assign( res, binop(Iop_Add64, |
| 22550 | mkexpr(old), |
| 22551 | binop(mulOp, mkexpr(argL), mkexpr(argR))) ); |
| 22552 | assign( resHi, unop(Iop_64HIto32, mkexpr(res)) ); |
| 22553 | assign( resLo, unop(Iop_64to32, mkexpr(res)) ); |
| 22554 | putIRegT( rDhi, mkexpr(resHi), condT ); |
| 22555 | putIRegT( rDlo, mkexpr(resLo), condT ); |
| 22556 | DIP("%cmlal r%u, r%u, r%u, r%u\n", |
| 22557 | isS ? 's' : 'u', rDlo, rDhi, rN, rM); |
| 22558 | goto decode_success; |
| 22559 | } |
| 22560 | } |
| 22561 | |
sewardj | abf3945 | 2012-12-12 00:16:41 +0000 | [diff] [blame] | 22562 | /* ------------------ (T1) UMAAL ------------------ */ |
| 22563 | if (INSN0(15,4) == 0xFBE && INSN1(7,4) == BITS4(0,1,1,0)) { |
| 22564 | UInt rN = INSN0(3,0); |
| 22565 | UInt rDlo = INSN1(15,12); |
| 22566 | UInt rDhi = INSN1(11,8); |
| 22567 | UInt rM = INSN1(3,0); |
| 22568 | if (!isBadRegT(rDlo) && !isBadRegT(rDhi) && !isBadRegT(rN) |
| 22569 | && !isBadRegT(rM) && rDhi != rDlo) { |
| 22570 | IRTemp argN = newTemp(Ity_I32); |
| 22571 | IRTemp argM = newTemp(Ity_I32); |
| 22572 | IRTemp argDhi = newTemp(Ity_I32); |
| 22573 | IRTemp argDlo = newTemp(Ity_I32); |
| 22574 | IRTemp res = newTemp(Ity_I64); |
| 22575 | IRTemp resHi = newTemp(Ity_I32); |
| 22576 | IRTemp resLo = newTemp(Ity_I32); |
| 22577 | assign( argN, getIRegT(rN) ); |
| 22578 | assign( argM, getIRegT(rM) ); |
| 22579 | assign( argDhi, getIRegT(rDhi) ); |
| 22580 | assign( argDlo, getIRegT(rDlo) ); |
| 22581 | assign( res, |
| 22582 | binop(Iop_Add64, |
| 22583 | binop(Iop_Add64, |
| 22584 | binop(Iop_MullU32, mkexpr(argN), mkexpr(argM)), |
| 22585 | unop(Iop_32Uto64, mkexpr(argDhi))), |
| 22586 | unop(Iop_32Uto64, mkexpr(argDlo))) ); |
| 22587 | assign( resHi, unop(Iop_64HIto32, mkexpr(res)) ); |
| 22588 | assign( resLo, unop(Iop_64to32, mkexpr(res)) ); |
| 22589 | putIRegT( rDhi, mkexpr(resHi), condT ); |
| 22590 | putIRegT( rDlo, mkexpr(resLo), condT ); |
| 22591 | DIP("umaal r%u, r%u, r%u, r%u\n", rDlo, rDhi, rN, rM); |
| 22592 | goto decode_success; |
| 22593 | } |
| 22594 | } |
| 22595 | |
sewardj | ccff5f4 | 2013-04-23 08:56:43 +0000 | [diff] [blame] | 22596 | /* ------------------- (T1) SMMUL{R} ------------------ */ |
| 22597 | if (INSN0(15,7) == BITS9(1,1,1,1,1,0,1,1,0) |
| 22598 | && INSN0(6,4) == BITS3(1,0,1) |
| 22599 | && INSN1(15,12) == BITS4(1,1,1,1) |
| 22600 | && INSN1(7,5) == BITS3(0,0,0)) { |
| 22601 | UInt bitR = INSN1(4,4); |
| 22602 | UInt rD = INSN1(11,8); |
| 22603 | UInt rM = INSN1(3,0); |
| 22604 | UInt rN = INSN0(3,0); |
| 22605 | if (!isBadRegT(rD) && !isBadRegT(rN) && !isBadRegT(rM)) { |
| 22606 | IRExpr* res |
| 22607 | = unop(Iop_64HIto32, |
| 22608 | binop(Iop_Add64, |
| 22609 | binop(Iop_MullS32, getIRegT(rN), getIRegT(rM)), |
| 22610 | mkU64(bitR ? 0x80000000ULL : 0ULL))); |
| 22611 | putIRegT(rD, res, condT); |
| 22612 | DIP("smmul%s r%u, r%u, r%u\n", |
| 22613 | bitR ? "r" : "", rD, rN, rM); |
| 22614 | goto decode_success; |
| 22615 | } |
| 22616 | } |
| 22617 | |
| 22618 | /* ------------------- (T1) SMMLA{R} ------------------ */ |
| 22619 | if (INSN0(15,7) == BITS9(1,1,1,1,1,0,1,1,0) |
| 22620 | && INSN0(6,4) == BITS3(1,0,1) |
| 22621 | && INSN1(7,5) == BITS3(0,0,0)) { |
| 22622 | UInt bitR = INSN1(4,4); |
| 22623 | UInt rA = INSN1(15,12); |
| 22624 | UInt rD = INSN1(11,8); |
| 22625 | UInt rM = INSN1(3,0); |
| 22626 | UInt rN = INSN0(3,0); |
| 22627 | if (!isBadRegT(rD) && !isBadRegT(rN) && !isBadRegT(rM) && (rA != 13)) { |
| 22628 | IRExpr* res |
| 22629 | = unop(Iop_64HIto32, |
| 22630 | binop(Iop_Add64, |
| 22631 | binop(Iop_Add64, |
| 22632 | binop(Iop_32HLto64, getIRegT(rA), mkU32(0)), |
| 22633 | binop(Iop_MullS32, getIRegT(rN), getIRegT(rM))), |
| 22634 | mkU64(bitR ? 0x80000000ULL : 0ULL))); |
| 22635 | putIRegT(rD, res, condT); |
| 22636 | DIP("smmla%s r%u, r%u, r%u, r%u\n", |
| 22637 | bitR ? "r" : "", rD, rN, rM, rA); |
| 22638 | goto decode_success; |
| 22639 | } |
| 22640 | } |
| 22641 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22642 | /* ------------------ (T2) ADR ------------------ */ |
| 22643 | if ((INSN0(15,0) == 0xF2AF || INSN0(15,0) == 0xF6AF) |
| 22644 | && INSN1(15,15) == 0) { |
| 22645 | /* rD = align4(PC) - imm32 */ |
| 22646 | UInt rD = INSN1(11,8); |
| 22647 | if (!isBadRegT(rD)) { |
| 22648 | UInt imm32 = (INSN0(10,10) << 11) |
| 22649 | | (INSN1(14,12) << 8) | INSN1(7,0); |
| 22650 | putIRegT(rD, binop(Iop_Sub32, |
sewardj | df86f16 | 2010-08-22 18:47:30 +0000 | [diff] [blame] | 22651 | binop(Iop_And32, getIRegT(15), mkU32(~3U)), |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22652 | mkU32(imm32)), |
| 22653 | condT); |
| 22654 | DIP("sub r%u, pc, #%u\n", rD, imm32); |
| 22655 | goto decode_success; |
| 22656 | } |
| 22657 | } |
| 22658 | |
| 22659 | /* ------------------- (T1) BFI ------------------- */ |
| 22660 | /* ------------------- (T1) BFC ------------------- */ |
| 22661 | if (INSN0(15,4) == 0xF36 && INSN1(15,15) == 0 && INSN1(5,5) == 0) { |
| 22662 | UInt rD = INSN1(11,8); |
| 22663 | UInt rN = INSN0(3,0); |
| 22664 | UInt msb = INSN1(4,0); |
| 22665 | UInt lsb = (INSN1(14,12) << 2) | INSN1(7,6); |
| 22666 | if (isBadRegT(rD) || rN == 13 || msb < lsb) { |
| 22667 | /* undecodable; fall through */ |
| 22668 | } else { |
| 22669 | IRTemp src = newTemp(Ity_I32); |
| 22670 | IRTemp olddst = newTemp(Ity_I32); |
| 22671 | IRTemp newdst = newTemp(Ity_I32); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 22672 | UInt mask = ((UInt)1) << (msb - lsb); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22673 | mask = (mask - 1) + mask; |
| 22674 | vassert(mask != 0); // guaranteed by "msb < lsb" check above |
| 22675 | mask <<= lsb; |
| 22676 | |
| 22677 | assign(src, rN == 15 ? mkU32(0) : getIRegT(rN)); |
| 22678 | assign(olddst, getIRegT(rD)); |
| 22679 | assign(newdst, |
| 22680 | binop(Iop_Or32, |
| 22681 | binop(Iop_And32, |
| 22682 | binop(Iop_Shl32, mkexpr(src), mkU8(lsb)), |
| 22683 | mkU32(mask)), |
| 22684 | binop(Iop_And32, |
| 22685 | mkexpr(olddst), |
| 22686 | mkU32(~mask))) |
| 22687 | ); |
| 22688 | |
| 22689 | putIRegT(rD, mkexpr(newdst), condT); |
| 22690 | |
| 22691 | if (rN == 15) { |
| 22692 | DIP("bfc r%u, #%u, #%u\n", |
| 22693 | rD, lsb, msb-lsb+1); |
| 22694 | } else { |
| 22695 | DIP("bfi r%u, r%u, #%u, #%u\n", |
| 22696 | rD, rN, lsb, msb-lsb+1); |
| 22697 | } |
| 22698 | goto decode_success; |
| 22699 | } |
| 22700 | } |
| 22701 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22702 | /* ------------------- (T1) SXTAH ------------------- */ |
| 22703 | /* ------------------- (T1) UXTAH ------------------- */ |
| 22704 | if ((INSN0(15,4) == 0xFA1 // UXTAH |
| 22705 | || INSN0(15,4) == 0xFA0) // SXTAH |
| 22706 | && INSN1(15,12) == BITS4(1,1,1,1) |
| 22707 | && INSN1(7,6) == BITS2(1,0)) { |
| 22708 | Bool isU = INSN0(15,4) == 0xFA1; |
| 22709 | UInt rN = INSN0(3,0); |
| 22710 | UInt rD = INSN1(11,8); |
| 22711 | UInt rM = INSN1(3,0); |
| 22712 | UInt rot = INSN1(5,4); |
| 22713 | if (!isBadRegT(rD) && !isBadRegT(rN) && !isBadRegT(rM)) { |
| 22714 | IRTemp srcL = newTemp(Ity_I32); |
| 22715 | IRTemp srcR = newTemp(Ity_I32); |
| 22716 | IRTemp res = newTemp(Ity_I32); |
| 22717 | assign(srcR, getIRegT(rM)); |
| 22718 | assign(srcL, getIRegT(rN)); |
| 22719 | assign(res, binop(Iop_Add32, |
| 22720 | mkexpr(srcL), |
| 22721 | unop(isU ? Iop_16Uto32 : Iop_16Sto32, |
| 22722 | unop(Iop_32to16, |
| 22723 | genROR32(srcR, 8 * rot))))); |
| 22724 | putIRegT(rD, mkexpr(res), condT); |
| 22725 | DIP("%cxtah r%u, r%u, r%u, ror #%u\n", |
| 22726 | isU ? 'u' : 's', rD, rN, rM, rot); |
| 22727 | goto decode_success; |
| 22728 | } |
| 22729 | } |
| 22730 | |
| 22731 | /* ------------------- (T1) SXTAB ------------------- */ |
| 22732 | /* ------------------- (T1) UXTAB ------------------- */ |
| 22733 | if ((INSN0(15,4) == 0xFA5 // UXTAB |
| 22734 | || INSN0(15,4) == 0xFA4) // SXTAB |
| 22735 | && INSN1(15,12) == BITS4(1,1,1,1) |
| 22736 | && INSN1(7,6) == BITS2(1,0)) { |
| 22737 | Bool isU = INSN0(15,4) == 0xFA5; |
| 22738 | UInt rN = INSN0(3,0); |
| 22739 | UInt rD = INSN1(11,8); |
| 22740 | UInt rM = INSN1(3,0); |
| 22741 | UInt rot = INSN1(5,4); |
| 22742 | if (!isBadRegT(rD) && !isBadRegT(rN) && !isBadRegT(rM)) { |
| 22743 | IRTemp srcL = newTemp(Ity_I32); |
| 22744 | IRTemp srcR = newTemp(Ity_I32); |
| 22745 | IRTemp res = newTemp(Ity_I32); |
| 22746 | assign(srcR, getIRegT(rM)); |
| 22747 | assign(srcL, getIRegT(rN)); |
| 22748 | assign(res, binop(Iop_Add32, |
| 22749 | mkexpr(srcL), |
| 22750 | unop(isU ? Iop_8Uto32 : Iop_8Sto32, |
| 22751 | unop(Iop_32to8, |
| 22752 | genROR32(srcR, 8 * rot))))); |
| 22753 | putIRegT(rD, mkexpr(res), condT); |
| 22754 | DIP("%cxtab r%u, r%u, r%u, ror #%u\n", |
| 22755 | isU ? 'u' : 's', rD, rN, rM, rot); |
| 22756 | goto decode_success; |
| 22757 | } |
| 22758 | } |
| 22759 | |
| 22760 | /* ------------------- (T1) CLZ ------------------- */ |
| 22761 | if (INSN0(15,4) == 0xFAB |
| 22762 | && INSN1(15,12) == BITS4(1,1,1,1) |
| 22763 | && INSN1(7,4) == BITS4(1,0,0,0)) { |
| 22764 | UInt rM1 = INSN0(3,0); |
sewardj | 677e5af | 2010-09-02 21:02:47 +0000 | [diff] [blame] | 22765 | UInt rD = INSN1(11,8); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22766 | UInt rM2 = INSN1(3,0); |
| 22767 | if (!isBadRegT(rD) && !isBadRegT(rM1) && rM1 == rM2) { |
| 22768 | IRTemp arg = newTemp(Ity_I32); |
| 22769 | IRTemp res = newTemp(Ity_I32); |
| 22770 | assign(arg, getIRegT(rM1)); |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 22771 | assign(res, IRExpr_ITE( |
sewardj | 009230b | 2013-01-26 11:47:55 +0000 | [diff] [blame] | 22772 | binop(Iop_CmpEQ32, mkexpr(arg), mkU32(0)), |
florian | 99dd03e | 2013-01-29 03:56:06 +0000 | [diff] [blame] | 22773 | mkU32(32), |
| 22774 | unop(Iop_Clz32, mkexpr(arg)) |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22775 | )); |
| 22776 | putIRegT(rD, mkexpr(res), condT); |
| 22777 | DIP("clz r%u, r%u\n", rD, rM1); |
| 22778 | goto decode_success; |
| 22779 | } |
| 22780 | } |
| 22781 | |
sewardj | 677e5af | 2010-09-02 21:02:47 +0000 | [diff] [blame] | 22782 | /* ------------------- (T1) RBIT ------------------- */ |
| 22783 | if (INSN0(15,4) == 0xFA9 |
| 22784 | && INSN1(15,12) == BITS4(1,1,1,1) |
| 22785 | && INSN1(7,4) == BITS4(1,0,1,0)) { |
| 22786 | UInt rM1 = INSN0(3,0); |
| 22787 | UInt rD = INSN1(11,8); |
| 22788 | UInt rM2 = INSN1(3,0); |
| 22789 | if (!isBadRegT(rD) && !isBadRegT(rM1) && rM1 == rM2) { |
| 22790 | IRTemp arg = newTemp(Ity_I32); |
| 22791 | assign(arg, getIRegT(rM1)); |
| 22792 | IRTemp res = gen_BITREV(arg); |
| 22793 | putIRegT(rD, mkexpr(res), condT); |
| 22794 | DIP("rbit r%u, r%u\n", rD, rM1); |
| 22795 | goto decode_success; |
| 22796 | } |
| 22797 | } |
| 22798 | |
sewardj | 27312d3 | 2010-09-26 00:48:41 +0000 | [diff] [blame] | 22799 | /* ------------------- (T2) REV ------------------- */ |
| 22800 | /* ------------------- (T2) REV16 ------------------- */ |
| 22801 | if (INSN0(15,4) == 0xFA9 |
| 22802 | && INSN1(15,12) == BITS4(1,1,1,1) |
| 22803 | && ( INSN1(7,4) == BITS4(1,0,0,0) // REV |
| 22804 | || INSN1(7,4) == BITS4(1,0,0,1))) { // REV16 |
| 22805 | UInt rM1 = INSN0(3,0); |
| 22806 | UInt rD = INSN1(11,8); |
| 22807 | UInt rM2 = INSN1(3,0); |
| 22808 | Bool isREV = INSN1(7,4) == BITS4(1,0,0,0); |
| 22809 | if (!isBadRegT(rD) && !isBadRegT(rM1) && rM1 == rM2) { |
| 22810 | IRTemp arg = newTemp(Ity_I32); |
| 22811 | assign(arg, getIRegT(rM1)); |
| 22812 | IRTemp res = isREV ? gen_REV(arg) : gen_REV16(arg); |
| 22813 | putIRegT(rD, mkexpr(res), condT); |
| 22814 | DIP("rev%s r%u, r%u\n", isREV ? "" : "16", rD, rM1); |
| 22815 | goto decode_success; |
| 22816 | } |
| 22817 | } |
| 22818 | |
sewardj | a325ee1 | 2013-07-04 20:35:42 +0000 | [diff] [blame] | 22819 | /* ------------------- (T2) REVSH ------------------ */ |
| 22820 | if (INSN0(15,4) == 0xFA9 |
| 22821 | && INSN1(15,12) == BITS4(1,1,1,1) |
| 22822 | && INSN1(7,4) == BITS4(1,0,1,1)) { |
| 22823 | UInt rM1 = INSN0(3,0); |
| 22824 | UInt rM2 = INSN1(3,0); |
| 22825 | UInt rD = INSN1(11,8); |
| 22826 | if (!isBadRegT(rD) && !isBadRegT(rM1) && rM1 == rM2) { |
| 22827 | IRTemp irt_rM = newTemp(Ity_I32); |
| 22828 | IRTemp irt_hi = newTemp(Ity_I32); |
| 22829 | IRTemp irt_low = newTemp(Ity_I32); |
| 22830 | IRTemp irt_res = newTemp(Ity_I32); |
| 22831 | assign(irt_rM, getIRegT(rM1)); |
| 22832 | assign(irt_hi, |
| 22833 | binop(Iop_Sar32, |
| 22834 | binop(Iop_Shl32, mkexpr(irt_rM), mkU8(24)), |
| 22835 | mkU8(16) |
| 22836 | ) |
| 22837 | ); |
| 22838 | assign(irt_low, |
| 22839 | binop(Iop_And32, |
| 22840 | binop(Iop_Shr32, mkexpr(irt_rM), mkU8(8)), |
| 22841 | mkU32(0xFF) |
| 22842 | ) |
| 22843 | ); |
| 22844 | assign(irt_res, |
| 22845 | binop(Iop_Or32, mkexpr(irt_hi), mkexpr(irt_low)) |
| 22846 | ); |
| 22847 | putIRegT(rD, mkexpr(irt_res), condT); |
| 22848 | DIP("revsh r%u, r%u\n", rD, rM1); |
| 22849 | goto decode_success; |
| 22850 | } |
| 22851 | } |
| 22852 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22853 | /* -------------- (T1) MSR apsr, reg -------------- */ |
| 22854 | if (INSN0(15,4) == 0xF38 |
| 22855 | && INSN1(15,12) == BITS4(1,0,0,0) && INSN1(9,0) == 0x000) { |
| 22856 | UInt rN = INSN0(3,0); |
| 22857 | UInt write_ge = INSN1(10,10); |
| 22858 | UInt write_nzcvq = INSN1(11,11); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 22859 | if (!isBadRegT(rN) && (write_nzcvq || write_ge)) { |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22860 | IRTemp rNt = newTemp(Ity_I32); |
| 22861 | assign(rNt, getIRegT(rN)); |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 22862 | desynthesise_APSR( write_nzcvq, write_ge, rNt, condT ); |
| 22863 | DIP("msr cpsr_%s%s, r%u\n", |
| 22864 | write_nzcvq ? "f" : "", write_ge ? "g" : "", rN); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22865 | goto decode_success; |
| 22866 | } |
| 22867 | } |
| 22868 | |
| 22869 | /* -------------- (T1) MRS reg, apsr -------------- */ |
| 22870 | if (INSN0(15,0) == 0xF3EF |
| 22871 | && INSN1(15,12) == BITS4(1,0,0,0) && INSN1(7,0) == 0x00) { |
| 22872 | UInt rD = INSN1(11,8); |
| 22873 | if (!isBadRegT(rD)) { |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 22874 | IRTemp apsr = synthesise_APSR(); |
| 22875 | putIRegT( rD, mkexpr(apsr), condT ); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 22876 | DIP("mrs r%u, cpsr\n", rD); |
| 22877 | goto decode_success; |
| 22878 | } |
| 22879 | } |
| 22880 | |
sewardj | e1a9396 | 2010-09-24 23:35:59 +0000 | [diff] [blame] | 22881 | /* ----------------- (T1) LDREX ----------------- */ |
| 22882 | if (INSN0(15,4) == 0xE85 && INSN1(11,8) == BITS4(1,1,1,1)) { |
| 22883 | UInt rN = INSN0(3,0); |
| 22884 | UInt rT = INSN1(15,12); |
| 22885 | UInt imm8 = INSN1(7,0); |
| 22886 | if (!isBadRegT(rT) && rN != 15) { |
| 22887 | IRTemp res; |
| 22888 | // go uncond |
| 22889 | mk_skip_over_T32_if_cond_is_false( condT ); |
| 22890 | // now uncond |
| 22891 | res = newTemp(Ity_I32); |
| 22892 | stmt( IRStmt_LLSC(Iend_LE, |
| 22893 | res, |
| 22894 | binop(Iop_Add32, getIRegT(rN), mkU32(imm8 * 4)), |
| 22895 | NULL/*this is a load*/ )); |
| 22896 | putIRegT(rT, mkexpr(res), IRTemp_INVALID); |
| 22897 | DIP("ldrex r%u, [r%u, #+%u]\n", rT, rN, imm8 * 4); |
| 22898 | goto decode_success; |
| 22899 | } |
| 22900 | } |
| 22901 | |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 22902 | /* --------------- (T1) LDREX{B,H} --------------- */ |
| 22903 | if (INSN0(15,4) == 0xE8D |
| 22904 | && (INSN1(11,0) == 0xF4F || INSN1(11,0) == 0xF5F)) { |
| 22905 | UInt rN = INSN0(3,0); |
| 22906 | UInt rT = INSN1(15,12); |
| 22907 | Bool isH = INSN1(11,0) == 0xF5F; |
| 22908 | if (!isBadRegT(rT) && rN != 15) { |
| 22909 | IRTemp res; |
| 22910 | // go uncond |
| 22911 | mk_skip_over_T32_if_cond_is_false( condT ); |
| 22912 | // now uncond |
| 22913 | res = newTemp(isH ? Ity_I16 : Ity_I8); |
| 22914 | stmt( IRStmt_LLSC(Iend_LE, res, getIRegT(rN), |
| 22915 | NULL/*this is a load*/ )); |
| 22916 | putIRegT(rT, unop(isH ? Iop_16Uto32 : Iop_8Uto32, mkexpr(res)), |
| 22917 | IRTemp_INVALID); |
| 22918 | DIP("ldrex%c r%u, [r%u]\n", isH ? 'h' : 'b', rT, rN); |
| 22919 | goto decode_success; |
| 22920 | } |
| 22921 | } |
| 22922 | |
| 22923 | /* --------------- (T1) LDREXD --------------- */ |
| 22924 | if (INSN0(15,4) == 0xE8D && INSN1(7,0) == 0x7F) { |
| 22925 | UInt rN = INSN0(3,0); |
| 22926 | UInt rT = INSN1(15,12); |
| 22927 | UInt rT2 = INSN1(11,8); |
| 22928 | if (!isBadRegT(rT) && !isBadRegT(rT2) && rT != rT2 && rN != 15) { |
| 22929 | IRTemp res; |
| 22930 | // go uncond |
| 22931 | mk_skip_over_T32_if_cond_is_false( condT ); |
| 22932 | // now uncond |
| 22933 | res = newTemp(Ity_I64); |
| 22934 | // FIXME: assumes little-endian guest |
| 22935 | stmt( IRStmt_LLSC(Iend_LE, res, getIRegT(rN), |
| 22936 | NULL/*this is a load*/ )); |
| 22937 | // FIXME: assumes little-endian guest |
| 22938 | putIRegT(rT, unop(Iop_64to32, mkexpr(res)), IRTemp_INVALID); |
| 22939 | putIRegT(rT2, unop(Iop_64HIto32, mkexpr(res)), IRTemp_INVALID); |
| 22940 | DIP("ldrexd r%u, r%u, [r%u]\n", rT, rT2, rN); |
| 22941 | goto decode_success; |
| 22942 | } |
| 22943 | } |
| 22944 | |
sewardj | e1a9396 | 2010-09-24 23:35:59 +0000 | [diff] [blame] | 22945 | /* ----------------- (T1) STREX ----------------- */ |
| 22946 | if (INSN0(15,4) == 0xE84) { |
| 22947 | UInt rN = INSN0(3,0); |
| 22948 | UInt rT = INSN1(15,12); |
| 22949 | UInt rD = INSN1(11,8); |
| 22950 | UInt imm8 = INSN1(7,0); |
| 22951 | if (!isBadRegT(rD) && !isBadRegT(rT) && rN != 15 |
| 22952 | && rD != rN && rD != rT) { |
| 22953 | IRTemp resSC1, resSC32; |
sewardj | e1a9396 | 2010-09-24 23:35:59 +0000 | [diff] [blame] | 22954 | // go uncond |
| 22955 | mk_skip_over_T32_if_cond_is_false( condT ); |
| 22956 | // now uncond |
sewardj | e1a9396 | 2010-09-24 23:35:59 +0000 | [diff] [blame] | 22957 | /* Ok, now we're unconditional. Do the store. */ |
| 22958 | resSC1 = newTemp(Ity_I1); |
| 22959 | stmt( IRStmt_LLSC(Iend_LE, |
| 22960 | resSC1, |
| 22961 | binop(Iop_Add32, getIRegT(rN), mkU32(imm8 * 4)), |
| 22962 | getIRegT(rT)) ); |
sewardj | e1a9396 | 2010-09-24 23:35:59 +0000 | [diff] [blame] | 22963 | /* Set rD to 1 on failure, 0 on success. Currently we have |
| 22964 | resSC1 == 0 on failure, 1 on success. */ |
| 22965 | resSC32 = newTemp(Ity_I32); |
| 22966 | assign(resSC32, |
| 22967 | unop(Iop_1Uto32, unop(Iop_Not1, mkexpr(resSC1)))); |
sewardj | e1a9396 | 2010-09-24 23:35:59 +0000 | [diff] [blame] | 22968 | putIRegT(rD, mkexpr(resSC32), IRTemp_INVALID); |
| 22969 | DIP("strex r%u, r%u, [r%u, #+%u]\n", rD, rT, rN, imm8 * 4); |
| 22970 | goto decode_success; |
| 22971 | } |
| 22972 | } |
| 22973 | |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 22974 | /* --------------- (T1) STREX{B,H} --------------- */ |
| 22975 | if (INSN0(15,4) == 0xE8C |
| 22976 | && (INSN1(11,4) == 0xF4 || INSN1(11,4) == 0xF5)) { |
| 22977 | UInt rN = INSN0(3,0); |
| 22978 | UInt rT = INSN1(15,12); |
| 22979 | UInt rD = INSN1(3,0); |
| 22980 | Bool isH = INSN1(11,4) == 0xF5; |
| 22981 | if (!isBadRegT(rD) && !isBadRegT(rT) && rN != 15 |
| 22982 | && rD != rN && rD != rT) { |
| 22983 | IRTemp resSC1, resSC32; |
| 22984 | // go uncond |
| 22985 | mk_skip_over_T32_if_cond_is_false( condT ); |
| 22986 | // now uncond |
| 22987 | /* Ok, now we're unconditional. Do the store. */ |
| 22988 | resSC1 = newTemp(Ity_I1); |
| 22989 | stmt( IRStmt_LLSC(Iend_LE, resSC1, getIRegT(rN), |
| 22990 | unop(isH ? Iop_32to16 : Iop_32to8, |
| 22991 | getIRegT(rT))) ); |
| 22992 | /* Set rD to 1 on failure, 0 on success. Currently we have |
| 22993 | resSC1 == 0 on failure, 1 on success. */ |
| 22994 | resSC32 = newTemp(Ity_I32); |
| 22995 | assign(resSC32, |
| 22996 | unop(Iop_1Uto32, unop(Iop_Not1, mkexpr(resSC1)))); |
| 22997 | putIRegT(rD, mkexpr(resSC32), IRTemp_INVALID); |
| 22998 | DIP("strex%c r%u, r%u, [r%u]\n", isH ? 'h' : 'b', rD, rT, rN); |
| 22999 | goto decode_success; |
| 23000 | } |
| 23001 | } |
| 23002 | |
| 23003 | /* ---------------- (T1) STREXD ---------------- */ |
| 23004 | if (INSN0(15,4) == 0xE8C && INSN1(7,4) == BITS4(0,1,1,1)) { |
| 23005 | UInt rN = INSN0(3,0); |
| 23006 | UInt rT = INSN1(15,12); |
| 23007 | UInt rT2 = INSN1(11,8); |
| 23008 | UInt rD = INSN1(3,0); |
| 23009 | if (!isBadRegT(rD) && !isBadRegT(rT) && !isBadRegT(rT2) |
florian | bc2c60d | 2015-06-05 19:35:32 +0000 | [diff] [blame] | 23010 | && rN != 15 && rD != rN && rD != rT && rD != rT2) { |
sewardj | ff7f5b7 | 2011-07-11 11:43:38 +0000 | [diff] [blame] | 23011 | IRTemp resSC1, resSC32, data; |
| 23012 | // go uncond |
| 23013 | mk_skip_over_T32_if_cond_is_false( condT ); |
| 23014 | // now uncond |
| 23015 | /* Ok, now we're unconditional. Do the store. */ |
| 23016 | resSC1 = newTemp(Ity_I1); |
| 23017 | data = newTemp(Ity_I64); |
| 23018 | // FIXME: assumes little-endian guest |
| 23019 | assign(data, binop(Iop_32HLto64, getIRegT(rT2), getIRegT(rT))); |
| 23020 | // FIXME: assumes little-endian guest |
| 23021 | stmt( IRStmt_LLSC(Iend_LE, resSC1, getIRegT(rN), mkexpr(data))); |
| 23022 | /* Set rD to 1 on failure, 0 on success. Currently we have |
| 23023 | resSC1 == 0 on failure, 1 on success. */ |
| 23024 | resSC32 = newTemp(Ity_I32); |
| 23025 | assign(resSC32, |
| 23026 | unop(Iop_1Uto32, unop(Iop_Not1, mkexpr(resSC1)))); |
| 23027 | putIRegT(rD, mkexpr(resSC32), IRTemp_INVALID); |
| 23028 | DIP("strexd r%u, r%u, r%u, [r%u]\n", rD, rT, rT2, rN); |
| 23029 | goto decode_success; |
| 23030 | } |
| 23031 | } |
sewardj | ccff5f4 | 2013-04-23 08:56:43 +0000 | [diff] [blame] | 23032 | |
sewardj | 4aec376 | 2010-09-24 23:48:29 +0000 | [diff] [blame] | 23033 | /* -------------- v7 barrier insns -------------- */ |
sewardj | 5bb6ca2 | 2011-05-08 09:09:12 +0000 | [diff] [blame] | 23034 | if (INSN0(15,0) == 0xF3BF && (INSN1(15,0) & 0xFF00) == 0x8F00) { |
sewardj | 6d615ba | 2011-09-26 16:19:43 +0000 | [diff] [blame] | 23035 | /* FIXME: should this be unconditional? */ |
sewardj | 646bc00 | 2010-10-11 18:57:10 +0000 | [diff] [blame] | 23036 | /* XXX this isn't really right, is it? The generated IR does |
| 23037 | them unconditionally. I guess it doesn't matter since it |
| 23038 | doesn't do any harm to do them even when the guarding |
| 23039 | condition is false -- it's just a performance loss. */ |
sewardj | 5bb6ca2 | 2011-05-08 09:09:12 +0000 | [diff] [blame] | 23040 | switch (INSN1(7,0)) { |
| 23041 | case 0x4F: /* DSB sy */ |
| 23042 | case 0x4E: /* DSB st */ |
| 23043 | case 0x4B: /* DSB ish */ |
| 23044 | case 0x4A: /* DSB ishst */ |
| 23045 | case 0x47: /* DSB nsh */ |
| 23046 | case 0x46: /* DSB nshst */ |
| 23047 | case 0x43: /* DSB osh */ |
| 23048 | case 0x42: /* DSB oshst */ |
sewardj | 4aec376 | 2010-09-24 23:48:29 +0000 | [diff] [blame] | 23049 | stmt( IRStmt_MBE(Imbe_Fence) ); |
| 23050 | DIP("DSB\n"); |
| 23051 | goto decode_success; |
sewardj | 5bb6ca2 | 2011-05-08 09:09:12 +0000 | [diff] [blame] | 23052 | case 0x5F: /* DMB sy */ |
| 23053 | case 0x5E: /* DMB st */ |
| 23054 | case 0x5B: /* DMB ish */ |
| 23055 | case 0x5A: /* DMB ishst */ |
| 23056 | case 0x57: /* DMB nsh */ |
| 23057 | case 0x56: /* DMB nshst */ |
| 23058 | case 0x53: /* DMB osh */ |
| 23059 | case 0x52: /* DMB oshst */ |
sewardj | 4aec376 | 2010-09-24 23:48:29 +0000 | [diff] [blame] | 23060 | stmt( IRStmt_MBE(Imbe_Fence) ); |
| 23061 | DIP("DMB\n"); |
| 23062 | goto decode_success; |
sewardj | 5bb6ca2 | 2011-05-08 09:09:12 +0000 | [diff] [blame] | 23063 | case 0x6F: /* ISB */ |
sewardj | 4aec376 | 2010-09-24 23:48:29 +0000 | [diff] [blame] | 23064 | stmt( IRStmt_MBE(Imbe_Fence) ); |
| 23065 | DIP("ISB\n"); |
| 23066 | goto decode_success; |
| 23067 | default: |
| 23068 | break; |
| 23069 | } |
| 23070 | } |
| 23071 | |
sewardj | 7f5a841 | 2011-07-21 06:17:21 +0000 | [diff] [blame] | 23072 | /* ---------------------- PLD{,W} ---------------------- */ |
| 23073 | if ((INSN0(15,4) & 0xFFD) == 0xF89 && INSN1(15,12) == 0xF) { |
sewardj | 6d615ba | 2011-09-26 16:19:43 +0000 | [diff] [blame] | 23074 | /* FIXME: should this be unconditional? */ |
sewardj | 7f5a841 | 2011-07-21 06:17:21 +0000 | [diff] [blame] | 23075 | /* PLD/PLDW immediate, encoding T1 */ |
| 23076 | UInt rN = INSN0(3,0); |
| 23077 | UInt bW = INSN0(5,5); |
| 23078 | UInt imm12 = INSN1(11,0); |
| 23079 | DIP("pld%s [r%u, #%u]\n", bW ? "w" : "", rN, imm12); |
| 23080 | goto decode_success; |
| 23081 | } |
| 23082 | |
| 23083 | if ((INSN0(15,4) & 0xFFD) == 0xF81 && INSN1(15,8) == 0xFC) { |
sewardj | 6d615ba | 2011-09-26 16:19:43 +0000 | [diff] [blame] | 23084 | /* FIXME: should this be unconditional? */ |
sewardj | 7f5a841 | 2011-07-21 06:17:21 +0000 | [diff] [blame] | 23085 | /* PLD/PLDW immediate, encoding T2 */ |
| 23086 | UInt rN = INSN0(3,0); |
| 23087 | UInt bW = INSN0(5,5); |
| 23088 | UInt imm8 = INSN1(7,0); |
| 23089 | DIP("pld%s [r%u, #-%u]\n", bW ? "w" : "", rN, imm8); |
| 23090 | goto decode_success; |
| 23091 | } |
| 23092 | |
| 23093 | if ((INSN0(15,4) & 0xFFD) == 0xF81 && INSN1(15,6) == 0x3C0) { |
sewardj | 6d615ba | 2011-09-26 16:19:43 +0000 | [diff] [blame] | 23094 | /* FIXME: should this be unconditional? */ |
sewardj | 7f5a841 | 2011-07-21 06:17:21 +0000 | [diff] [blame] | 23095 | /* PLD/PLDW register, encoding T1 */ |
| 23096 | UInt rN = INSN0(3,0); |
| 23097 | UInt rM = INSN1(3,0); |
| 23098 | UInt bW = INSN0(5,5); |
| 23099 | UInt imm2 = INSN1(5,4); |
| 23100 | if (!isBadRegT(rM)) { |
florian | b173774 | 2015-08-03 16:03:13 +0000 | [diff] [blame] | 23101 | DIP("pld%s [r%u, r%u, lsl %u]\n", bW ? "w" : "", rN, rM, imm2); |
sewardj | 7f5a841 | 2011-07-21 06:17:21 +0000 | [diff] [blame] | 23102 | goto decode_success; |
| 23103 | } |
| 23104 | /* fall through */ |
| 23105 | } |
| 23106 | |
sewardj | ed75a68 | 2011-02-09 14:21:45 +0000 | [diff] [blame] | 23107 | /* -------------- read CP15 TPIDRURO register ------------- */ |
| 23108 | /* mrc p15, 0, r0, c13, c0, 3 up to |
| 23109 | mrc p15, 0, r14, c13, c0, 3 |
| 23110 | */ |
| 23111 | /* I don't know whether this is really v7-only. But anyway, we |
| 23112 | have to support it since arm-linux uses TPIDRURO as a thread |
| 23113 | state register. */ |
sewardj | ed75a68 | 2011-02-09 14:21:45 +0000 | [diff] [blame] | 23114 | if ((INSN0(15,0) == 0xEE1D) && (INSN1(11,0) == 0x0F70)) { |
| 23115 | UInt rD = INSN1(15,12); |
| 23116 | if (!isBadRegT(rD)) { |
sewardj | 543ad88 | 2015-08-20 16:53:18 +0000 | [diff] [blame] | 23117 | putIRegT(rD, IRExpr_Get(OFFB_TPIDRURO, Ity_I32), condT); |
sewardj | ed75a68 | 2011-02-09 14:21:45 +0000 | [diff] [blame] | 23118 | DIP("mrc p15,0, r%u, c13, c0, 3\n", rD); |
| 23119 | goto decode_success; |
| 23120 | } |
| 23121 | /* fall through */ |
| 23122 | } |
| 23123 | |
sewardj | 61d7d98 | 2015-08-20 16:17:11 +0000 | [diff] [blame] | 23124 | /* -------------- read CP15 PMUSRENR register ------------- */ |
| 23125 | /* mrc p15, 0, r0, c9, c14, 0 up to |
| 23126 | mrc p15, 0, r14, c9, c14, 0 |
| 23127 | See comment on the ARM equivalent of this (above) for details. |
| 23128 | */ |
| 23129 | if ((INSN0(15,0) == 0xEE19) && (INSN1(11,0) == 0x0F1E)) { |
sewardj | 61d7d98 | 2015-08-20 16:17:11 +0000 | [diff] [blame] | 23130 | UInt rD = INSN1(15,12); |
| 23131 | if (!isBadRegT(rD)) { |
sewardj | 543ad88 | 2015-08-20 16:53:18 +0000 | [diff] [blame] | 23132 | putIRegT(rD, mkU32(0), condT); |
sewardj | 61d7d98 | 2015-08-20 16:17:11 +0000 | [diff] [blame] | 23133 | DIP("mrc p15,0, r%u, c9, c14, 0\n", rD); |
| 23134 | goto decode_success; |
| 23135 | } |
| 23136 | /* fall through */ |
| 23137 | } |
| 23138 | |
sewardj | 6d615ba | 2011-09-26 16:19:43 +0000 | [diff] [blame] | 23139 | /* ------------------- CLREX ------------------ */ |
| 23140 | if (INSN0(15,0) == 0xF3BF && INSN1(15,0) == 0x8F2F) { |
| 23141 | /* AFAICS, this simply cancels a (all?) reservations made by a |
| 23142 | (any?) preceding LDREX(es). Arrange to hand it through to |
| 23143 | the back end. */ |
| 23144 | mk_skip_over_T32_if_cond_is_false( condT ); |
| 23145 | stmt( IRStmt_MBE(Imbe_CancelReservation) ); |
| 23146 | DIP("clrex\n"); |
| 23147 | goto decode_success; |
| 23148 | } |
| 23149 | |
sewardj | 646bc00 | 2010-10-11 18:57:10 +0000 | [diff] [blame] | 23150 | /* ------------------- NOP ------------------ */ |
| 23151 | if (INSN0(15,0) == 0xF3AF && INSN1(15,0) == 0x8000) { |
| 23152 | DIP("nop\n"); |
| 23153 | goto decode_success; |
| 23154 | } |
| 23155 | |
sewardj | 7af7687 | 2013-04-11 10:56:42 +0000 | [diff] [blame] | 23156 | /* -------------- (T1) LDRT reg+#imm8 -------------- */ |
| 23157 | /* Load Register Unprivileged: |
| 23158 | ldrt Rt, [Rn, #imm8] |
| 23159 | */ |
| 23160 | if (INSN0(15,6) == BITS10(1,1,1,1,1,0,0,0,0,1) && INSN0(5,4) == BITS2(0,1) |
| 23161 | && INSN1(11,8) == BITS4(1,1,1,0)) { |
| 23162 | UInt rT = INSN1(15,12); |
| 23163 | UInt rN = INSN0(3,0); |
| 23164 | UInt imm8 = INSN1(7,0); |
| 23165 | Bool valid = True; |
| 23166 | if (rN == 15 || isBadRegT(rT)) valid = False; |
| 23167 | if (valid) { |
| 23168 | put_ITSTATE(old_itstate); |
| 23169 | IRExpr* ea = binop(Iop_Add32, getIRegT(rN), mkU32(imm8)); |
| 23170 | IRTemp newRt = newTemp(Ity_I32); |
| 23171 | loadGuardedLE( newRt, ILGop_Ident32, ea, llGetIReg(rT), condT ); |
| 23172 | putIRegT(rT, mkexpr(newRt), IRTemp_INVALID); |
| 23173 | put_ITSTATE(new_itstate); |
| 23174 | DIP("ldrt r%u, [r%u, #%u]\n", rT, rN, imm8); |
| 23175 | goto decode_success; |
| 23176 | } |
| 23177 | } |
| 23178 | |
sewardj | 80f674a | 2013-05-13 10:28:59 +0000 | [diff] [blame] | 23179 | /* -------------- (T1) STRT reg+#imm8 -------------- */ |
| 23180 | /* Store Register Unprivileged: |
| 23181 | strt Rt, [Rn, #imm8] |
| 23182 | */ |
| 23183 | if (INSN0(15,6) == BITS10(1,1,1,1,1,0,0,0,0,1) && INSN0(5,4) == BITS2(0,0) |
| 23184 | && INSN1(11,8) == BITS4(1,1,1,0)) { |
| 23185 | UInt rT = INSN1(15,12); |
| 23186 | UInt rN = INSN0(3,0); |
| 23187 | UInt imm8 = INSN1(7,0); |
| 23188 | Bool valid = True; |
| 23189 | if (rN == 15 || isBadRegT(rT)) valid = False; |
| 23190 | if (valid) { |
| 23191 | put_ITSTATE(old_itstate); |
| 23192 | IRExpr* address = binop(Iop_Add32, getIRegT(rN), mkU32(imm8)); |
| 23193 | storeGuardedLE( address, llGetIReg(rT), condT ); |
| 23194 | put_ITSTATE(new_itstate); |
| 23195 | DIP("strt r%u, [r%u, #%u]\n", rT, rN, imm8); |
| 23196 | goto decode_success; |
| 23197 | } |
| 23198 | } |
| 23199 | |
sewardj | 6ed5c34 | 2013-07-04 15:25:36 +0000 | [diff] [blame] | 23200 | /* -------------- (T1) STRBT reg+#imm8 -------------- */ |
| 23201 | /* Store Register Byte Unprivileged: |
| 23202 | strbt Rt, [Rn, #imm8] |
| 23203 | */ |
| 23204 | if (INSN0(15,6) == BITS10(1,1,1,1,1,0,0,0,0,0) && INSN0(5,4) == BITS2(0,0) |
| 23205 | && INSN1(11,8) == BITS4(1,1,1,0)) { |
| 23206 | UInt rT = INSN1(15,12); |
| 23207 | UInt rN = INSN0(3,0); |
| 23208 | UInt imm8 = INSN1(7,0); |
| 23209 | Bool valid = True; |
| 23210 | if (rN == 15 || isBadRegT(rT)) valid = False; |
| 23211 | if (valid) { |
| 23212 | put_ITSTATE(old_itstate); |
| 23213 | IRExpr* address = binop(Iop_Add32, getIRegT(rN), mkU32(imm8)); |
| 23214 | IRExpr* data = unop(Iop_32to8, llGetIReg(rT)); |
| 23215 | storeGuardedLE( address, data, condT ); |
| 23216 | put_ITSTATE(new_itstate); |
| 23217 | DIP("strbt r%u, [r%u, #%u]\n", rT, rN, imm8); |
| 23218 | goto decode_success; |
| 23219 | } |
| 23220 | } |
| 23221 | |
sewardj | 3b22fc8 | 2013-09-29 18:24:15 +0000 | [diff] [blame] | 23222 | /* -------------- (T1) LDRHT reg+#imm8 -------------- */ |
| 23223 | /* Load Register Halfword Unprivileged: |
| 23224 | ldrht Rt, [Rn, #imm8] |
| 23225 | */ |
| 23226 | if (INSN0(15,6) == BITS10(1,1,1,1,1,0,0,0,0,0) && INSN0(5,4) == BITS2(1,1) |
| 23227 | && INSN1(11,8) == BITS4(1,1,1,0)) { |
| 23228 | UInt rN = INSN0(3,0); |
| 23229 | Bool valid = True; |
| 23230 | if (rN == 15) { |
| 23231 | /* In this case our instruction is LDRH (literal), in fact: |
| 23232 | LDRH (literal) was realized earlier, so we don't want to |
| 23233 | make it twice. */ |
| 23234 | valid = False; |
| 23235 | } |
| 23236 | UInt rT = INSN1(15,12); |
| 23237 | UInt imm8 = INSN1(7,0); |
| 23238 | if (isBadRegT(rT)) valid = False; |
| 23239 | if (valid) { |
| 23240 | put_ITSTATE(old_itstate); |
| 23241 | IRExpr* ea = binop(Iop_Add32, getIRegT(rN), mkU32(imm8)); |
| 23242 | IRTemp newRt = newTemp(Ity_I32); |
| 23243 | loadGuardedLE( newRt, ILGop_16Uto32, ea, llGetIReg(rT), condT ); |
| 23244 | putIRegT(rT, mkexpr(newRt), IRTemp_INVALID); |
| 23245 | put_ITSTATE(new_itstate); |
| 23246 | DIP("ldrht r%u, [r%u, #%u]\n", rT, rN, imm8); |
| 23247 | goto decode_success; |
| 23248 | } |
| 23249 | } |
| 23250 | |
| 23251 | /* -------------- (T1) LDRSHT reg+#imm8 -------------- */ |
| 23252 | /* Load Register Signed Halfword Unprivileged: |
| 23253 | ldrsht Rt, [Rn, #imm8] |
| 23254 | */ |
| 23255 | if (INSN0(15,6) == BITS10(1,1,1,1,1,0,0,1,0,0) && INSN0(5,4) == BITS2(1,1) |
| 23256 | && INSN1(11,8) == BITS4(1,1,1,0)) { |
| 23257 | UInt rN = INSN0(3,0); |
| 23258 | Bool valid = True; |
| 23259 | if (rN == 15) { |
| 23260 | /* In this case our instruction is LDRSH (literal), in fact: |
| 23261 | LDRSH (literal) was realized earlier, so we don't want to |
| 23262 | make it twice. */ |
| 23263 | valid = False; |
| 23264 | } |
| 23265 | UInt rT = INSN1(15,12); |
| 23266 | UInt imm8 = INSN1(7,0); |
| 23267 | if (isBadRegT(rT)) valid = False; |
| 23268 | if (valid) { |
| 23269 | put_ITSTATE(old_itstate); |
| 23270 | IRExpr* ea = binop(Iop_Add32, getIRegT(rN), mkU32(imm8)); |
| 23271 | IRTemp newRt = newTemp(Ity_I32); |
| 23272 | loadGuardedLE( newRt, ILGop_16Sto32, ea, llGetIReg(rT), condT ); |
| 23273 | putIRegT(rT, mkexpr(newRt), IRTemp_INVALID); |
| 23274 | put_ITSTATE(new_itstate); |
| 23275 | DIP("ldrsht r%u, [r%u, #%u]\n", rT, rN, imm8); |
| 23276 | goto decode_success; |
| 23277 | } |
| 23278 | } |
| 23279 | |
sewardj | 6ed5c34 | 2013-07-04 15:25:36 +0000 | [diff] [blame] | 23280 | /* -------------- (T1) STRHT reg+#imm8 -------------- */ |
| 23281 | /* Store Register Halfword Unprivileged: |
| 23282 | strht Rt, [Rn, #imm8] |
| 23283 | */ |
| 23284 | if (INSN0(15,6) == BITS10(1,1,1,1,1,0,0,0,0,0) && INSN0(5,4) == BITS2(1,0) |
| 23285 | && INSN1(11,8) == BITS4(1,1,1,0)) { |
| 23286 | UInt rT = INSN1(15,12); |
| 23287 | UInt rN = INSN0(3,0); |
| 23288 | UInt imm8 = INSN1(7,0); |
| 23289 | Bool valid = True; |
| 23290 | if (rN == 15 || isBadRegT(rT)) valid = False; |
| 23291 | if (valid) { |
| 23292 | put_ITSTATE(old_itstate); |
| 23293 | IRExpr* address = binop(Iop_Add32, getIRegT(rN), mkU32(imm8)); |
| 23294 | IRExpr* data = unop(Iop_32to16, llGetIReg(rT)); |
| 23295 | storeGuardedLE( address, data, condT ); |
| 23296 | put_ITSTATE(new_itstate); |
| 23297 | DIP("strht r%u, [r%u, #%u]\n", rT, rN, imm8); |
| 23298 | goto decode_success; |
| 23299 | } |
| 23300 | } |
| 23301 | |
| 23302 | /* -------------- (T1) LDRBT reg+#imm8 -------------- */ |
| 23303 | /* Load Register Byte Unprivileged: |
| 23304 | ldrbt Rt, [Rn, #imm8] |
| 23305 | */ |
| 23306 | if (INSN0(15,6) == BITS10(1,1,1,1,1,0,0,0,0,0) && INSN0(5,4) == BITS2(0,1) |
| 23307 | && INSN1(11,8) == BITS4(1,1,1,0)) { |
| 23308 | UInt rN = INSN0(3,0); |
| 23309 | UInt rT = INSN1(15,12); |
| 23310 | UInt imm8 = INSN1(7,0); |
| 23311 | Bool valid = True; |
| 23312 | if (rN == 15 /* insn is LDRB (literal) */) valid = False; |
| 23313 | if (isBadRegT(rT)) valid = False; |
| 23314 | if (valid) { |
| 23315 | put_ITSTATE(old_itstate); |
| 23316 | IRExpr* ea = binop(Iop_Add32, getIRegT(rN), mkU32(imm8)); |
| 23317 | IRTemp newRt = newTemp(Ity_I32); |
| 23318 | loadGuardedLE( newRt, ILGop_8Uto32, ea, llGetIReg(rT), condT ); |
| 23319 | putIRegT(rT, mkexpr(newRt), IRTemp_INVALID); |
| 23320 | put_ITSTATE(new_itstate); |
| 23321 | DIP("ldrbt r%u, [r%u, #%u]\n", rT, rN, imm8); |
| 23322 | goto decode_success; |
| 23323 | } |
| 23324 | } |
| 23325 | |
| 23326 | /* -------------- (T1) LDRSBT reg+#imm8 -------------- */ |
| 23327 | /* Load Register Signed Byte Unprivileged: |
| 23328 | ldrsbt Rt, [Rn, #imm8] |
| 23329 | */ |
| 23330 | if (INSN0(15,6) == BITS10(1,1,1,1,1,0,0,1,0,0) && INSN0(5,4) == BITS2(0,1) |
| 23331 | && INSN1(11,8) == BITS4(1,1,1,0)) { |
| 23332 | UInt rN = INSN0(3,0); |
| 23333 | Bool valid = True; |
| 23334 | UInt rT = INSN1(15,12); |
| 23335 | UInt imm8 = INSN1(7,0); |
| 23336 | if (rN == 15 /* insn is LDRSB (literal) */) valid = False; |
| 23337 | if (isBadRegT(rT)) valid = False; |
| 23338 | if (valid) { |
| 23339 | put_ITSTATE(old_itstate); |
| 23340 | IRExpr* ea = binop(Iop_Add32, getIRegT(rN), mkU32(imm8)); |
| 23341 | IRTemp newRt = newTemp(Ity_I32); |
| 23342 | loadGuardedLE( newRt, ILGop_8Sto32, ea, llGetIReg(rT), condT ); |
| 23343 | putIRegT(rT, mkexpr(newRt), IRTemp_INVALID); |
| 23344 | put_ITSTATE(new_itstate); |
| 23345 | DIP("ldrsbt r%u, [r%u, #%u]\n", rT, rN, imm8); |
| 23346 | goto decode_success; |
| 23347 | } |
| 23348 | } |
| 23349 | |
| 23350 | /* -------------- (T1) PLI reg+#imm12 -------------- */ |
| 23351 | /* Preload Instruction: |
| 23352 | pli [Rn, #imm12] |
| 23353 | */ |
| 23354 | if (INSN0(15,6) == BITS10(1,1,1,1,1,0,0,1,1,0) && INSN0(5,4) == BITS2(0,1) |
| 23355 | && INSN1(15,12) == BITS4(1,1,1,1)) { |
| 23356 | UInt rN = INSN0(3,0); |
| 23357 | UInt imm12 = INSN1(11,0); |
| 23358 | if (rN != 15) { |
| 23359 | DIP("pli [r%u, #%u]\n", rN, imm12); |
| 23360 | goto decode_success; |
| 23361 | } |
| 23362 | } |
| 23363 | |
| 23364 | /* -------------- (T2) PLI reg-#imm8 -------------- */ |
| 23365 | /* Preload Instruction: |
| 23366 | pli [Rn, #-imm8] |
| 23367 | */ |
| 23368 | if (INSN0(15,6) == BITS10(1,1,1,1,1,0,0,1,0,0) && INSN0(5,4) == BITS2(0,1) |
| 23369 | && INSN1(15,8) == BITS8(1,1,1,1,1,1,0,0)) { |
| 23370 | UInt rN = INSN0(3,0); |
| 23371 | UInt imm8 = INSN1(7,0); |
| 23372 | if (rN != 15) { |
| 23373 | DIP("pli [r%u, #-%u]\n", rN, imm8); |
| 23374 | goto decode_success; |
| 23375 | } |
| 23376 | } |
| 23377 | |
| 23378 | /* -------------- (T3) PLI PC+/-#imm12 -------------- */ |
| 23379 | /* Preload Instruction: |
| 23380 | pli [PC, #+/-imm12] |
| 23381 | */ |
| 23382 | if (INSN0(15,8) == BITS8(1,1,1,1,1,0,0,1) |
| 23383 | && INSN0(6,0) == BITS7(0,0,1,1,1,1,1) |
| 23384 | && INSN1(15,12) == BITS4(1,1,1,1)) { |
| 23385 | UInt imm12 = INSN1(11,0); |
| 23386 | UInt bU = INSN0(7,7); |
| 23387 | DIP("pli [pc, #%c%u]\n", bU == 1 ? '+' : '-', imm12); |
| 23388 | goto decode_success; |
| 23389 | } |
sewardj | 80f674a | 2013-05-13 10:28:59 +0000 | [diff] [blame] | 23390 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23391 | /* ----------------------------------------------------------- */ |
| 23392 | /* -- VFP (CP 10, CP 11) instructions (in Thumb mode) -- */ |
| 23393 | /* ----------------------------------------------------------- */ |
| 23394 | |
| 23395 | if (INSN0(15,12) == BITS4(1,1,1,0)) { |
| 23396 | UInt insn28 = (INSN0(11,0) << 16) | INSN1(15,0); |
| 23397 | Bool ok_vfp = decode_CP10_CP11_instruction ( |
| 23398 | &dres, insn28, condT, ARMCondAL/*bogus*/, |
| 23399 | True/*isT*/ |
| 23400 | ); |
| 23401 | if (ok_vfp) |
| 23402 | goto decode_success; |
| 23403 | } |
| 23404 | |
| 23405 | /* ----------------------------------------------------------- */ |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 23406 | /* -- NEON instructions (only v7 and below, in Thumb mode) -- */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23407 | /* ----------------------------------------------------------- */ |
| 23408 | |
sewardj | 1fce8de | 2010-09-09 07:27:24 +0000 | [diff] [blame] | 23409 | if (archinfo->hwcaps & VEX_HWCAPS_ARM_NEON) { |
| 23410 | UInt insn32 = (INSN0(15,0) << 16) | INSN1(15,0); |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 23411 | Bool ok_neon = decode_NEON_instruction_ARMv7_and_below( |
sewardj | 1fce8de | 2010-09-09 07:27:24 +0000 | [diff] [blame] | 23412 | &dres, insn32, condT, True/*isT*/ |
| 23413 | ); |
| 23414 | if (ok_neon) |
| 23415 | goto decode_success; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23416 | } |
| 23417 | |
| 23418 | /* ----------------------------------------------------------- */ |
sewardj | 1f139f5 | 2010-08-29 12:33:02 +0000 | [diff] [blame] | 23419 | /* -- v6 media instructions (in Thumb mode) -- */ |
| 23420 | /* ----------------------------------------------------------- */ |
| 23421 | |
| 23422 | { UInt insn32 = (INSN0(15,0) << 16) | INSN1(15,0); |
| 23423 | Bool ok_v6m = decode_V6MEDIA_instruction( |
| 23424 | &dres, insn32, condT, ARMCondAL/*bogus*/, |
| 23425 | True/*isT*/ |
| 23426 | ); |
| 23427 | if (ok_v6m) |
| 23428 | goto decode_success; |
| 23429 | } |
| 23430 | |
| 23431 | /* ----------------------------------------------------------- */ |
Elliott Hughes | a0664b9 | 2017-04-18 17:46:52 -0700 | [diff] [blame] | 23432 | /* -- v8 instructions (in Thumb mode) -- */ |
| 23433 | /* ----------------------------------------------------------- */ |
| 23434 | |
| 23435 | /* If we get here, it means that all attempts to decode the |
| 23436 | instruction as ARMv7 or earlier have failed. So, if we're doing |
| 23437 | ARMv8 or later, here is the point to try for it. */ |
| 23438 | |
| 23439 | if (VEX_ARM_ARCHLEVEL(archinfo->hwcaps) >= 8) { |
| 23440 | UInt insn32 = (INSN0(15,0) << 16) | INSN1(15,0); |
| 23441 | Bool ok_v8 |
| 23442 | = decode_V8_instruction( &dres, insn32, condT, True/*isT*/, |
| 23443 | old_itstate, new_itstate ); |
| 23444 | if (ok_v8) |
| 23445 | goto decode_success; |
| 23446 | } |
| 23447 | |
| 23448 | /* ----------------------------------------------------------- */ |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23449 | /* -- Undecodable -- */ |
| 23450 | /* ----------------------------------------------------------- */ |
| 23451 | |
| 23452 | goto decode_failure; |
| 23453 | /*NOTREACHED*/ |
| 23454 | |
| 23455 | decode_failure: |
| 23456 | /* All decode failures end up here. */ |
sewardj | 442e51a | 2012-12-06 18:08:04 +0000 | [diff] [blame] | 23457 | if (sigill_diag) |
| 23458 | vex_printf("disInstr(thumb): unhandled instruction: " |
| 23459 | "0x%04x 0x%04x\n", (UInt)insn0, (UInt)insn1); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23460 | |
| 23461 | /* Back up ITSTATE to the initial value for this instruction. |
| 23462 | If we don't do that, any subsequent restart of the instruction |
| 23463 | will restart with the wrong value. */ |
sewardj | 5276ff5 | 2012-07-14 14:21:56 +0000 | [diff] [blame] | 23464 | if (old_itstate != IRTemp_INVALID) |
| 23465 | put_ITSTATE(old_itstate); |
| 23466 | |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23467 | /* Tell the dispatcher that this insn cannot be decoded, and so has |
| 23468 | not been executed, and (is currently) the next to be executed. |
| 23469 | R15 should be up-to-date since it made so at the start of each |
| 23470 | insn, but nevertheless be paranoid and update it again right |
| 23471 | now. */ |
| 23472 | vassert(0 == (guest_R15_curr_instr_notENC & 1)); |
| 23473 | llPutIReg( 15, mkU32(guest_R15_curr_instr_notENC | 1) ); |
philippe | 2faf591 | 2014-08-11 22:45:47 +0000 | [diff] [blame] | 23474 | dres.len = 0; |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 23475 | dres.whatNext = Dis_StopHere; |
| 23476 | dres.jk_StopHere = Ijk_NoDecode; |
philippe | 2faf591 | 2014-08-11 22:45:47 +0000 | [diff] [blame] | 23477 | dres.continueAt = 0; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23478 | return dres; |
| 23479 | |
| 23480 | decode_success: |
| 23481 | /* All decode successes end up here. */ |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 23482 | vassert(dres.len == 4 || dres.len == 2 || dres.len == 20); |
| 23483 | switch (dres.whatNext) { |
| 23484 | case Dis_Continue: |
| 23485 | llPutIReg(15, mkU32(dres.len + (guest_R15_curr_instr_notENC | 1))); |
| 23486 | break; |
| 23487 | case Dis_ResteerU: |
| 23488 | case Dis_ResteerC: |
| 23489 | llPutIReg(15, mkU32(dres.continueAt)); |
| 23490 | break; |
| 23491 | case Dis_StopHere: |
| 23492 | break; |
| 23493 | default: |
| 23494 | vassert(0); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23495 | } |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 23496 | |
| 23497 | DIP("\n"); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23498 | |
| 23499 | return dres; |
| 23500 | |
| 23501 | # undef INSN0 |
| 23502 | # undef INSN1 |
| 23503 | } |
| 23504 | |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 23505 | #undef DIP |
| 23506 | #undef DIS |
| 23507 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 23508 | |
sewardj | 66c8c9b | 2011-07-04 16:58:40 +0000 | [diff] [blame] | 23509 | /* Helper table for figuring out how many insns an IT insn |
| 23510 | conditionalises. |
| 23511 | |
| 23512 | An ITxyz instruction of the format "1011 1111 firstcond mask" |
| 23513 | conditionalises some number of instructions, as indicated by the |
| 23514 | following table. A value of zero indicates the instruction is |
| 23515 | invalid in some way. |
| 23516 | |
| 23517 | mask = 0 means this isn't an IT instruction |
| 23518 | fc = 15 (NV) means unpredictable |
| 23519 | |
| 23520 | The line fc = 14 (AL) is different from the others; there are |
| 23521 | additional constraints in this case. |
| 23522 | |
| 23523 | mask(0 .. 15) |
| 23524 | +-------------------------------- |
| 23525 | fc(0 | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23526 | .. | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23527 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23528 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23529 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23530 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23531 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23532 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23533 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23534 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23535 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23536 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23537 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23538 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23539 | | 0 4 3 0 2 0 0 0 1 0 0 0 0 0 0 0 |
| 23540 | 15) | 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 |
| 23541 | |
| 23542 | To be conservative with the analysis, let's rule out the mask = 0 |
| 23543 | case, since that isn't an IT insn at all. But for all the other |
| 23544 | cases where the table contains zero, that means unpredictable, so |
| 23545 | let's say 4 to be conservative. Hence we have a safe value for any |
| 23546 | IT (mask,fc) pair that the CPU would actually identify as an IT |
| 23547 | instruction. The final table is |
| 23548 | |
| 23549 | mask(0 .. 15) |
| 23550 | +-------------------------------- |
| 23551 | fc(0 | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23552 | .. | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23553 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23554 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23555 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23556 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23557 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23558 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23559 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23560 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23561 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23562 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23563 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23564 | | 0 4 3 4 2 4 3 4 1 4 3 4 2 4 3 4 |
| 23565 | | 0 4 3 4 2 4 4 4 1 4 4 4 4 4 4 4 |
| 23566 | 15) | 0 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 |
| 23567 | */ |
| 23568 | static const UChar it_length_table[256] |
| 23569 | = { 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23570 | 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23571 | 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23572 | 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23573 | 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23574 | 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23575 | 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23576 | 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23577 | 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23578 | 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23579 | 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23580 | 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23581 | 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23582 | 0, 4, 3, 4, 2, 4, 3, 4, 1, 4, 3, 4, 2, 4, 3, 4, |
| 23583 | 0, 4, 3, 4, 2, 4, 4, 4, 1, 4, 4, 4, 4, 4, 4, 4, |
| 23584 | 0, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4 |
| 23585 | }; |
| 23586 | |
| 23587 | |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 23588 | /*------------------------------------------------------------*/ |
| 23589 | /*--- Top-level fn ---*/ |
| 23590 | /*------------------------------------------------------------*/ |
| 23591 | |
| 23592 | /* Disassemble a single instruction into IR. The instruction |
| 23593 | is located in host memory at &guest_code[delta]. */ |
| 23594 | |
| 23595 | DisResult disInstr_ARM ( IRSB* irsb_IN, |
florian | beac530 | 2014-12-31 12:09:38 +0000 | [diff] [blame] | 23596 | Bool (*resteerOkFn) ( void*, Addr ), |
sewardj | 984d9b1 | 2010-01-15 10:53:21 +0000 | [diff] [blame] | 23597 | Bool resteerCisOk, |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 23598 | void* callback_opaque, |
florian | 8462d11 | 2014-09-24 15:18:09 +0000 | [diff] [blame] | 23599 | const UChar* guest_code_IN, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23600 | Long delta_ENCODED, |
florian | d4cc0de | 2015-01-02 11:44:12 +0000 | [diff] [blame] | 23601 | Addr guest_IP_ENCODED, |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 23602 | VexArch guest_arch, |
florian | cacba8e | 2014-12-15 18:58:07 +0000 | [diff] [blame] | 23603 | const VexArchInfo* archinfo, |
| 23604 | const VexAbiInfo* abiinfo, |
sewardj | 9b76916 | 2014-07-24 12:42:03 +0000 | [diff] [blame] | 23605 | VexEndness host_endness_IN, |
sewardj | 442e51a | 2012-12-06 18:08:04 +0000 | [diff] [blame] | 23606 | Bool sigill_diag_IN ) |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 23607 | { |
| 23608 | DisResult dres; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23609 | Bool isThumb = (Bool)(guest_IP_ENCODED & 1); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 23610 | |
| 23611 | /* Set globals (see top of this file) */ |
| 23612 | vassert(guest_arch == VexArchARM); |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 23613 | |
sewardj | 9b76916 | 2014-07-24 12:42:03 +0000 | [diff] [blame] | 23614 | irsb = irsb_IN; |
| 23615 | host_endness = host_endness_IN; |
| 23616 | __curr_is_Thumb = isThumb; |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23617 | |
| 23618 | if (isThumb) { |
| 23619 | guest_R15_curr_instr_notENC = (Addr32)guest_IP_ENCODED - 1; |
| 23620 | } else { |
| 23621 | guest_R15_curr_instr_notENC = (Addr32)guest_IP_ENCODED; |
| 23622 | } |
| 23623 | |
| 23624 | if (isThumb) { |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 23625 | dres = disInstr_THUMB_WRK ( resteerOkFn, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23626 | resteerCisOk, callback_opaque, |
| 23627 | &guest_code_IN[delta_ENCODED - 1], |
sewardj | 442e51a | 2012-12-06 18:08:04 +0000 | [diff] [blame] | 23628 | archinfo, abiinfo, sigill_diag_IN ); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23629 | } else { |
sewardj | c6f970f | 2012-04-02 21:54:49 +0000 | [diff] [blame] | 23630 | dres = disInstr_ARM_WRK ( resteerOkFn, |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23631 | resteerCisOk, callback_opaque, |
| 23632 | &guest_code_IN[delta_ENCODED], |
sewardj | 442e51a | 2012-12-06 18:08:04 +0000 | [diff] [blame] | 23633 | archinfo, abiinfo, sigill_diag_IN ); |
sewardj | d266447 | 2010-08-22 12:44:20 +0000 | [diff] [blame] | 23634 | } |
sewardj | 6c299f3 | 2009-12-31 18:00:12 +0000 | [diff] [blame] | 23635 | |
| 23636 | return dres; |
| 23637 | } |
| 23638 | |
| 23639 | /* Test program for the conversion of IRCmpF64Result values to VFP |
| 23640 | nzcv values. See handling of FCMPD et al above. */ |
| 23641 | /* |
| 23642 | UInt foo ( UInt x ) |
| 23643 | { |
| 23644 | UInt ix = ((x >> 5) & 3) | (x & 1); |
| 23645 | UInt termL = (((((ix ^ 1) << 30) - 1) >> 29) + 1); |
| 23646 | UInt termR = (ix & (ix >> 1) & 1); |
| 23647 | return termL - termR; |
| 23648 | } |
| 23649 | |
| 23650 | void try ( char* s, UInt ir, UInt req ) |
| 23651 | { |
| 23652 | UInt act = foo(ir); |
| 23653 | printf("%s 0x%02x -> req %d%d%d%d act %d%d%d%d (0x%x)\n", |
| 23654 | s, ir, (req >> 3) & 1, (req >> 2) & 1, |
| 23655 | (req >> 1) & 1, (req >> 0) & 1, |
| 23656 | (act >> 3) & 1, (act >> 2) & 1, |
| 23657 | (act >> 1) & 1, (act >> 0) & 1, act); |
| 23658 | |
| 23659 | } |
| 23660 | |
| 23661 | int main ( void ) |
| 23662 | { |
| 23663 | printf("\n"); |
| 23664 | try("UN", 0x45, 0b0011); |
| 23665 | try("LT", 0x01, 0b1000); |
| 23666 | try("GT", 0x00, 0b0010); |
| 23667 | try("EQ", 0x40, 0b0110); |
| 23668 | printf("\n"); |
| 23669 | return 0; |
| 23670 | } |
| 23671 | */ |
| 23672 | |
sewardj | 1df8096 | 2013-04-18 11:50:58 +0000 | [diff] [blame] | 23673 | /* Spare code for doing reference implementations of various 64-bit |
| 23674 | SIMD interleaves/deinterleaves/concatenation ops. */ |
| 23675 | /* |
| 23676 | // Split a 64 bit value into 4 16 bit ones, in 32-bit IRTemps with |
| 23677 | // the top halves guaranteed to be zero. |
| 23678 | static void break64to16s ( IRTemp* out3, IRTemp* out2, IRTemp* out1, |
| 23679 | IRTemp* out0, IRTemp v64 ) |
| 23680 | { |
| 23681 | if (out3) *out3 = newTemp(Ity_I32); |
| 23682 | if (out2) *out2 = newTemp(Ity_I32); |
| 23683 | if (out1) *out1 = newTemp(Ity_I32); |
| 23684 | if (out0) *out0 = newTemp(Ity_I32); |
| 23685 | IRTemp hi32 = newTemp(Ity_I32); |
| 23686 | IRTemp lo32 = newTemp(Ity_I32); |
| 23687 | assign(hi32, unop(Iop_64HIto32, mkexpr(v64)) ); |
| 23688 | assign(lo32, unop(Iop_64to32, mkexpr(v64)) ); |
| 23689 | if (out3) assign(*out3, binop(Iop_Shr32, mkexpr(hi32), mkU8(16))); |
| 23690 | if (out2) assign(*out2, binop(Iop_And32, mkexpr(hi32), mkU32(0xFFFF))); |
| 23691 | if (out1) assign(*out1, binop(Iop_Shr32, mkexpr(lo32), mkU8(16))); |
| 23692 | if (out0) assign(*out0, binop(Iop_And32, mkexpr(lo32), mkU32(0xFFFF))); |
| 23693 | } |
| 23694 | |
| 23695 | // Make a 64 bit value from 4 16 bit ones, each of which is in a 32 bit |
| 23696 | // IRTemp. |
| 23697 | static IRTemp mk64from16s ( IRTemp in3, IRTemp in2, IRTemp in1, IRTemp in0 ) |
| 23698 | { |
| 23699 | IRTemp hi32 = newTemp(Ity_I32); |
| 23700 | IRTemp lo32 = newTemp(Ity_I32); |
| 23701 | assign(hi32, |
| 23702 | binop(Iop_Or32, |
| 23703 | binop(Iop_Shl32, mkexpr(in3), mkU8(16)), |
| 23704 | binop(Iop_And32, mkexpr(in2), mkU32(0xFFFF)))); |
| 23705 | assign(lo32, |
| 23706 | binop(Iop_Or32, |
| 23707 | binop(Iop_Shl32, mkexpr(in1), mkU8(16)), |
| 23708 | binop(Iop_And32, mkexpr(in0), mkU32(0xFFFF)))); |
| 23709 | IRTemp res = newTemp(Ity_I64); |
| 23710 | assign(res, binop(Iop_32HLto64, mkexpr(hi32), mkexpr(lo32))); |
| 23711 | return res; |
| 23712 | } |
| 23713 | |
| 23714 | static IRExpr* mk_InterleaveLO16x4 ( IRTemp a3210, IRTemp b3210 ) |
| 23715 | { |
| 23716 | // returns a1 b1 a0 b0 |
| 23717 | IRTemp a1, a0, b1, b0; |
| 23718 | break64to16s(NULL, NULL, &a1, &a0, a3210); |
| 23719 | break64to16s(NULL, NULL, &b1, &b0, b3210); |
| 23720 | return mkexpr(mk64from16s(a1, b1, a0, b0)); |
| 23721 | } |
| 23722 | |
| 23723 | static IRExpr* mk_InterleaveHI16x4 ( IRTemp a3210, IRTemp b3210 ) |
| 23724 | { |
| 23725 | // returns a3 b3 a2 b2 |
| 23726 | IRTemp a3, a2, b3, b2; |
| 23727 | break64to16s(&a3, &a2, NULL, NULL, a3210); |
| 23728 | break64to16s(&b3, &b2, NULL, NULL, b3210); |
| 23729 | return mkexpr(mk64from16s(a3, b3, a2, b2)); |
| 23730 | } |
| 23731 | |
| 23732 | static IRExpr* mk_CatEvenLanes16x4 ( IRTemp a3210, IRTemp b3210 ) |
| 23733 | { |
| 23734 | // returns a2 a0 b2 b0 |
| 23735 | IRTemp a2, a0, b2, b0; |
| 23736 | break64to16s(NULL, &a2, NULL, &a0, a3210); |
| 23737 | break64to16s(NULL, &b2, NULL, &b0, b3210); |
| 23738 | return mkexpr(mk64from16s(a2, a0, b2, b0)); |
| 23739 | } |
| 23740 | |
| 23741 | static IRExpr* mk_CatOddLanes16x4 ( IRTemp a3210, IRTemp b3210 ) |
| 23742 | { |
| 23743 | // returns a3 a1 b3 b1 |
| 23744 | IRTemp a3, a1, b3, b1; |
| 23745 | break64to16s(&a3, NULL, &a1, NULL, a3210); |
| 23746 | break64to16s(&b3, NULL, &b1, NULL, b3210); |
| 23747 | return mkexpr(mk64from16s(a3, a1, b3, b1)); |
| 23748 | } |
| 23749 | |
| 23750 | static IRExpr* mk_InterleaveOddLanes16x4 ( IRTemp a3210, IRTemp b3210 ) |
| 23751 | { |
| 23752 | // returns a3 b3 a1 b1 |
| 23753 | IRTemp a3, b3, a1, b1; |
| 23754 | break64to16s(&a3, NULL, &a1, NULL, a3210); |
| 23755 | break64to16s(&b3, NULL, &b1, NULL, b3210); |
| 23756 | return mkexpr(mk64from16s(a3, b3, a1, b1)); |
| 23757 | } |
| 23758 | |
| 23759 | static IRExpr* mk_InterleaveEvenLanes16x4 ( IRTemp a3210, IRTemp b3210 ) |
| 23760 | { |
| 23761 | // returns a2 b2 a0 b0 |
| 23762 | IRTemp a2, b2, a0, b0; |
| 23763 | break64to16s(NULL, &a2, NULL, &a0, a3210); |
| 23764 | break64to16s(NULL, &b2, NULL, &b0, b3210); |
| 23765 | return mkexpr(mk64from16s(a2, b2, a0, b0)); |
| 23766 | } |
| 23767 | |
| 23768 | static void break64to8s ( IRTemp* out7, IRTemp* out6, IRTemp* out5, |
| 23769 | IRTemp* out4, IRTemp* out3, IRTemp* out2, |
| 23770 | IRTemp* out1,IRTemp* out0, IRTemp v64 ) |
| 23771 | { |
| 23772 | if (out7) *out7 = newTemp(Ity_I32); |
| 23773 | if (out6) *out6 = newTemp(Ity_I32); |
| 23774 | if (out5) *out5 = newTemp(Ity_I32); |
| 23775 | if (out4) *out4 = newTemp(Ity_I32); |
| 23776 | if (out3) *out3 = newTemp(Ity_I32); |
| 23777 | if (out2) *out2 = newTemp(Ity_I32); |
| 23778 | if (out1) *out1 = newTemp(Ity_I32); |
| 23779 | if (out0) *out0 = newTemp(Ity_I32); |
| 23780 | IRTemp hi32 = newTemp(Ity_I32); |
| 23781 | IRTemp lo32 = newTemp(Ity_I32); |
| 23782 | assign(hi32, unop(Iop_64HIto32, mkexpr(v64)) ); |
| 23783 | assign(lo32, unop(Iop_64to32, mkexpr(v64)) ); |
| 23784 | if (out7) |
| 23785 | assign(*out7, binop(Iop_And32, |
| 23786 | binop(Iop_Shr32, mkexpr(hi32), mkU8(24)), |
| 23787 | mkU32(0xFF))); |
| 23788 | if (out6) |
| 23789 | assign(*out6, binop(Iop_And32, |
| 23790 | binop(Iop_Shr32, mkexpr(hi32), mkU8(16)), |
| 23791 | mkU32(0xFF))); |
| 23792 | if (out5) |
| 23793 | assign(*out5, binop(Iop_And32, |
| 23794 | binop(Iop_Shr32, mkexpr(hi32), mkU8(8)), |
| 23795 | mkU32(0xFF))); |
| 23796 | if (out4) |
| 23797 | assign(*out4, binop(Iop_And32, mkexpr(hi32), mkU32(0xFF))); |
| 23798 | if (out3) |
| 23799 | assign(*out3, binop(Iop_And32, |
| 23800 | binop(Iop_Shr32, mkexpr(lo32), mkU8(24)), |
| 23801 | mkU32(0xFF))); |
| 23802 | if (out2) |
| 23803 | assign(*out2, binop(Iop_And32, |
| 23804 | binop(Iop_Shr32, mkexpr(lo32), mkU8(16)), |
| 23805 | mkU32(0xFF))); |
| 23806 | if (out1) |
| 23807 | assign(*out1, binop(Iop_And32, |
| 23808 | binop(Iop_Shr32, mkexpr(lo32), mkU8(8)), |
| 23809 | mkU32(0xFF))); |
| 23810 | if (out0) |
| 23811 | assign(*out0, binop(Iop_And32, mkexpr(lo32), mkU32(0xFF))); |
| 23812 | } |
| 23813 | |
| 23814 | static IRTemp mk64from8s ( IRTemp in7, IRTemp in6, IRTemp in5, IRTemp in4, |
| 23815 | IRTemp in3, IRTemp in2, IRTemp in1, IRTemp in0 ) |
| 23816 | { |
| 23817 | IRTemp hi32 = newTemp(Ity_I32); |
| 23818 | IRTemp lo32 = newTemp(Ity_I32); |
| 23819 | assign(hi32, |
| 23820 | binop(Iop_Or32, |
| 23821 | binop(Iop_Or32, |
| 23822 | binop(Iop_Shl32, |
| 23823 | binop(Iop_And32, mkexpr(in7), mkU32(0xFF)), |
| 23824 | mkU8(24)), |
| 23825 | binop(Iop_Shl32, |
| 23826 | binop(Iop_And32, mkexpr(in6), mkU32(0xFF)), |
| 23827 | mkU8(16))), |
| 23828 | binop(Iop_Or32, |
| 23829 | binop(Iop_Shl32, |
| 23830 | binop(Iop_And32, mkexpr(in5), mkU32(0xFF)), mkU8(8)), |
| 23831 | binop(Iop_And32, |
| 23832 | mkexpr(in4), mkU32(0xFF))))); |
| 23833 | assign(lo32, |
| 23834 | binop(Iop_Or32, |
| 23835 | binop(Iop_Or32, |
| 23836 | binop(Iop_Shl32, |
| 23837 | binop(Iop_And32, mkexpr(in3), mkU32(0xFF)), |
| 23838 | mkU8(24)), |
| 23839 | binop(Iop_Shl32, |
| 23840 | binop(Iop_And32, mkexpr(in2), mkU32(0xFF)), |
| 23841 | mkU8(16))), |
| 23842 | binop(Iop_Or32, |
| 23843 | binop(Iop_Shl32, |
| 23844 | binop(Iop_And32, mkexpr(in1), mkU32(0xFF)), mkU8(8)), |
| 23845 | binop(Iop_And32, |
| 23846 | mkexpr(in0), mkU32(0xFF))))); |
| 23847 | IRTemp res = newTemp(Ity_I64); |
| 23848 | assign(res, binop(Iop_32HLto64, mkexpr(hi32), mkexpr(lo32))); |
| 23849 | return res; |
| 23850 | } |
| 23851 | |
| 23852 | static IRExpr* mk_InterleaveLO8x8 ( IRTemp a76543210, IRTemp b76543210 ) |
| 23853 | { |
| 23854 | // returns a3 b3 a2 b2 a1 b1 a0 b0 |
| 23855 | IRTemp a3, b3, a2, b2, a1, a0, b1, b0; |
| 23856 | break64to8s(NULL, NULL, NULL, NULL, &a3, &a2, &a1, &a0, a76543210); |
| 23857 | break64to8s(NULL, NULL, NULL, NULL, &b3, &b2, &b1, &b0, b76543210); |
| 23858 | return mkexpr(mk64from8s(a3, b3, a2, b2, a1, b1, a0, b0)); |
| 23859 | } |
| 23860 | |
| 23861 | static IRExpr* mk_InterleaveHI8x8 ( IRTemp a76543210, IRTemp b76543210 ) |
| 23862 | { |
| 23863 | // returns a7 b7 a6 b6 a5 b5 a4 b4 |
| 23864 | IRTemp a7, b7, a6, b6, a5, b5, a4, b4; |
| 23865 | break64to8s(&a7, &a6, &a5, &a4, NULL, NULL, NULL, NULL, a76543210); |
| 23866 | break64to8s(&b7, &b6, &b5, &b4, NULL, NULL, NULL, NULL, b76543210); |
| 23867 | return mkexpr(mk64from8s(a7, b7, a6, b6, a5, b5, a4, b4)); |
| 23868 | } |
| 23869 | |
| 23870 | static IRExpr* mk_CatEvenLanes8x8 ( IRTemp a76543210, IRTemp b76543210 ) |
| 23871 | { |
| 23872 | // returns a6 a4 a2 a0 b6 b4 b2 b0 |
| 23873 | IRTemp a6, a4, a2, a0, b6, b4, b2, b0; |
| 23874 | break64to8s(NULL, &a6, NULL, &a4, NULL, &a2, NULL, &a0, a76543210); |
| 23875 | break64to8s(NULL, &b6, NULL, &b4, NULL, &b2, NULL, &b0, b76543210); |
| 23876 | return mkexpr(mk64from8s(a6, a4, a2, a0, b6, b4, b2, b0)); |
| 23877 | } |
| 23878 | |
| 23879 | static IRExpr* mk_CatOddLanes8x8 ( IRTemp a76543210, IRTemp b76543210 ) |
| 23880 | { |
| 23881 | // returns a7 a5 a3 a1 b7 b5 b3 b1 |
| 23882 | IRTemp a7, a5, a3, a1, b7, b5, b3, b1; |
| 23883 | break64to8s(&a7, NULL, &a5, NULL, &a3, NULL, &a1, NULL, a76543210); |
| 23884 | break64to8s(&b7, NULL, &b5, NULL, &b3, NULL, &b1, NULL, b76543210); |
| 23885 | return mkexpr(mk64from8s(a7, a5, a3, a1, b7, b5, b3, b1)); |
| 23886 | } |
| 23887 | |
| 23888 | static IRExpr* mk_InterleaveEvenLanes8x8 ( IRTemp a76543210, IRTemp b76543210 ) |
| 23889 | { |
| 23890 | // returns a6 b6 a4 b4 a2 b2 a0 b0 |
| 23891 | IRTemp a6, b6, a4, b4, a2, b2, a0, b0; |
| 23892 | break64to8s(NULL, &a6, NULL, &a4, NULL, &a2, NULL, &a0, a76543210); |
| 23893 | break64to8s(NULL, &b6, NULL, &b4, NULL, &b2, NULL, &b0, b76543210); |
| 23894 | return mkexpr(mk64from8s(a6, b6, a4, b4, a2, b2, a0, b0)); |
| 23895 | } |
| 23896 | |
| 23897 | static IRExpr* mk_InterleaveOddLanes8x8 ( IRTemp a76543210, IRTemp b76543210 ) |
| 23898 | { |
| 23899 | // returns a7 b7 a5 b5 a3 b3 a1 b1 |
| 23900 | IRTemp a7, b7, a5, b5, a3, b3, a1, b1; |
| 23901 | break64to8s(&a7, NULL, &a5, NULL, &a3, NULL, &a1, NULL, a76543210); |
| 23902 | break64to8s(&b7, NULL, &b5, NULL, &b3, NULL, &b1, NULL, b76543210); |
| 23903 | return mkexpr(mk64from8s(a7, b7, a5, b5, a3, b3, a1, b1)); |
| 23904 | } |
| 23905 | |
| 23906 | static IRExpr* mk_InterleaveLO32x2 ( IRTemp a10, IRTemp b10 ) |
| 23907 | { |
| 23908 | // returns a0 b0 |
| 23909 | return binop(Iop_32HLto64, unop(Iop_64to32, mkexpr(a10)), |
| 23910 | unop(Iop_64to32, mkexpr(b10))); |
| 23911 | } |
| 23912 | |
| 23913 | static IRExpr* mk_InterleaveHI32x2 ( IRTemp a10, IRTemp b10 ) |
| 23914 | { |
| 23915 | // returns a1 b1 |
| 23916 | return binop(Iop_32HLto64, unop(Iop_64HIto32, mkexpr(a10)), |
| 23917 | unop(Iop_64HIto32, mkexpr(b10))); |
| 23918 | } |
| 23919 | */ |
| 23920 | |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 23921 | /*--------------------------------------------------------------------*/ |
sewardj | cef7d3e | 2009-07-02 12:21:59 +0000 | [diff] [blame] | 23922 | /*--- end guest_arm_toIR.c ---*/ |
sewardj | c2c8716 | 2004-11-25 13:07:02 +0000 | [diff] [blame] | 23923 | /*--------------------------------------------------------------------*/ |