sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1 | /* -*- mode: C; c-basic-offset: 3; -*- */ |
| 2 | |
| 3 | /*--------------------------------------------------------------------*/ |
| 4 | /*--- begin guest_arm64_toIR.c ---*/ |
| 5 | /*--------------------------------------------------------------------*/ |
| 6 | |
| 7 | /* |
| 8 | This file is part of Valgrind, a dynamic binary instrumentation |
| 9 | framework. |
| 10 | |
| 11 | Copyright (C) 2013-2013 OpenWorks |
| 12 | info@open-works.net |
| 13 | |
| 14 | This program is free software; you can redistribute it and/or |
| 15 | modify it under the terms of the GNU General Public License as |
| 16 | published by the Free Software Foundation; either version 2 of the |
| 17 | License, or (at your option) any later version. |
| 18 | |
| 19 | This program is distributed in the hope that it will be useful, but |
| 20 | WITHOUT ANY WARRANTY; without even the implied warranty of |
| 21 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 22 | General Public License for more details. |
| 23 | |
| 24 | You should have received a copy of the GNU General Public License |
| 25 | along with this program; if not, write to the Free Software |
| 26 | Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA |
| 27 | 02110-1301, USA. |
| 28 | |
| 29 | The GNU General Public License is contained in the file COPYING. |
| 30 | */ |
| 31 | |
| 32 | //ZZ /* XXXX thumb to check: |
| 33 | //ZZ that all cases where putIRegT writes r15, we generate a jump. |
| 34 | //ZZ |
| 35 | //ZZ All uses of newTemp assign to an IRTemp and not a UInt |
| 36 | //ZZ |
| 37 | //ZZ For all thumb loads and stores, including VFP ones, new-ITSTATE is |
| 38 | //ZZ backed out before the memory op, and restored afterwards. This |
| 39 | //ZZ needs to happen even after we go uncond. (and for sure it doesn't |
| 40 | //ZZ happen for VFP loads/stores right now). |
| 41 | //ZZ |
| 42 | //ZZ VFP on thumb: check that we exclude all r13/r15 cases that we |
| 43 | //ZZ should. |
| 44 | //ZZ |
| 45 | //ZZ XXXX thumb to do: improve the ITSTATE-zeroing optimisation by |
| 46 | //ZZ taking into account the number of insns guarded by an IT. |
| 47 | //ZZ |
| 48 | //ZZ remove the nasty hack, in the spechelper, of looking for Or32(..., |
| 49 | //ZZ 0xE0) in as the first arg to armg_calculate_condition, and instead |
| 50 | //ZZ use Slice44 as specified in comments in the spechelper. |
| 51 | //ZZ |
| 52 | //ZZ add specialisations for armg_calculate_flag_c and _v, as they |
| 53 | //ZZ are moderately often needed in Thumb code. |
| 54 | //ZZ |
| 55 | //ZZ Correctness: ITSTATE handling in Thumb SVCs is wrong. |
| 56 | //ZZ |
| 57 | //ZZ Correctness (obscure): in m_transtab, when invalidating code |
| 58 | //ZZ address ranges, invalidate up to 18 bytes after the end of the |
| 59 | //ZZ range. This is because the ITSTATE optimisation at the top of |
| 60 | //ZZ _THUMB_WRK below analyses up to 18 bytes before the start of any |
| 61 | //ZZ given instruction, and so might depend on the invalidated area. |
| 62 | //ZZ */ |
| 63 | //ZZ |
| 64 | //ZZ /* Limitations, etc |
| 65 | //ZZ |
| 66 | //ZZ - pretty dodgy exception semantics for {LD,ST}Mxx and {LD,ST}RD. |
| 67 | //ZZ These instructions are non-restartable in the case where the |
| 68 | //ZZ transfer(s) fault. |
| 69 | //ZZ |
| 70 | //ZZ - SWP: the restart jump back is Ijk_Boring; it should be |
| 71 | //ZZ Ijk_NoRedir but that's expensive. See comments on casLE() in |
| 72 | //ZZ guest_x86_toIR.c. |
| 73 | //ZZ */ |
| 74 | |
| 75 | /* "Special" instructions. |
| 76 | |
| 77 | This instruction decoder can decode four special instructions |
| 78 | which mean nothing natively (are no-ops as far as regs/mem are |
| 79 | concerned) but have meaning for supporting Valgrind. A special |
| 80 | instruction is flagged by a 16-byte preamble: |
| 81 | |
| 82 | 93CC0D8C 93CC358C 93CCCD8C 93CCF58C |
| 83 | (ror x12, x12, #3; ror x12, x12, #13 |
| 84 | ror x12, x12, #51; ror x12, x12, #61) |
| 85 | |
| 86 | Following that, one of the following 3 are allowed |
| 87 | (standard interpretation in parentheses): |
| 88 | |
| 89 | AA0A014A (orr x10,x10,x10) X3 = client_request ( X4 ) |
| 90 | AA0B016B (orr x11,x11,x11) X3 = guest_NRADDR |
| 91 | AA0C018C (orr x12,x12,x12) branch-and-link-to-noredir X8 |
| 92 | AA090129 (orr x9,x9,x9) IR injection |
| 93 | |
| 94 | Any other bytes following the 16-byte preamble are illegal and |
| 95 | constitute a failure in instruction decoding. This all assumes |
| 96 | that the preamble will never occur except in specific code |
| 97 | fragments designed for Valgrind to catch. |
| 98 | */ |
| 99 | |
| 100 | /* Translates ARM64 code to IR. */ |
| 101 | |
| 102 | #include "libvex_basictypes.h" |
| 103 | #include "libvex_ir.h" |
| 104 | #include "libvex.h" |
| 105 | #include "libvex_guest_arm64.h" |
| 106 | |
| 107 | #include "main_util.h" |
| 108 | #include "main_globals.h" |
| 109 | #include "guest_generic_bb_to_IR.h" |
| 110 | #include "guest_arm64_defs.h" |
| 111 | |
| 112 | |
| 113 | /*------------------------------------------------------------*/ |
| 114 | /*--- Globals ---*/ |
| 115 | /*------------------------------------------------------------*/ |
| 116 | |
| 117 | /* These are set at the start of the translation of a instruction, so |
| 118 | that we don't have to pass them around endlessly. CONST means does |
| 119 | not change during translation of the instruction. |
| 120 | */ |
| 121 | |
sewardj | 9b76916 | 2014-07-24 12:42:03 +0000 | [diff] [blame] | 122 | /* CONST: what is the host's endianness? We need to know this in |
| 123 | order to do sub-register accesses to the SIMD/FP registers |
| 124 | correctly. */ |
| 125 | static VexEndness host_endness; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 126 | |
| 127 | /* CONST: The guest address for the instruction currently being |
| 128 | translated. */ |
| 129 | static Addr64 guest_PC_curr_instr; |
| 130 | |
| 131 | /* MOD: The IRSB* into which we're generating code. */ |
| 132 | static IRSB* irsb; |
| 133 | |
| 134 | |
| 135 | /*------------------------------------------------------------*/ |
| 136 | /*--- Debugging output ---*/ |
| 137 | /*------------------------------------------------------------*/ |
| 138 | |
| 139 | #define DIP(format, args...) \ |
| 140 | if (vex_traceflags & VEX_TRACE_FE) \ |
| 141 | vex_printf(format, ## args) |
| 142 | |
| 143 | #define DIS(buf, format, args...) \ |
| 144 | if (vex_traceflags & VEX_TRACE_FE) \ |
| 145 | vex_sprintf(buf, format, ## args) |
| 146 | |
| 147 | |
| 148 | /*------------------------------------------------------------*/ |
| 149 | /*--- Helper bits and pieces for deconstructing the ---*/ |
| 150 | /*--- arm insn stream. ---*/ |
| 151 | /*------------------------------------------------------------*/ |
| 152 | |
| 153 | /* Do a little-endian load of a 32-bit word, regardless of the |
| 154 | endianness of the underlying host. */ |
| 155 | static inline UInt getUIntLittleEndianly ( UChar* p ) |
| 156 | { |
| 157 | UInt w = 0; |
| 158 | w = (w << 8) | p[3]; |
| 159 | w = (w << 8) | p[2]; |
| 160 | w = (w << 8) | p[1]; |
| 161 | w = (w << 8) | p[0]; |
| 162 | return w; |
| 163 | } |
| 164 | |
| 165 | /* Sign extend a N-bit value up to 64 bits, by copying |
| 166 | bit N-1 into all higher positions. */ |
| 167 | static ULong sx_to_64 ( ULong x, UInt n ) |
| 168 | { |
| 169 | vassert(n > 1 && n < 64); |
| 170 | Long r = (Long)x; |
| 171 | r = (r << (64-n)) >> (64-n); |
| 172 | return (ULong)r; |
| 173 | } |
| 174 | |
| 175 | //ZZ /* Do a little-endian load of a 16-bit word, regardless of the |
| 176 | //ZZ endianness of the underlying host. */ |
| 177 | //ZZ static inline UShort getUShortLittleEndianly ( UChar* p ) |
| 178 | //ZZ { |
| 179 | //ZZ UShort w = 0; |
| 180 | //ZZ w = (w << 8) | p[1]; |
| 181 | //ZZ w = (w << 8) | p[0]; |
| 182 | //ZZ return w; |
| 183 | //ZZ } |
| 184 | //ZZ |
| 185 | //ZZ static UInt ROR32 ( UInt x, UInt sh ) { |
| 186 | //ZZ vassert(sh >= 0 && sh < 32); |
| 187 | //ZZ if (sh == 0) |
| 188 | //ZZ return x; |
| 189 | //ZZ else |
| 190 | //ZZ return (x << (32-sh)) | (x >> sh); |
| 191 | //ZZ } |
| 192 | //ZZ |
| 193 | //ZZ static Int popcount32 ( UInt x ) |
| 194 | //ZZ { |
| 195 | //ZZ Int res = 0, i; |
| 196 | //ZZ for (i = 0; i < 32; i++) { |
| 197 | //ZZ res += (x & 1); |
| 198 | //ZZ x >>= 1; |
| 199 | //ZZ } |
| 200 | //ZZ return res; |
| 201 | //ZZ } |
| 202 | //ZZ |
| 203 | //ZZ static UInt setbit32 ( UInt x, Int ix, UInt b ) |
| 204 | //ZZ { |
| 205 | //ZZ UInt mask = 1 << ix; |
| 206 | //ZZ x &= ~mask; |
| 207 | //ZZ x |= ((b << ix) & mask); |
| 208 | //ZZ return x; |
| 209 | //ZZ } |
| 210 | |
| 211 | #define BITS2(_b1,_b0) \ |
| 212 | (((_b1) << 1) | (_b0)) |
| 213 | |
| 214 | #define BITS3(_b2,_b1,_b0) \ |
| 215 | (((_b2) << 2) | ((_b1) << 1) | (_b0)) |
| 216 | |
| 217 | #define BITS4(_b3,_b2,_b1,_b0) \ |
| 218 | (((_b3) << 3) | ((_b2) << 2) | ((_b1) << 1) | (_b0)) |
| 219 | |
| 220 | #define BITS8(_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 221 | ((BITS4((_b7),(_b6),(_b5),(_b4)) << 4) \ |
| 222 | | BITS4((_b3),(_b2),(_b1),(_b0))) |
| 223 | |
| 224 | #define BITS5(_b4,_b3,_b2,_b1,_b0) \ |
| 225 | (BITS8(0,0,0,(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 226 | #define BITS6(_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 227 | (BITS8(0,0,(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 228 | #define BITS7(_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 229 | (BITS8(0,(_b6),(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 230 | |
| 231 | #define BITS9(_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 232 | (((_b8) << 8) \ |
| 233 | | BITS8((_b7),(_b6),(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 234 | |
| 235 | #define BITS10(_b9,_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 236 | (((_b9) << 9) | ((_b8) << 8) \ |
| 237 | | BITS8((_b7),(_b6),(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 238 | |
| 239 | #define BITS11(_b10,_b9,_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 240 | (((_b10) << 10) \ |
| 241 | | BITS10(_b9,_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0)) |
| 242 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 243 | #define BITS12(_b11, _b10,_b9,_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 244 | (((_b11) << 11) \ |
| 245 | | BITS11(_b10,_b9,_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0)) |
| 246 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 247 | #define X00 BITS2(0,0) |
| 248 | #define X01 BITS2(0,1) |
| 249 | #define X10 BITS2(1,0) |
| 250 | #define X11 BITS2(1,1) |
| 251 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 252 | // produces _uint[_bMax:_bMin] |
| 253 | #define SLICE_UInt(_uint,_bMax,_bMin) \ |
| 254 | (( ((UInt)(_uint)) >> (_bMin)) \ |
| 255 | & (UInt)((1ULL << ((_bMax) - (_bMin) + 1)) - 1ULL)) |
| 256 | |
| 257 | |
| 258 | /*------------------------------------------------------------*/ |
| 259 | /*--- Helper bits and pieces for creating IR fragments. ---*/ |
| 260 | /*------------------------------------------------------------*/ |
| 261 | |
| 262 | static IRExpr* mkV128 ( UShort w ) |
| 263 | { |
| 264 | return IRExpr_Const(IRConst_V128(w)); |
| 265 | } |
| 266 | |
| 267 | static IRExpr* mkU64 ( ULong i ) |
| 268 | { |
| 269 | return IRExpr_Const(IRConst_U64(i)); |
| 270 | } |
| 271 | |
| 272 | static IRExpr* mkU32 ( UInt i ) |
| 273 | { |
| 274 | return IRExpr_Const(IRConst_U32(i)); |
| 275 | } |
| 276 | |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 277 | static IRExpr* mkU16 ( UInt i ) |
| 278 | { |
| 279 | vassert(i < 65536); |
| 280 | return IRExpr_Const(IRConst_U16(i)); |
| 281 | } |
| 282 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 283 | static IRExpr* mkU8 ( UInt i ) |
| 284 | { |
| 285 | vassert(i < 256); |
| 286 | return IRExpr_Const(IRConst_U8( (UChar)i )); |
| 287 | } |
| 288 | |
| 289 | static IRExpr* mkexpr ( IRTemp tmp ) |
| 290 | { |
| 291 | return IRExpr_RdTmp(tmp); |
| 292 | } |
| 293 | |
| 294 | static IRExpr* unop ( IROp op, IRExpr* a ) |
| 295 | { |
| 296 | return IRExpr_Unop(op, a); |
| 297 | } |
| 298 | |
| 299 | static IRExpr* binop ( IROp op, IRExpr* a1, IRExpr* a2 ) |
| 300 | { |
| 301 | return IRExpr_Binop(op, a1, a2); |
| 302 | } |
| 303 | |
| 304 | static IRExpr* triop ( IROp op, IRExpr* a1, IRExpr* a2, IRExpr* a3 ) |
| 305 | { |
| 306 | return IRExpr_Triop(op, a1, a2, a3); |
| 307 | } |
| 308 | |
| 309 | static IRExpr* loadLE ( IRType ty, IRExpr* addr ) |
| 310 | { |
| 311 | return IRExpr_Load(Iend_LE, ty, addr); |
| 312 | } |
| 313 | |
| 314 | /* Add a statement to the list held by "irbb". */ |
| 315 | static void stmt ( IRStmt* st ) |
| 316 | { |
| 317 | addStmtToIRSB( irsb, st ); |
| 318 | } |
| 319 | |
| 320 | static void assign ( IRTemp dst, IRExpr* e ) |
| 321 | { |
| 322 | stmt( IRStmt_WrTmp(dst, e) ); |
| 323 | } |
| 324 | |
| 325 | static void storeLE ( IRExpr* addr, IRExpr* data ) |
| 326 | { |
| 327 | stmt( IRStmt_Store(Iend_LE, addr, data) ); |
| 328 | } |
| 329 | |
| 330 | //ZZ static void storeGuardedLE ( IRExpr* addr, IRExpr* data, IRTemp guardT ) |
| 331 | //ZZ { |
| 332 | //ZZ if (guardT == IRTemp_INVALID) { |
| 333 | //ZZ /* unconditional */ |
| 334 | //ZZ storeLE(addr, data); |
| 335 | //ZZ } else { |
| 336 | //ZZ stmt( IRStmt_StoreG(Iend_LE, addr, data, |
| 337 | //ZZ binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0))) ); |
| 338 | //ZZ } |
| 339 | //ZZ } |
| 340 | //ZZ |
| 341 | //ZZ static void loadGuardedLE ( IRTemp dst, IRLoadGOp cvt, |
| 342 | //ZZ IRExpr* addr, IRExpr* alt, |
| 343 | //ZZ IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
| 344 | //ZZ { |
| 345 | //ZZ if (guardT == IRTemp_INVALID) { |
| 346 | //ZZ /* unconditional */ |
| 347 | //ZZ IRExpr* loaded = NULL; |
| 348 | //ZZ switch (cvt) { |
| 349 | //ZZ case ILGop_Ident32: |
| 350 | //ZZ loaded = loadLE(Ity_I32, addr); break; |
| 351 | //ZZ case ILGop_8Uto32: |
| 352 | //ZZ loaded = unop(Iop_8Uto32, loadLE(Ity_I8, addr)); break; |
| 353 | //ZZ case ILGop_8Sto32: |
| 354 | //ZZ loaded = unop(Iop_8Sto32, loadLE(Ity_I8, addr)); break; |
| 355 | //ZZ case ILGop_16Uto32: |
| 356 | //ZZ loaded = unop(Iop_16Uto32, loadLE(Ity_I16, addr)); break; |
| 357 | //ZZ case ILGop_16Sto32: |
| 358 | //ZZ loaded = unop(Iop_16Sto32, loadLE(Ity_I16, addr)); break; |
| 359 | //ZZ default: |
| 360 | //ZZ vassert(0); |
| 361 | //ZZ } |
| 362 | //ZZ vassert(loaded != NULL); |
| 363 | //ZZ assign(dst, loaded); |
| 364 | //ZZ } else { |
| 365 | //ZZ /* Generate a guarded load into 'dst', but apply 'cvt' to the |
| 366 | //ZZ loaded data before putting the data in 'dst'. If the load |
| 367 | //ZZ does not take place, 'alt' is placed directly in 'dst'. */ |
| 368 | //ZZ stmt( IRStmt_LoadG(Iend_LE, cvt, dst, addr, alt, |
| 369 | //ZZ binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0))) ); |
| 370 | //ZZ } |
| 371 | //ZZ } |
| 372 | |
| 373 | /* Generate a new temporary of the given type. */ |
| 374 | static IRTemp newTemp ( IRType ty ) |
| 375 | { |
| 376 | vassert(isPlausibleIRType(ty)); |
| 377 | return newIRTemp( irsb->tyenv, ty ); |
| 378 | } |
| 379 | |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 380 | /* This is used in many places, so the brevity is an advantage. */ |
| 381 | static IRTemp newTempV128(void) |
| 382 | { |
| 383 | return newTemp(Ity_V128); |
| 384 | } |
| 385 | |
| 386 | /* Initialise V128 temporaries en masse. */ |
| 387 | static |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 388 | void newTempsV128_2(IRTemp* t1, IRTemp* t2) |
| 389 | { |
| 390 | vassert(t1 && *t1 == IRTemp_INVALID); |
| 391 | vassert(t2 && *t2 == IRTemp_INVALID); |
| 392 | *t1 = newTempV128(); |
| 393 | *t2 = newTempV128(); |
| 394 | } |
| 395 | |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 396 | static |
| 397 | void newTempsV128_3(IRTemp* t1, IRTemp* t2, IRTemp* t3) |
| 398 | { |
| 399 | vassert(t1 && *t1 == IRTemp_INVALID); |
| 400 | vassert(t2 && *t2 == IRTemp_INVALID); |
| 401 | vassert(t3 && *t3 == IRTemp_INVALID); |
| 402 | *t1 = newTempV128(); |
| 403 | *t2 = newTempV128(); |
| 404 | *t3 = newTempV128(); |
| 405 | } |
| 406 | |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 407 | //static |
| 408 | //void newTempsV128_4(IRTemp* t1, IRTemp* t2, IRTemp* t3, IRTemp* t4) |
| 409 | //{ |
| 410 | // vassert(t1 && *t1 == IRTemp_INVALID); |
| 411 | // vassert(t2 && *t2 == IRTemp_INVALID); |
| 412 | // vassert(t3 && *t3 == IRTemp_INVALID); |
| 413 | // vassert(t4 && *t4 == IRTemp_INVALID); |
| 414 | // *t1 = newTempV128(); |
| 415 | // *t2 = newTempV128(); |
| 416 | // *t3 = newTempV128(); |
| 417 | // *t4 = newTempV128(); |
| 418 | //} |
| 419 | |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 420 | static |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 421 | void newTempsV128_7(IRTemp* t1, IRTemp* t2, IRTemp* t3, |
| 422 | IRTemp* t4, IRTemp* t5, IRTemp* t6, IRTemp* t7) |
| 423 | { |
| 424 | vassert(t1 && *t1 == IRTemp_INVALID); |
| 425 | vassert(t2 && *t2 == IRTemp_INVALID); |
| 426 | vassert(t3 && *t3 == IRTemp_INVALID); |
| 427 | vassert(t4 && *t4 == IRTemp_INVALID); |
| 428 | vassert(t5 && *t5 == IRTemp_INVALID); |
| 429 | vassert(t6 && *t6 == IRTemp_INVALID); |
| 430 | vassert(t7 && *t7 == IRTemp_INVALID); |
| 431 | *t1 = newTempV128(); |
| 432 | *t2 = newTempV128(); |
| 433 | *t3 = newTempV128(); |
| 434 | *t4 = newTempV128(); |
| 435 | *t5 = newTempV128(); |
| 436 | *t6 = newTempV128(); |
| 437 | *t7 = newTempV128(); |
| 438 | } |
| 439 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 440 | //ZZ /* Produces a value in 0 .. 3, which is encoded as per the type |
| 441 | //ZZ IRRoundingMode. */ |
| 442 | //ZZ static IRExpr* /* :: Ity_I32 */ get_FAKE_roundingmode ( void ) |
| 443 | //ZZ { |
| 444 | //ZZ return mkU32(Irrm_NEAREST); |
| 445 | //ZZ } |
| 446 | //ZZ |
| 447 | //ZZ /* Generate an expression for SRC rotated right by ROT. */ |
| 448 | //ZZ static IRExpr* genROR32( IRTemp src, Int rot ) |
| 449 | //ZZ { |
| 450 | //ZZ vassert(rot >= 0 && rot < 32); |
| 451 | //ZZ if (rot == 0) |
| 452 | //ZZ return mkexpr(src); |
| 453 | //ZZ return |
| 454 | //ZZ binop(Iop_Or32, |
| 455 | //ZZ binop(Iop_Shl32, mkexpr(src), mkU8(32 - rot)), |
| 456 | //ZZ binop(Iop_Shr32, mkexpr(src), mkU8(rot))); |
| 457 | //ZZ } |
| 458 | //ZZ |
| 459 | //ZZ static IRExpr* mkU128 ( ULong i ) |
| 460 | //ZZ { |
| 461 | //ZZ return binop(Iop_64HLtoV128, mkU64(i), mkU64(i)); |
| 462 | //ZZ } |
| 463 | //ZZ |
| 464 | //ZZ /* Generate a 4-aligned version of the given expression if |
| 465 | //ZZ the given condition is true. Else return it unchanged. */ |
| 466 | //ZZ static IRExpr* align4if ( IRExpr* e, Bool b ) |
| 467 | //ZZ { |
| 468 | //ZZ if (b) |
| 469 | //ZZ return binop(Iop_And32, e, mkU32(~3)); |
| 470 | //ZZ else |
| 471 | //ZZ return e; |
| 472 | //ZZ } |
| 473 | |
| 474 | /* Other IR construction helpers. */ |
| 475 | static IROp mkAND ( IRType ty ) { |
| 476 | switch (ty) { |
| 477 | case Ity_I32: return Iop_And32; |
| 478 | case Ity_I64: return Iop_And64; |
| 479 | default: vpanic("mkAND"); |
| 480 | } |
| 481 | } |
| 482 | |
| 483 | static IROp mkOR ( IRType ty ) { |
| 484 | switch (ty) { |
| 485 | case Ity_I32: return Iop_Or32; |
| 486 | case Ity_I64: return Iop_Or64; |
| 487 | default: vpanic("mkOR"); |
| 488 | } |
| 489 | } |
| 490 | |
| 491 | static IROp mkXOR ( IRType ty ) { |
| 492 | switch (ty) { |
| 493 | case Ity_I32: return Iop_Xor32; |
| 494 | case Ity_I64: return Iop_Xor64; |
| 495 | default: vpanic("mkXOR"); |
| 496 | } |
| 497 | } |
| 498 | |
| 499 | static IROp mkSHL ( IRType ty ) { |
| 500 | switch (ty) { |
| 501 | case Ity_I32: return Iop_Shl32; |
| 502 | case Ity_I64: return Iop_Shl64; |
| 503 | default: vpanic("mkSHL"); |
| 504 | } |
| 505 | } |
| 506 | |
| 507 | static IROp mkSHR ( IRType ty ) { |
| 508 | switch (ty) { |
| 509 | case Ity_I32: return Iop_Shr32; |
| 510 | case Ity_I64: return Iop_Shr64; |
| 511 | default: vpanic("mkSHR"); |
| 512 | } |
| 513 | } |
| 514 | |
| 515 | static IROp mkSAR ( IRType ty ) { |
| 516 | switch (ty) { |
| 517 | case Ity_I32: return Iop_Sar32; |
| 518 | case Ity_I64: return Iop_Sar64; |
| 519 | default: vpanic("mkSAR"); |
| 520 | } |
| 521 | } |
| 522 | |
| 523 | static IROp mkNOT ( IRType ty ) { |
| 524 | switch (ty) { |
| 525 | case Ity_I32: return Iop_Not32; |
| 526 | case Ity_I64: return Iop_Not64; |
| 527 | default: vpanic("mkNOT"); |
| 528 | } |
| 529 | } |
| 530 | |
| 531 | static IROp mkADD ( IRType ty ) { |
| 532 | switch (ty) { |
| 533 | case Ity_I32: return Iop_Add32; |
| 534 | case Ity_I64: return Iop_Add64; |
| 535 | default: vpanic("mkADD"); |
| 536 | } |
| 537 | } |
| 538 | |
| 539 | static IROp mkSUB ( IRType ty ) { |
| 540 | switch (ty) { |
| 541 | case Ity_I32: return Iop_Sub32; |
| 542 | case Ity_I64: return Iop_Sub64; |
| 543 | default: vpanic("mkSUB"); |
| 544 | } |
| 545 | } |
| 546 | |
| 547 | static IROp mkADDF ( IRType ty ) { |
| 548 | switch (ty) { |
| 549 | case Ity_F32: return Iop_AddF32; |
| 550 | case Ity_F64: return Iop_AddF64; |
| 551 | default: vpanic("mkADDF"); |
| 552 | } |
| 553 | } |
| 554 | |
| 555 | static IROp mkSUBF ( IRType ty ) { |
| 556 | switch (ty) { |
| 557 | case Ity_F32: return Iop_SubF32; |
| 558 | case Ity_F64: return Iop_SubF64; |
| 559 | default: vpanic("mkSUBF"); |
| 560 | } |
| 561 | } |
| 562 | |
| 563 | static IROp mkMULF ( IRType ty ) { |
| 564 | switch (ty) { |
| 565 | case Ity_F32: return Iop_MulF32; |
| 566 | case Ity_F64: return Iop_MulF64; |
| 567 | default: vpanic("mkMULF"); |
| 568 | } |
| 569 | } |
| 570 | |
| 571 | static IROp mkDIVF ( IRType ty ) { |
| 572 | switch (ty) { |
| 573 | case Ity_F32: return Iop_DivF32; |
| 574 | case Ity_F64: return Iop_DivF64; |
| 575 | default: vpanic("mkMULF"); |
| 576 | } |
| 577 | } |
| 578 | |
| 579 | static IROp mkNEGF ( IRType ty ) { |
| 580 | switch (ty) { |
| 581 | case Ity_F32: return Iop_NegF32; |
| 582 | case Ity_F64: return Iop_NegF64; |
| 583 | default: vpanic("mkNEGF"); |
| 584 | } |
| 585 | } |
| 586 | |
| 587 | static IROp mkABSF ( IRType ty ) { |
| 588 | switch (ty) { |
| 589 | case Ity_F32: return Iop_AbsF32; |
| 590 | case Ity_F64: return Iop_AbsF64; |
| 591 | default: vpanic("mkNEGF"); |
| 592 | } |
| 593 | } |
| 594 | |
| 595 | static IROp mkSQRTF ( IRType ty ) { |
| 596 | switch (ty) { |
| 597 | case Ity_F32: return Iop_SqrtF32; |
| 598 | case Ity_F64: return Iop_SqrtF64; |
| 599 | default: vpanic("mkNEGF"); |
| 600 | } |
| 601 | } |
| 602 | |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 603 | static IROp mkVecADD ( UInt size ) { |
| 604 | const IROp ops[4] |
| 605 | = { Iop_Add8x16, Iop_Add16x8, Iop_Add32x4, Iop_Add64x2 }; |
| 606 | vassert(size < 4); |
| 607 | return ops[size]; |
| 608 | } |
| 609 | |
| 610 | static IROp mkVecQADDU ( UInt size ) { |
| 611 | const IROp ops[4] |
| 612 | = { Iop_QAdd8Ux16, Iop_QAdd16Ux8, Iop_QAdd32Ux4, Iop_QAdd64Ux2 }; |
| 613 | vassert(size < 4); |
| 614 | return ops[size]; |
| 615 | } |
| 616 | |
| 617 | static IROp mkVecQADDS ( UInt size ) { |
| 618 | const IROp ops[4] |
| 619 | = { Iop_QAdd8Sx16, Iop_QAdd16Sx8, Iop_QAdd32Sx4, Iop_QAdd64Sx2 }; |
| 620 | vassert(size < 4); |
| 621 | return ops[size]; |
| 622 | } |
| 623 | |
| 624 | static IROp mkVecSUB ( UInt size ) { |
| 625 | const IROp ops[4] |
| 626 | = { Iop_Sub8x16, Iop_Sub16x8, Iop_Sub32x4, Iop_Sub64x2 }; |
| 627 | vassert(size < 4); |
| 628 | return ops[size]; |
| 629 | } |
| 630 | |
| 631 | static IROp mkVecQSUBU ( UInt size ) { |
| 632 | const IROp ops[4] |
| 633 | = { Iop_QSub8Ux16, Iop_QSub16Ux8, Iop_QSub32Ux4, Iop_QSub64Ux2 }; |
| 634 | vassert(size < 4); |
| 635 | return ops[size]; |
| 636 | } |
| 637 | |
| 638 | static IROp mkVecQSUBS ( UInt size ) { |
| 639 | const IROp ops[4] |
| 640 | = { Iop_QSub8Sx16, Iop_QSub16Sx8, Iop_QSub32Sx4, Iop_QSub64Sx2 }; |
| 641 | vassert(size < 4); |
| 642 | return ops[size]; |
| 643 | } |
| 644 | |
| 645 | static IROp mkVecSARN ( UInt size ) { |
| 646 | const IROp ops[4] |
| 647 | = { Iop_SarN8x16, Iop_SarN16x8, Iop_SarN32x4, Iop_SarN64x2 }; |
| 648 | vassert(size < 4); |
| 649 | return ops[size]; |
| 650 | } |
| 651 | |
| 652 | static IROp mkVecSHRN ( UInt size ) { |
| 653 | const IROp ops[4] |
| 654 | = { Iop_ShrN8x16, Iop_ShrN16x8, Iop_ShrN32x4, Iop_ShrN64x2 }; |
| 655 | vassert(size < 4); |
| 656 | return ops[size]; |
| 657 | } |
| 658 | |
| 659 | static IROp mkVecSHLN ( UInt size ) { |
| 660 | const IROp ops[4] |
| 661 | = { Iop_ShlN8x16, Iop_ShlN16x8, Iop_ShlN32x4, Iop_ShlN64x2 }; |
| 662 | vassert(size < 4); |
| 663 | return ops[size]; |
| 664 | } |
| 665 | |
| 666 | static IROp mkVecCATEVENLANES ( UInt size ) { |
| 667 | const IROp ops[4] |
| 668 | = { Iop_CatEvenLanes8x16, Iop_CatEvenLanes16x8, |
| 669 | Iop_CatEvenLanes32x4, Iop_InterleaveLO64x2 }; |
| 670 | vassert(size < 4); |
| 671 | return ops[size]; |
| 672 | } |
| 673 | |
| 674 | static IROp mkVecCATODDLANES ( UInt size ) { |
| 675 | const IROp ops[4] |
| 676 | = { Iop_CatOddLanes8x16, Iop_CatOddLanes16x8, |
| 677 | Iop_CatOddLanes32x4, Iop_InterleaveHI64x2 }; |
| 678 | vassert(size < 4); |
| 679 | return ops[size]; |
| 680 | } |
| 681 | |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 682 | static IROp mkVecINTERLEAVELO ( UInt size ) { |
| 683 | const IROp ops[4] |
| 684 | = { Iop_InterleaveLO8x16, Iop_InterleaveLO16x8, |
| 685 | Iop_InterleaveLO32x4, Iop_InterleaveLO64x2 }; |
| 686 | vassert(size < 4); |
| 687 | return ops[size]; |
| 688 | } |
| 689 | |
| 690 | static IROp mkVecINTERLEAVEHI ( UInt size ) { |
| 691 | const IROp ops[4] |
| 692 | = { Iop_InterleaveHI8x16, Iop_InterleaveHI16x8, |
| 693 | Iop_InterleaveHI32x4, Iop_InterleaveHI64x2 }; |
| 694 | vassert(size < 4); |
| 695 | return ops[size]; |
| 696 | } |
| 697 | |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 698 | static IROp mkVecMAXU ( UInt size ) { |
| 699 | const IROp ops[4] |
| 700 | = { Iop_Max8Ux16, Iop_Max16Ux8, Iop_Max32Ux4, Iop_Max64Ux2 }; |
| 701 | vassert(size < 4); |
| 702 | return ops[size]; |
| 703 | } |
| 704 | |
| 705 | static IROp mkVecMAXS ( UInt size ) { |
| 706 | const IROp ops[4] |
| 707 | = { Iop_Max8Sx16, Iop_Max16Sx8, Iop_Max32Sx4, Iop_Max64Sx2 }; |
| 708 | vassert(size < 4); |
| 709 | return ops[size]; |
| 710 | } |
| 711 | |
| 712 | static IROp mkVecMINU ( UInt size ) { |
| 713 | const IROp ops[4] |
| 714 | = { Iop_Min8Ux16, Iop_Min16Ux8, Iop_Min32Ux4, Iop_Min64Ux2 }; |
| 715 | vassert(size < 4); |
| 716 | return ops[size]; |
| 717 | } |
| 718 | |
| 719 | static IROp mkVecMINS ( UInt size ) { |
| 720 | const IROp ops[4] |
| 721 | = { Iop_Min8Sx16, Iop_Min16Sx8, Iop_Min32Sx4, Iop_Min64Sx2 }; |
| 722 | vassert(size < 4); |
| 723 | return ops[size]; |
| 724 | } |
| 725 | |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 726 | static IROp mkVecMUL ( UInt size ) { |
| 727 | const IROp ops[4] |
| 728 | = { Iop_Mul8x16, Iop_Mul16x8, Iop_Mul32x4, Iop_INVALID }; |
| 729 | vassert(size < 3); |
| 730 | return ops[size]; |
| 731 | } |
| 732 | |
| 733 | static IROp mkVecMULLU ( UInt sizeNarrow ) { |
| 734 | const IROp ops[4] |
| 735 | = { Iop_Mull8Ux8, Iop_Mull16Ux4, Iop_Mull32Ux2, Iop_INVALID }; |
| 736 | vassert(sizeNarrow < 3); |
| 737 | return ops[sizeNarrow]; |
| 738 | } |
| 739 | |
| 740 | static IROp mkVecMULLS ( UInt sizeNarrow ) { |
| 741 | const IROp ops[4] |
| 742 | = { Iop_Mull8Sx8, Iop_Mull16Sx4, Iop_Mull32Sx2, Iop_INVALID }; |
| 743 | vassert(sizeNarrow < 3); |
| 744 | return ops[sizeNarrow]; |
| 745 | } |
| 746 | |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 747 | static IROp mkVecQDMULLS ( UInt sizeNarrow ) { |
| 748 | const IROp ops[4] |
| 749 | = { Iop_INVALID, Iop_QDMull16Sx4, Iop_QDMull32Sx2, Iop_INVALID }; |
| 750 | vassert(sizeNarrow < 3); |
| 751 | return ops[sizeNarrow]; |
| 752 | } |
| 753 | |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 754 | static IROp mkVecCMPEQ ( UInt size ) { |
| 755 | const IROp ops[4] |
| 756 | = { Iop_CmpEQ8x16, Iop_CmpEQ16x8, Iop_CmpEQ32x4, Iop_CmpEQ64x2 }; |
| 757 | vassert(size < 4); |
| 758 | return ops[size]; |
| 759 | } |
| 760 | |
| 761 | static IROp mkVecCMPGTU ( UInt size ) { |
| 762 | const IROp ops[4] |
| 763 | = { Iop_CmpGT8Ux16, Iop_CmpGT16Ux8, Iop_CmpGT32Ux4, Iop_CmpGT64Ux2 }; |
| 764 | vassert(size < 4); |
| 765 | return ops[size]; |
| 766 | } |
| 767 | |
| 768 | static IROp mkVecCMPGTS ( UInt size ) { |
| 769 | const IROp ops[4] |
| 770 | = { Iop_CmpGT8Sx16, Iop_CmpGT16Sx8, Iop_CmpGT32Sx4, Iop_CmpGT64Sx2 }; |
| 771 | vassert(size < 4); |
| 772 | return ops[size]; |
| 773 | } |
| 774 | |
| 775 | static IROp mkVecABS ( UInt size ) { |
| 776 | const IROp ops[4] |
| 777 | = { Iop_Abs8x16, Iop_Abs16x8, Iop_Abs32x4, Iop_Abs64x2 }; |
| 778 | vassert(size < 4); |
| 779 | return ops[size]; |
| 780 | } |
| 781 | |
| 782 | static IROp mkVecZEROHIxxOFV128 ( UInt size ) { |
| 783 | const IROp ops[4] |
| 784 | = { Iop_ZeroHI120ofV128, Iop_ZeroHI112ofV128, |
| 785 | Iop_ZeroHI96ofV128, Iop_ZeroHI64ofV128 }; |
| 786 | vassert(size < 4); |
| 787 | return ops[size]; |
| 788 | } |
| 789 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 790 | static IRExpr* mkU ( IRType ty, ULong imm ) { |
| 791 | switch (ty) { |
| 792 | case Ity_I32: return mkU32((UInt)(imm & 0xFFFFFFFFULL)); |
| 793 | case Ity_I64: return mkU64(imm); |
| 794 | default: vpanic("mkU"); |
| 795 | } |
| 796 | } |
| 797 | |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 798 | static IROp mkVecQDMULHIS ( UInt size ) { |
| 799 | const IROp ops[4] |
| 800 | = { Iop_INVALID, Iop_QDMulHi16Sx8, Iop_QDMulHi32Sx4, Iop_INVALID }; |
| 801 | vassert(size < 4); |
| 802 | return ops[size]; |
| 803 | } |
| 804 | |
| 805 | static IROp mkVecQRDMULHIS ( UInt size ) { |
| 806 | const IROp ops[4] |
| 807 | = { Iop_INVALID, Iop_QRDMulHi16Sx8, Iop_QRDMulHi32Sx4, Iop_INVALID }; |
| 808 | vassert(size < 4); |
| 809 | return ops[size]; |
| 810 | } |
| 811 | |
sewardj | ecedd98 | 2014-08-11 14:02:47 +0000 | [diff] [blame] | 812 | static IROp mkVecQANDUQSH ( UInt size ) { |
sewardj | 1297218 | 2014-08-04 08:09:47 +0000 | [diff] [blame] | 813 | const IROp ops[4] |
| 814 | = { Iop_QandUQsh8x16, Iop_QandUQsh16x8, |
| 815 | Iop_QandUQsh32x4, Iop_QandUQsh64x2 }; |
| 816 | vassert(size < 4); |
| 817 | return ops[size]; |
| 818 | } |
| 819 | |
sewardj | ecedd98 | 2014-08-11 14:02:47 +0000 | [diff] [blame] | 820 | static IROp mkVecQANDSQSH ( UInt size ) { |
sewardj | 1297218 | 2014-08-04 08:09:47 +0000 | [diff] [blame] | 821 | const IROp ops[4] |
| 822 | = { Iop_QandSQsh8x16, Iop_QandSQsh16x8, |
| 823 | Iop_QandSQsh32x4, Iop_QandSQsh64x2 }; |
| 824 | vassert(size < 4); |
| 825 | return ops[size]; |
| 826 | } |
| 827 | |
sewardj | ecedd98 | 2014-08-11 14:02:47 +0000 | [diff] [blame] | 828 | static IROp mkVecQANDUQRSH ( UInt size ) { |
sewardj | 1297218 | 2014-08-04 08:09:47 +0000 | [diff] [blame] | 829 | const IROp ops[4] |
| 830 | = { Iop_QandUQRsh8x16, Iop_QandUQRsh16x8, |
| 831 | Iop_QandUQRsh32x4, Iop_QandUQRsh64x2 }; |
| 832 | vassert(size < 4); |
| 833 | return ops[size]; |
| 834 | } |
| 835 | |
sewardj | ecedd98 | 2014-08-11 14:02:47 +0000 | [diff] [blame] | 836 | static IROp mkVecQANDSQRSH ( UInt size ) { |
sewardj | 1297218 | 2014-08-04 08:09:47 +0000 | [diff] [blame] | 837 | const IROp ops[4] |
| 838 | = { Iop_QandSQRsh8x16, Iop_QandSQRsh16x8, |
| 839 | Iop_QandSQRsh32x4, Iop_QandSQRsh64x2 }; |
| 840 | vassert(size < 4); |
| 841 | return ops[size]; |
| 842 | } |
| 843 | |
sewardj | ecedd98 | 2014-08-11 14:02:47 +0000 | [diff] [blame] | 844 | static IROp mkVecNARROWUN ( UInt sizeNarrow ) { |
| 845 | const IROp ops[4] |
| 846 | = { Iop_NarrowUn16to8x8, Iop_NarrowUn32to16x4, |
| 847 | Iop_NarrowUn64to32x2, Iop_INVALID }; |
| 848 | vassert(sizeNarrow < 4); |
| 849 | return ops[sizeNarrow]; |
| 850 | } |
| 851 | |
| 852 | static IROp mkVecQNARROWUNSU ( UInt sizeNarrow ) { |
| 853 | const IROp ops[4] |
| 854 | = { Iop_QNarrowUn16Sto8Ux8, Iop_QNarrowUn32Sto16Ux4, |
| 855 | Iop_QNarrowUn64Sto32Ux2, Iop_INVALID }; |
| 856 | vassert(sizeNarrow < 4); |
| 857 | return ops[sizeNarrow]; |
| 858 | } |
| 859 | |
| 860 | static IROp mkVecQNARROWUNSS ( UInt sizeNarrow ) { |
| 861 | const IROp ops[4] |
| 862 | = { Iop_QNarrowUn16Sto8Sx8, Iop_QNarrowUn32Sto16Sx4, |
| 863 | Iop_QNarrowUn64Sto32Sx2, Iop_INVALID }; |
| 864 | vassert(sizeNarrow < 4); |
| 865 | return ops[sizeNarrow]; |
| 866 | } |
| 867 | |
| 868 | static IROp mkVecQNARROWUNUU ( UInt sizeNarrow ) { |
| 869 | const IROp ops[4] |
| 870 | = { Iop_QNarrowUn16Uto8Ux8, Iop_QNarrowUn32Uto16Ux4, |
| 871 | Iop_QNarrowUn64Uto32Ux2, Iop_INVALID }; |
| 872 | vassert(sizeNarrow < 4); |
| 873 | return ops[sizeNarrow]; |
| 874 | } |
| 875 | |
| 876 | static IROp mkVecQANDqshrNNARROWUU ( UInt sizeNarrow ) { |
| 877 | const IROp ops[4] |
| 878 | = { Iop_QandQShrNnarrow16Uto8Ux8, Iop_QandQShrNnarrow32Uto16Ux4, |
| 879 | Iop_QandQShrNnarrow64Uto32Ux2, Iop_INVALID }; |
| 880 | vassert(sizeNarrow < 4); |
| 881 | return ops[sizeNarrow]; |
| 882 | } |
| 883 | |
| 884 | static IROp mkVecQANDqsarNNARROWSS ( UInt sizeNarrow ) { |
| 885 | const IROp ops[4] |
| 886 | = { Iop_QandQSarNnarrow16Sto8Sx8, Iop_QandQSarNnarrow32Sto16Sx4, |
| 887 | Iop_QandQSarNnarrow64Sto32Sx2, Iop_INVALID }; |
| 888 | vassert(sizeNarrow < 4); |
| 889 | return ops[sizeNarrow]; |
| 890 | } |
| 891 | |
| 892 | static IROp mkVecQANDqsarNNARROWSU ( UInt sizeNarrow ) { |
| 893 | const IROp ops[4] |
| 894 | = { Iop_QandQSarNnarrow16Sto8Ux8, Iop_QandQSarNnarrow32Sto16Ux4, |
| 895 | Iop_QandQSarNnarrow64Sto32Ux2, Iop_INVALID }; |
| 896 | vassert(sizeNarrow < 4); |
| 897 | return ops[sizeNarrow]; |
| 898 | } |
| 899 | |
| 900 | static IROp mkVecQANDqrshrNNARROWUU ( UInt sizeNarrow ) { |
| 901 | const IROp ops[4] |
| 902 | = { Iop_QandQRShrNnarrow16Uto8Ux8, Iop_QandQRShrNnarrow32Uto16Ux4, |
| 903 | Iop_QandQRShrNnarrow64Uto32Ux2, Iop_INVALID }; |
| 904 | vassert(sizeNarrow < 4); |
| 905 | return ops[sizeNarrow]; |
| 906 | } |
| 907 | |
| 908 | static IROp mkVecQANDqrsarNNARROWSS ( UInt sizeNarrow ) { |
| 909 | const IROp ops[4] |
| 910 | = { Iop_QandQRSarNnarrow16Sto8Sx8, Iop_QandQRSarNnarrow32Sto16Sx4, |
| 911 | Iop_QandQRSarNnarrow64Sto32Sx2, Iop_INVALID }; |
| 912 | vassert(sizeNarrow < 4); |
| 913 | return ops[sizeNarrow]; |
| 914 | } |
| 915 | |
| 916 | static IROp mkVecQANDqrsarNNARROWSU ( UInt sizeNarrow ) { |
| 917 | const IROp ops[4] |
| 918 | = { Iop_QandQRSarNnarrow16Sto8Ux8, Iop_QandQRSarNnarrow32Sto16Ux4, |
| 919 | Iop_QandQRSarNnarrow64Sto32Ux2, Iop_INVALID }; |
| 920 | vassert(sizeNarrow < 4); |
| 921 | return ops[sizeNarrow]; |
| 922 | } |
| 923 | |
sewardj | a97dddf | 2014-08-14 22:26:52 +0000 | [diff] [blame^] | 924 | static IROp mkVecQSHLNSATU2U ( UInt size ) { |
| 925 | const IROp ops[4] |
| 926 | = { Iop_QShlN8x16, Iop_QShlN16x8, Iop_QShlN32x4, Iop_QShlN64x2 }; |
| 927 | vassert(size < 4); |
| 928 | return ops[size]; |
| 929 | } |
| 930 | |
| 931 | static IROp mkVecQSHLNSATS2S ( UInt size ) { |
| 932 | const IROp ops[4] |
| 933 | = { Iop_QSalN8x16, Iop_QSalN16x8, Iop_QSalN32x4, Iop_QSalN64x2 }; |
| 934 | vassert(size < 4); |
| 935 | return ops[size]; |
| 936 | } |
| 937 | |
| 938 | static IROp mkVecQSHLNSATS2U ( UInt size ) { |
| 939 | const IROp ops[4] |
| 940 | = { Iop_QShlN8Sx16, Iop_QShlN16Sx8, Iop_QShlN32Sx4, Iop_QShlN64Sx2 }; |
| 941 | vassert(size < 4); |
| 942 | return ops[size]; |
| 943 | } |
| 944 | |
| 945 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 946 | /* Generate IR to create 'arg rotated right by imm', for sane values |
| 947 | of 'ty' and 'imm'. */ |
| 948 | static IRTemp mathROR ( IRType ty, IRTemp arg, UInt imm ) |
| 949 | { |
| 950 | UInt w = 0; |
| 951 | if (ty == Ity_I64) { |
| 952 | w = 64; |
| 953 | } else { |
| 954 | vassert(ty == Ity_I32); |
| 955 | w = 32; |
| 956 | } |
| 957 | vassert(w != 0); |
| 958 | vassert(imm < w); |
| 959 | if (imm == 0) { |
| 960 | return arg; |
| 961 | } |
| 962 | IRTemp res = newTemp(ty); |
| 963 | assign(res, binop(mkOR(ty), |
| 964 | binop(mkSHL(ty), mkexpr(arg), mkU8(w - imm)), |
| 965 | binop(mkSHR(ty), mkexpr(arg), mkU8(imm)) )); |
| 966 | return res; |
| 967 | } |
| 968 | |
| 969 | /* Generate IR to set the returned temp to either all-zeroes or |
| 970 | all ones, as a copy of arg<imm>. */ |
| 971 | static IRTemp mathREPLICATE ( IRType ty, IRTemp arg, UInt imm ) |
| 972 | { |
| 973 | UInt w = 0; |
| 974 | if (ty == Ity_I64) { |
| 975 | w = 64; |
| 976 | } else { |
| 977 | vassert(ty == Ity_I32); |
| 978 | w = 32; |
| 979 | } |
| 980 | vassert(w != 0); |
| 981 | vassert(imm < w); |
| 982 | IRTemp res = newTemp(ty); |
| 983 | assign(res, binop(mkSAR(ty), |
| 984 | binop(mkSHL(ty), mkexpr(arg), mkU8(w - 1 - imm)), |
| 985 | mkU8(w - 1))); |
| 986 | return res; |
| 987 | } |
| 988 | |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 989 | /* U-widen 8/16/32/64 bit int expr to 64. */ |
| 990 | static IRExpr* widenUto64 ( IRType srcTy, IRExpr* e ) |
| 991 | { |
| 992 | switch (srcTy) { |
| 993 | case Ity_I64: return e; |
| 994 | case Ity_I32: return unop(Iop_32Uto64, e); |
| 995 | case Ity_I16: return unop(Iop_16Uto64, e); |
| 996 | case Ity_I8: return unop(Iop_8Uto64, e); |
| 997 | default: vpanic("widenUto64(arm64)"); |
| 998 | } |
| 999 | } |
| 1000 | |
| 1001 | /* Narrow 64 bit int expr to 8/16/32/64. Clearly only some |
| 1002 | of these combinations make sense. */ |
| 1003 | static IRExpr* narrowFrom64 ( IRType dstTy, IRExpr* e ) |
| 1004 | { |
| 1005 | switch (dstTy) { |
| 1006 | case Ity_I64: return e; |
| 1007 | case Ity_I32: return unop(Iop_64to32, e); |
| 1008 | case Ity_I16: return unop(Iop_64to16, e); |
| 1009 | case Ity_I8: return unop(Iop_64to8, e); |
| 1010 | default: vpanic("narrowFrom64(arm64)"); |
| 1011 | } |
| 1012 | } |
| 1013 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1014 | |
| 1015 | /*------------------------------------------------------------*/ |
| 1016 | /*--- Helpers for accessing guest registers. ---*/ |
| 1017 | /*------------------------------------------------------------*/ |
| 1018 | |
| 1019 | #define OFFB_X0 offsetof(VexGuestARM64State,guest_X0) |
| 1020 | #define OFFB_X1 offsetof(VexGuestARM64State,guest_X1) |
| 1021 | #define OFFB_X2 offsetof(VexGuestARM64State,guest_X2) |
| 1022 | #define OFFB_X3 offsetof(VexGuestARM64State,guest_X3) |
| 1023 | #define OFFB_X4 offsetof(VexGuestARM64State,guest_X4) |
| 1024 | #define OFFB_X5 offsetof(VexGuestARM64State,guest_X5) |
| 1025 | #define OFFB_X6 offsetof(VexGuestARM64State,guest_X6) |
| 1026 | #define OFFB_X7 offsetof(VexGuestARM64State,guest_X7) |
| 1027 | #define OFFB_X8 offsetof(VexGuestARM64State,guest_X8) |
| 1028 | #define OFFB_X9 offsetof(VexGuestARM64State,guest_X9) |
| 1029 | #define OFFB_X10 offsetof(VexGuestARM64State,guest_X10) |
| 1030 | #define OFFB_X11 offsetof(VexGuestARM64State,guest_X11) |
| 1031 | #define OFFB_X12 offsetof(VexGuestARM64State,guest_X12) |
| 1032 | #define OFFB_X13 offsetof(VexGuestARM64State,guest_X13) |
| 1033 | #define OFFB_X14 offsetof(VexGuestARM64State,guest_X14) |
| 1034 | #define OFFB_X15 offsetof(VexGuestARM64State,guest_X15) |
| 1035 | #define OFFB_X16 offsetof(VexGuestARM64State,guest_X16) |
| 1036 | #define OFFB_X17 offsetof(VexGuestARM64State,guest_X17) |
| 1037 | #define OFFB_X18 offsetof(VexGuestARM64State,guest_X18) |
| 1038 | #define OFFB_X19 offsetof(VexGuestARM64State,guest_X19) |
| 1039 | #define OFFB_X20 offsetof(VexGuestARM64State,guest_X20) |
| 1040 | #define OFFB_X21 offsetof(VexGuestARM64State,guest_X21) |
| 1041 | #define OFFB_X22 offsetof(VexGuestARM64State,guest_X22) |
| 1042 | #define OFFB_X23 offsetof(VexGuestARM64State,guest_X23) |
| 1043 | #define OFFB_X24 offsetof(VexGuestARM64State,guest_X24) |
| 1044 | #define OFFB_X25 offsetof(VexGuestARM64State,guest_X25) |
| 1045 | #define OFFB_X26 offsetof(VexGuestARM64State,guest_X26) |
| 1046 | #define OFFB_X27 offsetof(VexGuestARM64State,guest_X27) |
| 1047 | #define OFFB_X28 offsetof(VexGuestARM64State,guest_X28) |
| 1048 | #define OFFB_X29 offsetof(VexGuestARM64State,guest_X29) |
| 1049 | #define OFFB_X30 offsetof(VexGuestARM64State,guest_X30) |
| 1050 | |
sewardj | 6068788 | 2014-01-15 10:25:21 +0000 | [diff] [blame] | 1051 | #define OFFB_XSP offsetof(VexGuestARM64State,guest_XSP) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1052 | #define OFFB_PC offsetof(VexGuestARM64State,guest_PC) |
| 1053 | |
| 1054 | #define OFFB_CC_OP offsetof(VexGuestARM64State,guest_CC_OP) |
| 1055 | #define OFFB_CC_DEP1 offsetof(VexGuestARM64State,guest_CC_DEP1) |
| 1056 | #define OFFB_CC_DEP2 offsetof(VexGuestARM64State,guest_CC_DEP2) |
| 1057 | #define OFFB_CC_NDEP offsetof(VexGuestARM64State,guest_CC_NDEP) |
| 1058 | |
| 1059 | #define OFFB_TPIDR_EL0 offsetof(VexGuestARM64State,guest_TPIDR_EL0) |
| 1060 | #define OFFB_NRADDR offsetof(VexGuestARM64State,guest_NRADDR) |
| 1061 | |
| 1062 | #define OFFB_Q0 offsetof(VexGuestARM64State,guest_Q0) |
| 1063 | #define OFFB_Q1 offsetof(VexGuestARM64State,guest_Q1) |
| 1064 | #define OFFB_Q2 offsetof(VexGuestARM64State,guest_Q2) |
| 1065 | #define OFFB_Q3 offsetof(VexGuestARM64State,guest_Q3) |
| 1066 | #define OFFB_Q4 offsetof(VexGuestARM64State,guest_Q4) |
| 1067 | #define OFFB_Q5 offsetof(VexGuestARM64State,guest_Q5) |
| 1068 | #define OFFB_Q6 offsetof(VexGuestARM64State,guest_Q6) |
| 1069 | #define OFFB_Q7 offsetof(VexGuestARM64State,guest_Q7) |
| 1070 | #define OFFB_Q8 offsetof(VexGuestARM64State,guest_Q8) |
| 1071 | #define OFFB_Q9 offsetof(VexGuestARM64State,guest_Q9) |
| 1072 | #define OFFB_Q10 offsetof(VexGuestARM64State,guest_Q10) |
| 1073 | #define OFFB_Q11 offsetof(VexGuestARM64State,guest_Q11) |
| 1074 | #define OFFB_Q12 offsetof(VexGuestARM64State,guest_Q12) |
| 1075 | #define OFFB_Q13 offsetof(VexGuestARM64State,guest_Q13) |
| 1076 | #define OFFB_Q14 offsetof(VexGuestARM64State,guest_Q14) |
| 1077 | #define OFFB_Q15 offsetof(VexGuestARM64State,guest_Q15) |
| 1078 | #define OFFB_Q16 offsetof(VexGuestARM64State,guest_Q16) |
| 1079 | #define OFFB_Q17 offsetof(VexGuestARM64State,guest_Q17) |
| 1080 | #define OFFB_Q18 offsetof(VexGuestARM64State,guest_Q18) |
| 1081 | #define OFFB_Q19 offsetof(VexGuestARM64State,guest_Q19) |
| 1082 | #define OFFB_Q20 offsetof(VexGuestARM64State,guest_Q20) |
| 1083 | #define OFFB_Q21 offsetof(VexGuestARM64State,guest_Q21) |
| 1084 | #define OFFB_Q22 offsetof(VexGuestARM64State,guest_Q22) |
| 1085 | #define OFFB_Q23 offsetof(VexGuestARM64State,guest_Q23) |
| 1086 | #define OFFB_Q24 offsetof(VexGuestARM64State,guest_Q24) |
| 1087 | #define OFFB_Q25 offsetof(VexGuestARM64State,guest_Q25) |
| 1088 | #define OFFB_Q26 offsetof(VexGuestARM64State,guest_Q26) |
| 1089 | #define OFFB_Q27 offsetof(VexGuestARM64State,guest_Q27) |
| 1090 | #define OFFB_Q28 offsetof(VexGuestARM64State,guest_Q28) |
| 1091 | #define OFFB_Q29 offsetof(VexGuestARM64State,guest_Q29) |
| 1092 | #define OFFB_Q30 offsetof(VexGuestARM64State,guest_Q30) |
| 1093 | #define OFFB_Q31 offsetof(VexGuestARM64State,guest_Q31) |
| 1094 | |
| 1095 | #define OFFB_FPCR offsetof(VexGuestARM64State,guest_FPCR) |
sewardj | a0645d5 | 2014-06-28 22:11:16 +0000 | [diff] [blame] | 1096 | #define OFFB_QCFLAG offsetof(VexGuestARM64State,guest_QCFLAG) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1097 | |
sewardj | 05f5e01 | 2014-05-04 10:52:11 +0000 | [diff] [blame] | 1098 | #define OFFB_CMSTART offsetof(VexGuestARM64State,guest_CMSTART) |
| 1099 | #define OFFB_CMLEN offsetof(VexGuestARM64State,guest_CMLEN) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1100 | |
| 1101 | |
| 1102 | /* ---------------- Integer registers ---------------- */ |
| 1103 | |
| 1104 | static Int offsetIReg64 ( UInt iregNo ) |
| 1105 | { |
| 1106 | /* Do we care about endianness here? We do if sub-parts of integer |
| 1107 | registers are accessed. */ |
| 1108 | switch (iregNo) { |
| 1109 | case 0: return OFFB_X0; |
| 1110 | case 1: return OFFB_X1; |
| 1111 | case 2: return OFFB_X2; |
| 1112 | case 3: return OFFB_X3; |
| 1113 | case 4: return OFFB_X4; |
| 1114 | case 5: return OFFB_X5; |
| 1115 | case 6: return OFFB_X6; |
| 1116 | case 7: return OFFB_X7; |
| 1117 | case 8: return OFFB_X8; |
| 1118 | case 9: return OFFB_X9; |
| 1119 | case 10: return OFFB_X10; |
| 1120 | case 11: return OFFB_X11; |
| 1121 | case 12: return OFFB_X12; |
| 1122 | case 13: return OFFB_X13; |
| 1123 | case 14: return OFFB_X14; |
| 1124 | case 15: return OFFB_X15; |
| 1125 | case 16: return OFFB_X16; |
| 1126 | case 17: return OFFB_X17; |
| 1127 | case 18: return OFFB_X18; |
| 1128 | case 19: return OFFB_X19; |
| 1129 | case 20: return OFFB_X20; |
| 1130 | case 21: return OFFB_X21; |
| 1131 | case 22: return OFFB_X22; |
| 1132 | case 23: return OFFB_X23; |
| 1133 | case 24: return OFFB_X24; |
| 1134 | case 25: return OFFB_X25; |
| 1135 | case 26: return OFFB_X26; |
| 1136 | case 27: return OFFB_X27; |
| 1137 | case 28: return OFFB_X28; |
| 1138 | case 29: return OFFB_X29; |
| 1139 | case 30: return OFFB_X30; |
| 1140 | /* but not 31 */ |
| 1141 | default: vassert(0); |
| 1142 | } |
| 1143 | } |
| 1144 | |
| 1145 | static Int offsetIReg64orSP ( UInt iregNo ) |
| 1146 | { |
sewardj | 6068788 | 2014-01-15 10:25:21 +0000 | [diff] [blame] | 1147 | return iregNo == 31 ? OFFB_XSP : offsetIReg64(iregNo); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1148 | } |
| 1149 | |
| 1150 | static const HChar* nameIReg64orZR ( UInt iregNo ) |
| 1151 | { |
| 1152 | vassert(iregNo < 32); |
| 1153 | static const HChar* names[32] |
| 1154 | = { "x0", "x1", "x2", "x3", "x4", "x5", "x6", "x7", |
| 1155 | "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", |
| 1156 | "x16", "x17", "x18", "x19", "x20", "x21", "x22", "x23", |
| 1157 | "x24", "x25", "x26", "x27", "x28", "x29", "x30", "xzr" }; |
| 1158 | return names[iregNo]; |
| 1159 | } |
| 1160 | |
| 1161 | static const HChar* nameIReg64orSP ( UInt iregNo ) |
| 1162 | { |
| 1163 | if (iregNo == 31) { |
| 1164 | return "sp"; |
| 1165 | } |
| 1166 | vassert(iregNo < 31); |
| 1167 | return nameIReg64orZR(iregNo); |
| 1168 | } |
| 1169 | |
| 1170 | static IRExpr* getIReg64orSP ( UInt iregNo ) |
| 1171 | { |
| 1172 | vassert(iregNo < 32); |
| 1173 | return IRExpr_Get( offsetIReg64orSP(iregNo), Ity_I64 ); |
| 1174 | } |
| 1175 | |
| 1176 | static IRExpr* getIReg64orZR ( UInt iregNo ) |
| 1177 | { |
| 1178 | if (iregNo == 31) { |
| 1179 | return mkU64(0); |
| 1180 | } |
| 1181 | vassert(iregNo < 31); |
| 1182 | return IRExpr_Get( offsetIReg64orSP(iregNo), Ity_I64 ); |
| 1183 | } |
| 1184 | |
| 1185 | static void putIReg64orSP ( UInt iregNo, IRExpr* e ) |
| 1186 | { |
| 1187 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I64); |
| 1188 | stmt( IRStmt_Put(offsetIReg64orSP(iregNo), e) ); |
| 1189 | } |
| 1190 | |
| 1191 | static void putIReg64orZR ( UInt iregNo, IRExpr* e ) |
| 1192 | { |
| 1193 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I64); |
| 1194 | if (iregNo == 31) { |
| 1195 | return; |
| 1196 | } |
| 1197 | vassert(iregNo < 31); |
| 1198 | stmt( IRStmt_Put(offsetIReg64orSP(iregNo), e) ); |
| 1199 | } |
| 1200 | |
| 1201 | static const HChar* nameIReg32orZR ( UInt iregNo ) |
| 1202 | { |
| 1203 | vassert(iregNo < 32); |
| 1204 | static const HChar* names[32] |
| 1205 | = { "w0", "w1", "w2", "w3", "w4", "w5", "w6", "w7", |
| 1206 | "w8", "w9", "w10", "w11", "w12", "w13", "w14", "w15", |
| 1207 | "w16", "w17", "w18", "w19", "w20", "w21", "w22", "w23", |
| 1208 | "w24", "w25", "w26", "w27", "w28", "w29", "w30", "wzr" }; |
| 1209 | return names[iregNo]; |
| 1210 | } |
| 1211 | |
| 1212 | static const HChar* nameIReg32orSP ( UInt iregNo ) |
| 1213 | { |
| 1214 | if (iregNo == 31) { |
| 1215 | return "wsp"; |
| 1216 | } |
| 1217 | vassert(iregNo < 31); |
| 1218 | return nameIReg32orZR(iregNo); |
| 1219 | } |
| 1220 | |
| 1221 | static IRExpr* getIReg32orSP ( UInt iregNo ) |
| 1222 | { |
| 1223 | vassert(iregNo < 32); |
| 1224 | return unop(Iop_64to32, |
| 1225 | IRExpr_Get( offsetIReg64orSP(iregNo), Ity_I64 )); |
| 1226 | } |
| 1227 | |
| 1228 | static IRExpr* getIReg32orZR ( UInt iregNo ) |
| 1229 | { |
| 1230 | if (iregNo == 31) { |
| 1231 | return mkU32(0); |
| 1232 | } |
| 1233 | vassert(iregNo < 31); |
| 1234 | return unop(Iop_64to32, |
| 1235 | IRExpr_Get( offsetIReg64orSP(iregNo), Ity_I64 )); |
| 1236 | } |
| 1237 | |
| 1238 | static void putIReg32orSP ( UInt iregNo, IRExpr* e ) |
| 1239 | { |
| 1240 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I32); |
| 1241 | stmt( IRStmt_Put(offsetIReg64orSP(iregNo), unop(Iop_32Uto64, e)) ); |
| 1242 | } |
| 1243 | |
| 1244 | static void putIReg32orZR ( UInt iregNo, IRExpr* e ) |
| 1245 | { |
| 1246 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I32); |
| 1247 | if (iregNo == 31) { |
| 1248 | return; |
| 1249 | } |
| 1250 | vassert(iregNo < 31); |
| 1251 | stmt( IRStmt_Put(offsetIReg64orSP(iregNo), unop(Iop_32Uto64, e)) ); |
| 1252 | } |
| 1253 | |
| 1254 | static const HChar* nameIRegOrSP ( Bool is64, UInt iregNo ) |
| 1255 | { |
| 1256 | vassert(is64 == True || is64 == False); |
| 1257 | return is64 ? nameIReg64orSP(iregNo) : nameIReg32orSP(iregNo); |
| 1258 | } |
| 1259 | |
| 1260 | static const HChar* nameIRegOrZR ( Bool is64, UInt iregNo ) |
| 1261 | { |
| 1262 | vassert(is64 == True || is64 == False); |
| 1263 | return is64 ? nameIReg64orZR(iregNo) : nameIReg32orZR(iregNo); |
| 1264 | } |
| 1265 | |
| 1266 | static IRExpr* getIRegOrZR ( Bool is64, UInt iregNo ) |
| 1267 | { |
| 1268 | vassert(is64 == True || is64 == False); |
| 1269 | return is64 ? getIReg64orZR(iregNo) : getIReg32orZR(iregNo); |
| 1270 | } |
| 1271 | |
| 1272 | static void putIRegOrZR ( Bool is64, UInt iregNo, IRExpr* e ) |
| 1273 | { |
| 1274 | vassert(is64 == True || is64 == False); |
| 1275 | if (is64) putIReg64orZR(iregNo, e); else putIReg32orZR(iregNo, e); |
| 1276 | } |
| 1277 | |
| 1278 | static void putPC ( IRExpr* e ) |
| 1279 | { |
| 1280 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I64); |
| 1281 | stmt( IRStmt_Put(OFFB_PC, e) ); |
| 1282 | } |
| 1283 | |
| 1284 | |
| 1285 | /* ---------------- Vector (Q) registers ---------------- */ |
| 1286 | |
| 1287 | static Int offsetQReg128 ( UInt qregNo ) |
| 1288 | { |
| 1289 | /* We don't care about endianness at this point. It only becomes |
| 1290 | relevant when dealing with sections of these registers.*/ |
| 1291 | switch (qregNo) { |
| 1292 | case 0: return OFFB_Q0; |
| 1293 | case 1: return OFFB_Q1; |
| 1294 | case 2: return OFFB_Q2; |
| 1295 | case 3: return OFFB_Q3; |
| 1296 | case 4: return OFFB_Q4; |
| 1297 | case 5: return OFFB_Q5; |
| 1298 | case 6: return OFFB_Q6; |
| 1299 | case 7: return OFFB_Q7; |
| 1300 | case 8: return OFFB_Q8; |
| 1301 | case 9: return OFFB_Q9; |
| 1302 | case 10: return OFFB_Q10; |
| 1303 | case 11: return OFFB_Q11; |
| 1304 | case 12: return OFFB_Q12; |
| 1305 | case 13: return OFFB_Q13; |
| 1306 | case 14: return OFFB_Q14; |
| 1307 | case 15: return OFFB_Q15; |
| 1308 | case 16: return OFFB_Q16; |
| 1309 | case 17: return OFFB_Q17; |
| 1310 | case 18: return OFFB_Q18; |
| 1311 | case 19: return OFFB_Q19; |
| 1312 | case 20: return OFFB_Q20; |
| 1313 | case 21: return OFFB_Q21; |
| 1314 | case 22: return OFFB_Q22; |
| 1315 | case 23: return OFFB_Q23; |
| 1316 | case 24: return OFFB_Q24; |
| 1317 | case 25: return OFFB_Q25; |
| 1318 | case 26: return OFFB_Q26; |
| 1319 | case 27: return OFFB_Q27; |
| 1320 | case 28: return OFFB_Q28; |
| 1321 | case 29: return OFFB_Q29; |
| 1322 | case 30: return OFFB_Q30; |
| 1323 | case 31: return OFFB_Q31; |
| 1324 | default: vassert(0); |
| 1325 | } |
| 1326 | } |
| 1327 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1328 | /* Write to a complete Qreg. */ |
| 1329 | static void putQReg128 ( UInt qregNo, IRExpr* e ) |
| 1330 | { |
| 1331 | vassert(qregNo < 32); |
| 1332 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_V128); |
| 1333 | stmt( IRStmt_Put(offsetQReg128(qregNo), e) ); |
| 1334 | } |
| 1335 | |
| 1336 | /* Read a complete Qreg. */ |
| 1337 | static IRExpr* getQReg128 ( UInt qregNo ) |
| 1338 | { |
| 1339 | vassert(qregNo < 32); |
| 1340 | return IRExpr_Get(offsetQReg128(qregNo), Ity_V128); |
| 1341 | } |
| 1342 | |
| 1343 | /* Produce the IR type for some sub-part of a vector. For 32- and 64- |
| 1344 | bit sub-parts we can choose either integer or float types, and |
| 1345 | choose float on the basis that that is the common use case and so |
| 1346 | will give least interference with Put-to-Get forwarding later |
| 1347 | on. */ |
| 1348 | static IRType preferredVectorSubTypeFromSize ( UInt szB ) |
| 1349 | { |
| 1350 | switch (szB) { |
| 1351 | case 1: return Ity_I8; |
| 1352 | case 2: return Ity_I16; |
| 1353 | case 4: return Ity_I32; //Ity_F32; |
| 1354 | case 8: return Ity_F64; |
| 1355 | case 16: return Ity_V128; |
| 1356 | default: vassert(0); |
| 1357 | } |
| 1358 | } |
| 1359 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1360 | /* Find the offset of the laneNo'th lane of type laneTy in the given |
| 1361 | Qreg. Since the host is little-endian, the least significant lane |
| 1362 | has the lowest offset. */ |
| 1363 | static Int offsetQRegLane ( UInt qregNo, IRType laneTy, UInt laneNo ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1364 | { |
sewardj | 9b76916 | 2014-07-24 12:42:03 +0000 | [diff] [blame] | 1365 | vassert(host_endness == VexEndnessLE); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1366 | Int base = offsetQReg128(qregNo); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1367 | /* Since the host is little-endian, the least significant lane |
| 1368 | will be at the lowest address. */ |
| 1369 | /* Restrict this to known types, so as to avoid silently accepting |
| 1370 | stupid types. */ |
| 1371 | UInt laneSzB = 0; |
| 1372 | switch (laneTy) { |
sewardj | 5860ec7 | 2014-03-01 11:19:45 +0000 | [diff] [blame] | 1373 | case Ity_I8: laneSzB = 1; break; |
| 1374 | case Ity_I16: laneSzB = 2; break; |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1375 | case Ity_F32: case Ity_I32: laneSzB = 4; break; |
| 1376 | case Ity_F64: case Ity_I64: laneSzB = 8; break; |
| 1377 | case Ity_V128: laneSzB = 16; break; |
| 1378 | default: break; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1379 | } |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1380 | vassert(laneSzB > 0); |
| 1381 | UInt minOff = laneNo * laneSzB; |
| 1382 | UInt maxOff = minOff + laneSzB - 1; |
| 1383 | vassert(maxOff < 16); |
| 1384 | return base + minOff; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1385 | } |
| 1386 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1387 | /* Put to the least significant lane of a Qreg. */ |
| 1388 | static void putQRegLO ( UInt qregNo, IRExpr* e ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1389 | { |
| 1390 | IRType ty = typeOfIRExpr(irsb->tyenv, e); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1391 | Int off = offsetQRegLane(qregNo, ty, 0); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1392 | switch (ty) { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1393 | case Ity_I8: case Ity_I16: case Ity_I32: case Ity_I64: |
| 1394 | case Ity_F32: case Ity_F64: case Ity_V128: |
| 1395 | break; |
| 1396 | default: |
| 1397 | vassert(0); // Other cases are probably invalid |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1398 | } |
| 1399 | stmt(IRStmt_Put(off, e)); |
| 1400 | } |
| 1401 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1402 | /* Get from the least significant lane of a Qreg. */ |
| 1403 | static IRExpr* getQRegLO ( UInt qregNo, IRType ty ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1404 | { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1405 | Int off = offsetQRegLane(qregNo, ty, 0); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1406 | switch (ty) { |
sewardj | b355347 | 2014-05-15 16:49:21 +0000 | [diff] [blame] | 1407 | case Ity_I8: |
| 1408 | case Ity_I16: |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1409 | case Ity_I32: case Ity_I64: |
| 1410 | case Ity_F32: case Ity_F64: case Ity_V128: |
| 1411 | break; |
| 1412 | default: |
| 1413 | vassert(0); // Other cases are ATC |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1414 | } |
| 1415 | return IRExpr_Get(off, ty); |
| 1416 | } |
| 1417 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1418 | static const HChar* nameQRegLO ( UInt qregNo, IRType laneTy ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1419 | { |
| 1420 | static const HChar* namesQ[32] |
| 1421 | = { "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7", |
| 1422 | "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15", |
| 1423 | "q16", "q17", "q18", "q19", "q20", "q21", "q22", "q23", |
| 1424 | "q24", "q25", "q26", "q27", "q28", "q29", "q30", "q31" }; |
| 1425 | static const HChar* namesD[32] |
| 1426 | = { "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", |
| 1427 | "d8", "d9", "d10", "d11", "d12", "d13", "d14", "d15", |
| 1428 | "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23", |
| 1429 | "d24", "d25", "d26", "d27", "d28", "d29", "d30", "d31" }; |
| 1430 | static const HChar* namesS[32] |
| 1431 | = { "s0", "s1", "s2", "s3", "s4", "s5", "s6", "s7", |
| 1432 | "s8", "s9", "s10", "s11", "s12", "s13", "s14", "s15", |
| 1433 | "s16", "s17", "s18", "s19", "s20", "s21", "s22", "s23", |
| 1434 | "s24", "s25", "s26", "s27", "s28", "s29", "s30", "s31" }; |
| 1435 | static const HChar* namesH[32] |
| 1436 | = { "h0", "h1", "h2", "h3", "h4", "h5", "h6", "h7", |
| 1437 | "h8", "h9", "h10", "h11", "h12", "h13", "h14", "h15", |
| 1438 | "h16", "h17", "h18", "h19", "h20", "h21", "h22", "h23", |
| 1439 | "h24", "h25", "h26", "h27", "h28", "h29", "h30", "h31" }; |
| 1440 | static const HChar* namesB[32] |
| 1441 | = { "b0", "b1", "b2", "b3", "b4", "b5", "b6", "b7", |
| 1442 | "b8", "b9", "b10", "b11", "b12", "b13", "b14", "b15", |
| 1443 | "b16", "b17", "b18", "b19", "b20", "b21", "b22", "b23", |
| 1444 | "b24", "b25", "b26", "b27", "b28", "b29", "b30", "b31" }; |
| 1445 | vassert(qregNo < 32); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1446 | switch (sizeofIRType(laneTy)) { |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1447 | case 1: return namesB[qregNo]; |
| 1448 | case 2: return namesH[qregNo]; |
| 1449 | case 4: return namesS[qregNo]; |
| 1450 | case 8: return namesD[qregNo]; |
| 1451 | case 16: return namesQ[qregNo]; |
| 1452 | default: vassert(0); |
| 1453 | } |
| 1454 | /*NOTREACHED*/ |
| 1455 | } |
| 1456 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1457 | static const HChar* nameQReg128 ( UInt qregNo ) |
| 1458 | { |
| 1459 | return nameQRegLO(qregNo, Ity_V128); |
| 1460 | } |
| 1461 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1462 | /* Find the offset of the most significant half (8 bytes) of the given |
| 1463 | Qreg. This requires knowing the endianness of the host. */ |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1464 | static Int offsetQRegHI64 ( UInt qregNo ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1465 | { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1466 | return offsetQRegLane(qregNo, Ity_I64, 1); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1467 | } |
| 1468 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1469 | static IRExpr* getQRegHI64 ( UInt qregNo ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1470 | { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1471 | return IRExpr_Get(offsetQRegHI64(qregNo), Ity_I64); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1472 | } |
| 1473 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1474 | static void putQRegHI64 ( UInt qregNo, IRExpr* e ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1475 | { |
| 1476 | IRType ty = typeOfIRExpr(irsb->tyenv, e); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1477 | Int off = offsetQRegHI64(qregNo); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1478 | switch (ty) { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1479 | case Ity_I64: case Ity_F64: |
| 1480 | break; |
| 1481 | default: |
| 1482 | vassert(0); // Other cases are plain wrong |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1483 | } |
| 1484 | stmt(IRStmt_Put(off, e)); |
| 1485 | } |
| 1486 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1487 | /* Put to a specified lane of a Qreg. */ |
| 1488 | static void putQRegLane ( UInt qregNo, UInt laneNo, IRExpr* e ) |
| 1489 | { |
| 1490 | IRType laneTy = typeOfIRExpr(irsb->tyenv, e); |
| 1491 | Int off = offsetQRegLane(qregNo, laneTy, laneNo); |
| 1492 | switch (laneTy) { |
| 1493 | case Ity_F64: case Ity_I64: |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 1494 | case Ity_I32: case Ity_F32: |
sewardj | 5860ec7 | 2014-03-01 11:19:45 +0000 | [diff] [blame] | 1495 | case Ity_I16: |
| 1496 | case Ity_I8: |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1497 | break; |
| 1498 | default: |
| 1499 | vassert(0); // Other cases are ATC |
| 1500 | } |
| 1501 | stmt(IRStmt_Put(off, e)); |
| 1502 | } |
| 1503 | |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 1504 | /* Get from a specified lane of a Qreg. */ |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1505 | static IRExpr* getQRegLane ( UInt qregNo, UInt laneNo, IRType laneTy ) |
| 1506 | { |
| 1507 | Int off = offsetQRegLane(qregNo, laneTy, laneNo); |
| 1508 | switch (laneTy) { |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 1509 | case Ity_I64: case Ity_I32: case Ity_I16: case Ity_I8: |
sewardj | 85fbb02 | 2014-06-12 13:16:01 +0000 | [diff] [blame] | 1510 | case Ity_F64: case Ity_F32: |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1511 | break; |
| 1512 | default: |
| 1513 | vassert(0); // Other cases are ATC |
| 1514 | } |
| 1515 | return IRExpr_Get(off, laneTy); |
| 1516 | } |
| 1517 | |
| 1518 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1519 | //ZZ /* ---------------- Misc registers ---------------- */ |
| 1520 | //ZZ |
| 1521 | //ZZ static void putMiscReg32 ( UInt gsoffset, |
| 1522 | //ZZ IRExpr* e, /* :: Ity_I32 */ |
| 1523 | //ZZ IRTemp guardT /* :: Ity_I32, 0 or 1 */) |
| 1524 | //ZZ { |
| 1525 | //ZZ switch (gsoffset) { |
| 1526 | //ZZ case OFFB_FPSCR: break; |
| 1527 | //ZZ case OFFB_QFLAG32: break; |
| 1528 | //ZZ case OFFB_GEFLAG0: break; |
| 1529 | //ZZ case OFFB_GEFLAG1: break; |
| 1530 | //ZZ case OFFB_GEFLAG2: break; |
| 1531 | //ZZ case OFFB_GEFLAG3: break; |
| 1532 | //ZZ default: vassert(0); /* awaiting more cases */ |
| 1533 | //ZZ } |
| 1534 | //ZZ vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I32); |
| 1535 | //ZZ |
| 1536 | //ZZ if (guardT == IRTemp_INVALID) { |
| 1537 | //ZZ /* unconditional write */ |
| 1538 | //ZZ stmt(IRStmt_Put(gsoffset, e)); |
| 1539 | //ZZ } else { |
| 1540 | //ZZ stmt(IRStmt_Put( |
| 1541 | //ZZ gsoffset, |
| 1542 | //ZZ IRExpr_ITE( binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0)), |
| 1543 | //ZZ e, IRExpr_Get(gsoffset, Ity_I32) ) |
| 1544 | //ZZ )); |
| 1545 | //ZZ } |
| 1546 | //ZZ } |
| 1547 | //ZZ |
| 1548 | //ZZ static IRTemp get_ITSTATE ( void ) |
| 1549 | //ZZ { |
| 1550 | //ZZ ASSERT_IS_THUMB; |
| 1551 | //ZZ IRTemp t = newTemp(Ity_I32); |
| 1552 | //ZZ assign(t, IRExpr_Get( OFFB_ITSTATE, Ity_I32)); |
| 1553 | //ZZ return t; |
| 1554 | //ZZ } |
| 1555 | //ZZ |
| 1556 | //ZZ static void put_ITSTATE ( IRTemp t ) |
| 1557 | //ZZ { |
| 1558 | //ZZ ASSERT_IS_THUMB; |
| 1559 | //ZZ stmt( IRStmt_Put( OFFB_ITSTATE, mkexpr(t)) ); |
| 1560 | //ZZ } |
| 1561 | //ZZ |
| 1562 | //ZZ static IRTemp get_QFLAG32 ( void ) |
| 1563 | //ZZ { |
| 1564 | //ZZ IRTemp t = newTemp(Ity_I32); |
| 1565 | //ZZ assign(t, IRExpr_Get( OFFB_QFLAG32, Ity_I32)); |
| 1566 | //ZZ return t; |
| 1567 | //ZZ } |
| 1568 | //ZZ |
| 1569 | //ZZ static void put_QFLAG32 ( IRTemp t, IRTemp condT ) |
| 1570 | //ZZ { |
| 1571 | //ZZ putMiscReg32( OFFB_QFLAG32, mkexpr(t), condT ); |
| 1572 | //ZZ } |
| 1573 | //ZZ |
| 1574 | //ZZ /* Stickily set the 'Q' flag (APSR bit 27) of the APSR (Application Program |
| 1575 | //ZZ Status Register) to indicate that overflow or saturation occurred. |
| 1576 | //ZZ Nb: t must be zero to denote no saturation, and any nonzero |
| 1577 | //ZZ value to indicate saturation. */ |
| 1578 | //ZZ static void or_into_QFLAG32 ( IRExpr* e, IRTemp condT ) |
| 1579 | //ZZ { |
| 1580 | //ZZ IRTemp old = get_QFLAG32(); |
| 1581 | //ZZ IRTemp nyu = newTemp(Ity_I32); |
| 1582 | //ZZ assign(nyu, binop(Iop_Or32, mkexpr(old), e) ); |
| 1583 | //ZZ put_QFLAG32(nyu, condT); |
| 1584 | //ZZ } |
| 1585 | |
| 1586 | |
| 1587 | /* ---------------- FPCR stuff ---------------- */ |
| 1588 | |
| 1589 | /* Generate IR to get hold of the rounding mode bits in FPCR, and |
| 1590 | convert them to IR format. Bind the final result to the |
| 1591 | returned temp. */ |
| 1592 | static IRTemp /* :: Ity_I32 */ mk_get_IR_rounding_mode ( void ) |
| 1593 | { |
| 1594 | /* The ARMvfp encoding for rounding mode bits is: |
| 1595 | 00 to nearest |
| 1596 | 01 to +infinity |
| 1597 | 10 to -infinity |
| 1598 | 11 to zero |
| 1599 | We need to convert that to the IR encoding: |
| 1600 | 00 to nearest (the default) |
| 1601 | 10 to +infinity |
| 1602 | 01 to -infinity |
| 1603 | 11 to zero |
| 1604 | Which can be done by swapping bits 0 and 1. |
| 1605 | The rmode bits are at 23:22 in FPSCR. |
| 1606 | */ |
| 1607 | IRTemp armEncd = newTemp(Ity_I32); |
| 1608 | IRTemp swapped = newTemp(Ity_I32); |
| 1609 | /* Fish FPCR[23:22] out, and slide to bottom. Doesn't matter that |
| 1610 | we don't zero out bits 24 and above, since the assignment to |
| 1611 | 'swapped' will mask them out anyway. */ |
| 1612 | assign(armEncd, |
| 1613 | binop(Iop_Shr32, IRExpr_Get(OFFB_FPCR, Ity_I32), mkU8(22))); |
| 1614 | /* Now swap them. */ |
| 1615 | assign(swapped, |
| 1616 | binop(Iop_Or32, |
| 1617 | binop(Iop_And32, |
| 1618 | binop(Iop_Shl32, mkexpr(armEncd), mkU8(1)), |
| 1619 | mkU32(2)), |
| 1620 | binop(Iop_And32, |
| 1621 | binop(Iop_Shr32, mkexpr(armEncd), mkU8(1)), |
| 1622 | mkU32(1)) |
| 1623 | )); |
| 1624 | return swapped; |
| 1625 | } |
| 1626 | |
| 1627 | |
| 1628 | /*------------------------------------------------------------*/ |
| 1629 | /*--- Helpers for flag handling and conditional insns ---*/ |
| 1630 | /*------------------------------------------------------------*/ |
| 1631 | |
| 1632 | static const HChar* nameARM64Condcode ( ARM64Condcode cond ) |
| 1633 | { |
| 1634 | switch (cond) { |
| 1635 | case ARM64CondEQ: return "eq"; |
| 1636 | case ARM64CondNE: return "ne"; |
| 1637 | case ARM64CondCS: return "cs"; // or 'hs' |
| 1638 | case ARM64CondCC: return "cc"; // or 'lo' |
| 1639 | case ARM64CondMI: return "mi"; |
| 1640 | case ARM64CondPL: return "pl"; |
| 1641 | case ARM64CondVS: return "vs"; |
| 1642 | case ARM64CondVC: return "vc"; |
| 1643 | case ARM64CondHI: return "hi"; |
| 1644 | case ARM64CondLS: return "ls"; |
| 1645 | case ARM64CondGE: return "ge"; |
| 1646 | case ARM64CondLT: return "lt"; |
| 1647 | case ARM64CondGT: return "gt"; |
| 1648 | case ARM64CondLE: return "le"; |
| 1649 | case ARM64CondAL: return "al"; |
| 1650 | case ARM64CondNV: return "nv"; |
| 1651 | default: vpanic("name_ARM64Condcode"); |
| 1652 | } |
| 1653 | } |
| 1654 | |
| 1655 | /* and a handy shorthand for it */ |
| 1656 | static const HChar* nameCC ( ARM64Condcode cond ) { |
| 1657 | return nameARM64Condcode(cond); |
| 1658 | } |
| 1659 | |
| 1660 | |
| 1661 | /* Build IR to calculate some particular condition from stored |
| 1662 | CC_OP/CC_DEP1/CC_DEP2/CC_NDEP. Returns an expression of type |
| 1663 | Ity_I64, suitable for narrowing. Although the return type is |
| 1664 | Ity_I64, the returned value is either 0 or 1. 'cond' must be |
| 1665 | :: Ity_I64 and must denote the condition to compute in |
| 1666 | bits 7:4, and be zero everywhere else. |
| 1667 | */ |
| 1668 | static IRExpr* mk_arm64g_calculate_condition_dyn ( IRExpr* cond ) |
| 1669 | { |
| 1670 | vassert(typeOfIRExpr(irsb->tyenv, cond) == Ity_I64); |
| 1671 | /* And 'cond' had better produce a value in which only bits 7:4 are |
| 1672 | nonzero. However, obviously we can't assert for that. */ |
| 1673 | |
| 1674 | /* So what we're constructing for the first argument is |
| 1675 | "(cond << 4) | stored-operation". |
| 1676 | However, as per comments above, 'cond' must be supplied |
| 1677 | pre-shifted to this function. |
| 1678 | |
| 1679 | This pairing scheme requires that the ARM64_CC_OP_ values all fit |
| 1680 | in 4 bits. Hence we are passing a (COND, OP) pair in the lowest |
| 1681 | 8 bits of the first argument. */ |
| 1682 | IRExpr** args |
| 1683 | = mkIRExprVec_4( |
| 1684 | binop(Iop_Or64, IRExpr_Get(OFFB_CC_OP, Ity_I64), cond), |
| 1685 | IRExpr_Get(OFFB_CC_DEP1, Ity_I64), |
| 1686 | IRExpr_Get(OFFB_CC_DEP2, Ity_I64), |
| 1687 | IRExpr_Get(OFFB_CC_NDEP, Ity_I64) |
| 1688 | ); |
| 1689 | IRExpr* call |
| 1690 | = mkIRExprCCall( |
| 1691 | Ity_I64, |
| 1692 | 0/*regparm*/, |
| 1693 | "arm64g_calculate_condition", &arm64g_calculate_condition, |
| 1694 | args |
| 1695 | ); |
| 1696 | |
| 1697 | /* Exclude the requested condition, OP and NDEP from definedness |
| 1698 | checking. We're only interested in DEP1 and DEP2. */ |
| 1699 | call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3); |
| 1700 | return call; |
| 1701 | } |
| 1702 | |
| 1703 | |
| 1704 | /* Build IR to calculate some particular condition from stored |
| 1705 | CC_OP/CC_DEP1/CC_DEP2/CC_NDEP. Returns an expression of type |
| 1706 | Ity_I64, suitable for narrowing. Although the return type is |
| 1707 | Ity_I64, the returned value is either 0 or 1. |
| 1708 | */ |
| 1709 | static IRExpr* mk_arm64g_calculate_condition ( ARM64Condcode cond ) |
| 1710 | { |
| 1711 | /* First arg is "(cond << 4) | condition". This requires that the |
| 1712 | ARM64_CC_OP_ values all fit in 4 bits. Hence we are passing a |
| 1713 | (COND, OP) pair in the lowest 8 bits of the first argument. */ |
| 1714 | vassert(cond >= 0 && cond <= 15); |
| 1715 | return mk_arm64g_calculate_condition_dyn( mkU64(cond << 4) ); |
| 1716 | } |
| 1717 | |
| 1718 | |
sewardj | dee3050 | 2014-06-04 13:09:44 +0000 | [diff] [blame] | 1719 | /* Build IR to calculate just the carry flag from stored |
| 1720 | CC_OP/CC_DEP1/CC_DEP2/CC_NDEP. Returns an expression :: |
| 1721 | Ity_I64. */ |
| 1722 | static IRExpr* mk_arm64g_calculate_flag_c ( void ) |
| 1723 | { |
| 1724 | IRExpr** args |
| 1725 | = mkIRExprVec_4( IRExpr_Get(OFFB_CC_OP, Ity_I64), |
| 1726 | IRExpr_Get(OFFB_CC_DEP1, Ity_I64), |
| 1727 | IRExpr_Get(OFFB_CC_DEP2, Ity_I64), |
| 1728 | IRExpr_Get(OFFB_CC_NDEP, Ity_I64) ); |
| 1729 | IRExpr* call |
| 1730 | = mkIRExprCCall( |
| 1731 | Ity_I64, |
| 1732 | 0/*regparm*/, |
| 1733 | "arm64g_calculate_flag_c", &arm64g_calculate_flag_c, |
| 1734 | args |
| 1735 | ); |
| 1736 | /* Exclude OP and NDEP from definedness checking. We're only |
| 1737 | interested in DEP1 and DEP2. */ |
| 1738 | call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3); |
| 1739 | return call; |
| 1740 | } |
| 1741 | |
| 1742 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1743 | //ZZ /* Build IR to calculate just the overflow flag from stored |
| 1744 | //ZZ CC_OP/CC_DEP1/CC_DEP2/CC_NDEP. Returns an expression :: |
| 1745 | //ZZ Ity_I32. */ |
| 1746 | //ZZ static IRExpr* mk_armg_calculate_flag_v ( void ) |
| 1747 | //ZZ { |
| 1748 | //ZZ IRExpr** args |
| 1749 | //ZZ = mkIRExprVec_4( IRExpr_Get(OFFB_CC_OP, Ity_I32), |
| 1750 | //ZZ IRExpr_Get(OFFB_CC_DEP1, Ity_I32), |
| 1751 | //ZZ IRExpr_Get(OFFB_CC_DEP2, Ity_I32), |
| 1752 | //ZZ IRExpr_Get(OFFB_CC_NDEP, Ity_I32) ); |
| 1753 | //ZZ IRExpr* call |
| 1754 | //ZZ = mkIRExprCCall( |
| 1755 | //ZZ Ity_I32, |
| 1756 | //ZZ 0/*regparm*/, |
| 1757 | //ZZ "armg_calculate_flag_v", &armg_calculate_flag_v, |
| 1758 | //ZZ args |
| 1759 | //ZZ ); |
| 1760 | //ZZ /* Exclude OP and NDEP from definedness checking. We're only |
| 1761 | //ZZ interested in DEP1 and DEP2. */ |
| 1762 | //ZZ call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3); |
| 1763 | //ZZ return call; |
| 1764 | //ZZ } |
| 1765 | |
| 1766 | |
| 1767 | /* Build IR to calculate N Z C V in bits 31:28 of the |
| 1768 | returned word. */ |
| 1769 | static IRExpr* mk_arm64g_calculate_flags_nzcv ( void ) |
| 1770 | { |
| 1771 | IRExpr** args |
| 1772 | = mkIRExprVec_4( IRExpr_Get(OFFB_CC_OP, Ity_I64), |
| 1773 | IRExpr_Get(OFFB_CC_DEP1, Ity_I64), |
| 1774 | IRExpr_Get(OFFB_CC_DEP2, Ity_I64), |
| 1775 | IRExpr_Get(OFFB_CC_NDEP, Ity_I64) ); |
| 1776 | IRExpr* call |
| 1777 | = mkIRExprCCall( |
| 1778 | Ity_I64, |
| 1779 | 0/*regparm*/, |
| 1780 | "arm64g_calculate_flags_nzcv", &arm64g_calculate_flags_nzcv, |
| 1781 | args |
| 1782 | ); |
| 1783 | /* Exclude OP and NDEP from definedness checking. We're only |
| 1784 | interested in DEP1 and DEP2. */ |
| 1785 | call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3); |
| 1786 | return call; |
| 1787 | } |
| 1788 | |
| 1789 | |
| 1790 | /* Build IR to set the flags thunk, in the most general case. */ |
| 1791 | static |
| 1792 | void setFlags_D1_D2_ND ( UInt cc_op, |
| 1793 | IRTemp t_dep1, IRTemp t_dep2, IRTemp t_ndep ) |
| 1794 | { |
| 1795 | vassert(typeOfIRTemp(irsb->tyenv, t_dep1 == Ity_I64)); |
| 1796 | vassert(typeOfIRTemp(irsb->tyenv, t_dep2 == Ity_I64)); |
| 1797 | vassert(typeOfIRTemp(irsb->tyenv, t_ndep == Ity_I64)); |
| 1798 | vassert(cc_op >= ARM64G_CC_OP_COPY && cc_op < ARM64G_CC_OP_NUMBER); |
| 1799 | stmt( IRStmt_Put( OFFB_CC_OP, mkU64(cc_op) )); |
| 1800 | stmt( IRStmt_Put( OFFB_CC_DEP1, mkexpr(t_dep1) )); |
| 1801 | stmt( IRStmt_Put( OFFB_CC_DEP2, mkexpr(t_dep2) )); |
| 1802 | stmt( IRStmt_Put( OFFB_CC_NDEP, mkexpr(t_ndep) )); |
| 1803 | } |
| 1804 | |
| 1805 | /* Build IR to set the flags thunk after ADD or SUB. */ |
| 1806 | static |
| 1807 | void setFlags_ADD_SUB ( Bool is64, Bool isSUB, IRTemp argL, IRTemp argR ) |
| 1808 | { |
| 1809 | IRTemp argL64 = IRTemp_INVALID; |
| 1810 | IRTemp argR64 = IRTemp_INVALID; |
| 1811 | IRTemp z64 = newTemp(Ity_I64); |
| 1812 | if (is64) { |
| 1813 | argL64 = argL; |
| 1814 | argR64 = argR; |
| 1815 | } else { |
| 1816 | argL64 = newTemp(Ity_I64); |
| 1817 | argR64 = newTemp(Ity_I64); |
| 1818 | assign(argL64, unop(Iop_32Uto64, mkexpr(argL))); |
| 1819 | assign(argR64, unop(Iop_32Uto64, mkexpr(argR))); |
| 1820 | } |
| 1821 | assign(z64, mkU64(0)); |
| 1822 | UInt cc_op = ARM64G_CC_OP_NUMBER; |
| 1823 | /**/ if ( isSUB && is64) { cc_op = ARM64G_CC_OP_SUB64; } |
| 1824 | else if ( isSUB && !is64) { cc_op = ARM64G_CC_OP_SUB32; } |
| 1825 | else if (!isSUB && is64) { cc_op = ARM64G_CC_OP_ADD64; } |
| 1826 | else if (!isSUB && !is64) { cc_op = ARM64G_CC_OP_ADD32; } |
| 1827 | else { vassert(0); } |
| 1828 | setFlags_D1_D2_ND(cc_op, argL64, argR64, z64); |
| 1829 | } |
| 1830 | |
sewardj | dee3050 | 2014-06-04 13:09:44 +0000 | [diff] [blame] | 1831 | /* Build IR to set the flags thunk after ADC or SBC. */ |
| 1832 | static |
| 1833 | void setFlags_ADC_SBC ( Bool is64, Bool isSBC, |
| 1834 | IRTemp argL, IRTemp argR, IRTemp oldC ) |
| 1835 | { |
| 1836 | IRTemp argL64 = IRTemp_INVALID; |
| 1837 | IRTemp argR64 = IRTemp_INVALID; |
| 1838 | IRTemp oldC64 = IRTemp_INVALID; |
| 1839 | if (is64) { |
| 1840 | argL64 = argL; |
| 1841 | argR64 = argR; |
| 1842 | oldC64 = oldC; |
| 1843 | } else { |
| 1844 | argL64 = newTemp(Ity_I64); |
| 1845 | argR64 = newTemp(Ity_I64); |
| 1846 | oldC64 = newTemp(Ity_I64); |
| 1847 | assign(argL64, unop(Iop_32Uto64, mkexpr(argL))); |
| 1848 | assign(argR64, unop(Iop_32Uto64, mkexpr(argR))); |
| 1849 | assign(oldC64, unop(Iop_32Uto64, mkexpr(oldC))); |
| 1850 | } |
| 1851 | UInt cc_op = ARM64G_CC_OP_NUMBER; |
| 1852 | /**/ if ( isSBC && is64) { cc_op = ARM64G_CC_OP_SBC64; } |
| 1853 | else if ( isSBC && !is64) { cc_op = ARM64G_CC_OP_SBC32; } |
| 1854 | else if (!isSBC && is64) { cc_op = ARM64G_CC_OP_ADC64; } |
| 1855 | else if (!isSBC && !is64) { cc_op = ARM64G_CC_OP_ADC32; } |
| 1856 | else { vassert(0); } |
| 1857 | setFlags_D1_D2_ND(cc_op, argL64, argR64, oldC64); |
| 1858 | } |
| 1859 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1860 | /* Build IR to set the flags thunk after ADD or SUB, if the given |
| 1861 | condition evaluates to True at run time. If not, the flags are set |
| 1862 | to the specified NZCV value. */ |
| 1863 | static |
| 1864 | void setFlags_ADD_SUB_conditionally ( |
| 1865 | Bool is64, Bool isSUB, |
| 1866 | IRTemp cond, IRTemp argL, IRTemp argR, UInt nzcv |
| 1867 | ) |
| 1868 | { |
| 1869 | /* Generate IR as follows: |
| 1870 | CC_OP = ITE(cond, OP_{ADD,SUB}{32,64}, OP_COPY) |
| 1871 | CC_DEP1 = ITE(cond, argL64, nzcv << 28) |
| 1872 | CC_DEP2 = ITE(cond, argR64, 0) |
| 1873 | CC_NDEP = 0 |
| 1874 | */ |
| 1875 | |
| 1876 | IRTemp z64 = newTemp(Ity_I64); |
| 1877 | assign(z64, mkU64(0)); |
| 1878 | |
| 1879 | /* Establish the operation and operands for the True case. */ |
| 1880 | IRTemp t_dep1 = IRTemp_INVALID; |
| 1881 | IRTemp t_dep2 = IRTemp_INVALID; |
| 1882 | UInt t_op = ARM64G_CC_OP_NUMBER; |
| 1883 | /**/ if ( isSUB && is64) { t_op = ARM64G_CC_OP_SUB64; } |
| 1884 | else if ( isSUB && !is64) { t_op = ARM64G_CC_OP_SUB32; } |
| 1885 | else if (!isSUB && is64) { t_op = ARM64G_CC_OP_ADD64; } |
| 1886 | else if (!isSUB && !is64) { t_op = ARM64G_CC_OP_ADD32; } |
| 1887 | else { vassert(0); } |
| 1888 | /* */ |
| 1889 | if (is64) { |
| 1890 | t_dep1 = argL; |
| 1891 | t_dep2 = argR; |
| 1892 | } else { |
| 1893 | t_dep1 = newTemp(Ity_I64); |
| 1894 | t_dep2 = newTemp(Ity_I64); |
| 1895 | assign(t_dep1, unop(Iop_32Uto64, mkexpr(argL))); |
| 1896 | assign(t_dep2, unop(Iop_32Uto64, mkexpr(argR))); |
| 1897 | } |
| 1898 | |
| 1899 | /* Establish the operation and operands for the False case. */ |
| 1900 | IRTemp f_dep1 = newTemp(Ity_I64); |
| 1901 | IRTemp f_dep2 = z64; |
| 1902 | UInt f_op = ARM64G_CC_OP_COPY; |
| 1903 | assign(f_dep1, mkU64(nzcv << 28)); |
| 1904 | |
| 1905 | /* Final thunk values */ |
| 1906 | IRTemp dep1 = newTemp(Ity_I64); |
| 1907 | IRTemp dep2 = newTemp(Ity_I64); |
| 1908 | IRTemp op = newTemp(Ity_I64); |
| 1909 | |
| 1910 | assign(op, IRExpr_ITE(mkexpr(cond), mkU64(t_op), mkU64(f_op))); |
| 1911 | assign(dep1, IRExpr_ITE(mkexpr(cond), mkexpr(t_dep1), mkexpr(f_dep1))); |
| 1912 | assign(dep2, IRExpr_ITE(mkexpr(cond), mkexpr(t_dep2), mkexpr(f_dep2))); |
| 1913 | |
| 1914 | /* finally .. */ |
| 1915 | stmt( IRStmt_Put( OFFB_CC_OP, mkexpr(op) )); |
| 1916 | stmt( IRStmt_Put( OFFB_CC_DEP1, mkexpr(dep1) )); |
| 1917 | stmt( IRStmt_Put( OFFB_CC_DEP2, mkexpr(dep2) )); |
| 1918 | stmt( IRStmt_Put( OFFB_CC_NDEP, mkexpr(z64) )); |
| 1919 | } |
| 1920 | |
| 1921 | /* Build IR to set the flags thunk after AND/OR/XOR or variants thereof. */ |
| 1922 | static |
| 1923 | void setFlags_LOGIC ( Bool is64, IRTemp res ) |
| 1924 | { |
| 1925 | IRTemp res64 = IRTemp_INVALID; |
| 1926 | IRTemp z64 = newTemp(Ity_I64); |
| 1927 | UInt cc_op = ARM64G_CC_OP_NUMBER; |
| 1928 | if (is64) { |
| 1929 | res64 = res; |
| 1930 | cc_op = ARM64G_CC_OP_LOGIC64; |
| 1931 | } else { |
| 1932 | res64 = newTemp(Ity_I64); |
| 1933 | assign(res64, unop(Iop_32Uto64, mkexpr(res))); |
| 1934 | cc_op = ARM64G_CC_OP_LOGIC32; |
| 1935 | } |
| 1936 | assign(z64, mkU64(0)); |
| 1937 | setFlags_D1_D2_ND(cc_op, res64, z64, z64); |
| 1938 | } |
| 1939 | |
| 1940 | /* Build IR to set the flags thunk to a given NZCV value. NZCV is |
| 1941 | located in bits 31:28 of the supplied value. */ |
| 1942 | static |
| 1943 | void setFlags_COPY ( IRTemp nzcv_28x0 ) |
| 1944 | { |
| 1945 | IRTemp z64 = newTemp(Ity_I64); |
| 1946 | assign(z64, mkU64(0)); |
| 1947 | setFlags_D1_D2_ND(ARM64G_CC_OP_COPY, nzcv_28x0, z64, z64); |
| 1948 | } |
| 1949 | |
| 1950 | |
| 1951 | //ZZ /* Minor variant of the above that sets NDEP to zero (if it |
| 1952 | //ZZ sets it at all) */ |
| 1953 | //ZZ static void setFlags_D1_D2 ( UInt cc_op, IRTemp t_dep1, |
| 1954 | //ZZ IRTemp t_dep2, |
| 1955 | //ZZ IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
| 1956 | //ZZ { |
| 1957 | //ZZ IRTemp z32 = newTemp(Ity_I32); |
| 1958 | //ZZ assign( z32, mkU32(0) ); |
| 1959 | //ZZ setFlags_D1_D2_ND( cc_op, t_dep1, t_dep2, z32, guardT ); |
| 1960 | //ZZ } |
| 1961 | //ZZ |
| 1962 | //ZZ |
| 1963 | //ZZ /* Minor variant of the above that sets DEP2 to zero (if it |
| 1964 | //ZZ sets it at all) */ |
| 1965 | //ZZ static void setFlags_D1_ND ( UInt cc_op, IRTemp t_dep1, |
| 1966 | //ZZ IRTemp t_ndep, |
| 1967 | //ZZ IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
| 1968 | //ZZ { |
| 1969 | //ZZ IRTemp z32 = newTemp(Ity_I32); |
| 1970 | //ZZ assign( z32, mkU32(0) ); |
| 1971 | //ZZ setFlags_D1_D2_ND( cc_op, t_dep1, z32, t_ndep, guardT ); |
| 1972 | //ZZ } |
| 1973 | //ZZ |
| 1974 | //ZZ |
| 1975 | //ZZ /* Minor variant of the above that sets DEP2 and NDEP to zero (if it |
| 1976 | //ZZ sets them at all) */ |
| 1977 | //ZZ static void setFlags_D1 ( UInt cc_op, IRTemp t_dep1, |
| 1978 | //ZZ IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
| 1979 | //ZZ { |
| 1980 | //ZZ IRTemp z32 = newTemp(Ity_I32); |
| 1981 | //ZZ assign( z32, mkU32(0) ); |
| 1982 | //ZZ setFlags_D1_D2_ND( cc_op, t_dep1, z32, z32, guardT ); |
| 1983 | //ZZ } |
| 1984 | |
| 1985 | |
| 1986 | /*------------------------------------------------------------*/ |
| 1987 | /*--- Misc math helpers ---*/ |
| 1988 | /*------------------------------------------------------------*/ |
| 1989 | |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 1990 | /* Generate IR for ((x & mask) >>u sh) | ((x << sh) & mask) */ |
| 1991 | static IRTemp math_SWAPHELPER ( IRTemp x, ULong mask, Int sh ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1992 | { |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 1993 | IRTemp maskT = newTemp(Ity_I64); |
| 1994 | IRTemp res = newTemp(Ity_I64); |
| 1995 | vassert(sh >= 1 && sh <= 63); |
| 1996 | assign(maskT, mkU64(mask)); |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 1997 | assign( res, |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1998 | binop(Iop_Or64, |
| 1999 | binop(Iop_Shr64, |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 2000 | binop(Iop_And64,mkexpr(x),mkexpr(maskT)), |
| 2001 | mkU8(sh)), |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 2002 | binop(Iop_And64, |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 2003 | binop(Iop_Shl64,mkexpr(x),mkU8(sh)), |
| 2004 | mkexpr(maskT)) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 2005 | ) |
| 2006 | ); |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 2007 | return res; |
| 2008 | } |
| 2009 | |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 2010 | /* Generates byte swaps within 32-bit lanes. */ |
| 2011 | static IRTemp math_UINTSWAP64 ( IRTemp src ) |
| 2012 | { |
| 2013 | IRTemp res; |
| 2014 | res = math_SWAPHELPER(src, 0xFF00FF00FF00FF00ULL, 8); |
| 2015 | res = math_SWAPHELPER(res, 0xFFFF0000FFFF0000ULL, 16); |
| 2016 | return res; |
| 2017 | } |
| 2018 | |
| 2019 | /* Generates byte swaps within 16-bit lanes. */ |
| 2020 | static IRTemp math_USHORTSWAP64 ( IRTemp src ) |
| 2021 | { |
| 2022 | IRTemp res; |
| 2023 | res = math_SWAPHELPER(src, 0xFF00FF00FF00FF00ULL, 8); |
| 2024 | return res; |
| 2025 | } |
| 2026 | |
| 2027 | /* Generates a 64-bit byte swap. */ |
| 2028 | static IRTemp math_BYTESWAP64 ( IRTemp src ) |
| 2029 | { |
| 2030 | IRTemp res; |
| 2031 | res = math_SWAPHELPER(src, 0xFF00FF00FF00FF00ULL, 8); |
| 2032 | res = math_SWAPHELPER(res, 0xFFFF0000FFFF0000ULL, 16); |
| 2033 | res = math_SWAPHELPER(res, 0xFFFFFFFF00000000ULL, 32); |
| 2034 | return res; |
| 2035 | } |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 2036 | |
| 2037 | /* Generates a 64-bit bit swap. */ |
| 2038 | static IRTemp math_BITSWAP64 ( IRTemp src ) |
| 2039 | { |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 2040 | IRTemp res; |
| 2041 | res = math_SWAPHELPER(src, 0xAAAAAAAAAAAAAAAAULL, 1); |
| 2042 | res = math_SWAPHELPER(res, 0xCCCCCCCCCCCCCCCCULL, 2); |
| 2043 | res = math_SWAPHELPER(res, 0xF0F0F0F0F0F0F0F0ULL, 4); |
| 2044 | return math_BYTESWAP64(res); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 2045 | } |
| 2046 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 2047 | /* Duplicates the bits at the bottom of the given word to fill the |
| 2048 | whole word. src :: Ity_I64 is assumed to have zeroes everywhere |
| 2049 | except for the bottom bits. */ |
| 2050 | static IRTemp math_DUP_TO_64 ( IRTemp src, IRType srcTy ) |
| 2051 | { |
| 2052 | if (srcTy == Ity_I8) { |
| 2053 | IRTemp t16 = newTemp(Ity_I64); |
| 2054 | assign(t16, binop(Iop_Or64, mkexpr(src), |
| 2055 | binop(Iop_Shl64, mkexpr(src), mkU8(8)))); |
| 2056 | IRTemp t32 = newTemp(Ity_I64); |
| 2057 | assign(t32, binop(Iop_Or64, mkexpr(t16), |
| 2058 | binop(Iop_Shl64, mkexpr(t16), mkU8(16)))); |
| 2059 | IRTemp t64 = newTemp(Ity_I64); |
| 2060 | assign(t64, binop(Iop_Or64, mkexpr(t32), |
| 2061 | binop(Iop_Shl64, mkexpr(t32), mkU8(32)))); |
| 2062 | return t64; |
| 2063 | } |
| 2064 | if (srcTy == Ity_I16) { |
| 2065 | IRTemp t32 = newTemp(Ity_I64); |
| 2066 | assign(t32, binop(Iop_Or64, mkexpr(src), |
| 2067 | binop(Iop_Shl64, mkexpr(src), mkU8(16)))); |
| 2068 | IRTemp t64 = newTemp(Ity_I64); |
| 2069 | assign(t64, binop(Iop_Or64, mkexpr(t32), |
| 2070 | binop(Iop_Shl64, mkexpr(t32), mkU8(32)))); |
| 2071 | return t64; |
| 2072 | } |
| 2073 | if (srcTy == Ity_I32) { |
| 2074 | IRTemp t64 = newTemp(Ity_I64); |
| 2075 | assign(t64, binop(Iop_Or64, mkexpr(src), |
| 2076 | binop(Iop_Shl64, mkexpr(src), mkU8(32)))); |
| 2077 | return t64; |
| 2078 | } |
| 2079 | if (srcTy == Ity_I64) { |
| 2080 | return src; |
| 2081 | } |
| 2082 | vassert(0); |
| 2083 | } |
| 2084 | |
| 2085 | |
sewardj | 18bf517 | 2014-06-14 18:05:30 +0000 | [diff] [blame] | 2086 | /* Duplicates the src element exactly so as to fill a V128 value. */ |
sewardj | 85fbb02 | 2014-06-12 13:16:01 +0000 | [diff] [blame] | 2087 | static IRTemp math_DUP_TO_V128 ( IRTemp src, IRType srcTy ) |
| 2088 | { |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 2089 | IRTemp res = newTempV128(); |
sewardj | 85fbb02 | 2014-06-12 13:16:01 +0000 | [diff] [blame] | 2090 | if (srcTy == Ity_F64) { |
| 2091 | IRTemp i64 = newTemp(Ity_I64); |
| 2092 | assign(i64, unop(Iop_ReinterpF64asI64, mkexpr(src))); |
| 2093 | assign(res, binop(Iop_64HLtoV128, mkexpr(i64), mkexpr(i64))); |
| 2094 | return res; |
| 2095 | } |
| 2096 | if (srcTy == Ity_F32) { |
| 2097 | IRTemp i64a = newTemp(Ity_I64); |
| 2098 | assign(i64a, unop(Iop_32Uto64, unop(Iop_ReinterpF32asI32, mkexpr(src)))); |
| 2099 | IRTemp i64b = newTemp(Ity_I64); |
| 2100 | assign(i64b, binop(Iop_Or64, binop(Iop_Shl64, mkexpr(i64a), mkU8(32)), |
| 2101 | mkexpr(i64a))); |
| 2102 | assign(res, binop(Iop_64HLtoV128, mkexpr(i64b), mkexpr(i64b))); |
| 2103 | return res; |
| 2104 | } |
sewardj | 18bf517 | 2014-06-14 18:05:30 +0000 | [diff] [blame] | 2105 | if (srcTy == Ity_I64) { |
| 2106 | assign(res, binop(Iop_64HLtoV128, mkexpr(src), mkexpr(src))); |
| 2107 | return res; |
| 2108 | } |
| 2109 | if (srcTy == Ity_I32 || srcTy == Ity_I16 || srcTy == Ity_I8) { |
| 2110 | IRTemp t1 = newTemp(Ity_I64); |
| 2111 | assign(t1, widenUto64(srcTy, mkexpr(src))); |
| 2112 | IRTemp t2 = math_DUP_TO_64(t1, srcTy); |
| 2113 | assign(res, binop(Iop_64HLtoV128, mkexpr(t2), mkexpr(t2))); |
| 2114 | return res; |
| 2115 | } |
sewardj | 85fbb02 | 2014-06-12 13:16:01 +0000 | [diff] [blame] | 2116 | vassert(0); |
| 2117 | } |
| 2118 | |
| 2119 | |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 2120 | /* |fullWidth| is a full V128 width result. Depending on bitQ, |
| 2121 | zero out the upper half. */ |
| 2122 | static IRExpr* math_MAYBE_ZERO_HI64 ( UInt bitQ, IRTemp fullWidth ) |
| 2123 | { |
| 2124 | if (bitQ == 1) return mkexpr(fullWidth); |
| 2125 | if (bitQ == 0) return unop(Iop_ZeroHI64ofV128, mkexpr(fullWidth)); |
| 2126 | vassert(0); |
| 2127 | } |
| 2128 | |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 2129 | /* The same, but from an expression instead. */ |
| 2130 | static IRExpr* math_MAYBE_ZERO_HI64_fromE ( UInt bitQ, IRExpr* fullWidth ) |
| 2131 | { |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 2132 | IRTemp fullWidthT = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 2133 | assign(fullWidthT, fullWidth); |
| 2134 | return math_MAYBE_ZERO_HI64(bitQ, fullWidthT); |
| 2135 | } |
| 2136 | |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 2137 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 2138 | /*------------------------------------------------------------*/ |
| 2139 | /*--- FP comparison helpers ---*/ |
| 2140 | /*------------------------------------------------------------*/ |
| 2141 | |
| 2142 | /* irRes :: Ity_I32 holds a floating point comparison result encoded |
| 2143 | as an IRCmpF64Result. Generate code to convert it to an |
| 2144 | ARM64-encoded (N,Z,C,V) group in the lowest 4 bits of an I64 value. |
| 2145 | Assign a new temp to hold that value, and return the temp. */ |
| 2146 | static |
| 2147 | IRTemp mk_convert_IRCmpF64Result_to_NZCV ( IRTemp irRes32 ) |
| 2148 | { |
| 2149 | IRTemp ix = newTemp(Ity_I64); |
| 2150 | IRTemp termL = newTemp(Ity_I64); |
| 2151 | IRTemp termR = newTemp(Ity_I64); |
| 2152 | IRTemp nzcv = newTemp(Ity_I64); |
| 2153 | IRTemp irRes = newTemp(Ity_I64); |
| 2154 | |
| 2155 | /* This is where the fun starts. We have to convert 'irRes' from |
| 2156 | an IR-convention return result (IRCmpF64Result) to an |
| 2157 | ARM-encoded (N,Z,C,V) group. The final result is in the bottom |
| 2158 | 4 bits of 'nzcv'. */ |
| 2159 | /* Map compare result from IR to ARM(nzcv) */ |
| 2160 | /* |
| 2161 | FP cmp result | IR | ARM(nzcv) |
| 2162 | -------------------------------- |
| 2163 | UN 0x45 0011 |
| 2164 | LT 0x01 1000 |
| 2165 | GT 0x00 0010 |
| 2166 | EQ 0x40 0110 |
| 2167 | */ |
| 2168 | /* Now since you're probably wondering WTF .. |
| 2169 | |
| 2170 | ix fishes the useful bits out of the IR value, bits 6 and 0, and |
| 2171 | places them side by side, giving a number which is 0, 1, 2 or 3. |
| 2172 | |
| 2173 | termL is a sequence cooked up by GNU superopt. It converts ix |
| 2174 | into an almost correct value NZCV value (incredibly), except |
| 2175 | for the case of UN, where it produces 0100 instead of the |
| 2176 | required 0011. |
| 2177 | |
| 2178 | termR is therefore a correction term, also computed from ix. It |
| 2179 | is 1 in the UN case and 0 for LT, GT and UN. Hence, to get |
| 2180 | the final correct value, we subtract termR from termL. |
| 2181 | |
| 2182 | Don't take my word for it. There's a test program at the bottom |
| 2183 | of guest_arm_toIR.c, to try this out with. |
| 2184 | */ |
| 2185 | assign(irRes, unop(Iop_32Uto64, mkexpr(irRes32))); |
| 2186 | |
| 2187 | assign( |
| 2188 | ix, |
| 2189 | binop(Iop_Or64, |
| 2190 | binop(Iop_And64, |
| 2191 | binop(Iop_Shr64, mkexpr(irRes), mkU8(5)), |
| 2192 | mkU64(3)), |
| 2193 | binop(Iop_And64, mkexpr(irRes), mkU64(1)))); |
| 2194 | |
| 2195 | assign( |
| 2196 | termL, |
| 2197 | binop(Iop_Add64, |
| 2198 | binop(Iop_Shr64, |
| 2199 | binop(Iop_Sub64, |
| 2200 | binop(Iop_Shl64, |
| 2201 | binop(Iop_Xor64, mkexpr(ix), mkU64(1)), |
| 2202 | mkU8(62)), |
| 2203 | mkU64(1)), |
| 2204 | mkU8(61)), |
| 2205 | mkU64(1))); |
| 2206 | |
| 2207 | assign( |
| 2208 | termR, |
| 2209 | binop(Iop_And64, |
| 2210 | binop(Iop_And64, |
| 2211 | mkexpr(ix), |
| 2212 | binop(Iop_Shr64, mkexpr(ix), mkU8(1))), |
| 2213 | mkU64(1))); |
| 2214 | |
| 2215 | assign(nzcv, binop(Iop_Sub64, mkexpr(termL), mkexpr(termR))); |
| 2216 | return nzcv; |
| 2217 | } |
| 2218 | |
| 2219 | |
| 2220 | /*------------------------------------------------------------*/ |
| 2221 | /*--- Data processing (immediate) ---*/ |
| 2222 | /*------------------------------------------------------------*/ |
| 2223 | |
| 2224 | /* Helper functions for supporting "DecodeBitMasks" */ |
| 2225 | |
| 2226 | static ULong dbm_ROR ( Int width, ULong x, Int rot ) |
| 2227 | { |
| 2228 | vassert(width > 0 && width <= 64); |
| 2229 | vassert(rot >= 0 && rot < width); |
| 2230 | if (rot == 0) return x; |
| 2231 | ULong res = x >> rot; |
| 2232 | res |= (x << (width - rot)); |
| 2233 | if (width < 64) |
| 2234 | res &= ((1ULL << width) - 1); |
| 2235 | return res; |
| 2236 | } |
| 2237 | |
| 2238 | static ULong dbm_RepTo64( Int esize, ULong x ) |
| 2239 | { |
| 2240 | switch (esize) { |
| 2241 | case 64: |
| 2242 | return x; |
| 2243 | case 32: |
| 2244 | x &= 0xFFFFFFFF; x |= (x << 32); |
| 2245 | return x; |
| 2246 | case 16: |
| 2247 | x &= 0xFFFF; x |= (x << 16); x |= (x << 32); |
| 2248 | return x; |
| 2249 | case 8: |
| 2250 | x &= 0xFF; x |= (x << 8); x |= (x << 16); x |= (x << 32); |
| 2251 | return x; |
| 2252 | case 4: |
| 2253 | x &= 0xF; x |= (x << 4); x |= (x << 8); |
| 2254 | x |= (x << 16); x |= (x << 32); |
| 2255 | return x; |
| 2256 | case 2: |
| 2257 | x &= 0x3; x |= (x << 2); x |= (x << 4); x |= (x << 8); |
| 2258 | x |= (x << 16); x |= (x << 32); |
| 2259 | return x; |
| 2260 | default: |
| 2261 | break; |
| 2262 | } |
| 2263 | vpanic("dbm_RepTo64"); |
| 2264 | /*NOTREACHED*/ |
| 2265 | return 0; |
| 2266 | } |
| 2267 | |
| 2268 | static Int dbm_highestSetBit ( ULong x ) |
| 2269 | { |
| 2270 | Int i; |
| 2271 | for (i = 63; i >= 0; i--) { |
| 2272 | if (x & (1ULL << i)) |
| 2273 | return i; |
| 2274 | } |
| 2275 | vassert(x == 0); |
| 2276 | return -1; |
| 2277 | } |
| 2278 | |
| 2279 | static |
| 2280 | Bool dbm_DecodeBitMasks ( /*OUT*/ULong* wmask, /*OUT*/ULong* tmask, |
| 2281 | ULong immN, ULong imms, ULong immr, Bool immediate, |
| 2282 | UInt M /*32 or 64*/) |
| 2283 | { |
| 2284 | vassert(immN < (1ULL << 1)); |
| 2285 | vassert(imms < (1ULL << 6)); |
| 2286 | vassert(immr < (1ULL << 6)); |
| 2287 | vassert(immediate == False || immediate == True); |
| 2288 | vassert(M == 32 || M == 64); |
| 2289 | |
| 2290 | Int len = dbm_highestSetBit( ((immN << 6) & 64) | ((~imms) & 63) ); |
| 2291 | if (len < 1) { /* printf("fail1\n"); */ return False; } |
| 2292 | vassert(len <= 6); |
| 2293 | vassert(M >= (1 << len)); |
| 2294 | |
| 2295 | vassert(len >= 1 && len <= 6); |
| 2296 | ULong levels = // (zeroes(6 - len) << (6-len)) | ones(len); |
| 2297 | (1 << len) - 1; |
| 2298 | vassert(levels >= 1 && levels <= 63); |
| 2299 | |
| 2300 | if (immediate && ((imms & levels) == levels)) { |
| 2301 | /* printf("fail2 imms %llu levels %llu len %d\n", imms, levels, len); */ |
| 2302 | return False; |
| 2303 | } |
| 2304 | |
| 2305 | ULong S = imms & levels; |
| 2306 | ULong R = immr & levels; |
| 2307 | Int diff = S - R; |
| 2308 | diff &= 63; |
| 2309 | Int esize = 1 << len; |
| 2310 | vassert(2 <= esize && esize <= 64); |
| 2311 | |
| 2312 | /* Be careful of these (1ULL << (S+1)) - 1 expressions, and the |
| 2313 | same below with d. S can be 63 in which case we have an out of |
| 2314 | range and hence undefined shift. */ |
| 2315 | vassert(S >= 0 && S <= 63); |
| 2316 | vassert(esize >= (S+1)); |
| 2317 | ULong elem_s = // Zeroes(esize-(S+1)):Ones(S+1) |
| 2318 | //(1ULL << (S+1)) - 1; |
| 2319 | ((1ULL << S) - 1) + (1ULL << S); |
| 2320 | |
| 2321 | Int d = // diff<len-1:0> |
| 2322 | diff & ((1 << len)-1); |
| 2323 | vassert(esize >= (d+1)); |
| 2324 | vassert(d >= 0 && d <= 63); |
| 2325 | |
| 2326 | ULong elem_d = // Zeroes(esize-(d+1)):Ones(d+1) |
| 2327 | //(1ULL << (d+1)) - 1; |
| 2328 | ((1ULL << d) - 1) + (1ULL << d); |
| 2329 | |
| 2330 | if (esize != 64) vassert(elem_s < (1ULL << esize)); |
| 2331 | if (esize != 64) vassert(elem_d < (1ULL << esize)); |
| 2332 | |
| 2333 | if (wmask) *wmask = dbm_RepTo64(esize, dbm_ROR(esize, elem_s, R)); |
| 2334 | if (tmask) *tmask = dbm_RepTo64(esize, elem_d); |
| 2335 | |
| 2336 | return True; |
| 2337 | } |
| 2338 | |
| 2339 | |
| 2340 | static |
| 2341 | Bool dis_ARM64_data_processing_immediate(/*MB_OUT*/DisResult* dres, |
| 2342 | UInt insn) |
| 2343 | { |
| 2344 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 2345 | |
| 2346 | /* insn[28:23] |
| 2347 | 10000x PC-rel addressing |
| 2348 | 10001x Add/subtract (immediate) |
| 2349 | 100100 Logical (immediate) |
| 2350 | 100101 Move Wide (immediate) |
| 2351 | 100110 Bitfield |
| 2352 | 100111 Extract |
| 2353 | */ |
| 2354 | |
| 2355 | /* ------------------ ADD/SUB{,S} imm12 ------------------ */ |
| 2356 | if (INSN(28,24) == BITS5(1,0,0,0,1)) { |
| 2357 | Bool is64 = INSN(31,31) == 1; |
| 2358 | Bool isSub = INSN(30,30) == 1; |
| 2359 | Bool setCC = INSN(29,29) == 1; |
| 2360 | UInt sh = INSN(23,22); |
| 2361 | UInt uimm12 = INSN(21,10); |
| 2362 | UInt nn = INSN(9,5); |
| 2363 | UInt dd = INSN(4,0); |
| 2364 | const HChar* nm = isSub ? "sub" : "add"; |
| 2365 | if (sh >= 2) { |
| 2366 | /* Invalid; fall through */ |
| 2367 | } else { |
| 2368 | vassert(sh <= 1); |
| 2369 | uimm12 <<= (12 * sh); |
| 2370 | if (is64) { |
| 2371 | IRTemp argL = newTemp(Ity_I64); |
| 2372 | IRTemp argR = newTemp(Ity_I64); |
| 2373 | IRTemp res = newTemp(Ity_I64); |
| 2374 | assign(argL, getIReg64orSP(nn)); |
| 2375 | assign(argR, mkU64(uimm12)); |
| 2376 | assign(res, binop(isSub ? Iop_Sub64 : Iop_Add64, |
| 2377 | mkexpr(argL), mkexpr(argR))); |
| 2378 | if (setCC) { |
| 2379 | putIReg64orZR(dd, mkexpr(res)); |
| 2380 | setFlags_ADD_SUB(True/*is64*/, isSub, argL, argR); |
| 2381 | DIP("%ss %s, %s, 0x%x\n", |
| 2382 | nm, nameIReg64orZR(dd), nameIReg64orSP(nn), uimm12); |
| 2383 | } else { |
| 2384 | putIReg64orSP(dd, mkexpr(res)); |
| 2385 | DIP("%s %s, %s, 0x%x\n", |
| 2386 | nm, nameIReg64orSP(dd), nameIReg64orSP(nn), uimm12); |
| 2387 | } |
| 2388 | } else { |
| 2389 | IRTemp argL = newTemp(Ity_I32); |
| 2390 | IRTemp argR = newTemp(Ity_I32); |
| 2391 | IRTemp res = newTemp(Ity_I32); |
| 2392 | assign(argL, getIReg32orSP(nn)); |
| 2393 | assign(argR, mkU32(uimm12)); |
| 2394 | assign(res, binop(isSub ? Iop_Sub32 : Iop_Add32, |
| 2395 | mkexpr(argL), mkexpr(argR))); |
| 2396 | if (setCC) { |
| 2397 | putIReg32orZR(dd, mkexpr(res)); |
| 2398 | setFlags_ADD_SUB(False/*!is64*/, isSub, argL, argR); |
| 2399 | DIP("%ss %s, %s, 0x%x\n", |
| 2400 | nm, nameIReg32orZR(dd), nameIReg32orSP(nn), uimm12); |
| 2401 | } else { |
| 2402 | putIReg32orSP(dd, mkexpr(res)); |
| 2403 | DIP("%s %s, %s, 0x%x\n", |
| 2404 | nm, nameIReg32orSP(dd), nameIReg32orSP(nn), uimm12); |
| 2405 | } |
| 2406 | } |
| 2407 | return True; |
| 2408 | } |
| 2409 | } |
| 2410 | |
| 2411 | /* -------------------- ADR/ADRP -------------------- */ |
| 2412 | if (INSN(28,24) == BITS5(1,0,0,0,0)) { |
| 2413 | UInt bP = INSN(31,31); |
| 2414 | UInt immLo = INSN(30,29); |
| 2415 | UInt immHi = INSN(23,5); |
| 2416 | UInt rD = INSN(4,0); |
| 2417 | ULong uimm = (immHi << 2) | immLo; |
| 2418 | ULong simm = sx_to_64(uimm, 21); |
| 2419 | ULong val; |
| 2420 | if (bP) { |
| 2421 | val = (guest_PC_curr_instr & 0xFFFFFFFFFFFFF000ULL) + (simm << 12); |
| 2422 | } else { |
| 2423 | val = guest_PC_curr_instr + simm; |
| 2424 | } |
| 2425 | putIReg64orZR(rD, mkU64(val)); |
| 2426 | DIP("adr%s %s, 0x%llx\n", bP ? "p" : "", nameIReg64orZR(rD), val); |
| 2427 | return True; |
| 2428 | } |
| 2429 | |
| 2430 | /* -------------------- LOGIC(imm) -------------------- */ |
| 2431 | if (INSN(28,23) == BITS6(1,0,0,1,0,0)) { |
| 2432 | /* 31 30 28 22 21 15 9 4 |
| 2433 | sf op 100100 N immr imms Rn Rd |
| 2434 | op=00: AND Rd|SP, Rn, #imm |
| 2435 | op=01: ORR Rd|SP, Rn, #imm |
| 2436 | op=10: EOR Rd|SP, Rn, #imm |
| 2437 | op=11: ANDS Rd|ZR, Rn, #imm |
| 2438 | */ |
| 2439 | Bool is64 = INSN(31,31) == 1; |
| 2440 | UInt op = INSN(30,29); |
| 2441 | UInt N = INSN(22,22); |
| 2442 | UInt immR = INSN(21,16); |
| 2443 | UInt immS = INSN(15,10); |
| 2444 | UInt nn = INSN(9,5); |
| 2445 | UInt dd = INSN(4,0); |
| 2446 | ULong imm = 0; |
| 2447 | Bool ok; |
| 2448 | if (N == 1 && !is64) |
| 2449 | goto after_logic_imm; /* not allowed; fall through */ |
| 2450 | ok = dbm_DecodeBitMasks(&imm, NULL, |
| 2451 | N, immS, immR, True, is64 ? 64 : 32); |
| 2452 | if (!ok) |
| 2453 | goto after_logic_imm; |
| 2454 | |
| 2455 | const HChar* names[4] = { "and", "orr", "eor", "ands" }; |
| 2456 | const IROp ops64[4] = { Iop_And64, Iop_Or64, Iop_Xor64, Iop_And64 }; |
| 2457 | const IROp ops32[4] = { Iop_And32, Iop_Or32, Iop_Xor32, Iop_And32 }; |
| 2458 | |
| 2459 | vassert(op < 4); |
| 2460 | if (is64) { |
| 2461 | IRExpr* argL = getIReg64orZR(nn); |
| 2462 | IRExpr* argR = mkU64(imm); |
| 2463 | IRTemp res = newTemp(Ity_I64); |
| 2464 | assign(res, binop(ops64[op], argL, argR)); |
| 2465 | if (op < 3) { |
| 2466 | putIReg64orSP(dd, mkexpr(res)); |
| 2467 | DIP("%s %s, %s, 0x%llx\n", names[op], |
| 2468 | nameIReg64orSP(dd), nameIReg64orZR(nn), imm); |
| 2469 | } else { |
| 2470 | putIReg64orZR(dd, mkexpr(res)); |
| 2471 | setFlags_LOGIC(True/*is64*/, res); |
| 2472 | DIP("%s %s, %s, 0x%llx\n", names[op], |
| 2473 | nameIReg64orZR(dd), nameIReg64orZR(nn), imm); |
| 2474 | } |
| 2475 | } else { |
| 2476 | IRExpr* argL = getIReg32orZR(nn); |
| 2477 | IRExpr* argR = mkU32((UInt)imm); |
| 2478 | IRTemp res = newTemp(Ity_I32); |
| 2479 | assign(res, binop(ops32[op], argL, argR)); |
| 2480 | if (op < 3) { |
| 2481 | putIReg32orSP(dd, mkexpr(res)); |
| 2482 | DIP("%s %s, %s, 0x%x\n", names[op], |
| 2483 | nameIReg32orSP(dd), nameIReg32orZR(nn), (UInt)imm); |
| 2484 | } else { |
| 2485 | putIReg32orZR(dd, mkexpr(res)); |
| 2486 | setFlags_LOGIC(False/*!is64*/, res); |
| 2487 | DIP("%s %s, %s, 0x%x\n", names[op], |
| 2488 | nameIReg32orZR(dd), nameIReg32orZR(nn), (UInt)imm); |
| 2489 | } |
| 2490 | } |
| 2491 | return True; |
| 2492 | } |
| 2493 | after_logic_imm: |
| 2494 | |
| 2495 | /* -------------------- MOV{Z,N,K} -------------------- */ |
| 2496 | if (INSN(28,23) == BITS6(1,0,0,1,0,1)) { |
| 2497 | /* 31 30 28 22 20 4 |
| 2498 | | | | | | | |
| 2499 | sf 10 100 101 hw imm16 Rd MOV(Z) Rd, (imm16 << (16*hw)) |
| 2500 | sf 00 100 101 hw imm16 Rd MOV(N) Rd, ~(imm16 << (16*hw)) |
| 2501 | sf 11 100 101 hw imm16 Rd MOV(K) Rd, (imm16 << (16*hw)) |
| 2502 | */ |
| 2503 | Bool is64 = INSN(31,31) == 1; |
| 2504 | UInt subopc = INSN(30,29); |
| 2505 | UInt hw = INSN(22,21); |
| 2506 | UInt imm16 = INSN(20,5); |
| 2507 | UInt dd = INSN(4,0); |
| 2508 | if (subopc == BITS2(0,1) || (!is64 && hw >= 2)) { |
| 2509 | /* invalid; fall through */ |
| 2510 | } else { |
| 2511 | ULong imm64 = ((ULong)imm16) << (16 * hw); |
| 2512 | if (!is64) |
| 2513 | vassert(imm64 < 0x100000000ULL); |
| 2514 | switch (subopc) { |
| 2515 | case BITS2(1,0): // MOVZ |
| 2516 | putIRegOrZR(is64, dd, is64 ? mkU64(imm64) : mkU32((UInt)imm64)); |
| 2517 | DIP("movz %s, 0x%llx\n", nameIRegOrZR(is64, dd), imm64); |
| 2518 | break; |
| 2519 | case BITS2(0,0): // MOVN |
| 2520 | imm64 = ~imm64; |
| 2521 | if (!is64) |
| 2522 | imm64 &= 0xFFFFFFFFULL; |
| 2523 | putIRegOrZR(is64, dd, is64 ? mkU64(imm64) : mkU32((UInt)imm64)); |
| 2524 | DIP("movn %s, 0x%llx\n", nameIRegOrZR(is64, dd), imm64); |
| 2525 | break; |
| 2526 | case BITS2(1,1): // MOVK |
| 2527 | /* This is more complex. We are inserting a slice into |
| 2528 | the destination register, so we need to have the old |
| 2529 | value of it. */ |
| 2530 | if (is64) { |
| 2531 | IRTemp old = newTemp(Ity_I64); |
| 2532 | assign(old, getIReg64orZR(dd)); |
| 2533 | ULong mask = 0xFFFFULL << (16 * hw); |
| 2534 | IRExpr* res |
| 2535 | = binop(Iop_Or64, |
| 2536 | binop(Iop_And64, mkexpr(old), mkU64(~mask)), |
| 2537 | mkU64(imm64)); |
| 2538 | putIReg64orZR(dd, res); |
| 2539 | DIP("movk %s, 0x%x, lsl %u\n", |
| 2540 | nameIReg64orZR(dd), imm16, 16*hw); |
| 2541 | } else { |
| 2542 | IRTemp old = newTemp(Ity_I32); |
| 2543 | assign(old, getIReg32orZR(dd)); |
| 2544 | vassert(hw <= 1); |
| 2545 | UInt mask = 0xFFFF << (16 * hw); |
| 2546 | IRExpr* res |
| 2547 | = binop(Iop_Or32, |
| 2548 | binop(Iop_And32, mkexpr(old), mkU32(~mask)), |
| 2549 | mkU32((UInt)imm64)); |
| 2550 | putIReg32orZR(dd, res); |
| 2551 | DIP("movk %s, 0x%x, lsl %u\n", |
| 2552 | nameIReg32orZR(dd), imm16, 16*hw); |
| 2553 | } |
| 2554 | break; |
| 2555 | default: |
| 2556 | vassert(0); |
| 2557 | } |
| 2558 | return True; |
| 2559 | } |
| 2560 | } |
| 2561 | |
| 2562 | /* -------------------- {U,S,}BFM -------------------- */ |
| 2563 | /* 30 28 22 21 15 9 4 |
| 2564 | |
| 2565 | sf 10 100110 N immr imms nn dd |
| 2566 | UBFM Wd, Wn, #immr, #imms when sf=0, N=0, immr[5]=0, imms[5]=0 |
| 2567 | UBFM Xd, Xn, #immr, #imms when sf=1, N=1 |
| 2568 | |
| 2569 | sf 00 100110 N immr imms nn dd |
| 2570 | SBFM Wd, Wn, #immr, #imms when sf=0, N=0, immr[5]=0, imms[5]=0 |
| 2571 | SBFM Xd, Xn, #immr, #imms when sf=1, N=1 |
| 2572 | |
| 2573 | sf 01 100110 N immr imms nn dd |
| 2574 | BFM Wd, Wn, #immr, #imms when sf=0, N=0, immr[5]=0, imms[5]=0 |
| 2575 | BFM Xd, Xn, #immr, #imms when sf=1, N=1 |
| 2576 | */ |
| 2577 | if (INSN(28,23) == BITS6(1,0,0,1,1,0)) { |
| 2578 | UInt sf = INSN(31,31); |
| 2579 | UInt opc = INSN(30,29); |
| 2580 | UInt N = INSN(22,22); |
| 2581 | UInt immR = INSN(21,16); |
| 2582 | UInt immS = INSN(15,10); |
| 2583 | UInt nn = INSN(9,5); |
| 2584 | UInt dd = INSN(4,0); |
| 2585 | Bool inZero = False; |
| 2586 | Bool extend = False; |
| 2587 | const HChar* nm = "???"; |
| 2588 | /* skip invalid combinations */ |
| 2589 | switch (opc) { |
| 2590 | case BITS2(0,0): |
| 2591 | inZero = True; extend = True; nm = "sbfm"; break; |
| 2592 | case BITS2(0,1): |
| 2593 | inZero = False; extend = False; nm = "bfm"; break; |
| 2594 | case BITS2(1,0): |
| 2595 | inZero = True; extend = False; nm = "ubfm"; break; |
| 2596 | case BITS2(1,1): |
| 2597 | goto after_bfm; /* invalid */ |
| 2598 | default: |
| 2599 | vassert(0); |
| 2600 | } |
| 2601 | if (sf == 1 && N != 1) goto after_bfm; |
| 2602 | if (sf == 0 && (N != 0 || ((immR >> 5) & 1) != 0 |
| 2603 | || ((immS >> 5) & 1) != 0)) goto after_bfm; |
| 2604 | ULong wmask = 0, tmask = 0; |
| 2605 | Bool ok = dbm_DecodeBitMasks(&wmask, &tmask, |
| 2606 | N, immS, immR, False, sf == 1 ? 64 : 32); |
| 2607 | if (!ok) goto after_bfm; /* hmmm */ |
| 2608 | |
| 2609 | Bool is64 = sf == 1; |
| 2610 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2611 | |
| 2612 | IRTemp dst = newTemp(ty); |
| 2613 | IRTemp src = newTemp(ty); |
| 2614 | IRTemp bot = newTemp(ty); |
| 2615 | IRTemp top = newTemp(ty); |
| 2616 | IRTemp res = newTemp(ty); |
| 2617 | assign(dst, inZero ? mkU(ty,0) : getIRegOrZR(is64, dd)); |
| 2618 | assign(src, getIRegOrZR(is64, nn)); |
| 2619 | /* perform bitfield move on low bits */ |
| 2620 | assign(bot, binop(mkOR(ty), |
| 2621 | binop(mkAND(ty), mkexpr(dst), mkU(ty, ~wmask)), |
| 2622 | binop(mkAND(ty), mkexpr(mathROR(ty, src, immR)), |
| 2623 | mkU(ty, wmask)))); |
| 2624 | /* determine extension bits (sign, zero or dest register) */ |
| 2625 | assign(top, mkexpr(extend ? mathREPLICATE(ty, src, immS) : dst)); |
| 2626 | /* combine extension bits and result bits */ |
| 2627 | assign(res, binop(mkOR(ty), |
| 2628 | binop(mkAND(ty), mkexpr(top), mkU(ty, ~tmask)), |
| 2629 | binop(mkAND(ty), mkexpr(bot), mkU(ty, tmask)))); |
| 2630 | putIRegOrZR(is64, dd, mkexpr(res)); |
| 2631 | DIP("%s %s, %s, immR=%u, immS=%u\n", |
| 2632 | nm, nameIRegOrZR(is64, dd), nameIRegOrZR(is64, nn), immR, immS); |
| 2633 | return True; |
| 2634 | } |
| 2635 | after_bfm: |
| 2636 | |
| 2637 | /* ---------------------- EXTR ---------------------- */ |
| 2638 | /* 30 28 22 20 15 9 4 |
| 2639 | 1 00 100111 10 m imm6 n d EXTR Xd, Xn, Xm, #imm6 |
| 2640 | 0 00 100111 00 m imm6 n d EXTR Wd, Wn, Wm, #imm6 when #imm6 < 32 |
| 2641 | */ |
| 2642 | if (INSN(30,23) == BITS8(0,0,1,0,0,1,1,1) && INSN(21,21) == 0) { |
| 2643 | Bool is64 = INSN(31,31) == 1; |
| 2644 | UInt mm = INSN(20,16); |
| 2645 | UInt imm6 = INSN(15,10); |
| 2646 | UInt nn = INSN(9,5); |
| 2647 | UInt dd = INSN(4,0); |
| 2648 | Bool valid = True; |
| 2649 | if (INSN(31,31) != INSN(22,22)) |
| 2650 | valid = False; |
| 2651 | if (!is64 && imm6 >= 32) |
| 2652 | valid = False; |
| 2653 | if (!valid) goto after_extr; |
| 2654 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2655 | IRTemp srcHi = newTemp(ty); |
| 2656 | IRTemp srcLo = newTemp(ty); |
| 2657 | IRTemp res = newTemp(ty); |
| 2658 | assign(srcHi, getIRegOrZR(is64, nn)); |
| 2659 | assign(srcLo, getIRegOrZR(is64, mm)); |
| 2660 | if (imm6 == 0) { |
| 2661 | assign(res, mkexpr(srcLo)); |
| 2662 | } else { |
| 2663 | UInt szBits = 8 * sizeofIRType(ty); |
| 2664 | vassert(imm6 > 0 && imm6 < szBits); |
| 2665 | assign(res, binop(mkOR(ty), |
| 2666 | binop(mkSHL(ty), mkexpr(srcHi), mkU8(szBits-imm6)), |
| 2667 | binop(mkSHR(ty), mkexpr(srcLo), mkU8(imm6)))); |
| 2668 | } |
| 2669 | putIRegOrZR(is64, dd, mkexpr(res)); |
| 2670 | DIP("extr %s, %s, %s, #%u\n", |
| 2671 | nameIRegOrZR(is64,dd), |
| 2672 | nameIRegOrZR(is64,nn), nameIRegOrZR(is64,mm), imm6); |
| 2673 | return True; |
| 2674 | } |
| 2675 | after_extr: |
| 2676 | |
| 2677 | vex_printf("ARM64 front end: data_processing_immediate\n"); |
| 2678 | return False; |
| 2679 | # undef INSN |
| 2680 | } |
| 2681 | |
| 2682 | |
| 2683 | /*------------------------------------------------------------*/ |
| 2684 | /*--- Data processing (register) instructions ---*/ |
| 2685 | /*------------------------------------------------------------*/ |
| 2686 | |
| 2687 | static const HChar* nameSH ( UInt sh ) { |
| 2688 | switch (sh) { |
| 2689 | case 0: return "lsl"; |
| 2690 | case 1: return "lsr"; |
| 2691 | case 2: return "asr"; |
| 2692 | case 3: return "ror"; |
| 2693 | default: vassert(0); |
| 2694 | } |
| 2695 | } |
| 2696 | |
| 2697 | /* Generate IR to get a register value, possibly shifted by an |
| 2698 | immediate. Returns either a 32- or 64-bit temporary holding the |
| 2699 | result. After the shift, the value can optionally be NOT-ed |
| 2700 | too. |
| 2701 | |
| 2702 | sh_how coding: 00=SHL, 01=SHR, 10=SAR, 11=ROR. sh_amt may only be |
| 2703 | in the range 0 to (is64 ? 64 : 32)-1. For some instructions, ROR |
| 2704 | isn't allowed, but it's the job of the caller to check that. |
| 2705 | */ |
| 2706 | static IRTemp getShiftedIRegOrZR ( Bool is64, |
| 2707 | UInt sh_how, UInt sh_amt, UInt regNo, |
| 2708 | Bool invert ) |
| 2709 | { |
| 2710 | vassert(sh_how < 4); |
| 2711 | vassert(sh_amt < (is64 ? 64 : 32)); |
| 2712 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2713 | IRTemp t0 = newTemp(ty); |
| 2714 | assign(t0, getIRegOrZR(is64, regNo)); |
| 2715 | IRTemp t1 = newTemp(ty); |
| 2716 | switch (sh_how) { |
| 2717 | case BITS2(0,0): |
| 2718 | assign(t1, binop(mkSHL(ty), mkexpr(t0), mkU8(sh_amt))); |
| 2719 | break; |
| 2720 | case BITS2(0,1): |
| 2721 | assign(t1, binop(mkSHR(ty), mkexpr(t0), mkU8(sh_amt))); |
| 2722 | break; |
| 2723 | case BITS2(1,0): |
| 2724 | assign(t1, binop(mkSAR(ty), mkexpr(t0), mkU8(sh_amt))); |
| 2725 | break; |
| 2726 | case BITS2(1,1): |
| 2727 | assign(t1, mkexpr(mathROR(ty, t0, sh_amt))); |
| 2728 | break; |
| 2729 | default: |
| 2730 | vassert(0); |
| 2731 | } |
| 2732 | if (invert) { |
| 2733 | IRTemp t2 = newTemp(ty); |
| 2734 | assign(t2, unop(mkNOT(ty), mkexpr(t1))); |
| 2735 | return t2; |
| 2736 | } else { |
| 2737 | return t1; |
| 2738 | } |
| 2739 | } |
| 2740 | |
| 2741 | |
| 2742 | static |
| 2743 | Bool dis_ARM64_data_processing_register(/*MB_OUT*/DisResult* dres, |
| 2744 | UInt insn) |
| 2745 | { |
| 2746 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 2747 | |
| 2748 | /* ------------------- ADD/SUB(reg) ------------------- */ |
| 2749 | /* x==0 => 32 bit op x==1 => 64 bit op |
| 2750 | sh: 00=LSL, 01=LSR, 10=ASR, 11=ROR(NOT ALLOWED) |
| 2751 | |
| 2752 | 31 30 29 28 23 21 20 15 9 4 |
| 2753 | | | | | | | | | | | |
| 2754 | x 0 0 01011 sh 0 Rm imm6 Rn Rd ADD Rd,Rn, sh(Rm,imm6) |
| 2755 | x 0 1 01011 sh 0 Rm imm6 Rn Rd ADDS Rd,Rn, sh(Rm,imm6) |
| 2756 | x 1 0 01011 sh 0 Rm imm6 Rn Rd SUB Rd,Rn, sh(Rm,imm6) |
| 2757 | x 1 1 01011 sh 0 Rm imm6 Rn Rd SUBS Rd,Rn, sh(Rm,imm6) |
| 2758 | */ |
| 2759 | if (INSN(28,24) == BITS5(0,1,0,1,1) && INSN(21,21) == 0) { |
| 2760 | UInt bX = INSN(31,31); |
| 2761 | UInt bOP = INSN(30,30); /* 0: ADD, 1: SUB */ |
| 2762 | UInt bS = INSN(29, 29); /* set flags? */ |
| 2763 | UInt sh = INSN(23,22); |
| 2764 | UInt rM = INSN(20,16); |
| 2765 | UInt imm6 = INSN(15,10); |
| 2766 | UInt rN = INSN(9,5); |
| 2767 | UInt rD = INSN(4,0); |
| 2768 | Bool isSUB = bOP == 1; |
| 2769 | Bool is64 = bX == 1; |
| 2770 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2771 | if ((!is64 && imm6 > 31) || sh == BITS2(1,1)) { |
| 2772 | /* invalid; fall through */ |
| 2773 | } else { |
| 2774 | IRTemp argL = newTemp(ty); |
| 2775 | assign(argL, getIRegOrZR(is64, rN)); |
| 2776 | IRTemp argR = getShiftedIRegOrZR(is64, sh, imm6, rM, False); |
| 2777 | IROp op = isSUB ? mkSUB(ty) : mkADD(ty); |
| 2778 | IRTemp res = newTemp(ty); |
| 2779 | assign(res, binop(op, mkexpr(argL), mkexpr(argR))); |
| 2780 | if (rD != 31) putIRegOrZR(is64, rD, mkexpr(res)); |
| 2781 | if (bS) { |
| 2782 | setFlags_ADD_SUB(is64, isSUB, argL, argR); |
| 2783 | } |
| 2784 | DIP("%s%s %s, %s, %s, %s #%u\n", |
| 2785 | bOP ? "sub" : "add", bS ? "s" : "", |
| 2786 | nameIRegOrZR(is64, rD), nameIRegOrZR(is64, rN), |
| 2787 | nameIRegOrZR(is64, rM), nameSH(sh), imm6); |
| 2788 | return True; |
| 2789 | } |
| 2790 | } |
| 2791 | |
sewardj | dee3050 | 2014-06-04 13:09:44 +0000 | [diff] [blame] | 2792 | /* ------------------- ADC/SBC(reg) ------------------- */ |
| 2793 | /* x==0 => 32 bit op x==1 => 64 bit op |
| 2794 | |
| 2795 | 31 30 29 28 23 21 20 15 9 4 |
| 2796 | | | | | | | | | | | |
| 2797 | x 0 0 11010 00 0 Rm 000000 Rn Rd ADC Rd,Rn,Rm |
| 2798 | x 0 1 11010 00 0 Rm 000000 Rn Rd ADCS Rd,Rn,Rm |
| 2799 | x 1 0 11010 00 0 Rm 000000 Rn Rd SBC Rd,Rn,Rm |
| 2800 | x 1 1 11010 00 0 Rm 000000 Rn Rd SBCS Rd,Rn,Rm |
| 2801 | */ |
| 2802 | |
| 2803 | if (INSN(28,21) == BITS8(1,1,0,1,0,0,0,0) && INSN(15,10) == 0 ) { |
| 2804 | UInt bX = INSN(31,31); |
| 2805 | UInt bOP = INSN(30,30); /* 0: ADC, 1: SBC */ |
| 2806 | UInt bS = INSN(29,29); /* set flags */ |
| 2807 | UInt rM = INSN(20,16); |
| 2808 | UInt rN = INSN(9,5); |
| 2809 | UInt rD = INSN(4,0); |
| 2810 | |
| 2811 | Bool isSUB = bOP == 1; |
| 2812 | Bool is64 = bX == 1; |
| 2813 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2814 | |
| 2815 | IRTemp oldC = newTemp(ty); |
| 2816 | assign(oldC, |
| 2817 | is64 ? mk_arm64g_calculate_flag_c() |
| 2818 | : unop(Iop_64to32, mk_arm64g_calculate_flag_c()) ); |
| 2819 | |
| 2820 | IRTemp argL = newTemp(ty); |
| 2821 | assign(argL, getIRegOrZR(is64, rN)); |
| 2822 | IRTemp argR = newTemp(ty); |
| 2823 | assign(argR, getIRegOrZR(is64, rM)); |
| 2824 | |
| 2825 | IROp op = isSUB ? mkSUB(ty) : mkADD(ty); |
| 2826 | IRTemp res = newTemp(ty); |
| 2827 | if (isSUB) { |
| 2828 | IRExpr* one = is64 ? mkU64(1) : mkU32(1); |
| 2829 | IROp xorOp = is64 ? Iop_Xor64 : Iop_Xor32; |
| 2830 | assign(res, |
| 2831 | binop(op, |
| 2832 | binop(op, mkexpr(argL), mkexpr(argR)), |
| 2833 | binop(xorOp, mkexpr(oldC), one))); |
| 2834 | } else { |
| 2835 | assign(res, |
| 2836 | binop(op, |
| 2837 | binop(op, mkexpr(argL), mkexpr(argR)), |
| 2838 | mkexpr(oldC))); |
| 2839 | } |
| 2840 | |
| 2841 | if (rD != 31) putIRegOrZR(is64, rD, mkexpr(res)); |
| 2842 | |
| 2843 | if (bS) { |
| 2844 | setFlags_ADC_SBC(is64, isSUB, argL, argR, oldC); |
| 2845 | } |
| 2846 | |
| 2847 | DIP("%s%s %s, %s, %s\n", |
| 2848 | bOP ? "sbc" : "adc", bS ? "s" : "", |
| 2849 | nameIRegOrZR(is64, rD), nameIRegOrZR(is64, rN), |
| 2850 | nameIRegOrZR(is64, rM)); |
| 2851 | return True; |
| 2852 | } |
| 2853 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 2854 | /* -------------------- LOGIC(reg) -------------------- */ |
| 2855 | /* x==0 => 32 bit op x==1 => 64 bit op |
| 2856 | N==0 => inv? is no-op (no inversion) |
| 2857 | N==1 => inv? is NOT |
| 2858 | sh: 00=LSL, 01=LSR, 10=ASR, 11=ROR |
| 2859 | |
| 2860 | 31 30 28 23 21 20 15 9 4 |
| 2861 | | | | | | | | | | |
| 2862 | x 00 01010 sh N Rm imm6 Rn Rd AND Rd,Rn, inv?(sh(Rm,imm6)) |
| 2863 | x 01 01010 sh N Rm imm6 Rn Rd ORR Rd,Rn, inv?(sh(Rm,imm6)) |
| 2864 | x 10 01010 sh N Rm imm6 Rn Rd EOR Rd,Rn, inv?(sh(Rm,imm6)) |
| 2865 | x 11 01010 sh N Rm imm6 Rn Rd ANDS Rd,Rn, inv?(sh(Rm,imm6)) |
| 2866 | With N=1, the names are: BIC ORN EON BICS |
| 2867 | */ |
| 2868 | if (INSN(28,24) == BITS5(0,1,0,1,0)) { |
| 2869 | UInt bX = INSN(31,31); |
| 2870 | UInt sh = INSN(23,22); |
| 2871 | UInt bN = INSN(21,21); |
| 2872 | UInt rM = INSN(20,16); |
| 2873 | UInt imm6 = INSN(15,10); |
| 2874 | UInt rN = INSN(9,5); |
| 2875 | UInt rD = INSN(4,0); |
| 2876 | Bool is64 = bX == 1; |
| 2877 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2878 | if (!is64 && imm6 > 31) { |
| 2879 | /* invalid; fall though */ |
| 2880 | } else { |
| 2881 | IRTemp argL = newTemp(ty); |
| 2882 | assign(argL, getIRegOrZR(is64, rN)); |
| 2883 | IRTemp argR = getShiftedIRegOrZR(is64, sh, imm6, rM, bN == 1); |
| 2884 | IROp op = Iop_INVALID; |
| 2885 | switch (INSN(30,29)) { |
| 2886 | case BITS2(0,0): case BITS2(1,1): op = mkAND(ty); break; |
| 2887 | case BITS2(0,1): op = mkOR(ty); break; |
| 2888 | case BITS2(1,0): op = mkXOR(ty); break; |
| 2889 | default: vassert(0); |
| 2890 | } |
| 2891 | IRTemp res = newTemp(ty); |
| 2892 | assign(res, binop(op, mkexpr(argL), mkexpr(argR))); |
| 2893 | if (INSN(30,29) == BITS2(1,1)) { |
| 2894 | setFlags_LOGIC(is64, res); |
| 2895 | } |
| 2896 | putIRegOrZR(is64, rD, mkexpr(res)); |
| 2897 | |
| 2898 | static const HChar* names_op[8] |
| 2899 | = { "and", "orr", "eor", "ands", "bic", "orn", "eon", "bics" }; |
| 2900 | vassert(((bN << 2) | INSN(30,29)) < 8); |
| 2901 | const HChar* nm_op = names_op[(bN << 2) | INSN(30,29)]; |
| 2902 | /* Special-case the printing of "MOV" */ |
| 2903 | if (rN == 31/*zr*/ && sh == 0/*LSL*/ && imm6 == 0 && bN == 0) { |
| 2904 | DIP("mov %s, %s\n", nameIRegOrZR(is64, rD), |
| 2905 | nameIRegOrZR(is64, rM)); |
| 2906 | } else { |
| 2907 | DIP("%s %s, %s, %s, %s #%u\n", nm_op, |
| 2908 | nameIRegOrZR(is64, rD), nameIRegOrZR(is64, rN), |
| 2909 | nameIRegOrZR(is64, rM), nameSH(sh), imm6); |
| 2910 | } |
| 2911 | return True; |
| 2912 | } |
| 2913 | } |
| 2914 | |
| 2915 | /* -------------------- {U,S}MULH -------------------- */ |
| 2916 | /* 31 23 22 20 15 9 4 |
| 2917 | 10011011 1 10 Rm 011111 Rn Rd UMULH Xd,Xn,Xm |
| 2918 | 10011011 0 10 Rm 011111 Rn Rd SMULH Xd,Xn,Xm |
| 2919 | */ |
| 2920 | if (INSN(31,24) == BITS8(1,0,0,1,1,0,1,1) |
sewardj | 7fce7cc | 2014-05-07 09:41:40 +0000 | [diff] [blame] | 2921 | && INSN(22,21) == BITS2(1,0) && INSN(15,10) == BITS6(0,1,1,1,1,1)) { |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 2922 | Bool isU = INSN(23,23) == 1; |
| 2923 | UInt mm = INSN(20,16); |
| 2924 | UInt nn = INSN(9,5); |
| 2925 | UInt dd = INSN(4,0); |
| 2926 | putIReg64orZR(dd, unop(Iop_128HIto64, |
| 2927 | binop(isU ? Iop_MullU64 : Iop_MullS64, |
| 2928 | getIReg64orZR(nn), getIReg64orZR(mm)))); |
| 2929 | DIP("%cmulh %s, %s, %s\n", |
| 2930 | isU ? 'u' : 's', |
| 2931 | nameIReg64orZR(dd), nameIReg64orZR(nn), nameIReg64orZR(mm)); |
| 2932 | return True; |
| 2933 | } |
| 2934 | |
| 2935 | /* -------------------- M{ADD,SUB} -------------------- */ |
| 2936 | /* 31 30 20 15 14 9 4 |
| 2937 | sf 00 11011 000 m 0 a n r MADD Rd,Rn,Rm,Ra d = a+m*n |
| 2938 | sf 00 11011 000 m 1 a n r MADD Rd,Rn,Rm,Ra d = a-m*n |
| 2939 | */ |
| 2940 | if (INSN(30,21) == BITS10(0,0,1,1,0,1,1,0,0,0)) { |
| 2941 | Bool is64 = INSN(31,31) == 1; |
| 2942 | UInt mm = INSN(20,16); |
| 2943 | Bool isAdd = INSN(15,15) == 0; |
| 2944 | UInt aa = INSN(14,10); |
| 2945 | UInt nn = INSN(9,5); |
| 2946 | UInt dd = INSN(4,0); |
| 2947 | if (is64) { |
| 2948 | putIReg64orZR( |
| 2949 | dd, |
| 2950 | binop(isAdd ? Iop_Add64 : Iop_Sub64, |
| 2951 | getIReg64orZR(aa), |
| 2952 | binop(Iop_Mul64, getIReg64orZR(mm), getIReg64orZR(nn)))); |
| 2953 | } else { |
| 2954 | putIReg32orZR( |
| 2955 | dd, |
| 2956 | binop(isAdd ? Iop_Add32 : Iop_Sub32, |
| 2957 | getIReg32orZR(aa), |
| 2958 | binop(Iop_Mul32, getIReg32orZR(mm), getIReg32orZR(nn)))); |
| 2959 | } |
| 2960 | DIP("%s %s, %s, %s, %s\n", |
| 2961 | isAdd ? "madd" : "msub", |
| 2962 | nameIRegOrZR(is64, dd), nameIRegOrZR(is64, nn), |
| 2963 | nameIRegOrZR(is64, mm), nameIRegOrZR(is64, aa)); |
| 2964 | return True; |
| 2965 | } |
| 2966 | |
| 2967 | /* ---------------- CS{EL,INC,INV,NEG} ---------------- */ |
| 2968 | /* 31 30 28 20 15 11 9 4 |
| 2969 | sf 00 1101 0100 mm cond 00 nn dd CSEL Rd,Rn,Rm |
| 2970 | sf 00 1101 0100 mm cond 01 nn dd CSINC Rd,Rn,Rm |
| 2971 | sf 10 1101 0100 mm cond 00 nn dd CSINV Rd,Rn,Rm |
| 2972 | sf 10 1101 0100 mm cond 01 nn dd CSNEG Rd,Rn,Rm |
| 2973 | In all cases, the operation is: Rd = if cond then Rn else OP(Rm) |
| 2974 | */ |
| 2975 | if (INSN(29,21) == BITS9(0, 1,1,0,1, 0,1,0,0) && INSN(11,11) == 0) { |
| 2976 | Bool is64 = INSN(31,31) == 1; |
| 2977 | UInt b30 = INSN(30,30); |
| 2978 | UInt mm = INSN(20,16); |
| 2979 | UInt cond = INSN(15,12); |
| 2980 | UInt b10 = INSN(10,10); |
| 2981 | UInt nn = INSN(9,5); |
| 2982 | UInt dd = INSN(4,0); |
| 2983 | UInt op = (b30 << 1) | b10; /* 00=id 01=inc 10=inv 11=neg */ |
| 2984 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2985 | IRExpr* argL = getIRegOrZR(is64, nn); |
| 2986 | IRExpr* argR = getIRegOrZR(is64, mm); |
| 2987 | switch (op) { |
| 2988 | case BITS2(0,0): |
| 2989 | break; |
| 2990 | case BITS2(0,1): |
| 2991 | argR = binop(mkADD(ty), argR, mkU(ty,1)); |
| 2992 | break; |
| 2993 | case BITS2(1,0): |
| 2994 | argR = unop(mkNOT(ty), argR); |
| 2995 | break; |
| 2996 | case BITS2(1,1): |
| 2997 | argR = binop(mkSUB(ty), mkU(ty,0), argR); |
| 2998 | break; |
| 2999 | default: |
| 3000 | vassert(0); |
| 3001 | } |
| 3002 | putIRegOrZR( |
| 3003 | is64, dd, |
| 3004 | IRExpr_ITE(unop(Iop_64to1, mk_arm64g_calculate_condition(cond)), |
| 3005 | argL, argR) |
| 3006 | ); |
| 3007 | const HChar* op_nm[4] = { "csel", "csinc", "csinv", "csneg" }; |
| 3008 | DIP("%s %s, %s, %s, %s\n", op_nm[op], |
| 3009 | nameIRegOrZR(is64, dd), nameIRegOrZR(is64, nn), |
| 3010 | nameIRegOrZR(is64, mm), nameCC(cond)); |
| 3011 | return True; |
| 3012 | } |
| 3013 | |
| 3014 | /* -------------- ADD/SUB(extended reg) -------------- */ |
| 3015 | /* 28 20 15 12 9 4 |
| 3016 | 000 01011 00 1 m opt imm3 n d ADD Wd|SP, Wn|SP, Wm ext&lsld |
| 3017 | 100 01011 00 1 m opt imm3 n d ADD Xd|SP, Xn|SP, Rm ext&lsld |
| 3018 | |
| 3019 | 001 01011 00 1 m opt imm3 n d ADDS Wd, Wn|SP, Wm ext&lsld |
| 3020 | 101 01011 00 1 m opt imm3 n d ADDS Xd, Xn|SP, Rm ext&lsld |
| 3021 | |
| 3022 | 010 01011 00 1 m opt imm3 n d SUB Wd|SP, Wn|SP, Wm ext&lsld |
| 3023 | 110 01011 00 1 m opt imm3 n d SUB Xd|SP, Xn|SP, Rm ext&lsld |
| 3024 | |
| 3025 | 011 01011 00 1 m opt imm3 n d SUBS Wd, Wn|SP, Wm ext&lsld |
| 3026 | 111 01011 00 1 m opt imm3 n d SUBS Xd, Xn|SP, Rm ext&lsld |
| 3027 | |
| 3028 | The 'm' operand is extended per opt, thusly: |
| 3029 | |
| 3030 | 000 Xm & 0xFF UXTB |
| 3031 | 001 Xm & 0xFFFF UXTH |
| 3032 | 010 Xm & (2^32)-1 UXTW |
| 3033 | 011 Xm UXTX |
| 3034 | |
| 3035 | 100 Xm sx from bit 7 SXTB |
| 3036 | 101 Xm sx from bit 15 SXTH |
| 3037 | 110 Xm sx from bit 31 SXTW |
| 3038 | 111 Xm SXTX |
| 3039 | |
| 3040 | In the 64 bit case (bit31 == 1), UXTX and SXTX are the identity |
| 3041 | operation on Xm. In the 32 bit case, UXTW, UXTX, SXTW and SXTX |
| 3042 | are the identity operation on Wm. |
| 3043 | |
| 3044 | After extension, the value is shifted left by imm3 bits, which |
| 3045 | may only be in the range 0 .. 4 inclusive. |
| 3046 | */ |
| 3047 | if (INSN(28,21) == BITS8(0,1,0,1,1,0,0,1) && INSN(12,10) <= 4) { |
| 3048 | Bool is64 = INSN(31,31) == 1; |
| 3049 | Bool isSub = INSN(30,30) == 1; |
| 3050 | Bool setCC = INSN(29,29) == 1; |
| 3051 | UInt mm = INSN(20,16); |
| 3052 | UInt opt = INSN(15,13); |
| 3053 | UInt imm3 = INSN(12,10); |
| 3054 | UInt nn = INSN(9,5); |
| 3055 | UInt dd = INSN(4,0); |
| 3056 | const HChar* nameExt[8] = { "uxtb", "uxth", "uxtw", "uxtx", |
| 3057 | "sxtb", "sxth", "sxtw", "sxtx" }; |
| 3058 | /* Do almost the same thing in the 32- and 64-bit cases. */ |
| 3059 | IRTemp xN = newTemp(Ity_I64); |
| 3060 | IRTemp xM = newTemp(Ity_I64); |
| 3061 | assign(xN, getIReg64orSP(nn)); |
| 3062 | assign(xM, getIReg64orZR(mm)); |
| 3063 | IRExpr* xMw = mkexpr(xM); /* "xM widened" */ |
| 3064 | Int shSX = 0; |
| 3065 | /* widen Xm .. */ |
| 3066 | switch (opt) { |
| 3067 | case BITS3(0,0,0): // UXTB |
| 3068 | xMw = binop(Iop_And64, xMw, mkU64(0xFF)); break; |
| 3069 | case BITS3(0,0,1): // UXTH |
| 3070 | xMw = binop(Iop_And64, xMw, mkU64(0xFFFF)); break; |
| 3071 | case BITS3(0,1,0): // UXTW -- noop for the 32bit case |
| 3072 | if (is64) { |
| 3073 | xMw = unop(Iop_32Uto64, unop(Iop_64to32, xMw)); |
| 3074 | } |
| 3075 | break; |
| 3076 | case BITS3(0,1,1): // UXTX -- always a noop |
| 3077 | break; |
| 3078 | case BITS3(1,0,0): // SXTB |
| 3079 | shSX = 56; goto sxTo64; |
| 3080 | case BITS3(1,0,1): // SXTH |
| 3081 | shSX = 48; goto sxTo64; |
| 3082 | case BITS3(1,1,0): // SXTW -- noop for the 32bit case |
| 3083 | if (is64) { |
| 3084 | shSX = 32; goto sxTo64; |
| 3085 | } |
| 3086 | break; |
| 3087 | case BITS3(1,1,1): // SXTX -- always a noop |
| 3088 | break; |
| 3089 | sxTo64: |
| 3090 | vassert(shSX >= 32); |
| 3091 | xMw = binop(Iop_Sar64, binop(Iop_Shl64, xMw, mkU8(shSX)), |
| 3092 | mkU8(shSX)); |
| 3093 | break; |
| 3094 | default: |
| 3095 | vassert(0); |
| 3096 | } |
| 3097 | /* and now shift */ |
| 3098 | IRTemp argL = xN; |
| 3099 | IRTemp argR = newTemp(Ity_I64); |
| 3100 | assign(argR, binop(Iop_Shl64, xMw, mkU8(imm3))); |
| 3101 | IRTemp res = newTemp(Ity_I64); |
| 3102 | assign(res, binop(isSub ? Iop_Sub64 : Iop_Add64, |
| 3103 | mkexpr(argL), mkexpr(argR))); |
| 3104 | if (is64) { |
| 3105 | if (setCC) { |
| 3106 | putIReg64orZR(dd, mkexpr(res)); |
| 3107 | setFlags_ADD_SUB(True/*is64*/, isSub, argL, argR); |
| 3108 | } else { |
| 3109 | putIReg64orSP(dd, mkexpr(res)); |
| 3110 | } |
| 3111 | } else { |
| 3112 | if (setCC) { |
| 3113 | IRTemp argL32 = newTemp(Ity_I32); |
| 3114 | IRTemp argR32 = newTemp(Ity_I32); |
| 3115 | putIReg32orZR(dd, unop(Iop_64to32, mkexpr(res))); |
| 3116 | assign(argL32, unop(Iop_64to32, mkexpr(argL))); |
| 3117 | assign(argR32, unop(Iop_64to32, mkexpr(argR))); |
| 3118 | setFlags_ADD_SUB(False/*!is64*/, isSub, argL32, argR32); |
| 3119 | } else { |
| 3120 | putIReg32orSP(dd, unop(Iop_64to32, mkexpr(res))); |
| 3121 | } |
| 3122 | } |
| 3123 | DIP("%s%s %s, %s, %s %s lsl %u\n", |
| 3124 | isSub ? "sub" : "add", setCC ? "s" : "", |
| 3125 | setCC ? nameIRegOrZR(is64, dd) : nameIRegOrSP(is64, dd), |
| 3126 | nameIRegOrSP(is64, nn), nameIRegOrSP(is64, mm), |
| 3127 | nameExt[opt], imm3); |
| 3128 | return True; |
| 3129 | } |
| 3130 | |
| 3131 | /* ---------------- CCMP/CCMN(imm) ---------------- */ |
| 3132 | /* Bizarrely, these appear in the "data processing register" |
| 3133 | category, even though they are operations against an |
| 3134 | immediate. */ |
| 3135 | /* 31 29 20 15 11 9 3 |
| 3136 | sf 1 111010010 imm5 cond 10 Rn 0 nzcv CCMP Rn, #imm5, #nzcv, cond |
| 3137 | sf 0 111010010 imm5 cond 10 Rn 0 nzcv CCMN Rn, #imm5, #nzcv, cond |
| 3138 | |
| 3139 | Operation is: |
| 3140 | (CCMP) flags = if cond then flags-after-sub(Rn,imm5) else nzcv |
| 3141 | (CCMN) flags = if cond then flags-after-add(Rn,imm5) else nzcv |
| 3142 | */ |
| 3143 | if (INSN(29,21) == BITS9(1,1,1,0,1,0,0,1,0) |
| 3144 | && INSN(11,10) == BITS2(1,0) && INSN(4,4) == 0) { |
| 3145 | Bool is64 = INSN(31,31) == 1; |
| 3146 | Bool isSUB = INSN(30,30) == 1; |
| 3147 | UInt imm5 = INSN(20,16); |
| 3148 | UInt cond = INSN(15,12); |
| 3149 | UInt nn = INSN(9,5); |
| 3150 | UInt nzcv = INSN(3,0); |
| 3151 | |
| 3152 | IRTemp condT = newTemp(Ity_I1); |
| 3153 | assign(condT, unop(Iop_64to1, mk_arm64g_calculate_condition(cond))); |
| 3154 | |
| 3155 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 3156 | IRTemp argL = newTemp(ty); |
| 3157 | IRTemp argR = newTemp(ty); |
| 3158 | |
| 3159 | if (is64) { |
| 3160 | assign(argL, getIReg64orZR(nn)); |
| 3161 | assign(argR, mkU64(imm5)); |
| 3162 | } else { |
| 3163 | assign(argL, getIReg32orZR(nn)); |
| 3164 | assign(argR, mkU32(imm5)); |
| 3165 | } |
| 3166 | setFlags_ADD_SUB_conditionally(is64, isSUB, condT, argL, argR, nzcv); |
| 3167 | |
| 3168 | DIP("ccm%c %s, #%u, #%u, %s\n", |
| 3169 | isSUB ? 'p' : 'n', nameIRegOrZR(is64, nn), |
| 3170 | imm5, nzcv, nameCC(cond)); |
| 3171 | return True; |
| 3172 | } |
| 3173 | |
| 3174 | /* ---------------- CCMP/CCMN(reg) ---------------- */ |
| 3175 | /* 31 29 20 15 11 9 3 |
| 3176 | sf 1 111010010 Rm cond 00 Rn 0 nzcv CCMP Rn, Rm, #nzcv, cond |
| 3177 | sf 0 111010010 Rm cond 00 Rn 0 nzcv CCMN Rn, Rm, #nzcv, cond |
| 3178 | Operation is: |
| 3179 | (CCMP) flags = if cond then flags-after-sub(Rn,Rm) else nzcv |
| 3180 | (CCMN) flags = if cond then flags-after-add(Rn,Rm) else nzcv |
| 3181 | */ |
| 3182 | if (INSN(29,21) == BITS9(1,1,1,0,1,0,0,1,0) |
| 3183 | && INSN(11,10) == BITS2(0,0) && INSN(4,4) == 0) { |
| 3184 | Bool is64 = INSN(31,31) == 1; |
| 3185 | Bool isSUB = INSN(30,30) == 1; |
| 3186 | UInt mm = INSN(20,16); |
| 3187 | UInt cond = INSN(15,12); |
| 3188 | UInt nn = INSN(9,5); |
| 3189 | UInt nzcv = INSN(3,0); |
| 3190 | |
| 3191 | IRTemp condT = newTemp(Ity_I1); |
| 3192 | assign(condT, unop(Iop_64to1, mk_arm64g_calculate_condition(cond))); |
| 3193 | |
| 3194 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 3195 | IRTemp argL = newTemp(ty); |
| 3196 | IRTemp argR = newTemp(ty); |
| 3197 | |
| 3198 | if (is64) { |
| 3199 | assign(argL, getIReg64orZR(nn)); |
| 3200 | assign(argR, getIReg64orZR(mm)); |
| 3201 | } else { |
| 3202 | assign(argL, getIReg32orZR(nn)); |
| 3203 | assign(argR, getIReg32orZR(mm)); |
| 3204 | } |
| 3205 | setFlags_ADD_SUB_conditionally(is64, isSUB, condT, argL, argR, nzcv); |
| 3206 | |
| 3207 | DIP("ccm%c %s, %s, #%u, %s\n", |
| 3208 | isSUB ? 'p' : 'n', nameIRegOrZR(is64, nn), |
| 3209 | nameIRegOrZR(is64, mm), nzcv, nameCC(cond)); |
| 3210 | return True; |
| 3211 | } |
| 3212 | |
| 3213 | |
| 3214 | /* -------------- REV/REV16/REV32/RBIT -------------- */ |
| 3215 | /* 31 30 28 20 15 11 9 4 |
| 3216 | |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 3217 | 1 10 11010110 00000 0000 11 n d (1) REV Xd, Xn |
| 3218 | 0 10 11010110 00000 0000 10 n d (2) REV Wd, Wn |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3219 | |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 3220 | 1 10 11010110 00000 0000 00 n d (3) RBIT Xd, Xn |
| 3221 | 0 10 11010110 00000 0000 00 n d (4) RBIT Wd, Wn |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3222 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 3223 | 1 10 11010110 00000 0000 01 n d (5) REV16 Xd, Xn |
| 3224 | 0 10 11010110 00000 0000 01 n d (6) REV16 Wd, Wn |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3225 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 3226 | 1 10 11010110 00000 0000 10 n d (7) REV32 Xd, Xn |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3227 | */ |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3228 | if (INSN(30,21) == BITS10(1,0,1,1,0,1,0,1,1,0) |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 3229 | && INSN(20,12) == BITS9(0,0,0,0,0,0,0,0,0)) { |
| 3230 | UInt b31 = INSN(31,31); |
| 3231 | UInt opc = INSN(11,10); |
| 3232 | |
| 3233 | UInt ix = 0; |
| 3234 | /**/ if (b31 == 1 && opc == BITS2(1,1)) ix = 1; |
| 3235 | else if (b31 == 0 && opc == BITS2(1,0)) ix = 2; |
| 3236 | else if (b31 == 1 && opc == BITS2(0,0)) ix = 3; |
| 3237 | else if (b31 == 0 && opc == BITS2(0,0)) ix = 4; |
| 3238 | else if (b31 == 1 && opc == BITS2(0,1)) ix = 5; |
| 3239 | else if (b31 == 0 && opc == BITS2(0,1)) ix = 6; |
| 3240 | else if (b31 == 1 && opc == BITS2(1,0)) ix = 7; |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 3241 | if (ix >= 1 && ix <= 7) { |
| 3242 | Bool is64 = ix == 1 || ix == 3 || ix == 5 || ix == 7; |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 3243 | UInt nn = INSN(9,5); |
| 3244 | UInt dd = INSN(4,0); |
| 3245 | IRTemp src = newTemp(Ity_I64); |
| 3246 | IRTemp dst = IRTemp_INVALID; |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 3247 | IRTemp (*math)(IRTemp) = NULL; |
| 3248 | switch (ix) { |
| 3249 | case 1: case 2: math = math_BYTESWAP64; break; |
| 3250 | case 3: case 4: math = math_BITSWAP64; break; |
| 3251 | case 5: case 6: math = math_USHORTSWAP64; break; |
| 3252 | case 7: math = math_UINTSWAP64; break; |
| 3253 | default: vassert(0); |
| 3254 | } |
| 3255 | const HChar* names[7] |
| 3256 | = { "rev", "rev", "rbit", "rbit", "rev16", "rev16", "rev32" }; |
| 3257 | const HChar* nm = names[ix-1]; |
| 3258 | vassert(math); |
| 3259 | if (ix == 6) { |
| 3260 | /* This has to be special cased, since the logic below doesn't |
| 3261 | handle it correctly. */ |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 3262 | assign(src, getIReg64orZR(nn)); |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 3263 | dst = math(src); |
| 3264 | putIReg64orZR(dd, |
| 3265 | unop(Iop_32Uto64, unop(Iop_64to32, mkexpr(dst)))); |
| 3266 | } else if (is64) { |
| 3267 | assign(src, getIReg64orZR(nn)); |
| 3268 | dst = math(src); |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 3269 | putIReg64orZR(dd, mkexpr(dst)); |
| 3270 | } else { |
| 3271 | assign(src, binop(Iop_Shl64, getIReg64orZR(nn), mkU8(32))); |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 3272 | dst = math(src); |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 3273 | putIReg32orZR(dd, unop(Iop_64to32, mkexpr(dst))); |
| 3274 | } |
sewardj | 32d8675 | 2014-03-02 12:47:18 +0000 | [diff] [blame] | 3275 | DIP("%s %s, %s\n", nm, |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 3276 | nameIRegOrZR(is64,dd), nameIRegOrZR(is64,nn)); |
| 3277 | return True; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3278 | } |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 3279 | /* else fall through */ |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3280 | } |
| 3281 | |
| 3282 | /* -------------------- CLZ/CLS -------------------- */ |
| 3283 | /* 30 28 24 20 15 9 4 |
| 3284 | sf 10 1101 0110 00000 00010 0 n d CLZ Rd, Rn |
| 3285 | sf 10 1101 0110 00000 00010 1 n d CLS Rd, Rn |
| 3286 | */ |
| 3287 | if (INSN(30,21) == BITS10(1,0,1,1,0,1,0,1,1,0) |
| 3288 | && INSN(20,11) == BITS10(0,0,0,0,0,0,0,0,1,0)) { |
| 3289 | Bool is64 = INSN(31,31) == 1; |
| 3290 | Bool isCLS = INSN(10,10) == 1; |
| 3291 | UInt nn = INSN(9,5); |
| 3292 | UInt dd = INSN(4,0); |
| 3293 | IRTemp src = newTemp(Ity_I64); |
| 3294 | IRTemp dst = newTemp(Ity_I64); |
| 3295 | if (!isCLS) { // CLS not yet supported |
| 3296 | if (is64) { |
| 3297 | assign(src, getIReg64orZR(nn)); |
| 3298 | assign(dst, IRExpr_ITE(binop(Iop_CmpEQ64, mkexpr(src), mkU64(0)), |
| 3299 | mkU64(64), |
| 3300 | unop(Iop_Clz64, mkexpr(src)))); |
| 3301 | putIReg64orZR(dd, mkexpr(dst)); |
| 3302 | } else { |
| 3303 | assign(src, binop(Iop_Shl64, |
| 3304 | unop(Iop_32Uto64, getIReg32orZR(nn)), mkU8(32))); |
| 3305 | assign(dst, IRExpr_ITE(binop(Iop_CmpEQ64, mkexpr(src), mkU64(0)), |
| 3306 | mkU64(32), |
| 3307 | unop(Iop_Clz64, mkexpr(src)))); |
| 3308 | putIReg32orZR(dd, unop(Iop_64to32, mkexpr(dst))); |
| 3309 | } |
| 3310 | DIP("cl%c %s, %s\n", |
| 3311 | isCLS ? 's' : 'z', nameIRegOrZR(is64, dd), nameIRegOrZR(is64, nn)); |
| 3312 | return True; |
| 3313 | } |
| 3314 | } |
| 3315 | |
| 3316 | /* -------------------- LSLV/LSRV/ASRV -------------------- */ |
| 3317 | /* 30 28 20 15 11 9 4 |
| 3318 | sf 00 1101 0110 m 0010 00 n d LSLV Rd,Rn,Rm |
| 3319 | sf 00 1101 0110 m 0010 01 n d LSRV Rd,Rn,Rm |
| 3320 | sf 00 1101 0110 m 0010 10 n d ASRV Rd,Rn,Rm |
| 3321 | */ |
| 3322 | if (INSN(30,21) == BITS10(0,0,1,1,0,1,0,1,1,0) |
| 3323 | && INSN(15,12) == BITS4(0,0,1,0) && INSN(11,10) < BITS2(1,1)) { |
| 3324 | Bool is64 = INSN(31,31) == 1; |
| 3325 | UInt mm = INSN(20,16); |
| 3326 | UInt op = INSN(11,10); |
| 3327 | UInt nn = INSN(9,5); |
| 3328 | UInt dd = INSN(4,0); |
| 3329 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 3330 | IRTemp srcL = newTemp(ty); |
| 3331 | IRTemp srcR = newTemp(Ity_I8); |
| 3332 | IRTemp res = newTemp(ty); |
| 3333 | IROp iop = Iop_INVALID; |
| 3334 | assign(srcL, getIRegOrZR(is64, nn)); |
| 3335 | assign(srcR, |
| 3336 | unop(Iop_64to8, |
| 3337 | binop(Iop_And64, |
| 3338 | getIReg64orZR(mm), mkU64(is64 ? 63 : 31)))); |
| 3339 | switch (op) { |
| 3340 | case BITS2(0,0): iop = mkSHL(ty); break; |
| 3341 | case BITS2(0,1): iop = mkSHR(ty); break; |
| 3342 | case BITS2(1,0): iop = mkSAR(ty); break; |
| 3343 | default: vassert(0); |
| 3344 | } |
| 3345 | assign(res, binop(iop, mkexpr(srcL), mkexpr(srcR))); |
| 3346 | putIRegOrZR(is64, dd, mkexpr(res)); |
| 3347 | vassert(op < 3); |
| 3348 | const HChar* names[3] = { "lslv", "lsrv", "asrv" }; |
| 3349 | DIP("%s %s, %s, %s\n", |
| 3350 | names[op], nameIRegOrZR(is64,dd), |
| 3351 | nameIRegOrZR(is64,nn), nameIRegOrZR(is64,mm)); |
| 3352 | return True; |
| 3353 | } |
| 3354 | |
| 3355 | /* -------------------- SDIV/UDIV -------------------- */ |
| 3356 | /* 30 28 20 15 10 9 4 |
| 3357 | sf 00 1101 0110 m 00001 1 n d SDIV Rd,Rn,Rm |
| 3358 | sf 00 1101 0110 m 00001 0 n d UDIV Rd,Rn,Rm |
| 3359 | */ |
| 3360 | if (INSN(30,21) == BITS10(0,0,1,1,0,1,0,1,1,0) |
| 3361 | && INSN(15,11) == BITS5(0,0,0,0,1)) { |
| 3362 | Bool is64 = INSN(31,31) == 1; |
| 3363 | UInt mm = INSN(20,16); |
| 3364 | Bool isS = INSN(10,10) == 1; |
| 3365 | UInt nn = INSN(9,5); |
| 3366 | UInt dd = INSN(4,0); |
| 3367 | if (isS) { |
| 3368 | putIRegOrZR(is64, dd, binop(is64 ? Iop_DivS64 : Iop_DivS32, |
| 3369 | getIRegOrZR(is64, nn), |
| 3370 | getIRegOrZR(is64, mm))); |
| 3371 | } else { |
| 3372 | putIRegOrZR(is64, dd, binop(is64 ? Iop_DivU64 : Iop_DivU32, |
| 3373 | getIRegOrZR(is64, nn), |
| 3374 | getIRegOrZR(is64, mm))); |
| 3375 | } |
| 3376 | DIP("%cdiv %s, %s, %s\n", isS ? 's' : 'u', |
| 3377 | nameIRegOrZR(is64, dd), |
| 3378 | nameIRegOrZR(is64, nn), nameIRegOrZR(is64, mm)); |
| 3379 | return True; |
| 3380 | } |
| 3381 | |
| 3382 | /* ------------------ {S,U}M{ADD,SUB}L ------------------ */ |
| 3383 | /* 31 23 20 15 14 9 4 |
| 3384 | 1001 1011 101 m 0 a n d UMADDL Xd,Wn,Wm,Xa |
| 3385 | 1001 1011 001 m 0 a n d SMADDL Xd,Wn,Wm,Xa |
| 3386 | 1001 1011 101 m 1 a n d UMSUBL Xd,Wn,Wm,Xa |
| 3387 | 1001 1011 001 m 1 a n d SMSUBL Xd,Wn,Wm,Xa |
| 3388 | with operation |
| 3389 | Xd = Xa +/- (Wn *u/s Wm) |
| 3390 | */ |
| 3391 | if (INSN(31,24) == BITS8(1,0,0,1,1,0,1,1) && INSN(22,21) == BITS2(0,1)) { |
| 3392 | Bool isU = INSN(23,23) == 1; |
| 3393 | UInt mm = INSN(20,16); |
| 3394 | Bool isAdd = INSN(15,15) == 0; |
| 3395 | UInt aa = INSN(14,10); |
| 3396 | UInt nn = INSN(9,5); |
| 3397 | UInt dd = INSN(4,0); |
| 3398 | IRTemp wN = newTemp(Ity_I32); |
| 3399 | IRTemp wM = newTemp(Ity_I32); |
| 3400 | IRTemp xA = newTemp(Ity_I64); |
| 3401 | IRTemp muld = newTemp(Ity_I64); |
| 3402 | IRTemp res = newTemp(Ity_I64); |
| 3403 | assign(wN, getIReg32orZR(nn)); |
| 3404 | assign(wM, getIReg32orZR(mm)); |
| 3405 | assign(xA, getIReg64orZR(aa)); |
| 3406 | assign(muld, binop(isU ? Iop_MullU32 : Iop_MullS32, |
| 3407 | mkexpr(wN), mkexpr(wM))); |
| 3408 | assign(res, binop(isAdd ? Iop_Add64 : Iop_Sub64, |
| 3409 | mkexpr(xA), mkexpr(muld))); |
| 3410 | putIReg64orZR(dd, mkexpr(res)); |
| 3411 | DIP("%cm%sl %s, %s, %s, %s\n", isU ? 'u' : 's', isAdd ? "add" : "sub", |
| 3412 | nameIReg64orZR(dd), nameIReg32orZR(nn), |
| 3413 | nameIReg32orZR(mm), nameIReg64orZR(aa)); |
| 3414 | return True; |
| 3415 | } |
| 3416 | vex_printf("ARM64 front end: data_processing_register\n"); |
| 3417 | return False; |
| 3418 | # undef INSN |
| 3419 | } |
| 3420 | |
| 3421 | |
| 3422 | /*------------------------------------------------------------*/ |
| 3423 | /*--- Load and Store instructions ---*/ |
| 3424 | /*------------------------------------------------------------*/ |
| 3425 | |
| 3426 | /* Generate the EA for a "reg + reg" style amode. This is done from |
| 3427 | parts of the insn, but for sanity checking sake it takes the whole |
| 3428 | insn. This appears to depend on insn[15:12], with opt=insn[15:13] |
| 3429 | and S=insn[12]: |
| 3430 | |
| 3431 | The possible forms, along with their opt:S values, are: |
| 3432 | 011:0 Xn|SP + Xm |
| 3433 | 111:0 Xn|SP + Xm |
| 3434 | 011:1 Xn|SP + Xm * transfer_szB |
| 3435 | 111:1 Xn|SP + Xm * transfer_szB |
| 3436 | 010:0 Xn|SP + 32Uto64(Wm) |
| 3437 | 010:1 Xn|SP + 32Uto64(Wm) * transfer_szB |
| 3438 | 110:0 Xn|SP + 32Sto64(Wm) |
| 3439 | 110:1 Xn|SP + 32Sto64(Wm) * transfer_szB |
| 3440 | |
| 3441 | Rm is insn[20:16]. Rn is insn[9:5]. Rt is insn[4:0]. Log2 of |
| 3442 | the transfer size is insn[23,31,30]. For integer loads/stores, |
| 3443 | insn[23] is zero, hence szLg2 can be at most 3 in such cases. |
| 3444 | |
| 3445 | If the decoding fails, it returns IRTemp_INVALID. |
| 3446 | |
| 3447 | isInt is True iff this is decoding is for transfers to/from integer |
| 3448 | registers. If False it is for transfers to/from vector registers. |
| 3449 | */ |
| 3450 | static IRTemp gen_indexed_EA ( /*OUT*/HChar* buf, UInt insn, Bool isInt ) |
| 3451 | { |
| 3452 | UInt optS = SLICE_UInt(insn, 15, 12); |
| 3453 | UInt mm = SLICE_UInt(insn, 20, 16); |
| 3454 | UInt nn = SLICE_UInt(insn, 9, 5); |
| 3455 | UInt szLg2 = (isInt ? 0 : (SLICE_UInt(insn, 23, 23) << 2)) |
| 3456 | | SLICE_UInt(insn, 31, 30); // Log2 of the size |
| 3457 | |
| 3458 | buf[0] = 0; |
| 3459 | |
| 3460 | /* Sanity checks, that this really is a load/store insn. */ |
| 3461 | if (SLICE_UInt(insn, 11, 10) != BITS2(1,0)) |
| 3462 | goto fail; |
| 3463 | |
| 3464 | if (isInt |
| 3465 | && SLICE_UInt(insn, 29, 21) != BITS9(1,1,1,0,0,0,0,1,1)/*LDR*/ |
| 3466 | && SLICE_UInt(insn, 29, 21) != BITS9(1,1,1,0,0,0,0,0,1)/*STR*/ |
| 3467 | && SLICE_UInt(insn, 29, 21) != BITS9(1,1,1,0,0,0,1,0,1)/*LDRSbhw Xt*/ |
| 3468 | && SLICE_UInt(insn, 29, 21) != BITS9(1,1,1,0,0,0,1,1,1))/*LDRSbhw Wt*/ |
| 3469 | goto fail; |
| 3470 | |
| 3471 | if (!isInt |
| 3472 | && SLICE_UInt(insn, 29, 24) != BITS6(1,1,1,1,0,0)) /*LDR/STR*/ |
| 3473 | goto fail; |
| 3474 | |
| 3475 | /* Throw out non-verified but possibly valid cases. */ |
| 3476 | switch (szLg2) { |
| 3477 | case BITS3(0,0,0): break; // 8 bit, valid for both int and vec |
| 3478 | case BITS3(0,0,1): break; // 16 bit, valid for both int and vec |
| 3479 | case BITS3(0,1,0): break; // 32 bit, valid for both int and vec |
| 3480 | case BITS3(0,1,1): break; // 64 bit, valid for both int and vec |
| 3481 | case BITS3(1,0,0): // can only ever be valid for the vector case |
| 3482 | if (isInt) goto fail; else goto fail; |
| 3483 | case BITS3(1,0,1): // these sizes are never valid |
| 3484 | case BITS3(1,1,0): |
| 3485 | case BITS3(1,1,1): goto fail; |
| 3486 | |
| 3487 | default: vassert(0); |
| 3488 | } |
| 3489 | |
| 3490 | IRExpr* rhs = NULL; |
| 3491 | switch (optS) { |
| 3492 | case BITS4(1,1,1,0): goto fail; //ATC |
| 3493 | case BITS4(0,1,1,0): |
| 3494 | rhs = getIReg64orZR(mm); |
| 3495 | vex_sprintf(buf, "[%s, %s]", |
| 3496 | nameIReg64orZR(nn), nameIReg64orZR(mm)); |
| 3497 | break; |
| 3498 | case BITS4(1,1,1,1): goto fail; //ATC |
| 3499 | case BITS4(0,1,1,1): |
| 3500 | rhs = binop(Iop_Shl64, getIReg64orZR(mm), mkU8(szLg2)); |
| 3501 | vex_sprintf(buf, "[%s, %s lsl %u]", |
| 3502 | nameIReg64orZR(nn), nameIReg64orZR(mm), szLg2); |
| 3503 | break; |
| 3504 | case BITS4(0,1,0,0): |
| 3505 | rhs = unop(Iop_32Uto64, getIReg32orZR(mm)); |
| 3506 | vex_sprintf(buf, "[%s, %s uxtx]", |
| 3507 | nameIReg64orZR(nn), nameIReg32orZR(mm)); |
| 3508 | break; |
| 3509 | case BITS4(0,1,0,1): |
| 3510 | rhs = binop(Iop_Shl64, |
| 3511 | unop(Iop_32Uto64, getIReg32orZR(mm)), mkU8(szLg2)); |
| 3512 | vex_sprintf(buf, "[%s, %s uxtx, lsl %u]", |
| 3513 | nameIReg64orZR(nn), nameIReg32orZR(mm), szLg2); |
| 3514 | break; |
| 3515 | case BITS4(1,1,0,0): |
| 3516 | rhs = unop(Iop_32Sto64, getIReg32orZR(mm)); |
| 3517 | vex_sprintf(buf, "[%s, %s sxtx]", |
| 3518 | nameIReg64orZR(nn), nameIReg32orZR(mm)); |
| 3519 | break; |
| 3520 | case BITS4(1,1,0,1): |
| 3521 | rhs = binop(Iop_Shl64, |
| 3522 | unop(Iop_32Sto64, getIReg32orZR(mm)), mkU8(szLg2)); |
| 3523 | vex_sprintf(buf, "[%s, %s sxtx, lsl %u]", |
| 3524 | nameIReg64orZR(nn), nameIReg32orZR(mm), szLg2); |
| 3525 | break; |
| 3526 | default: |
| 3527 | /* The rest appear to be genuinely invalid */ |
| 3528 | goto fail; |
| 3529 | } |
| 3530 | |
| 3531 | vassert(rhs); |
| 3532 | IRTemp res = newTemp(Ity_I64); |
| 3533 | assign(res, binop(Iop_Add64, getIReg64orSP(nn), rhs)); |
| 3534 | return res; |
| 3535 | |
| 3536 | fail: |
| 3537 | vex_printf("gen_indexed_EA: unhandled case optS == 0x%x\n", optS); |
| 3538 | return IRTemp_INVALID; |
| 3539 | } |
| 3540 | |
| 3541 | |
| 3542 | /* Generate an 8/16/32/64 bit integer store to ADDR for the lowest |
| 3543 | bits of DATAE :: Ity_I64. */ |
| 3544 | static void gen_narrowing_store ( UInt szB, IRTemp addr, IRExpr* dataE ) |
| 3545 | { |
| 3546 | IRExpr* addrE = mkexpr(addr); |
| 3547 | switch (szB) { |
| 3548 | case 8: |
| 3549 | storeLE(addrE, dataE); |
| 3550 | break; |
| 3551 | case 4: |
| 3552 | storeLE(addrE, unop(Iop_64to32, dataE)); |
| 3553 | break; |
| 3554 | case 2: |
| 3555 | storeLE(addrE, unop(Iop_64to16, dataE)); |
| 3556 | break; |
| 3557 | case 1: |
| 3558 | storeLE(addrE, unop(Iop_64to8, dataE)); |
| 3559 | break; |
| 3560 | default: |
| 3561 | vassert(0); |
| 3562 | } |
| 3563 | } |
| 3564 | |
| 3565 | |
| 3566 | /* Generate an 8/16/32/64 bit unsigned widening load from ADDR, |
| 3567 | placing the result in an Ity_I64 temporary. */ |
| 3568 | static IRTemp gen_zwidening_load ( UInt szB, IRTemp addr ) |
| 3569 | { |
| 3570 | IRTemp res = newTemp(Ity_I64); |
| 3571 | IRExpr* addrE = mkexpr(addr); |
| 3572 | switch (szB) { |
| 3573 | case 8: |
| 3574 | assign(res, loadLE(Ity_I64,addrE)); |
| 3575 | break; |
| 3576 | case 4: |
| 3577 | assign(res, unop(Iop_32Uto64, loadLE(Ity_I32,addrE))); |
| 3578 | break; |
| 3579 | case 2: |
| 3580 | assign(res, unop(Iop_16Uto64, loadLE(Ity_I16,addrE))); |
| 3581 | break; |
| 3582 | case 1: |
| 3583 | assign(res, unop(Iop_8Uto64, loadLE(Ity_I8,addrE))); |
| 3584 | break; |
| 3585 | default: |
| 3586 | vassert(0); |
| 3587 | } |
| 3588 | return res; |
| 3589 | } |
| 3590 | |
| 3591 | |
sewardj | 18bf517 | 2014-06-14 18:05:30 +0000 | [diff] [blame] | 3592 | /* Generate a "standard 7" name, from bitQ and size. But also |
| 3593 | allow ".1d" since that's occasionally useful. */ |
| 3594 | static |
| 3595 | const HChar* nameArr_Q_SZ ( UInt bitQ, UInt size ) |
| 3596 | { |
| 3597 | vassert(bitQ <= 1 && size <= 3); |
| 3598 | const HChar* nms[8] |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 3599 | = { "8b", "4h", "2s", "1d", "16b", "8h", "4s", "2d" }; |
sewardj | 18bf517 | 2014-06-14 18:05:30 +0000 | [diff] [blame] | 3600 | UInt ix = (bitQ << 2) | size; |
| 3601 | vassert(ix < 8); |
| 3602 | return nms[ix]; |
| 3603 | } |
| 3604 | |
| 3605 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3606 | static |
| 3607 | Bool dis_ARM64_load_store(/*MB_OUT*/DisResult* dres, UInt insn) |
| 3608 | { |
| 3609 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 3610 | |
| 3611 | /* ------------ LDR,STR (immediate, uimm12) ----------- */ |
| 3612 | /* uimm12 is scaled by the transfer size |
| 3613 | |
| 3614 | 31 29 26 21 9 4 |
| 3615 | | | | | | | |
| 3616 | 11 111 00100 imm12 nn tt STR Xt, [Xn|SP, #imm12 * 8] |
| 3617 | 11 111 00101 imm12 nn tt LDR Xt, [Xn|SP, #imm12 * 8] |
| 3618 | |
| 3619 | 10 111 00100 imm12 nn tt STR Wt, [Xn|SP, #imm12 * 4] |
| 3620 | 10 111 00101 imm12 nn tt LDR Wt, [Xn|SP, #imm12 * 4] |
| 3621 | |
| 3622 | 01 111 00100 imm12 nn tt STRH Wt, [Xn|SP, #imm12 * 2] |
| 3623 | 01 111 00101 imm12 nn tt LDRH Wt, [Xn|SP, #imm12 * 2] |
| 3624 | |
| 3625 | 00 111 00100 imm12 nn tt STRB Wt, [Xn|SP, #imm12 * 1] |
| 3626 | 00 111 00101 imm12 nn tt LDRB Wt, [Xn|SP, #imm12 * 1] |
| 3627 | */ |
| 3628 | if (INSN(29,23) == BITS7(1,1,1,0,0,1,0)) { |
| 3629 | UInt szLg2 = INSN(31,30); |
| 3630 | UInt szB = 1 << szLg2; |
| 3631 | Bool isLD = INSN(22,22) == 1; |
| 3632 | UInt offs = INSN(21,10) * szB; |
| 3633 | UInt nn = INSN(9,5); |
| 3634 | UInt tt = INSN(4,0); |
| 3635 | IRTemp ta = newTemp(Ity_I64); |
| 3636 | assign(ta, binop(Iop_Add64, getIReg64orSP(nn), mkU64(offs))); |
| 3637 | if (nn == 31) { /* FIXME generate stack alignment check */ } |
| 3638 | vassert(szLg2 < 4); |
| 3639 | if (isLD) { |
| 3640 | putIReg64orZR(tt, mkexpr(gen_zwidening_load(szB, ta))); |
| 3641 | } else { |
| 3642 | gen_narrowing_store(szB, ta, getIReg64orZR(tt)); |
| 3643 | } |
| 3644 | const HChar* ld_name[4] = { "ldrb", "ldrh", "ldr", "ldr" }; |
| 3645 | const HChar* st_name[4] = { "strb", "strh", "str", "str" }; |
| 3646 | DIP("%s %s, [%s, #%u]\n", |
| 3647 | (isLD ? ld_name : st_name)[szLg2], nameIRegOrZR(szB == 8, tt), |
| 3648 | nameIReg64orSP(nn), offs); |
| 3649 | return True; |
| 3650 | } |
| 3651 | |
| 3652 | /* ------------ LDUR,STUR (immediate, simm9) ----------- */ |
| 3653 | /* |
| 3654 | 31 29 26 20 11 9 4 |
| 3655 | | | | | | | | |
| 3656 | (at-Rn-then-Rn=EA) | | | |
| 3657 | sz 111 00000 0 imm9 01 Rn Rt STR Rt, [Xn|SP], #simm9 |
| 3658 | sz 111 00001 0 imm9 01 Rn Rt LDR Rt, [Xn|SP], #simm9 |
| 3659 | |
| 3660 | (at-EA-then-Rn=EA) |
| 3661 | sz 111 00000 0 imm9 11 Rn Rt STR Rt, [Xn|SP, #simm9]! |
| 3662 | sz 111 00001 0 imm9 11 Rn Rt LDR Rt, [Xn|SP, #simm9]! |
| 3663 | |
| 3664 | (at-EA) |
| 3665 | sz 111 00000 0 imm9 00 Rn Rt STR Rt, [Xn|SP, #simm9] |
| 3666 | sz 111 00001 0 imm9 00 Rn Rt LDR Rt, [Xn|SP, #simm9] |
| 3667 | |
| 3668 | simm9 is unscaled. |
| 3669 | |
| 3670 | The case 'wback && Rn == Rt && Rt != 31' is disallowed. In the |
| 3671 | load case this is because would create two competing values for |
| 3672 | Rt. In the store case the reason is unclear, but the spec |
| 3673 | disallows it anyway. |
| 3674 | |
| 3675 | Stores are narrowing, loads are unsigned widening. sz encodes |
| 3676 | the transfer size in the normal way: 00=1, 01=2, 10=4, 11=8. |
| 3677 | */ |
| 3678 | if ((INSN(29,21) & BITS9(1,1,1, 1,1,1,1,0, 1)) |
| 3679 | == BITS9(1,1,1, 0,0,0,0,0, 0)) { |
| 3680 | UInt szLg2 = INSN(31,30); |
| 3681 | UInt szB = 1 << szLg2; |
| 3682 | Bool isLoad = INSN(22,22) == 1; |
| 3683 | UInt imm9 = INSN(20,12); |
| 3684 | UInt nn = INSN(9,5); |
| 3685 | UInt tt = INSN(4,0); |
| 3686 | Bool wBack = INSN(10,10) == 1; |
| 3687 | UInt how = INSN(11,10); |
| 3688 | if (how == BITS2(1,0) || (wBack && nn == tt && tt != 31)) { |
| 3689 | /* undecodable; fall through */ |
| 3690 | } else { |
| 3691 | if (nn == 31) { /* FIXME generate stack alignment check */ } |
| 3692 | |
| 3693 | // Compute the transfer address TA and the writeback address WA. |
| 3694 | IRTemp tRN = newTemp(Ity_I64); |
| 3695 | assign(tRN, getIReg64orSP(nn)); |
| 3696 | IRTemp tEA = newTemp(Ity_I64); |
| 3697 | Long simm9 = (Long)sx_to_64(imm9, 9); |
| 3698 | assign(tEA, binop(Iop_Add64, mkexpr(tRN), mkU64(simm9))); |
| 3699 | |
| 3700 | IRTemp tTA = newTemp(Ity_I64); |
| 3701 | IRTemp tWA = newTemp(Ity_I64); |
| 3702 | switch (how) { |
| 3703 | case BITS2(0,1): |
| 3704 | assign(tTA, mkexpr(tRN)); assign(tWA, mkexpr(tEA)); break; |
| 3705 | case BITS2(1,1): |
| 3706 | assign(tTA, mkexpr(tEA)); assign(tWA, mkexpr(tEA)); break; |
| 3707 | case BITS2(0,0): |
| 3708 | assign(tTA, mkexpr(tEA)); /* tWA is unused */ break; |
| 3709 | default: |
| 3710 | vassert(0); /* NOTREACHED */ |
| 3711 | } |
| 3712 | |
sewardj | e0bff8b | 2014-03-09 09:40:23 +0000 | [diff] [blame] | 3713 | /* Normally rN would be updated after the transfer. However, in |
| 3714 | the special case typifed by |
| 3715 | str x30, [sp,#-16]! |
| 3716 | it is necessary to update SP before the transfer, (1) |
| 3717 | because Memcheck will otherwise complain about a write |
| 3718 | below the stack pointer, and (2) because the segfault |
| 3719 | stack extension mechanism will otherwise extend the stack |
| 3720 | only down to SP before the instruction, which might not be |
| 3721 | far enough, if the -16 bit takes the actual access |
| 3722 | address to the next page. |
| 3723 | */ |
| 3724 | Bool earlyWBack |
| 3725 | = wBack && simm9 < 0 && szB == 8 |
| 3726 | && how == BITS2(1,1) && nn == 31 && !isLoad && tt != nn; |
| 3727 | |
| 3728 | if (wBack && earlyWBack) |
| 3729 | putIReg64orSP(nn, mkexpr(tEA)); |
| 3730 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3731 | if (isLoad) { |
| 3732 | putIReg64orZR(tt, mkexpr(gen_zwidening_load(szB, tTA))); |
| 3733 | } else { |
| 3734 | gen_narrowing_store(szB, tTA, getIReg64orZR(tt)); |
| 3735 | } |
| 3736 | |
sewardj | e0bff8b | 2014-03-09 09:40:23 +0000 | [diff] [blame] | 3737 | if (wBack && !earlyWBack) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3738 | putIReg64orSP(nn, mkexpr(tEA)); |
| 3739 | |
| 3740 | const HChar* ld_name[4] = { "ldurb", "ldurh", "ldur", "ldur" }; |
| 3741 | const HChar* st_name[4] = { "sturb", "sturh", "stur", "stur" }; |
| 3742 | const HChar* fmt_str = NULL; |
| 3743 | switch (how) { |
| 3744 | case BITS2(0,1): |
| 3745 | fmt_str = "%s %s, [%s], #%lld (at-Rn-then-Rn=EA)\n"; |
| 3746 | break; |
| 3747 | case BITS2(1,1): |
| 3748 | fmt_str = "%s %s, [%s, #%lld]! (at-EA-then-Rn=EA)\n"; |
| 3749 | break; |
| 3750 | case BITS2(0,0): |
| 3751 | fmt_str = "%s %s, [%s, #%lld] (at-Rn)\n"; |
| 3752 | break; |
| 3753 | default: |
| 3754 | vassert(0); |
| 3755 | } |
| 3756 | DIP(fmt_str, (isLoad ? ld_name : st_name)[szLg2], |
| 3757 | nameIRegOrZR(szB == 8, tt), |
| 3758 | nameIReg64orSP(nn), simm9); |
| 3759 | return True; |
| 3760 | } |
| 3761 | } |
| 3762 | |
| 3763 | /* -------- LDP,STP (immediate, simm7) (INT REGS) -------- */ |
| 3764 | /* L==1 => mm==LD |
| 3765 | L==0 => mm==ST |
| 3766 | x==0 => 32 bit transfers, and zero extended loads |
| 3767 | x==1 => 64 bit transfers |
| 3768 | simm7 is scaled by the (single-register) transfer size |
| 3769 | |
| 3770 | (at-Rn-then-Rn=EA) |
| 3771 | x0 101 0001 L imm7 Rt2 Rn Rt1 mmP Rt1,Rt2, [Xn|SP], #imm |
| 3772 | |
| 3773 | (at-EA-then-Rn=EA) |
| 3774 | x0 101 0011 L imm7 Rt2 Rn Rt1 mmP Rt1,Rt2, [Xn|SP, #imm]! |
| 3775 | |
| 3776 | (at-EA) |
| 3777 | x0 101 0010 L imm7 Rt2 Rn Rt1 mmP Rt1,Rt2, [Xn|SP, #imm] |
| 3778 | */ |
| 3779 | |
| 3780 | UInt insn_30_23 = INSN(30,23); |
| 3781 | if (insn_30_23 == BITS8(0,1,0,1,0,0,0,1) |
| 3782 | || insn_30_23 == BITS8(0,1,0,1,0,0,1,1) |
| 3783 | || insn_30_23 == BITS8(0,1,0,1,0,0,1,0)) { |
| 3784 | UInt bL = INSN(22,22); |
| 3785 | UInt bX = INSN(31,31); |
| 3786 | UInt bWBack = INSN(23,23); |
| 3787 | UInt rT1 = INSN(4,0); |
| 3788 | UInt rN = INSN(9,5); |
| 3789 | UInt rT2 = INSN(14,10); |
| 3790 | Long simm7 = (Long)sx_to_64(INSN(21,15), 7); |
| 3791 | if ((bWBack && (rT1 == rN || rT2 == rN) && rN != 31) |
| 3792 | || (bL && rT1 == rT2)) { |
| 3793 | /* undecodable; fall through */ |
| 3794 | } else { |
| 3795 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 3796 | |
| 3797 | // Compute the transfer address TA and the writeback address WA. |
| 3798 | IRTemp tRN = newTemp(Ity_I64); |
| 3799 | assign(tRN, getIReg64orSP(rN)); |
| 3800 | IRTemp tEA = newTemp(Ity_I64); |
| 3801 | simm7 = (bX ? 8 : 4) * simm7; |
| 3802 | assign(tEA, binop(Iop_Add64, mkexpr(tRN), mkU64(simm7))); |
| 3803 | |
| 3804 | IRTemp tTA = newTemp(Ity_I64); |
| 3805 | IRTemp tWA = newTemp(Ity_I64); |
| 3806 | switch (INSN(24,23)) { |
| 3807 | case BITS2(0,1): |
| 3808 | assign(tTA, mkexpr(tRN)); assign(tWA, mkexpr(tEA)); break; |
| 3809 | case BITS2(1,1): |
| 3810 | assign(tTA, mkexpr(tEA)); assign(tWA, mkexpr(tEA)); break; |
| 3811 | case BITS2(1,0): |
| 3812 | assign(tTA, mkexpr(tEA)); /* tWA is unused */ break; |
| 3813 | default: |
| 3814 | vassert(0); /* NOTREACHED */ |
| 3815 | } |
| 3816 | |
| 3817 | /* Normally rN would be updated after the transfer. However, in |
| 3818 | the special case typifed by |
| 3819 | stp x29, x30, [sp,#-112]! |
| 3820 | it is necessary to update SP before the transfer, (1) |
| 3821 | because Memcheck will otherwise complain about a write |
| 3822 | below the stack pointer, and (2) because the segfault |
| 3823 | stack extension mechanism will otherwise extend the stack |
| 3824 | only down to SP before the instruction, which might not be |
| 3825 | far enough, if the -112 bit takes the actual access |
| 3826 | address to the next page. |
| 3827 | */ |
| 3828 | Bool earlyWBack |
| 3829 | = bWBack && simm7 < 0 |
| 3830 | && INSN(24,23) == BITS2(1,1) && rN == 31 && bL == 0; |
| 3831 | |
| 3832 | if (bWBack && earlyWBack) |
| 3833 | putIReg64orSP(rN, mkexpr(tEA)); |
| 3834 | |
| 3835 | /**/ if (bL == 1 && bX == 1) { |
| 3836 | // 64 bit load |
| 3837 | putIReg64orZR(rT1, loadLE(Ity_I64, |
| 3838 | binop(Iop_Add64,mkexpr(tTA),mkU64(0)))); |
| 3839 | putIReg64orZR(rT2, loadLE(Ity_I64, |
| 3840 | binop(Iop_Add64,mkexpr(tTA),mkU64(8)))); |
| 3841 | } else if (bL == 1 && bX == 0) { |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3842 | // 32 bit load |
| 3843 | putIReg32orZR(rT1, loadLE(Ity_I32, |
| 3844 | binop(Iop_Add64,mkexpr(tTA),mkU64(0)))); |
| 3845 | putIReg32orZR(rT2, loadLE(Ity_I32, |
| 3846 | binop(Iop_Add64,mkexpr(tTA),mkU64(4)))); |
| 3847 | } else if (bL == 0 && bX == 1) { |
| 3848 | // 64 bit store |
| 3849 | storeLE(binop(Iop_Add64,mkexpr(tTA),mkU64(0)), |
| 3850 | getIReg64orZR(rT1)); |
| 3851 | storeLE(binop(Iop_Add64,mkexpr(tTA),mkU64(8)), |
| 3852 | getIReg64orZR(rT2)); |
| 3853 | } else { |
| 3854 | vassert(bL == 0 && bX == 0); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3855 | // 32 bit store |
| 3856 | storeLE(binop(Iop_Add64,mkexpr(tTA),mkU64(0)), |
| 3857 | getIReg32orZR(rT1)); |
| 3858 | storeLE(binop(Iop_Add64,mkexpr(tTA),mkU64(4)), |
| 3859 | getIReg32orZR(rT2)); |
| 3860 | } |
| 3861 | |
| 3862 | if (bWBack && !earlyWBack) |
| 3863 | putIReg64orSP(rN, mkexpr(tEA)); |
| 3864 | |
| 3865 | const HChar* fmt_str = NULL; |
| 3866 | switch (INSN(24,23)) { |
| 3867 | case BITS2(0,1): |
| 3868 | fmt_str = "%sp %s, %s, [%s], #%lld (at-Rn-then-Rn=EA)\n"; |
| 3869 | break; |
| 3870 | case BITS2(1,1): |
| 3871 | fmt_str = "%sp %s, %s, [%s, #%lld]! (at-EA-then-Rn=EA)\n"; |
| 3872 | break; |
| 3873 | case BITS2(1,0): |
| 3874 | fmt_str = "%sp %s, %s, [%s, #%lld] (at-Rn)\n"; |
| 3875 | break; |
| 3876 | default: |
| 3877 | vassert(0); |
| 3878 | } |
| 3879 | DIP(fmt_str, bL == 0 ? "st" : "ld", |
| 3880 | nameIRegOrZR(bX == 1, rT1), |
| 3881 | nameIRegOrZR(bX == 1, rT2), |
| 3882 | nameIReg64orSP(rN), simm7); |
| 3883 | return True; |
| 3884 | } |
| 3885 | } |
| 3886 | |
| 3887 | /* ---------------- LDR (literal, int reg) ---------------- */ |
| 3888 | /* 31 29 23 4 |
| 3889 | 00 011 000 imm19 Rt LDR Wt, [PC + sxTo64(imm19 << 2)] |
| 3890 | 01 011 000 imm19 Rt LDR Xt, [PC + sxTo64(imm19 << 2)] |
| 3891 | 10 011 000 imm19 Rt LDRSW Xt, [PC + sxTo64(imm19 << 2)] |
| 3892 | 11 011 000 imm19 Rt prefetch [PC + sxTo64(imm19 << 2)] |
| 3893 | Just handles the first two cases for now. |
| 3894 | */ |
| 3895 | if (INSN(29,24) == BITS6(0,1,1,0,0,0) && INSN(31,31) == 0) { |
| 3896 | UInt imm19 = INSN(23,5); |
| 3897 | UInt rT = INSN(4,0); |
| 3898 | UInt bX = INSN(30,30); |
| 3899 | ULong ea = guest_PC_curr_instr + sx_to_64(imm19 << 2, 21); |
| 3900 | if (bX) { |
| 3901 | putIReg64orZR(rT, loadLE(Ity_I64, mkU64(ea))); |
| 3902 | } else { |
| 3903 | putIReg32orZR(rT, loadLE(Ity_I32, mkU64(ea))); |
| 3904 | } |
| 3905 | DIP("ldr %s, 0x%llx (literal)\n", nameIRegOrZR(bX == 1, rT), ea); |
| 3906 | return True; |
| 3907 | } |
| 3908 | |
| 3909 | /* -------------- {LD,ST}R (integer register) --------------- */ |
| 3910 | /* 31 29 20 15 12 11 9 4 |
| 3911 | | | | | | | | | |
| 3912 | 11 111000011 Rm option S 10 Rn Rt LDR Xt, [Xn|SP, R<m>{ext/sh}] |
| 3913 | 10 111000011 Rm option S 10 Rn Rt LDR Wt, [Xn|SP, R<m>{ext/sh}] |
| 3914 | 01 111000011 Rm option S 10 Rn Rt LDRH Wt, [Xn|SP, R<m>{ext/sh}] |
| 3915 | 00 111000011 Rm option S 10 Rn Rt LDRB Wt, [Xn|SP, R<m>{ext/sh}] |
| 3916 | |
| 3917 | 11 111000001 Rm option S 10 Rn Rt STR Xt, [Xn|SP, R<m>{ext/sh}] |
| 3918 | 10 111000001 Rm option S 10 Rn Rt STR Wt, [Xn|SP, R<m>{ext/sh}] |
| 3919 | 01 111000001 Rm option S 10 Rn Rt STRH Wt, [Xn|SP, R<m>{ext/sh}] |
| 3920 | 00 111000001 Rm option S 10 Rn Rt STRB Wt, [Xn|SP, R<m>{ext/sh}] |
| 3921 | */ |
| 3922 | if (INSN(29,23) == BITS7(1,1,1,0,0,0,0) |
| 3923 | && INSN(21,21) == 1 && INSN(11,10) == BITS2(1,0)) { |
| 3924 | HChar dis_buf[64]; |
| 3925 | UInt szLg2 = INSN(31,30); |
| 3926 | Bool isLD = INSN(22,22) == 1; |
| 3927 | UInt tt = INSN(4,0); |
| 3928 | IRTemp ea = gen_indexed_EA(dis_buf, insn, True/*to/from int regs*/); |
| 3929 | if (ea != IRTemp_INVALID) { |
| 3930 | switch (szLg2) { |
| 3931 | case 3: /* 64 bit */ |
| 3932 | if (isLD) { |
| 3933 | putIReg64orZR(tt, loadLE(Ity_I64, mkexpr(ea))); |
| 3934 | DIP("ldr %s, %s\n", nameIReg64orZR(tt), dis_buf); |
| 3935 | } else { |
| 3936 | storeLE(mkexpr(ea), getIReg64orZR(tt)); |
| 3937 | DIP("str %s, %s\n", nameIReg64orZR(tt), dis_buf); |
| 3938 | } |
| 3939 | break; |
| 3940 | case 2: /* 32 bit */ |
| 3941 | if (isLD) { |
| 3942 | putIReg32orZR(tt, loadLE(Ity_I32, mkexpr(ea))); |
| 3943 | DIP("ldr %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3944 | } else { |
| 3945 | storeLE(mkexpr(ea), getIReg32orZR(tt)); |
| 3946 | DIP("str %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3947 | } |
| 3948 | break; |
| 3949 | case 1: /* 16 bit */ |
| 3950 | if (isLD) { |
| 3951 | putIReg64orZR(tt, unop(Iop_16Uto64, |
| 3952 | loadLE(Ity_I16, mkexpr(ea)))); |
| 3953 | DIP("ldruh %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3954 | } else { |
| 3955 | storeLE(mkexpr(ea), unop(Iop_64to16, getIReg64orZR(tt))); |
| 3956 | DIP("strh %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3957 | } |
| 3958 | break; |
| 3959 | case 0: /* 8 bit */ |
| 3960 | if (isLD) { |
| 3961 | putIReg64orZR(tt, unop(Iop_8Uto64, |
| 3962 | loadLE(Ity_I8, mkexpr(ea)))); |
| 3963 | DIP("ldrub %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3964 | } else { |
| 3965 | storeLE(mkexpr(ea), unop(Iop_64to8, getIReg64orZR(tt))); |
| 3966 | DIP("strb %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3967 | } |
| 3968 | break; |
| 3969 | default: |
| 3970 | vassert(0); |
| 3971 | } |
| 3972 | return True; |
| 3973 | } |
| 3974 | } |
| 3975 | |
| 3976 | /* -------------- LDRS{B,H,W} (uimm12) -------------- */ |
| 3977 | /* 31 29 26 23 21 9 4 |
| 3978 | 10 111 001 10 imm12 n t LDRSW Xt, [Xn|SP, #pimm12 * 4] |
| 3979 | 01 111 001 1x imm12 n t LDRSH Rt, [Xn|SP, #pimm12 * 2] |
| 3980 | 00 111 001 1x imm12 n t LDRSB Rt, [Xn|SP, #pimm12 * 1] |
| 3981 | where |
| 3982 | Rt is Wt when x==1, Xt when x==0 |
| 3983 | */ |
| 3984 | if (INSN(29,23) == BITS7(1,1,1,0,0,1,1)) { |
| 3985 | /* Further checks on bits 31:30 and 22 */ |
| 3986 | Bool valid = False; |
| 3987 | switch ((INSN(31,30) << 1) | INSN(22,22)) { |
| 3988 | case BITS3(1,0,0): |
| 3989 | case BITS3(0,1,0): case BITS3(0,1,1): |
| 3990 | case BITS3(0,0,0): case BITS3(0,0,1): |
| 3991 | valid = True; |
| 3992 | break; |
| 3993 | } |
| 3994 | if (valid) { |
| 3995 | UInt szLg2 = INSN(31,30); |
| 3996 | UInt bitX = INSN(22,22); |
| 3997 | UInt imm12 = INSN(21,10); |
| 3998 | UInt nn = INSN(9,5); |
| 3999 | UInt tt = INSN(4,0); |
| 4000 | UInt szB = 1 << szLg2; |
| 4001 | IRExpr* ea = binop(Iop_Add64, |
| 4002 | getIReg64orSP(nn), mkU64(imm12 * szB)); |
| 4003 | switch (szB) { |
| 4004 | case 4: |
| 4005 | vassert(bitX == 0); |
| 4006 | putIReg64orZR(tt, unop(Iop_32Sto64, loadLE(Ity_I32, ea))); |
| 4007 | DIP("ldrsw %s, [%s, #%u]\n", nameIReg64orZR(tt), |
| 4008 | nameIReg64orSP(nn), imm12 * szB); |
| 4009 | break; |
| 4010 | case 2: |
| 4011 | if (bitX == 1) { |
| 4012 | putIReg32orZR(tt, unop(Iop_16Sto32, loadLE(Ity_I16, ea))); |
| 4013 | } else { |
| 4014 | putIReg64orZR(tt, unop(Iop_16Sto64, loadLE(Ity_I16, ea))); |
| 4015 | } |
| 4016 | DIP("ldrsh %s, [%s, #%u]\n", |
| 4017 | nameIRegOrZR(bitX == 0, tt), |
| 4018 | nameIReg64orSP(nn), imm12 * szB); |
| 4019 | break; |
| 4020 | case 1: |
| 4021 | if (bitX == 1) { |
| 4022 | putIReg32orZR(tt, unop(Iop_8Sto32, loadLE(Ity_I8, ea))); |
| 4023 | } else { |
| 4024 | putIReg64orZR(tt, unop(Iop_8Sto64, loadLE(Ity_I8, ea))); |
| 4025 | } |
| 4026 | DIP("ldrsb %s, [%s, #%u]\n", |
| 4027 | nameIRegOrZR(bitX == 0, tt), |
| 4028 | nameIReg64orSP(nn), imm12 * szB); |
| 4029 | break; |
| 4030 | default: |
| 4031 | vassert(0); |
| 4032 | } |
| 4033 | return True; |
| 4034 | } |
| 4035 | /* else fall through */ |
| 4036 | } |
| 4037 | |
| 4038 | /* -------------- LDRS{B,H,W} (simm9, upd) -------------- */ |
| 4039 | /* (at-Rn-then-Rn=EA) |
| 4040 | 31 29 23 21 20 11 9 4 |
| 4041 | 00 111 000 1x 0 imm9 01 n t LDRSB Rt, [Xn|SP], #simm9 |
| 4042 | 01 111 000 1x 0 imm9 01 n t LDRSH Rt, [Xn|SP], #simm9 |
| 4043 | 10 111 000 10 0 imm9 01 n t LDRSW Xt, [Xn|SP], #simm9 |
| 4044 | |
| 4045 | (at-EA-then-Rn=EA) |
| 4046 | 00 111 000 1x 0 imm9 11 n t LDRSB Rt, [Xn|SP, #simm9]! |
| 4047 | 01 111 000 1x 0 imm9 11 n t LDRSH Rt, [Xn|SP, #simm9]! |
| 4048 | 10 111 000 10 0 imm9 11 n t LDRSW Xt, [Xn|SP, #simm9]! |
| 4049 | where |
| 4050 | Rt is Wt when x==1, Xt when x==0 |
| 4051 | transfer-at-Rn when [11]==0, at EA when [11]==1 |
| 4052 | */ |
| 4053 | if (INSN(29,23) == BITS7(1,1,1,0,0,0,1) |
| 4054 | && INSN(21,21) == 0 && INSN(10,10) == 1) { |
| 4055 | /* Further checks on bits 31:30 and 22 */ |
| 4056 | Bool valid = False; |
| 4057 | switch ((INSN(31,30) << 1) | INSN(22,22)) { |
| 4058 | case BITS3(1,0,0): // LDRSW Xt |
| 4059 | case BITS3(0,1,0): case BITS3(0,1,1): // LDRSH Xt, Wt |
| 4060 | case BITS3(0,0,0): case BITS3(0,0,1): // LDRSB Xt, Wt |
| 4061 | valid = True; |
| 4062 | break; |
| 4063 | } |
| 4064 | if (valid) { |
| 4065 | UInt szLg2 = INSN(31,30); |
| 4066 | UInt imm9 = INSN(20,12); |
| 4067 | Bool atRN = INSN(11,11) == 0; |
| 4068 | UInt nn = INSN(9,5); |
| 4069 | UInt tt = INSN(4,0); |
| 4070 | IRTemp tRN = newTemp(Ity_I64); |
| 4071 | IRTemp tEA = newTemp(Ity_I64); |
| 4072 | IRTemp tTA = IRTemp_INVALID; |
| 4073 | ULong simm9 = sx_to_64(imm9, 9); |
| 4074 | Bool is64 = INSN(22,22) == 0; |
| 4075 | assign(tRN, getIReg64orSP(nn)); |
| 4076 | assign(tEA, binop(Iop_Add64, mkexpr(tRN), mkU64(simm9))); |
| 4077 | tTA = atRN ? tRN : tEA; |
| 4078 | HChar ch = '?'; |
| 4079 | /* There are 5 cases: |
| 4080 | byte load, SX to 64 |
| 4081 | byte load, SX to 32, ZX to 64 |
| 4082 | halfword load, SX to 64 |
| 4083 | halfword load, SX to 32, ZX to 64 |
| 4084 | word load, SX to 64 |
| 4085 | The ifs below handle them in the listed order. |
| 4086 | */ |
| 4087 | if (szLg2 == 0) { |
| 4088 | ch = 'b'; |
| 4089 | if (is64) { |
| 4090 | putIReg64orZR(tt, unop(Iop_8Sto64, |
| 4091 | loadLE(Ity_I8, mkexpr(tTA)))); |
| 4092 | } else { |
| 4093 | putIReg32orZR(tt, unop(Iop_8Sto32, |
| 4094 | loadLE(Ity_I8, mkexpr(tTA)))); |
| 4095 | } |
| 4096 | } |
| 4097 | else if (szLg2 == 1) { |
| 4098 | ch = 'h'; |
| 4099 | if (is64) { |
| 4100 | putIReg64orZR(tt, unop(Iop_16Sto64, |
| 4101 | loadLE(Ity_I16, mkexpr(tTA)))); |
| 4102 | } else { |
| 4103 | putIReg32orZR(tt, unop(Iop_16Sto32, |
| 4104 | loadLE(Ity_I16, mkexpr(tTA)))); |
| 4105 | } |
| 4106 | } |
| 4107 | else if (szLg2 == 2 && is64) { |
| 4108 | ch = 'w'; |
| 4109 | putIReg64orZR(tt, unop(Iop_32Sto64, |
| 4110 | loadLE(Ity_I32, mkexpr(tTA)))); |
| 4111 | } |
| 4112 | else { |
| 4113 | vassert(0); |
| 4114 | } |
| 4115 | putIReg64orSP(nn, mkexpr(tEA)); |
| 4116 | DIP(atRN ? "ldrs%c %s, [%s], #%lld\n" : "ldrs%c %s, [%s, #%lld]!", |
| 4117 | ch, nameIRegOrZR(is64, tt), nameIReg64orSP(nn), simm9); |
| 4118 | return True; |
| 4119 | } |
| 4120 | /* else fall through */ |
| 4121 | } |
| 4122 | |
| 4123 | /* -------------- LDRS{B,H,W} (simm9, noUpd) -------------- */ |
| 4124 | /* 31 29 23 21 20 11 9 4 |
| 4125 | 00 111 000 1x 0 imm9 00 n t LDURSB Rt, [Xn|SP, #simm9] |
| 4126 | 01 111 000 1x 0 imm9 00 n t LDURSH Rt, [Xn|SP, #simm9] |
| 4127 | 10 111 000 10 0 imm9 00 n t LDURSW Xt, [Xn|SP, #simm9] |
| 4128 | where |
| 4129 | Rt is Wt when x==1, Xt when x==0 |
| 4130 | */ |
| 4131 | if (INSN(29,23) == BITS7(1,1,1,0,0,0,1) |
| 4132 | && INSN(21,21) == 0 && INSN(11,10) == BITS2(0,0)) { |
| 4133 | /* Further checks on bits 31:30 and 22 */ |
| 4134 | Bool valid = False; |
| 4135 | switch ((INSN(31,30) << 1) | INSN(22,22)) { |
| 4136 | case BITS3(1,0,0): // LDURSW Xt |
| 4137 | case BITS3(0,1,0): case BITS3(0,1,1): // LDURSH Xt, Wt |
| 4138 | case BITS3(0,0,0): case BITS3(0,0,1): // LDURSB Xt, Wt |
| 4139 | valid = True; |
| 4140 | break; |
| 4141 | } |
| 4142 | if (valid) { |
| 4143 | UInt szLg2 = INSN(31,30); |
| 4144 | UInt imm9 = INSN(20,12); |
| 4145 | UInt nn = INSN(9,5); |
| 4146 | UInt tt = INSN(4,0); |
| 4147 | IRTemp tRN = newTemp(Ity_I64); |
| 4148 | IRTemp tEA = newTemp(Ity_I64); |
| 4149 | ULong simm9 = sx_to_64(imm9, 9); |
| 4150 | Bool is64 = INSN(22,22) == 0; |
| 4151 | assign(tRN, getIReg64orSP(nn)); |
| 4152 | assign(tEA, binop(Iop_Add64, mkexpr(tRN), mkU64(simm9))); |
| 4153 | HChar ch = '?'; |
| 4154 | /* There are 5 cases: |
| 4155 | byte load, SX to 64 |
| 4156 | byte load, SX to 32, ZX to 64 |
| 4157 | halfword load, SX to 64 |
| 4158 | halfword load, SX to 32, ZX to 64 |
| 4159 | word load, SX to 64 |
| 4160 | The ifs below handle them in the listed order. |
| 4161 | */ |
| 4162 | if (szLg2 == 0) { |
| 4163 | ch = 'b'; |
| 4164 | if (is64) { |
| 4165 | putIReg64orZR(tt, unop(Iop_8Sto64, |
| 4166 | loadLE(Ity_I8, mkexpr(tEA)))); |
| 4167 | } else { |
| 4168 | putIReg32orZR(tt, unop(Iop_8Sto32, |
| 4169 | loadLE(Ity_I8, mkexpr(tEA)))); |
| 4170 | } |
| 4171 | } |
| 4172 | else if (szLg2 == 1) { |
| 4173 | ch = 'h'; |
| 4174 | if (is64) { |
| 4175 | putIReg64orZR(tt, unop(Iop_16Sto64, |
| 4176 | loadLE(Ity_I16, mkexpr(tEA)))); |
| 4177 | } else { |
| 4178 | putIReg32orZR(tt, unop(Iop_16Sto32, |
| 4179 | loadLE(Ity_I16, mkexpr(tEA)))); |
| 4180 | } |
| 4181 | } |
| 4182 | else if (szLg2 == 2 && is64) { |
| 4183 | ch = 'w'; |
| 4184 | putIReg64orZR(tt, unop(Iop_32Sto64, |
| 4185 | loadLE(Ity_I32, mkexpr(tEA)))); |
| 4186 | } |
| 4187 | else { |
| 4188 | vassert(0); |
| 4189 | } |
| 4190 | DIP("ldurs%c %s, [%s, #%lld]", |
| 4191 | ch, nameIRegOrZR(is64, tt), nameIReg64orSP(nn), simm9); |
| 4192 | return True; |
| 4193 | } |
| 4194 | /* else fall through */ |
| 4195 | } |
| 4196 | |
| 4197 | /* -------- LDP,STP (immediate, simm7) (FP&VEC) -------- */ |
| 4198 | /* L==1 => mm==LD |
| 4199 | L==0 => mm==ST |
| 4200 | sz==00 => 32 bit (S) transfers |
| 4201 | sz==01 => 64 bit (D) transfers |
| 4202 | sz==10 => 128 bit (Q) transfers |
| 4203 | sz==11 isn't allowed |
| 4204 | simm7 is scaled by the (single-register) transfer size |
| 4205 | |
| 4206 | 31 29 22 21 14 9 4 |
| 4207 | sz 101 1001 L imm7 t2 n t1 mmP SDQt1, SDQt2, [Xn|SP], #imm |
| 4208 | (at-Rn-then-Rn=EA) |
| 4209 | |
| 4210 | sz 101 1011 L imm7 t2 n t1 mmP SDQt1, SDQt2, [Xn|SP, #imm]! |
| 4211 | (at-EA-then-Rn=EA) |
| 4212 | |
| 4213 | sz 101 1010 L imm7 t2 n t1 mmP SDQt1, SDQt2, [Xn|SP, #imm] |
| 4214 | (at-EA) |
| 4215 | */ |
| 4216 | |
| 4217 | UInt insn_29_23 = INSN(29,23); |
| 4218 | if (insn_29_23 == BITS7(1,0,1,1,0,0,1) |
| 4219 | || insn_29_23 == BITS7(1,0,1,1,0,1,1) |
| 4220 | || insn_29_23 == BITS7(1,0,1,1,0,1,0)) { |
| 4221 | UInt szSlg2 = INSN(31,30); // log2 of the xfer size in 32-bit units |
| 4222 | Bool isLD = INSN(22,22) == 1; |
| 4223 | Bool wBack = INSN(23,23) == 1; |
| 4224 | Long simm7 = (Long)sx_to_64(INSN(21,15), 7); |
| 4225 | UInt tt2 = INSN(14,10); |
| 4226 | UInt nn = INSN(9,5); |
| 4227 | UInt tt1 = INSN(4,0); |
| 4228 | if (szSlg2 == BITS2(1,1) || (isLD && tt1 == tt2)) { |
| 4229 | /* undecodable; fall through */ |
| 4230 | } else { |
| 4231 | if (nn == 31) { /* FIXME generate stack alignment check */ } |
| 4232 | |
| 4233 | // Compute the transfer address TA and the writeback address WA. |
| 4234 | UInt szB = 4 << szSlg2; /* szB is the per-register size */ |
| 4235 | IRTemp tRN = newTemp(Ity_I64); |
| 4236 | assign(tRN, getIReg64orSP(nn)); |
| 4237 | IRTemp tEA = newTemp(Ity_I64); |
| 4238 | simm7 = szB * simm7; |
| 4239 | assign(tEA, binop(Iop_Add64, mkexpr(tRN), mkU64(simm7))); |
| 4240 | |
| 4241 | IRTemp tTA = newTemp(Ity_I64); |
| 4242 | IRTemp tWA = newTemp(Ity_I64); |
| 4243 | switch (INSN(24,23)) { |
| 4244 | case BITS2(0,1): |
| 4245 | assign(tTA, mkexpr(tRN)); assign(tWA, mkexpr(tEA)); break; |
| 4246 | case BITS2(1,1): |
| 4247 | assign(tTA, mkexpr(tEA)); assign(tWA, mkexpr(tEA)); break; |
| 4248 | case BITS2(1,0): |
| 4249 | assign(tTA, mkexpr(tEA)); /* tWA is unused */ break; |
| 4250 | default: |
| 4251 | vassert(0); /* NOTREACHED */ |
| 4252 | } |
| 4253 | |
| 4254 | IRType ty = Ity_INVALID; |
| 4255 | switch (szB) { |
| 4256 | case 4: ty = Ity_F32; break; |
| 4257 | case 8: ty = Ity_F64; break; |
| 4258 | case 16: ty = Ity_V128; break; |
| 4259 | default: vassert(0); |
| 4260 | } |
| 4261 | |
sewardj | e0bff8b | 2014-03-09 09:40:23 +0000 | [diff] [blame] | 4262 | /* Normally rN would be updated after the transfer. However, in |
sewardj | 1955143 | 2014-05-07 09:20:11 +0000 | [diff] [blame] | 4263 | the special cases typifed by |
sewardj | e0bff8b | 2014-03-09 09:40:23 +0000 | [diff] [blame] | 4264 | stp q0, q1, [sp,#-512]! |
sewardj | 1955143 | 2014-05-07 09:20:11 +0000 | [diff] [blame] | 4265 | stp d0, d1, [sp,#-512]! |
| 4266 | stp s0, s1, [sp,#-512]! |
sewardj | e0bff8b | 2014-03-09 09:40:23 +0000 | [diff] [blame] | 4267 | it is necessary to update SP before the transfer, (1) |
| 4268 | because Memcheck will otherwise complain about a write |
| 4269 | below the stack pointer, and (2) because the segfault |
| 4270 | stack extension mechanism will otherwise extend the stack |
| 4271 | only down to SP before the instruction, which might not be |
| 4272 | far enough, if the -512 bit takes the actual access |
| 4273 | address to the next page. |
| 4274 | */ |
| 4275 | Bool earlyWBack |
sewardj | 1955143 | 2014-05-07 09:20:11 +0000 | [diff] [blame] | 4276 | = wBack && simm7 < 0 |
sewardj | e0bff8b | 2014-03-09 09:40:23 +0000 | [diff] [blame] | 4277 | && INSN(24,23) == BITS2(1,1) && nn == 31 && !isLD; |
| 4278 | |
| 4279 | if (wBack && earlyWBack) |
| 4280 | putIReg64orSP(nn, mkexpr(tEA)); |
| 4281 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4282 | if (isLD) { |
sewardj | 5ba4130 | 2014-03-03 08:42:16 +0000 | [diff] [blame] | 4283 | if (szB < 16) { |
| 4284 | putQReg128(tt1, mkV128(0x0000)); |
| 4285 | } |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4286 | putQRegLO(tt1, |
| 4287 | loadLE(ty, binop(Iop_Add64, mkexpr(tTA), mkU64(0)))); |
sewardj | 5ba4130 | 2014-03-03 08:42:16 +0000 | [diff] [blame] | 4288 | if (szB < 16) { |
| 4289 | putQReg128(tt2, mkV128(0x0000)); |
| 4290 | } |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4291 | putQRegLO(tt2, |
| 4292 | loadLE(ty, binop(Iop_Add64, mkexpr(tTA), mkU64(szB)))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4293 | } else { |
| 4294 | storeLE(binop(Iop_Add64, mkexpr(tTA), mkU64(0)), |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4295 | getQRegLO(tt1, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4296 | storeLE(binop(Iop_Add64, mkexpr(tTA), mkU64(szB)), |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4297 | getQRegLO(tt2, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4298 | } |
| 4299 | |
sewardj | e0bff8b | 2014-03-09 09:40:23 +0000 | [diff] [blame] | 4300 | if (wBack && !earlyWBack) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4301 | putIReg64orSP(nn, mkexpr(tEA)); |
| 4302 | |
| 4303 | const HChar* fmt_str = NULL; |
| 4304 | switch (INSN(24,23)) { |
| 4305 | case BITS2(0,1): |
| 4306 | fmt_str = "%sp %s, %s, [%s], #%lld (at-Rn-then-Rn=EA)\n"; |
| 4307 | break; |
| 4308 | case BITS2(1,1): |
| 4309 | fmt_str = "%sp %s, %s, [%s, #%lld]! (at-EA-then-Rn=EA)\n"; |
| 4310 | break; |
| 4311 | case BITS2(1,0): |
| 4312 | fmt_str = "%sp %s, %s, [%s, #%lld] (at-Rn)\n"; |
| 4313 | break; |
| 4314 | default: |
| 4315 | vassert(0); |
| 4316 | } |
| 4317 | DIP(fmt_str, isLD ? "ld" : "st", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4318 | nameQRegLO(tt1, ty), nameQRegLO(tt2, ty), |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4319 | nameIReg64orSP(nn), simm7); |
| 4320 | return True; |
| 4321 | } |
| 4322 | } |
| 4323 | |
| 4324 | /* -------------- {LD,ST}R (vector register) --------------- */ |
| 4325 | /* 31 29 23 20 15 12 11 9 4 |
| 4326 | | | | | | | | | | |
| 4327 | 00 111100 011 Rm option S 10 Rn Rt LDR Bt, [Xn|SP, R<m>{ext/sh}] |
| 4328 | 01 111100 011 Rm option S 10 Rn Rt LDR Ht, [Xn|SP, R<m>{ext/sh}] |
| 4329 | 10 111100 011 Rm option S 10 Rn Rt LDR St, [Xn|SP, R<m>{ext/sh}] |
| 4330 | 11 111100 011 Rm option S 10 Rn Rt LDR Dt, [Xn|SP, R<m>{ext/sh}] |
| 4331 | 00 111100 111 Rm option S 10 Rn Rt LDR Qt, [Xn|SP, R<m>{ext/sh}] |
| 4332 | |
| 4333 | 00 111100 001 Rm option S 10 Rn Rt STR Bt, [Xn|SP, R<m>{ext/sh}] |
| 4334 | 01 111100 001 Rm option S 10 Rn Rt STR Ht, [Xn|SP, R<m>{ext/sh}] |
| 4335 | 10 111100 001 Rm option S 10 Rn Rt STR St, [Xn|SP, R<m>{ext/sh}] |
| 4336 | 11 111100 001 Rm option S 10 Rn Rt STR Dt, [Xn|SP, R<m>{ext/sh}] |
| 4337 | 00 111100 101 Rm option S 10 Rn Rt STR Qt, [Xn|SP, R<m>{ext/sh}] |
| 4338 | */ |
| 4339 | if (INSN(29,24) == BITS6(1,1,1,1,0,0) |
| 4340 | && INSN(21,21) == 1 && INSN(11,10) == BITS2(1,0)) { |
| 4341 | HChar dis_buf[64]; |
| 4342 | UInt szLg2 = (INSN(23,23) << 2) | INSN(31,30); |
| 4343 | Bool isLD = INSN(22,22) == 1; |
| 4344 | UInt tt = INSN(4,0); |
| 4345 | if (szLg2 >= 4) goto after_LDR_STR_vector_register; |
| 4346 | IRTemp ea = gen_indexed_EA(dis_buf, insn, False/*to/from vec regs*/); |
| 4347 | if (ea == IRTemp_INVALID) goto after_LDR_STR_vector_register; |
| 4348 | switch (szLg2) { |
| 4349 | case 0: /* 8 bit */ |
| 4350 | if (isLD) { |
| 4351 | putQReg128(tt, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4352 | putQRegLO(tt, loadLE(Ity_I8, mkexpr(ea))); |
| 4353 | DIP("ldr %s, %s\n", nameQRegLO(tt, Ity_I8), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4354 | } else { |
| 4355 | vassert(0); //ATC |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4356 | storeLE(mkexpr(ea), getQRegLO(tt, Ity_I8)); |
| 4357 | DIP("str %s, %s\n", nameQRegLO(tt, Ity_I8), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4358 | } |
| 4359 | break; |
| 4360 | case 1: |
| 4361 | if (isLD) { |
| 4362 | putQReg128(tt, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4363 | putQRegLO(tt, loadLE(Ity_I16, mkexpr(ea))); |
| 4364 | DIP("ldr %s, %s\n", nameQRegLO(tt, Ity_I16), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4365 | } else { |
| 4366 | vassert(0); //ATC |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4367 | storeLE(mkexpr(ea), getQRegLO(tt, Ity_I16)); |
| 4368 | DIP("str %s, %s\n", nameQRegLO(tt, Ity_I16), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4369 | } |
| 4370 | break; |
| 4371 | case 2: /* 32 bit */ |
| 4372 | if (isLD) { |
| 4373 | putQReg128(tt, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4374 | putQRegLO(tt, loadLE(Ity_I32, mkexpr(ea))); |
| 4375 | DIP("ldr %s, %s\n", nameQRegLO(tt, Ity_I32), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4376 | } else { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4377 | storeLE(mkexpr(ea), getQRegLO(tt, Ity_I32)); |
| 4378 | DIP("str %s, %s\n", nameQRegLO(tt, Ity_I32), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4379 | } |
| 4380 | break; |
| 4381 | case 3: /* 64 bit */ |
| 4382 | if (isLD) { |
| 4383 | putQReg128(tt, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4384 | putQRegLO(tt, loadLE(Ity_I64, mkexpr(ea))); |
| 4385 | DIP("ldr %s, %s\n", nameQRegLO(tt, Ity_I64), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4386 | } else { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4387 | storeLE(mkexpr(ea), getQRegLO(tt, Ity_I64)); |
| 4388 | DIP("str %s, %s\n", nameQRegLO(tt, Ity_I64), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4389 | } |
| 4390 | break; |
| 4391 | case 4: return False; //ATC |
| 4392 | default: vassert(0); |
| 4393 | } |
| 4394 | return True; |
| 4395 | } |
| 4396 | after_LDR_STR_vector_register: |
| 4397 | |
| 4398 | /* ---------- LDRS{B,H,W} (integer register, SX) ---------- */ |
| 4399 | /* 31 29 22 20 15 12 11 9 4 |
| 4400 | | | | | | | | | | |
| 4401 | 10 1110001 01 Rm opt S 10 Rn Rt LDRSW Xt, [Xn|SP, R<m>{ext/sh}] |
| 4402 | |
| 4403 | 01 1110001 01 Rm opt S 10 Rn Rt LDRSH Xt, [Xn|SP, R<m>{ext/sh}] |
| 4404 | 01 1110001 11 Rm opt S 10 Rn Rt LDRSH Wt, [Xn|SP, R<m>{ext/sh}] |
| 4405 | |
| 4406 | 00 1110001 01 Rm opt S 10 Rn Rt LDRSB Xt, [Xn|SP, R<m>{ext/sh}] |
| 4407 | 00 1110001 11 Rm opt S 10 Rn Rt LDRSB Wt, [Xn|SP, R<m>{ext/sh}] |
| 4408 | */ |
| 4409 | if (INSN(29,23) == BITS7(1,1,1,0,0,0,1) |
| 4410 | && INSN(21,21) == 1 && INSN(11,10) == BITS2(1,0)) { |
| 4411 | HChar dis_buf[64]; |
| 4412 | UInt szLg2 = INSN(31,30); |
| 4413 | Bool sxTo64 = INSN(22,22) == 0; // else sx to 32 and zx to 64 |
| 4414 | UInt tt = INSN(4,0); |
| 4415 | if (szLg2 == 3) goto after_LDRS_integer_register; |
| 4416 | IRTemp ea = gen_indexed_EA(dis_buf, insn, True/*to/from int regs*/); |
| 4417 | if (ea == IRTemp_INVALID) goto after_LDRS_integer_register; |
| 4418 | /* Enumerate the 5 variants explicitly. */ |
| 4419 | if (szLg2 == 2/*32 bit*/ && sxTo64) { |
| 4420 | putIReg64orZR(tt, unop(Iop_32Sto64, loadLE(Ity_I32, mkexpr(ea)))); |
| 4421 | DIP("ldrsw %s, %s\n", nameIReg64orZR(tt), dis_buf); |
| 4422 | return True; |
| 4423 | } |
| 4424 | else |
| 4425 | if (szLg2 == 1/*16 bit*/) { |
| 4426 | if (sxTo64) { |
| 4427 | putIReg64orZR(tt, unop(Iop_16Sto64, loadLE(Ity_I16, mkexpr(ea)))); |
| 4428 | DIP("ldrsh %s, %s\n", nameIReg64orZR(tt), dis_buf); |
| 4429 | } else { |
| 4430 | putIReg32orZR(tt, unop(Iop_16Sto32, loadLE(Ity_I16, mkexpr(ea)))); |
| 4431 | DIP("ldrsh %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 4432 | } |
| 4433 | return True; |
| 4434 | } |
| 4435 | else |
| 4436 | if (szLg2 == 0/*8 bit*/) { |
| 4437 | if (sxTo64) { |
| 4438 | putIReg64orZR(tt, unop(Iop_8Sto64, loadLE(Ity_I8, mkexpr(ea)))); |
| 4439 | DIP("ldrsb %s, %s\n", nameIReg64orZR(tt), dis_buf); |
| 4440 | } else { |
| 4441 | putIReg32orZR(tt, unop(Iop_8Sto32, loadLE(Ity_I8, mkexpr(ea)))); |
| 4442 | DIP("ldrsb %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 4443 | } |
| 4444 | return True; |
| 4445 | } |
| 4446 | /* else it's an invalid combination */ |
| 4447 | } |
| 4448 | after_LDRS_integer_register: |
| 4449 | |
| 4450 | /* -------- LDR/STR (immediate, SIMD&FP, unsigned offset) -------- */ |
| 4451 | /* This is the Unsigned offset variant only. The Post-Index and |
| 4452 | Pre-Index variants are below. |
| 4453 | |
| 4454 | 31 29 23 21 9 4 |
| 4455 | 00 111 101 01 imm12 n t LDR Bt, [Xn|SP + imm12 * 1] |
| 4456 | 01 111 101 01 imm12 n t LDR Ht, [Xn|SP + imm12 * 2] |
| 4457 | 10 111 101 01 imm12 n t LDR St, [Xn|SP + imm12 * 4] |
| 4458 | 11 111 101 01 imm12 n t LDR Dt, [Xn|SP + imm12 * 8] |
| 4459 | 00 111 101 11 imm12 n t LDR Qt, [Xn|SP + imm12 * 16] |
| 4460 | |
| 4461 | 00 111 101 00 imm12 n t STR Bt, [Xn|SP + imm12 * 1] |
| 4462 | 01 111 101 00 imm12 n t STR Ht, [Xn|SP + imm12 * 2] |
| 4463 | 10 111 101 00 imm12 n t STR St, [Xn|SP + imm12 * 4] |
| 4464 | 11 111 101 00 imm12 n t STR Dt, [Xn|SP + imm12 * 8] |
| 4465 | 00 111 101 10 imm12 n t STR Qt, [Xn|SP + imm12 * 16] |
| 4466 | */ |
| 4467 | if (INSN(29,24) == BITS6(1,1,1,1,0,1) |
| 4468 | && ((INSN(23,23) << 2) | INSN(31,30)) <= 4) { |
| 4469 | UInt szLg2 = (INSN(23,23) << 2) | INSN(31,30); |
| 4470 | Bool isLD = INSN(22,22) == 1; |
| 4471 | UInt pimm12 = INSN(21,10) << szLg2; |
| 4472 | UInt nn = INSN(9,5); |
| 4473 | UInt tt = INSN(4,0); |
| 4474 | IRTemp tEA = newTemp(Ity_I64); |
| 4475 | IRType ty = preferredVectorSubTypeFromSize(1 << szLg2); |
| 4476 | assign(tEA, binop(Iop_Add64, getIReg64orSP(nn), mkU64(pimm12))); |
| 4477 | if (isLD) { |
| 4478 | if (szLg2 < 4) { |
| 4479 | putQReg128(tt, mkV128(0x0000)); |
| 4480 | } |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4481 | putQRegLO(tt, loadLE(ty, mkexpr(tEA))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4482 | } else { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4483 | storeLE(mkexpr(tEA), getQRegLO(tt, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4484 | } |
| 4485 | DIP("%s %s, [%s, #%u]\n", |
| 4486 | isLD ? "ldr" : "str", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4487 | nameQRegLO(tt, ty), nameIReg64orSP(nn), pimm12); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4488 | return True; |
| 4489 | } |
| 4490 | |
| 4491 | /* -------- LDR/STR (immediate, SIMD&FP, pre/post index) -------- */ |
| 4492 | /* These are the Post-Index and Pre-Index variants. |
| 4493 | |
| 4494 | 31 29 23 20 11 9 4 |
| 4495 | (at-Rn-then-Rn=EA) |
| 4496 | 00 111 100 01 0 imm9 01 n t LDR Bt, [Xn|SP], #simm |
| 4497 | 01 111 100 01 0 imm9 01 n t LDR Ht, [Xn|SP], #simm |
| 4498 | 10 111 100 01 0 imm9 01 n t LDR St, [Xn|SP], #simm |
| 4499 | 11 111 100 01 0 imm9 01 n t LDR Dt, [Xn|SP], #simm |
| 4500 | 00 111 100 11 0 imm9 01 n t LDR Qt, [Xn|SP], #simm |
| 4501 | |
| 4502 | (at-EA-then-Rn=EA) |
| 4503 | 00 111 100 01 0 imm9 11 n t LDR Bt, [Xn|SP, #simm]! |
| 4504 | 01 111 100 01 0 imm9 11 n t LDR Ht, [Xn|SP, #simm]! |
| 4505 | 10 111 100 01 0 imm9 11 n t LDR St, [Xn|SP, #simm]! |
| 4506 | 11 111 100 01 0 imm9 11 n t LDR Dt, [Xn|SP, #simm]! |
| 4507 | 00 111 100 11 0 imm9 11 n t LDR Qt, [Xn|SP, #simm]! |
| 4508 | |
| 4509 | Stores are the same except with bit 22 set to 0. |
| 4510 | */ |
| 4511 | if (INSN(29,24) == BITS6(1,1,1,1,0,0) |
| 4512 | && ((INSN(23,23) << 2) | INSN(31,30)) <= 4 |
| 4513 | && INSN(21,21) == 0 && INSN(10,10) == 1) { |
| 4514 | UInt szLg2 = (INSN(23,23) << 2) | INSN(31,30); |
| 4515 | Bool isLD = INSN(22,22) == 1; |
| 4516 | UInt imm9 = INSN(20,12); |
| 4517 | Bool atRN = INSN(11,11) == 0; |
| 4518 | UInt nn = INSN(9,5); |
| 4519 | UInt tt = INSN(4,0); |
| 4520 | IRTemp tRN = newTemp(Ity_I64); |
| 4521 | IRTemp tEA = newTemp(Ity_I64); |
| 4522 | IRTemp tTA = IRTemp_INVALID; |
| 4523 | IRType ty = preferredVectorSubTypeFromSize(1 << szLg2); |
| 4524 | ULong simm9 = sx_to_64(imm9, 9); |
| 4525 | assign(tRN, getIReg64orSP(nn)); |
| 4526 | assign(tEA, binop(Iop_Add64, mkexpr(tRN), mkU64(simm9))); |
| 4527 | tTA = atRN ? tRN : tEA; |
| 4528 | if (isLD) { |
| 4529 | if (szLg2 < 4) { |
| 4530 | putQReg128(tt, mkV128(0x0000)); |
| 4531 | } |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4532 | putQRegLO(tt, loadLE(ty, mkexpr(tTA))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4533 | } else { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4534 | storeLE(mkexpr(tTA), getQRegLO(tt, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4535 | } |
| 4536 | putIReg64orSP(nn, mkexpr(tEA)); |
| 4537 | DIP(atRN ? "%s %s, [%s], #%lld\n" : "%s %s, [%s, #%lld]!\n", |
| 4538 | isLD ? "ldr" : "str", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4539 | nameQRegLO(tt, ty), nameIReg64orSP(nn), simm9); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4540 | return True; |
| 4541 | } |
| 4542 | |
| 4543 | /* -------- LDUR/STUR (unscaled offset, SIMD&FP) -------- */ |
| 4544 | /* 31 29 23 20 11 9 4 |
| 4545 | 00 111 100 01 0 imm9 00 n t LDR Bt, [Xn|SP, #simm] |
| 4546 | 01 111 100 01 0 imm9 00 n t LDR Ht, [Xn|SP, #simm] |
| 4547 | 10 111 100 01 0 imm9 00 n t LDR St, [Xn|SP, #simm] |
| 4548 | 11 111 100 01 0 imm9 00 n t LDR Dt, [Xn|SP, #simm] |
| 4549 | 00 111 100 11 0 imm9 00 n t LDR Qt, [Xn|SP, #simm] |
| 4550 | |
| 4551 | 00 111 100 00 0 imm9 00 n t STR Bt, [Xn|SP, #simm] |
| 4552 | 01 111 100 00 0 imm9 00 n t STR Ht, [Xn|SP, #simm] |
| 4553 | 10 111 100 00 0 imm9 00 n t STR St, [Xn|SP, #simm] |
| 4554 | 11 111 100 00 0 imm9 00 n t STR Dt, [Xn|SP, #simm] |
| 4555 | 00 111 100 10 0 imm9 00 n t STR Qt, [Xn|SP, #simm] |
| 4556 | */ |
| 4557 | if (INSN(29,24) == BITS6(1,1,1,1,0,0) |
| 4558 | && ((INSN(23,23) << 2) | INSN(31,30)) <= 4 |
| 4559 | && INSN(21,21) == 0 && INSN(11,10) == BITS2(0,0)) { |
| 4560 | UInt szLg2 = (INSN(23,23) << 2) | INSN(31,30); |
| 4561 | Bool isLD = INSN(22,22) == 1; |
| 4562 | UInt imm9 = INSN(20,12); |
| 4563 | UInt nn = INSN(9,5); |
| 4564 | UInt tt = INSN(4,0); |
| 4565 | ULong simm9 = sx_to_64(imm9, 9); |
| 4566 | IRTemp tEA = newTemp(Ity_I64); |
| 4567 | IRType ty = preferredVectorSubTypeFromSize(1 << szLg2); |
| 4568 | assign(tEA, binop(Iop_Add64, getIReg64orSP(nn), mkU64(simm9))); |
| 4569 | if (isLD) { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4570 | if (szLg2 < 4) { |
| 4571 | putQReg128(tt, mkV128(0x0000)); |
| 4572 | } |
| 4573 | putQRegLO(tt, loadLE(ty, mkexpr(tEA))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4574 | } else { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4575 | storeLE(mkexpr(tEA), getQRegLO(tt, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4576 | } |
| 4577 | DIP("%s %s, [%s, #%lld]\n", |
| 4578 | isLD ? "ldur" : "stur", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4579 | nameQRegLO(tt, ty), nameIReg64orSP(nn), (Long)simm9); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4580 | return True; |
| 4581 | } |
| 4582 | |
| 4583 | /* ---------------- LDR (literal, SIMD&FP) ---------------- */ |
| 4584 | /* 31 29 23 4 |
| 4585 | 00 011 100 imm19 t LDR St, [PC + sxTo64(imm19 << 2)] |
| 4586 | 01 011 100 imm19 t LDR Dt, [PC + sxTo64(imm19 << 2)] |
| 4587 | 10 011 100 imm19 t LDR Qt, [PC + sxTo64(imm19 << 2)] |
| 4588 | */ |
| 4589 | if (INSN(29,24) == BITS6(0,1,1,1,0,0) && INSN(31,30) < BITS2(1,1)) { |
| 4590 | UInt szB = 4 << INSN(31,30); |
| 4591 | UInt imm19 = INSN(23,5); |
| 4592 | UInt tt = INSN(4,0); |
| 4593 | ULong ea = guest_PC_curr_instr + sx_to_64(imm19 << 2, 21); |
| 4594 | IRType ty = preferredVectorSubTypeFromSize(szB); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4595 | putQReg128(tt, mkV128(0x0000)); |
| 4596 | putQRegLO(tt, loadLE(ty, mkU64(ea))); |
| 4597 | DIP("ldr %s, 0x%llx (literal)\n", nameQRegLO(tt, ty), ea); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4598 | return True; |
| 4599 | } |
| 4600 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4601 | /* ---------- LD1/ST1 (single structure, no offset) ---------- */ |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4602 | /* 31 23 |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4603 | 0100 1100 0100 0000 0111 11 N T LD1 {vT.2d}, [Xn|SP] |
| 4604 | 0100 1100 0000 0000 0111 11 N T ST1 {vT.2d}, [Xn|SP] |
| 4605 | 0100 1100 0100 0000 0111 10 N T LD1 {vT.4s}, [Xn|SP] |
| 4606 | 0100 1100 0000 0000 0111 10 N T ST1 {vT.4s}, [Xn|SP] |
| 4607 | 0100 1100 0100 0000 0111 01 N T LD1 {vT.8h}, [Xn|SP] |
| 4608 | 0100 1100 0000 0000 0111 01 N T ST1 {vT.8h}, [Xn|SP] |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4609 | 0100 1100 0100 0000 0111 00 N T LD1 {vT.16b}, [Xn|SP] |
| 4610 | 0100 1100 0000 0000 0111 00 N T ST1 {vT.16b}, [Xn|SP] |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4611 | FIXME does this assume that the host is little endian? |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4612 | */ |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4613 | if ( (insn & 0xFFFFF000) == 0x4C407000 // LD1 cases |
| 4614 | || (insn & 0xFFFFF000) == 0x4C007000 // ST1 cases |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4615 | ) { |
| 4616 | Bool isLD = INSN(22,22) == 1; |
| 4617 | UInt rN = INSN(9,5); |
| 4618 | UInt vT = INSN(4,0); |
| 4619 | IRTemp tEA = newTemp(Ity_I64); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4620 | const HChar* names[4] = { "2d", "4s", "8h", "16b" }; |
| 4621 | const HChar* name = names[INSN(11,10)]; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4622 | assign(tEA, getIReg64orSP(rN)); |
| 4623 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 4624 | if (isLD) { |
| 4625 | putQReg128(vT, loadLE(Ity_V128, mkexpr(tEA))); |
| 4626 | } else { |
| 4627 | storeLE(mkexpr(tEA), getQReg128(vT)); |
| 4628 | } |
| 4629 | DIP("%s {v%u.%s}, [%s]\n", isLD ? "ld1" : "st1", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4630 | vT, name, nameIReg64orSP(rN)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4631 | return True; |
| 4632 | } |
| 4633 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4634 | /* 31 23 |
| 4635 | 0000 1100 0100 0000 0111 11 N T LD1 {vT.1d}, [Xn|SP] |
| 4636 | 0000 1100 0000 0000 0111 11 N T ST1 {vT.1d}, [Xn|SP] |
| 4637 | 0000 1100 0100 0000 0111 10 N T LD1 {vT.2s}, [Xn|SP] |
| 4638 | 0000 1100 0000 0000 0111 10 N T ST1 {vT.2s}, [Xn|SP] |
| 4639 | 0000 1100 0100 0000 0111 01 N T LD1 {vT.4h}, [Xn|SP] |
| 4640 | 0000 1100 0000 0000 0111 01 N T ST1 {vT.4h}, [Xn|SP] |
| 4641 | 0000 1100 0100 0000 0111 00 N T LD1 {vT.8b}, [Xn|SP] |
| 4642 | 0000 1100 0000 0000 0111 00 N T ST1 {vT.8b}, [Xn|SP] |
| 4643 | FIXME does this assume that the host is little endian? |
| 4644 | */ |
| 4645 | if ( (insn & 0xFFFFF000) == 0x0C407000 // LD1 cases |
| 4646 | || (insn & 0xFFFFF000) == 0x0C007000 // ST1 cases |
| 4647 | ) { |
| 4648 | Bool isLD = INSN(22,22) == 1; |
| 4649 | UInt rN = INSN(9,5); |
| 4650 | UInt vT = INSN(4,0); |
| 4651 | IRTemp tEA = newTemp(Ity_I64); |
| 4652 | const HChar* names[4] = { "1d", "2s", "4h", "8b" }; |
| 4653 | const HChar* name = names[INSN(11,10)]; |
| 4654 | assign(tEA, getIReg64orSP(rN)); |
| 4655 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 4656 | if (isLD) { |
| 4657 | putQRegLane(vT, 0, loadLE(Ity_I64, mkexpr(tEA))); |
| 4658 | putQRegLane(vT, 1, mkU64(0)); |
| 4659 | } else { |
| 4660 | storeLE(mkexpr(tEA), getQRegLane(vT, 0, Ity_I64)); |
| 4661 | } |
| 4662 | DIP("%s {v%u.%s}, [%s]\n", isLD ? "ld1" : "st1", |
| 4663 | vT, name, nameIReg64orSP(rN)); |
| 4664 | return True; |
| 4665 | } |
| 4666 | |
| 4667 | /* ---------- LD1/ST1 (single structure, post index) ---------- */ |
| 4668 | /* 31 23 |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 4669 | 0100 1100 1001 1111 0111 11 N T ST1 {vT.2d}, [xN|SP], #16 |
| 4670 | 0100 1100 1101 1111 0111 11 N T LD1 {vT.2d}, [xN|SP], #16 |
| 4671 | 0100 1100 1001 1111 0111 10 N T ST1 {vT.4s}, [xN|SP], #16 |
| 4672 | 0100 1100 1101 1111 0111 10 N T LD1 {vT.4s}, [xN|SP], #16 |
| 4673 | 0100 1100 1001 1111 0111 01 N T ST1 {vT.8h}, [xN|SP], #16 |
| 4674 | 0100 1100 1101 1111 0111 01 N T LD1 {vT.8h}, [xN|SP], #16 |
| 4675 | 0100 1100 1001 1111 0111 00 N T ST1 {vT.16b}, [xN|SP], #16 |
sewardj | f5b0891 | 2014-02-06 12:57:58 +0000 | [diff] [blame] | 4676 | 0100 1100 1101 1111 0111 00 N T LD1 {vT.16b}, [xN|SP], #16 |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4677 | Note that #16 is implied and cannot be any other value. |
| 4678 | FIXME does this assume that the host is little endian? |
| 4679 | */ |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 4680 | if ( (insn & 0xFFFFF000) == 0x4CDF7000 // LD1 cases |
| 4681 | || (insn & 0xFFFFF000) == 0x4C9F7000 // ST1 cases |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4682 | ) { |
| 4683 | Bool isLD = INSN(22,22) == 1; |
| 4684 | UInt rN = INSN(9,5); |
| 4685 | UInt vT = INSN(4,0); |
| 4686 | IRTemp tEA = newTemp(Ity_I64); |
| 4687 | const HChar* names[4] = { "2d", "4s", "8h", "16b" }; |
| 4688 | const HChar* name = names[INSN(11,10)]; |
| 4689 | assign(tEA, getIReg64orSP(rN)); |
| 4690 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 4691 | if (isLD) { |
| 4692 | putQReg128(vT, loadLE(Ity_V128, mkexpr(tEA))); |
| 4693 | } else { |
| 4694 | storeLE(mkexpr(tEA), getQReg128(vT)); |
| 4695 | } |
| 4696 | putIReg64orSP(rN, binop(Iop_Add64, mkexpr(tEA), mkU64(16))); |
| 4697 | DIP("%s {v%u.%s}, [%s], #16\n", isLD ? "ld1" : "st1", |
| 4698 | vT, name, nameIReg64orSP(rN)); |
| 4699 | return True; |
| 4700 | } |
| 4701 | |
sewardj | 950ca7a | 2014-04-03 23:03:32 +0000 | [diff] [blame] | 4702 | /* 31 23 |
| 4703 | 0000 1100 1001 1111 0111 11 N T ST1 {vT.1d}, [xN|SP], #8 |
| 4704 | 0000 1100 1101 1111 0111 11 N T LD1 {vT.1d}, [xN|SP], #8 |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4705 | 0000 1100 1001 1111 0111 10 N T ST1 {vT.2s}, [xN|SP], #8 |
sewardj | 950ca7a | 2014-04-03 23:03:32 +0000 | [diff] [blame] | 4706 | 0000 1100 1101 1111 0111 10 N T LD1 {vT.2s}, [xN|SP], #8 |
sewardj | f5b0891 | 2014-02-06 12:57:58 +0000 | [diff] [blame] | 4707 | 0000 1100 1001 1111 0111 01 N T ST1 {vT.4h}, [xN|SP], #8 |
sewardj | 950ca7a | 2014-04-03 23:03:32 +0000 | [diff] [blame] | 4708 | 0000 1100 1101 1111 0111 01 N T LD1 {vT.4h}, [xN|SP], #8 |
| 4709 | 0000 1100 1001 1111 0111 00 N T ST1 {vT.8b}, [xN|SP], #8 |
| 4710 | 0000 1100 1101 1111 0111 00 N T LD1 {vT.8b}, [xN|SP], #8 |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4711 | Note that #8 is implied and cannot be any other value. |
| 4712 | FIXME does this assume that the host is little endian? |
| 4713 | */ |
sewardj | 950ca7a | 2014-04-03 23:03:32 +0000 | [diff] [blame] | 4714 | if ( (insn & 0xFFFFF000) == 0x0CDF7000 // LD1 cases |
| 4715 | || (insn & 0xFFFFF000) == 0x0C9F7000 // ST1 cases |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4716 | ) { |
sewardj | 950ca7a | 2014-04-03 23:03:32 +0000 | [diff] [blame] | 4717 | Bool isLD = INSN(22,22) == 1; |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4718 | UInt rN = INSN(9,5); |
| 4719 | UInt vT = INSN(4,0); |
| 4720 | IRTemp tEA = newTemp(Ity_I64); |
| 4721 | const HChar* names[4] = { "1d", "2s", "4h", "8b" }; |
| 4722 | const HChar* name = names[INSN(11,10)]; |
| 4723 | assign(tEA, getIReg64orSP(rN)); |
| 4724 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
sewardj | 950ca7a | 2014-04-03 23:03:32 +0000 | [diff] [blame] | 4725 | if (isLD) { |
| 4726 | putQRegLane(vT, 0, loadLE(Ity_I64, mkexpr(tEA))); |
| 4727 | putQRegLane(vT, 1, mkU64(0)); |
| 4728 | } else { |
| 4729 | storeLE(mkexpr(tEA), getQRegLane(vT, 0, Ity_I64)); |
| 4730 | } |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4731 | putIReg64orSP(rN, binop(Iop_Add64, mkexpr(tEA), mkU64(8))); |
sewardj | 950ca7a | 2014-04-03 23:03:32 +0000 | [diff] [blame] | 4732 | DIP("%s {v%u.%s}, [%s], #8\n", isLD ? "ld1" : "st1", |
| 4733 | vT, name, nameIReg64orSP(rN)); |
| 4734 | return True; |
| 4735 | } |
| 4736 | |
sewardj | 18bf517 | 2014-06-14 18:05:30 +0000 | [diff] [blame] | 4737 | /* ---------- LD1R (single structure, replicate) ---------- */ |
| 4738 | /* 31 29 22 20 15 11 9 4 |
| 4739 | 0q 001 1010 10 00000 110 0 sz n t LD1R Vt.T, [Xn|SP] |
| 4740 | 0q 001 1011 10 m 110 0 sz n t LD1R Vt.T, [Xn|SP], #sz (m=11111) |
| 4741 | , Xm (m!=11111) |
| 4742 | */ |
| 4743 | if (INSN(31,31) == 0 && INSN(29,24) == BITS6(0,0,1,1,0,1) |
| 4744 | && INSN(22,21) == BITS2(1,0) && INSN(15,12) == BITS4(1,1,0,0)) { |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 4745 | UInt bitQ = INSN(30,30); |
sewardj | 18bf517 | 2014-06-14 18:05:30 +0000 | [diff] [blame] | 4746 | Bool isPX = INSN(23,23) == 1; |
| 4747 | UInt mm = INSN(20,16); |
| 4748 | UInt sz = INSN(11,10); |
| 4749 | UInt nn = INSN(9,5); |
| 4750 | UInt tt = INSN(4,0); |
| 4751 | IRType ty = integerIRTypeOfSize(1 << sz); |
| 4752 | IRTemp tEA = newTemp(Ity_I64); |
| 4753 | assign(tEA, getIReg64orSP(nn)); |
| 4754 | if (nn == 31) { /* FIXME generate stack alignment check */ } |
| 4755 | IRTemp loaded = newTemp(ty); |
| 4756 | assign(loaded, loadLE(ty, mkexpr(tEA))); |
| 4757 | IRTemp dupd = math_DUP_TO_V128(loaded, ty); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 4758 | putQReg128(tt, math_MAYBE_ZERO_HI64(bitQ, dupd)); |
| 4759 | const HChar* arr = nameArr_Q_SZ(bitQ, sz); |
sewardj | 18bf517 | 2014-06-14 18:05:30 +0000 | [diff] [blame] | 4760 | /* Deal with the writeback, if any. */ |
| 4761 | if (!isPX && mm == BITS5(0,0,0,0,0)) { |
| 4762 | /* No writeback. */ |
| 4763 | DIP("ld1r v%u.%s, [%s]\n", tt, arr, nameIReg64orSP(nn)); |
| 4764 | return True; |
| 4765 | } |
| 4766 | if (isPX) { |
| 4767 | putIReg64orSP(nn, binop(Iop_Add64, mkexpr(tEA), |
| 4768 | mm == BITS5(1,1,1,1,1) ? mkU64(1 << sz) |
| 4769 | : getIReg64orZR(mm))); |
| 4770 | if (mm == BITS5(1,1,1,1,1)) { |
| 4771 | DIP("ld1r v%u.%s, [%s], %s\n", tt, arr, |
| 4772 | nameIReg64orSP(nn), nameIReg64orZR(mm)); |
| 4773 | } else { |
| 4774 | DIP("ld1r v%u.%s, [%s], #%u\n", tt, arr, |
| 4775 | nameIReg64orSP(nn), 1 << sz); |
| 4776 | } |
| 4777 | return True; |
| 4778 | } |
| 4779 | return False; |
| 4780 | } |
| 4781 | |
sewardj | 168c8bd | 2014-06-25 13:05:23 +0000 | [diff] [blame] | 4782 | /* -------- LD2/ST2 (multi 2-elem structs, 2 regs, post index) -------- */ |
sewardj | 950ca7a | 2014-04-03 23:03:32 +0000 | [diff] [blame] | 4783 | /* Only a very few cases. */ |
| 4784 | /* 31 23 11 9 4 |
| 4785 | 0100 1100 1101 1111 1000 11 n t LD2 {Vt.2d, V(t+1)%32.2d}, [Xn|SP], #32 |
| 4786 | 0100 1100 1001 1111 1000 11 n t ST2 {Vt.2d, V(t+1)%32.2d}, [Xn|SP], #32 |
| 4787 | 0100 1100 1101 1111 1000 10 n t LD2 {Vt.4s, V(t+1)%32.4s}, [Xn|SP], #32 |
| 4788 | 0100 1100 1001 1111 1000 10 n t ST2 {Vt.4s, V(t+1)%32.4s}, [Xn|SP], #32 |
| 4789 | */ |
| 4790 | if ( (insn & 0xFFFFFC00) == 0x4CDF8C00 // LD2 .2d |
| 4791 | || (insn & 0xFFFFFC00) == 0x4C9F8C00 // ST2 .2d |
| 4792 | || (insn & 0xFFFFFC00) == 0x4CDF8800 // LD2 .4s |
| 4793 | || (insn & 0xFFFFFC00) == 0x4C9F8800 // ST2 .4s |
| 4794 | ) { |
| 4795 | Bool isLD = INSN(22,22) == 1; |
| 4796 | UInt rN = INSN(9,5); |
| 4797 | UInt vT = INSN(4,0); |
| 4798 | IRTemp tEA = newTemp(Ity_I64); |
| 4799 | UInt sz = INSN(11,10); |
| 4800 | const HChar* name = "??"; |
| 4801 | assign(tEA, getIReg64orSP(rN)); |
| 4802 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 4803 | IRExpr* tEA_0 = binop(Iop_Add64, mkexpr(tEA), mkU64(0)); |
| 4804 | IRExpr* tEA_8 = binop(Iop_Add64, mkexpr(tEA), mkU64(8)); |
| 4805 | IRExpr* tEA_16 = binop(Iop_Add64, mkexpr(tEA), mkU64(16)); |
| 4806 | IRExpr* tEA_24 = binop(Iop_Add64, mkexpr(tEA), mkU64(24)); |
| 4807 | if (sz == BITS2(1,1)) { |
| 4808 | name = "2d"; |
| 4809 | if (isLD) { |
| 4810 | putQRegLane((vT+0) % 32, 0, loadLE(Ity_I64, tEA_0)); |
| 4811 | putQRegLane((vT+0) % 32, 1, loadLE(Ity_I64, tEA_16)); |
| 4812 | putQRegLane((vT+1) % 32, 0, loadLE(Ity_I64, tEA_8)); |
| 4813 | putQRegLane((vT+1) % 32, 1, loadLE(Ity_I64, tEA_24)); |
| 4814 | } else { |
| 4815 | storeLE(tEA_0, getQRegLane((vT+0) % 32, 0, Ity_I64)); |
| 4816 | storeLE(tEA_16, getQRegLane((vT+0) % 32, 1, Ity_I64)); |
| 4817 | storeLE(tEA_8, getQRegLane((vT+1) % 32, 0, Ity_I64)); |
| 4818 | storeLE(tEA_24, getQRegLane((vT+1) % 32, 1, Ity_I64)); |
| 4819 | } |
| 4820 | } |
| 4821 | else if (sz == BITS2(1,0)) { |
| 4822 | /* Uh, this is ugly. TODO: better. */ |
| 4823 | name = "4s"; |
| 4824 | IRExpr* tEA_4 = binop(Iop_Add64, mkexpr(tEA), mkU64(4)); |
| 4825 | IRExpr* tEA_12 = binop(Iop_Add64, mkexpr(tEA), mkU64(12)); |
| 4826 | IRExpr* tEA_20 = binop(Iop_Add64, mkexpr(tEA), mkU64(20)); |
| 4827 | IRExpr* tEA_28 = binop(Iop_Add64, mkexpr(tEA), mkU64(28)); |
| 4828 | if (isLD) { |
| 4829 | putQRegLane((vT+0) % 32, 0, loadLE(Ity_I32, tEA_0)); |
| 4830 | putQRegLane((vT+0) % 32, 1, loadLE(Ity_I32, tEA_8)); |
| 4831 | putQRegLane((vT+0) % 32, 2, loadLE(Ity_I32, tEA_16)); |
| 4832 | putQRegLane((vT+0) % 32, 3, loadLE(Ity_I32, tEA_24)); |
| 4833 | putQRegLane((vT+1) % 32, 0, loadLE(Ity_I32, tEA_4)); |
| 4834 | putQRegLane((vT+1) % 32, 1, loadLE(Ity_I32, tEA_12)); |
| 4835 | putQRegLane((vT+1) % 32, 2, loadLE(Ity_I32, tEA_20)); |
| 4836 | putQRegLane((vT+1) % 32, 3, loadLE(Ity_I32, tEA_28)); |
| 4837 | } else { |
| 4838 | storeLE(tEA_0, getQRegLane((vT+0) % 32, 0, Ity_I32)); |
| 4839 | storeLE(tEA_8, getQRegLane((vT+0) % 32, 1, Ity_I32)); |
| 4840 | storeLE(tEA_16, getQRegLane((vT+0) % 32, 2, Ity_I32)); |
| 4841 | storeLE(tEA_24, getQRegLane((vT+0) % 32, 3, Ity_I32)); |
| 4842 | storeLE(tEA_4, getQRegLane((vT+1) % 32, 0, Ity_I32)); |
| 4843 | storeLE(tEA_12, getQRegLane((vT+1) % 32, 1, Ity_I32)); |
| 4844 | storeLE(tEA_20, getQRegLane((vT+1) % 32, 2, Ity_I32)); |
| 4845 | storeLE(tEA_28, getQRegLane((vT+1) % 32, 3, Ity_I32)); |
| 4846 | } |
| 4847 | } |
| 4848 | else { |
| 4849 | vassert(0); // Can't happen. |
| 4850 | } |
| 4851 | putIReg64orSP(rN, binop(Iop_Add64, mkexpr(tEA), mkU64(32))); |
| 4852 | DIP("%s {v%u.%s, v%u.%s}, [%s], #32\n", isLD ? "ld2" : "st2", |
| 4853 | (vT+0) % 32, name, (vT+1) % 32, name, nameIReg64orSP(rN)); |
| 4854 | return True; |
| 4855 | } |
| 4856 | |
sewardj | 39f754d | 2014-06-24 10:26:52 +0000 | [diff] [blame] | 4857 | /* -------- LD1/ST1 (multi 1-elem structs, 2 regs, no offset) -------- */ |
sewardj | 950ca7a | 2014-04-03 23:03:32 +0000 | [diff] [blame] | 4858 | /* Only a very few cases. */ |
| 4859 | /* 31 23 |
| 4860 | 0100 1100 0100 0000 1010 00 n t LD1 {Vt.16b, V(t+1)%32.16b}, [Xn|SP] |
| 4861 | 0100 1100 0000 0000 1010 00 n t ST1 {Vt.16b, V(t+1)%32.16b}, [Xn|SP] |
| 4862 | */ |
| 4863 | if ( (insn & 0xFFFFFC00) == 0x4C40A000 // LD1 |
| 4864 | || (insn & 0xFFFFFC00) == 0x4C00A000 // ST1 |
| 4865 | ) { |
| 4866 | Bool isLD = INSN(22,22) == 1; |
| 4867 | UInt rN = INSN(9,5); |
| 4868 | UInt vT = INSN(4,0); |
| 4869 | IRTemp tEA = newTemp(Ity_I64); |
| 4870 | const HChar* name = "16b"; |
| 4871 | assign(tEA, getIReg64orSP(rN)); |
| 4872 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 4873 | IRExpr* tEA_0 = binop(Iop_Add64, mkexpr(tEA), mkU64(0)); |
| 4874 | IRExpr* tEA_16 = binop(Iop_Add64, mkexpr(tEA), mkU64(16)); |
| 4875 | if (isLD) { |
| 4876 | putQReg128((vT+0) % 32, loadLE(Ity_V128, tEA_0)); |
| 4877 | putQReg128((vT+1) % 32, loadLE(Ity_V128, tEA_16)); |
| 4878 | } else { |
| 4879 | storeLE(tEA_0, getQReg128((vT+0) % 32)); |
| 4880 | storeLE(tEA_16, getQReg128((vT+1) % 32)); |
| 4881 | } |
sewardj | 8a5ed54 | 2014-07-15 11:08:42 +0000 | [diff] [blame] | 4882 | DIP("%s {v%u.%s, v%u.%s}, [%s]\n", isLD ? "ld1" : "st1", |
| 4883 | (vT+0) % 32, name, (vT+1) % 32, name, nameIReg64orSP(rN)); |
| 4884 | return True; |
| 4885 | } |
| 4886 | |
| 4887 | /* -------- LD1/ST1 (multi 1-elem structs, 2 regs, post index) -------- */ |
| 4888 | /* Only a very few cases. */ |
| 4889 | /* 31 23 |
| 4890 | 0100 1100 1101 1111 1010 00 n t LD1 {Vt.16b, V(t+1)%32.16b}, [Xn|SP], #32 |
| 4891 | 0100 1100 1001 1111 1010 00 n t ST1 {Vt.16b, V(t+1)%32.16b}, [Xn|SP], #32 |
| 4892 | */ |
| 4893 | if ( (insn & 0xFFFFFC00) == 0x4CDFA000 // LD1 |
| 4894 | || (insn & 0xFFFFFC00) == 0x4C9FA000 // ST1 |
| 4895 | ) { |
| 4896 | Bool isLD = INSN(22,22) == 1; |
| 4897 | UInt rN = INSN(9,5); |
| 4898 | UInt vT = INSN(4,0); |
| 4899 | IRTemp tEA = newTemp(Ity_I64); |
| 4900 | const HChar* name = "16b"; |
| 4901 | assign(tEA, getIReg64orSP(rN)); |
| 4902 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 4903 | IRExpr* tEA_0 = binop(Iop_Add64, mkexpr(tEA), mkU64(0)); |
| 4904 | IRExpr* tEA_16 = binop(Iop_Add64, mkexpr(tEA), mkU64(16)); |
| 4905 | if (isLD) { |
| 4906 | putQReg128((vT+0) % 32, loadLE(Ity_V128, tEA_0)); |
| 4907 | putQReg128((vT+1) % 32, loadLE(Ity_V128, tEA_16)); |
| 4908 | } else { |
| 4909 | storeLE(tEA_0, getQReg128((vT+0) % 32)); |
| 4910 | storeLE(tEA_16, getQReg128((vT+1) % 32)); |
| 4911 | } |
| 4912 | putIReg64orSP(rN, binop(Iop_Add64, mkexpr(tEA), mkU64(32))); |
sewardj | 950ca7a | 2014-04-03 23:03:32 +0000 | [diff] [blame] | 4913 | DIP("%s {v%u.%s, v%u.%s}, [%s], #32\n", isLD ? "ld1" : "st1", |
| 4914 | (vT+0) % 32, name, (vT+1) % 32, name, nameIReg64orSP(rN)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4915 | return True; |
| 4916 | } |
| 4917 | |
sewardj | 39f754d | 2014-06-24 10:26:52 +0000 | [diff] [blame] | 4918 | /* -------- LD1/ST1 (multi 1-elem structs, 3 regs, no offset) -------- */ |
| 4919 | /* Only a very few cases. */ |
| 4920 | /* 31 23 |
| 4921 | 0100 1100 0100 0000 0110 00 n t LD1 {Vt.16b .. V(t+2)%32.16b}, [Xn|SP] |
| 4922 | 0100 1100 0000 0000 0110 00 n t ST1 {Vt.16b .. V(t+2)%32.16b}, [Xn|SP] |
| 4923 | */ |
| 4924 | if ( (insn & 0xFFFFFC00) == 0x4C406000 // LD1 |
| 4925 | || (insn & 0xFFFFFC00) == 0x4C006000 // ST1 |
| 4926 | ) { |
| 4927 | Bool isLD = INSN(22,22) == 1; |
| 4928 | UInt rN = INSN(9,5); |
| 4929 | UInt vT = INSN(4,0); |
| 4930 | IRTemp tEA = newTemp(Ity_I64); |
| 4931 | const HChar* name = "16b"; |
| 4932 | assign(tEA, getIReg64orSP(rN)); |
| 4933 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 4934 | IRExpr* tEA_0 = binop(Iop_Add64, mkexpr(tEA), mkU64(0)); |
| 4935 | IRExpr* tEA_16 = binop(Iop_Add64, mkexpr(tEA), mkU64(16)); |
| 4936 | IRExpr* tEA_32 = binop(Iop_Add64, mkexpr(tEA), mkU64(32)); |
| 4937 | if (isLD) { |
| 4938 | putQReg128((vT+0) % 32, loadLE(Ity_V128, tEA_0)); |
| 4939 | putQReg128((vT+1) % 32, loadLE(Ity_V128, tEA_16)); |
| 4940 | putQReg128((vT+2) % 32, loadLE(Ity_V128, tEA_32)); |
| 4941 | } else { |
| 4942 | storeLE(tEA_0, getQReg128((vT+0) % 32)); |
| 4943 | storeLE(tEA_16, getQReg128((vT+1) % 32)); |
| 4944 | storeLE(tEA_32, getQReg128((vT+2) % 32)); |
| 4945 | } |
| 4946 | DIP("%s {v%u.%s, v%u.%s, v%u.%s}, [%s], #32\n", |
| 4947 | isLD ? "ld1" : "st1", |
| 4948 | (vT+0) % 32, name, (vT+1) % 32, name, (vT+2) % 32, name, |
| 4949 | nameIReg64orSP(rN)); |
| 4950 | return True; |
| 4951 | } |
| 4952 | |
sewardj | 168c8bd | 2014-06-25 13:05:23 +0000 | [diff] [blame] | 4953 | /* -------- LD3/ST3 (multi 3-elem structs, 3 regs, post index) -------- */ |
| 4954 | /* Only a very few cases. */ |
| 4955 | /* 31 23 11 9 4 |
| 4956 | 0100 1100 1101 1111 0100 11 n t LD3 {Vt.2d .. V(t+2)%32.2d}, [Xn|SP], #48 |
| 4957 | 0100 1100 1001 1111 0100 11 n t ST3 {Vt.2d .. V(t+2)%32.2d}, [Xn|SP], #48 |
| 4958 | */ |
| 4959 | if ( (insn & 0xFFFFFC00) == 0x4CDF4C00 // LD3 .2d |
| 4960 | || (insn & 0xFFFFFC00) == 0x4C9F4C00 // ST3 .2d |
| 4961 | ) { |
| 4962 | Bool isLD = INSN(22,22) == 1; |
| 4963 | UInt rN = INSN(9,5); |
| 4964 | UInt vT = INSN(4,0); |
| 4965 | IRTemp tEA = newTemp(Ity_I64); |
| 4966 | UInt sz = INSN(11,10); |
| 4967 | const HChar* name = "??"; |
| 4968 | assign(tEA, getIReg64orSP(rN)); |
| 4969 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 4970 | IRExpr* tEA_0 = binop(Iop_Add64, mkexpr(tEA), mkU64(0)); |
| 4971 | IRExpr* tEA_8 = binop(Iop_Add64, mkexpr(tEA), mkU64(8)); |
| 4972 | IRExpr* tEA_16 = binop(Iop_Add64, mkexpr(tEA), mkU64(16)); |
| 4973 | IRExpr* tEA_24 = binop(Iop_Add64, mkexpr(tEA), mkU64(24)); |
| 4974 | IRExpr* tEA_32 = binop(Iop_Add64, mkexpr(tEA), mkU64(32)); |
| 4975 | IRExpr* tEA_40 = binop(Iop_Add64, mkexpr(tEA), mkU64(40)); |
| 4976 | if (sz == BITS2(1,1)) { |
| 4977 | name = "2d"; |
| 4978 | if (isLD) { |
| 4979 | putQRegLane((vT+0) % 32, 0, loadLE(Ity_I64, tEA_0)); |
| 4980 | putQRegLane((vT+0) % 32, 1, loadLE(Ity_I64, tEA_24)); |
| 4981 | putQRegLane((vT+1) % 32, 0, loadLE(Ity_I64, tEA_8)); |
| 4982 | putQRegLane((vT+1) % 32, 1, loadLE(Ity_I64, tEA_32)); |
| 4983 | putQRegLane((vT+2) % 32, 0, loadLE(Ity_I64, tEA_16)); |
| 4984 | putQRegLane((vT+2) % 32, 1, loadLE(Ity_I64, tEA_40)); |
| 4985 | } else { |
| 4986 | storeLE(tEA_0, getQRegLane((vT+0) % 32, 0, Ity_I64)); |
| 4987 | storeLE(tEA_24, getQRegLane((vT+0) % 32, 1, Ity_I64)); |
| 4988 | storeLE(tEA_8, getQRegLane((vT+1) % 32, 0, Ity_I64)); |
| 4989 | storeLE(tEA_32, getQRegLane((vT+1) % 32, 1, Ity_I64)); |
| 4990 | storeLE(tEA_16, getQRegLane((vT+2) % 32, 0, Ity_I64)); |
| 4991 | storeLE(tEA_40, getQRegLane((vT+2) % 32, 1, Ity_I64)); |
| 4992 | } |
| 4993 | } |
| 4994 | else { |
| 4995 | vassert(0); // Can't happen. |
| 4996 | } |
| 4997 | putIReg64orSP(rN, binop(Iop_Add64, mkexpr(tEA), mkU64(48))); |
| 4998 | DIP("%s {v%u.%s, v%u.%s, v%u.%s}, [%s], #32\n", |
| 4999 | isLD ? "ld3" : "st3", |
| 5000 | (vT+0) % 32, name, (vT+1) % 32, name, (vT+2) % 32, name, |
| 5001 | nameIReg64orSP(rN)); |
| 5002 | return True; |
| 5003 | } |
| 5004 | |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 5005 | /* ------------------ LD{,A}X{R,RH,RB} ------------------ */ |
| 5006 | /* ------------------ ST{,L}X{R,RH,RB} ------------------ */ |
| 5007 | /* 31 29 23 20 14 9 4 |
| 5008 | sz 001000 010 11111 0 11111 n t LDX{R,RH,RB} Rt, [Xn|SP] |
| 5009 | sz 001000 010 11111 1 11111 n t LDAX{R,RH,RB} Rt, [Xn|SP] |
| 5010 | sz 001000 000 s 0 11111 n t STX{R,RH,RB} Ws, Rt, [Xn|SP] |
| 5011 | sz 001000 000 s 1 11111 n t STLX{R,RH,RB} Ws, Rt, [Xn|SP] |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5012 | */ |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 5013 | if (INSN(29,23) == BITS7(0,0,1,0,0,0,0) |
| 5014 | && (INSN(23,21) & BITS3(1,0,1)) == BITS3(0,0,0) |
| 5015 | && INSN(14,10) == BITS5(1,1,1,1,1)) { |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 5016 | UInt szBlg2 = INSN(31,30); |
| 5017 | Bool isLD = INSN(22,22) == 1; |
| 5018 | Bool isAcqOrRel = INSN(15,15) == 1; |
| 5019 | UInt ss = INSN(20,16); |
| 5020 | UInt nn = INSN(9,5); |
| 5021 | UInt tt = INSN(4,0); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5022 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 5023 | vassert(szBlg2 < 4); |
| 5024 | UInt szB = 1 << szBlg2; /* 1, 2, 4 or 8 */ |
| 5025 | IRType ty = integerIRTypeOfSize(szB); |
| 5026 | const HChar* suffix[4] = { "rb", "rh", "r", "r" }; |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 5027 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 5028 | IRTemp ea = newTemp(Ity_I64); |
| 5029 | assign(ea, getIReg64orSP(nn)); |
| 5030 | /* FIXME generate check that ea is szB-aligned */ |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 5031 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 5032 | if (isLD && ss == BITS5(1,1,1,1,1)) { |
| 5033 | IRTemp res = newTemp(ty); |
| 5034 | stmt(IRStmt_LLSC(Iend_LE, res, mkexpr(ea), NULL/*LL*/)); |
| 5035 | putIReg64orZR(tt, widenUto64(ty, mkexpr(res))); |
| 5036 | if (isAcqOrRel) { |
| 5037 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 5038 | } |
| 5039 | DIP("ld%sx%s %s, [%s]\n", isAcqOrRel ? "a" : "", suffix[szBlg2], |
| 5040 | nameIRegOrZR(szB == 8, tt), nameIReg64orSP(nn)); |
| 5041 | return True; |
| 5042 | } |
| 5043 | if (!isLD) { |
| 5044 | if (isAcqOrRel) { |
| 5045 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 5046 | } |
| 5047 | IRTemp res = newTemp(Ity_I1); |
| 5048 | IRExpr* data = narrowFrom64(ty, getIReg64orZR(tt)); |
| 5049 | stmt(IRStmt_LLSC(Iend_LE, res, mkexpr(ea), data)); |
| 5050 | /* IR semantics: res is 1 if store succeeds, 0 if it fails. |
| 5051 | Need to set rS to 1 on failure, 0 on success. */ |
| 5052 | putIReg64orZR(ss, binop(Iop_Xor64, unop(Iop_1Uto64, mkexpr(res)), |
| 5053 | mkU64(1))); |
| 5054 | DIP("st%sx%s %s, %s, [%s]\n", isAcqOrRel ? "a" : "", suffix[szBlg2], |
| 5055 | nameIRegOrZR(False, ss), |
| 5056 | nameIRegOrZR(szB == 8, tt), nameIReg64orSP(nn)); |
| 5057 | return True; |
| 5058 | } |
| 5059 | /* else fall through */ |
| 5060 | } |
| 5061 | |
| 5062 | /* ------------------ LDA{R,RH,RB} ------------------ */ |
| 5063 | /* ------------------ STL{R,RH,RB} ------------------ */ |
| 5064 | /* 31 29 23 20 14 9 4 |
| 5065 | sz 001000 110 11111 1 11111 n t LDAR<sz> Rt, [Xn|SP] |
| 5066 | sz 001000 100 11111 1 11111 n t STLR<sz> Rt, [Xn|SP] |
| 5067 | */ |
| 5068 | if (INSN(29,23) == BITS7(0,0,1,0,0,0,1) |
| 5069 | && INSN(21,10) == BITS12(0,1,1,1,1,1,1,1,1,1,1,1)) { |
| 5070 | UInt szBlg2 = INSN(31,30); |
| 5071 | Bool isLD = INSN(22,22) == 1; |
| 5072 | UInt nn = INSN(9,5); |
| 5073 | UInt tt = INSN(4,0); |
| 5074 | |
| 5075 | vassert(szBlg2 < 4); |
| 5076 | UInt szB = 1 << szBlg2; /* 1, 2, 4 or 8 */ |
| 5077 | IRType ty = integerIRTypeOfSize(szB); |
| 5078 | const HChar* suffix[4] = { "rb", "rh", "r", "r" }; |
| 5079 | |
| 5080 | IRTemp ea = newTemp(Ity_I64); |
| 5081 | assign(ea, getIReg64orSP(nn)); |
| 5082 | /* FIXME generate check that ea is szB-aligned */ |
| 5083 | |
| 5084 | if (isLD) { |
| 5085 | IRTemp res = newTemp(ty); |
| 5086 | assign(res, loadLE(ty, mkexpr(ea))); |
| 5087 | putIReg64orZR(tt, widenUto64(ty, mkexpr(res))); |
| 5088 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 5089 | DIP("lda%s %s, [%s]\n", suffix[szBlg2], |
| 5090 | nameIRegOrZR(szB == 8, tt), nameIReg64orSP(nn)); |
| 5091 | } else { |
| 5092 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 5093 | IRExpr* data = narrowFrom64(ty, getIReg64orZR(tt)); |
| 5094 | storeLE(mkexpr(ea), data); |
| 5095 | DIP("stl%s %s, [%s]\n", suffix[szBlg2], |
| 5096 | nameIRegOrZR(szB == 8, tt), nameIReg64orSP(nn)); |
| 5097 | } |
| 5098 | return True; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5099 | } |
| 5100 | |
| 5101 | vex_printf("ARM64 front end: load_store\n"); |
| 5102 | return False; |
| 5103 | # undef INSN |
| 5104 | } |
| 5105 | |
| 5106 | |
| 5107 | /*------------------------------------------------------------*/ |
| 5108 | /*--- Control flow and misc instructions ---*/ |
| 5109 | /*------------------------------------------------------------*/ |
| 5110 | |
| 5111 | static |
sewardj | 6590299 | 2014-05-03 21:20:56 +0000 | [diff] [blame] | 5112 | Bool dis_ARM64_branch_etc(/*MB_OUT*/DisResult* dres, UInt insn, |
| 5113 | VexArchInfo* archinfo) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5114 | { |
| 5115 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 5116 | |
| 5117 | /* ---------------------- B cond ----------------------- */ |
| 5118 | /* 31 24 4 3 |
| 5119 | 0101010 0 imm19 0 cond */ |
| 5120 | if (INSN(31,24) == BITS8(0,1,0,1,0,1,0,0) && INSN(4,4) == 0) { |
| 5121 | UInt cond = INSN(3,0); |
| 5122 | ULong uimm64 = INSN(23,5) << 2; |
| 5123 | Long simm64 = (Long)sx_to_64(uimm64, 21); |
| 5124 | vassert(dres->whatNext == Dis_Continue); |
| 5125 | vassert(dres->len == 4); |
| 5126 | vassert(dres->continueAt == 0); |
| 5127 | vassert(dres->jk_StopHere == Ijk_INVALID); |
| 5128 | stmt( IRStmt_Exit(unop(Iop_64to1, mk_arm64g_calculate_condition(cond)), |
| 5129 | Ijk_Boring, |
| 5130 | IRConst_U64(guest_PC_curr_instr + simm64), |
| 5131 | OFFB_PC) ); |
| 5132 | putPC(mkU64(guest_PC_curr_instr + 4)); |
| 5133 | dres->whatNext = Dis_StopHere; |
| 5134 | dres->jk_StopHere = Ijk_Boring; |
| 5135 | DIP("b.%s 0x%llx\n", nameCC(cond), guest_PC_curr_instr + simm64); |
| 5136 | return True; |
| 5137 | } |
| 5138 | |
| 5139 | /* -------------------- B{L} uncond -------------------- */ |
| 5140 | if (INSN(30,26) == BITS5(0,0,1,0,1)) { |
| 5141 | /* 000101 imm26 B (PC + sxTo64(imm26 << 2)) |
| 5142 | 100101 imm26 B (PC + sxTo64(imm26 << 2)) |
| 5143 | */ |
| 5144 | UInt bLink = INSN(31,31); |
| 5145 | ULong uimm64 = INSN(25,0) << 2; |
| 5146 | Long simm64 = (Long)sx_to_64(uimm64, 28); |
| 5147 | if (bLink) { |
| 5148 | putIReg64orSP(30, mkU64(guest_PC_curr_instr + 4)); |
| 5149 | } |
| 5150 | putPC(mkU64(guest_PC_curr_instr + simm64)); |
| 5151 | dres->whatNext = Dis_StopHere; |
| 5152 | dres->jk_StopHere = Ijk_Call; |
| 5153 | DIP("b%s 0x%llx\n", bLink == 1 ? "l" : "", |
| 5154 | guest_PC_curr_instr + simm64); |
| 5155 | return True; |
| 5156 | } |
| 5157 | |
| 5158 | /* --------------------- B{L} reg --------------------- */ |
| 5159 | /* 31 24 22 20 15 9 4 |
| 5160 | 1101011 00 10 11111 000000 nn 00000 RET Rn |
| 5161 | 1101011 00 01 11111 000000 nn 00000 CALL Rn |
| 5162 | 1101011 00 00 11111 000000 nn 00000 JMP Rn |
| 5163 | */ |
| 5164 | if (INSN(31,23) == BITS9(1,1,0,1,0,1,1,0,0) |
| 5165 | && INSN(20,16) == BITS5(1,1,1,1,1) |
| 5166 | && INSN(15,10) == BITS6(0,0,0,0,0,0) |
| 5167 | && INSN(4,0) == BITS5(0,0,0,0,0)) { |
| 5168 | UInt branch_type = INSN(22,21); |
| 5169 | UInt nn = INSN(9,5); |
| 5170 | if (branch_type == BITS2(1,0) /* RET */) { |
| 5171 | putPC(getIReg64orZR(nn)); |
| 5172 | dres->whatNext = Dis_StopHere; |
| 5173 | dres->jk_StopHere = Ijk_Ret; |
| 5174 | DIP("ret %s\n", nameIReg64orZR(nn)); |
| 5175 | return True; |
| 5176 | } |
| 5177 | if (branch_type == BITS2(0,1) /* CALL */) { |
sewardj | 702054e | 2014-05-07 11:09:28 +0000 | [diff] [blame] | 5178 | IRTemp dst = newTemp(Ity_I64); |
| 5179 | assign(dst, getIReg64orZR(nn)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5180 | putIReg64orSP(30, mkU64(guest_PC_curr_instr + 4)); |
sewardj | 702054e | 2014-05-07 11:09:28 +0000 | [diff] [blame] | 5181 | putPC(mkexpr(dst)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5182 | dres->whatNext = Dis_StopHere; |
| 5183 | dres->jk_StopHere = Ijk_Call; |
| 5184 | DIP("blr %s\n", nameIReg64orZR(nn)); |
| 5185 | return True; |
| 5186 | } |
| 5187 | if (branch_type == BITS2(0,0) /* JMP */) { |
| 5188 | putPC(getIReg64orZR(nn)); |
| 5189 | dres->whatNext = Dis_StopHere; |
| 5190 | dres->jk_StopHere = Ijk_Boring; |
| 5191 | DIP("jmp %s\n", nameIReg64orZR(nn)); |
| 5192 | return True; |
| 5193 | } |
| 5194 | } |
| 5195 | |
| 5196 | /* -------------------- CB{N}Z -------------------- */ |
| 5197 | /* sf 011 010 1 imm19 Rt CBNZ Xt|Wt, (PC + sxTo64(imm19 << 2)) |
| 5198 | sf 011 010 0 imm19 Rt CBZ Xt|Wt, (PC + sxTo64(imm19 << 2)) |
| 5199 | */ |
| 5200 | if (INSN(30,25) == BITS6(0,1,1,0,1,0)) { |
| 5201 | Bool is64 = INSN(31,31) == 1; |
| 5202 | Bool bIfZ = INSN(24,24) == 0; |
| 5203 | ULong uimm64 = INSN(23,5) << 2; |
| 5204 | UInt rT = INSN(4,0); |
| 5205 | Long simm64 = (Long)sx_to_64(uimm64, 21); |
| 5206 | IRExpr* cond = NULL; |
| 5207 | if (is64) { |
| 5208 | cond = binop(bIfZ ? Iop_CmpEQ64 : Iop_CmpNE64, |
| 5209 | getIReg64orZR(rT), mkU64(0)); |
| 5210 | } else { |
| 5211 | cond = binop(bIfZ ? Iop_CmpEQ32 : Iop_CmpNE32, |
| 5212 | getIReg32orZR(rT), mkU32(0)); |
| 5213 | } |
| 5214 | stmt( IRStmt_Exit(cond, |
| 5215 | Ijk_Boring, |
| 5216 | IRConst_U64(guest_PC_curr_instr + simm64), |
| 5217 | OFFB_PC) ); |
| 5218 | putPC(mkU64(guest_PC_curr_instr + 4)); |
| 5219 | dres->whatNext = Dis_StopHere; |
| 5220 | dres->jk_StopHere = Ijk_Boring; |
| 5221 | DIP("cb%sz %s, 0x%llx\n", |
| 5222 | bIfZ ? "" : "n", nameIRegOrZR(is64, rT), |
| 5223 | guest_PC_curr_instr + simm64); |
| 5224 | return True; |
| 5225 | } |
| 5226 | |
| 5227 | /* -------------------- TB{N}Z -------------------- */ |
| 5228 | /* 31 30 24 23 18 5 4 |
| 5229 | b5 011 011 1 b40 imm14 t TBNZ Xt, #(b5:b40), (PC + sxTo64(imm14 << 2)) |
| 5230 | b5 011 011 0 b40 imm14 t TBZ Xt, #(b5:b40), (PC + sxTo64(imm14 << 2)) |
| 5231 | */ |
| 5232 | if (INSN(30,25) == BITS6(0,1,1,0,1,1)) { |
| 5233 | UInt b5 = INSN(31,31); |
| 5234 | Bool bIfZ = INSN(24,24) == 0; |
| 5235 | UInt b40 = INSN(23,19); |
| 5236 | UInt imm14 = INSN(18,5); |
| 5237 | UInt tt = INSN(4,0); |
| 5238 | UInt bitNo = (b5 << 5) | b40; |
| 5239 | ULong uimm64 = imm14 << 2; |
| 5240 | Long simm64 = sx_to_64(uimm64, 16); |
| 5241 | IRExpr* cond |
| 5242 | = binop(bIfZ ? Iop_CmpEQ64 : Iop_CmpNE64, |
| 5243 | binop(Iop_And64, |
| 5244 | binop(Iop_Shr64, getIReg64orZR(tt), mkU8(bitNo)), |
| 5245 | mkU64(1)), |
| 5246 | mkU64(0)); |
| 5247 | stmt( IRStmt_Exit(cond, |
| 5248 | Ijk_Boring, |
| 5249 | IRConst_U64(guest_PC_curr_instr + simm64), |
| 5250 | OFFB_PC) ); |
| 5251 | putPC(mkU64(guest_PC_curr_instr + 4)); |
| 5252 | dres->whatNext = Dis_StopHere; |
| 5253 | dres->jk_StopHere = Ijk_Boring; |
| 5254 | DIP("tb%sz %s, #%u, 0x%llx\n", |
| 5255 | bIfZ ? "" : "n", nameIReg64orZR(tt), bitNo, |
| 5256 | guest_PC_curr_instr + simm64); |
| 5257 | return True; |
| 5258 | } |
| 5259 | |
| 5260 | /* -------------------- SVC -------------------- */ |
| 5261 | /* 11010100 000 imm16 000 01 |
| 5262 | Don't bother with anything except the imm16==0 case. |
| 5263 | */ |
| 5264 | if (INSN(31,0) == 0xD4000001) { |
| 5265 | putPC(mkU64(guest_PC_curr_instr + 4)); |
| 5266 | dres->whatNext = Dis_StopHere; |
| 5267 | dres->jk_StopHere = Ijk_Sys_syscall; |
| 5268 | DIP("svc #0\n"); |
| 5269 | return True; |
| 5270 | } |
| 5271 | |
| 5272 | /* ------------------ M{SR,RS} ------------------ */ |
sewardj | 6eb5ef8 | 2014-07-14 20:39:23 +0000 | [diff] [blame] | 5273 | /* ---- Cases for TPIDR_EL0 ---- |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5274 | 0xD51BD0 010 Rt MSR tpidr_el0, rT |
| 5275 | 0xD53BD0 010 Rt MRS rT, tpidr_el0 |
| 5276 | */ |
| 5277 | if ( (INSN(31,0) & 0xFFFFFFE0) == 0xD51BD040 /*MSR*/ |
| 5278 | || (INSN(31,0) & 0xFFFFFFE0) == 0xD53BD040 /*MRS*/) { |
| 5279 | Bool toSys = INSN(21,21) == 0; |
| 5280 | UInt tt = INSN(4,0); |
| 5281 | if (toSys) { |
| 5282 | stmt( IRStmt_Put( OFFB_TPIDR_EL0, getIReg64orZR(tt)) ); |
| 5283 | DIP("msr tpidr_el0, %s\n", nameIReg64orZR(tt)); |
| 5284 | } else { |
| 5285 | putIReg64orZR(tt, IRExpr_Get( OFFB_TPIDR_EL0, Ity_I64 )); |
| 5286 | DIP("mrs %s, tpidr_el0\n", nameIReg64orZR(tt)); |
| 5287 | } |
| 5288 | return True; |
| 5289 | } |
sewardj | 6eb5ef8 | 2014-07-14 20:39:23 +0000 | [diff] [blame] | 5290 | /* ---- Cases for FPCR ---- |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5291 | 0xD51B44 000 Rt MSR fpcr, rT |
| 5292 | 0xD53B44 000 Rt MSR rT, fpcr |
| 5293 | */ |
| 5294 | if ( (INSN(31,0) & 0xFFFFFFE0) == 0xD51B4400 /*MSR*/ |
| 5295 | || (INSN(31,0) & 0xFFFFFFE0) == 0xD53B4400 /*MRS*/) { |
| 5296 | Bool toSys = INSN(21,21) == 0; |
| 5297 | UInt tt = INSN(4,0); |
| 5298 | if (toSys) { |
| 5299 | stmt( IRStmt_Put( OFFB_FPCR, getIReg32orZR(tt)) ); |
| 5300 | DIP("msr fpcr, %s\n", nameIReg64orZR(tt)); |
| 5301 | } else { |
| 5302 | putIReg32orZR(tt, IRExpr_Get(OFFB_FPCR, Ity_I32)); |
| 5303 | DIP("mrs %s, fpcr\n", nameIReg64orZR(tt)); |
| 5304 | } |
| 5305 | return True; |
| 5306 | } |
sewardj | 6eb5ef8 | 2014-07-14 20:39:23 +0000 | [diff] [blame] | 5307 | /* ---- Cases for FPSR ---- |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 5308 | 0xD51B44 001 Rt MSR fpsr, rT |
| 5309 | 0xD53B44 001 Rt MSR rT, fpsr |
sewardj | a0645d5 | 2014-06-28 22:11:16 +0000 | [diff] [blame] | 5310 | The only part of this we model is FPSR.QC. All other bits |
| 5311 | are ignored when writing to it and RAZ when reading from it. |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5312 | */ |
| 5313 | if ( (INSN(31,0) & 0xFFFFFFE0) == 0xD51B4420 /*MSR*/ |
| 5314 | || (INSN(31,0) & 0xFFFFFFE0) == 0xD53B4420 /*MRS*/) { |
| 5315 | Bool toSys = INSN(21,21) == 0; |
| 5316 | UInt tt = INSN(4,0); |
| 5317 | if (toSys) { |
sewardj | a0645d5 | 2014-06-28 22:11:16 +0000 | [diff] [blame] | 5318 | /* Just deal with FPSR.QC. Make up a V128 value which is |
| 5319 | zero if Xt[27] is zero and any other value if Xt[27] is |
| 5320 | nonzero. */ |
| 5321 | IRTemp qc64 = newTemp(Ity_I64); |
| 5322 | assign(qc64, binop(Iop_And64, |
| 5323 | binop(Iop_Shr64, getIReg64orZR(tt), mkU8(27)), |
| 5324 | mkU64(1))); |
| 5325 | IRExpr* qcV128 = binop(Iop_64HLtoV128, mkexpr(qc64), mkexpr(qc64)); |
| 5326 | stmt( IRStmt_Put( OFFB_QCFLAG, qcV128 ) ); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5327 | DIP("msr fpsr, %s\n", nameIReg64orZR(tt)); |
| 5328 | } else { |
sewardj | a0645d5 | 2014-06-28 22:11:16 +0000 | [diff] [blame] | 5329 | /* Generate a value which is all zeroes except for bit 27, |
| 5330 | which must be zero if QCFLAG is all zeroes and one otherwise. */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5331 | IRTemp qcV128 = newTempV128(); |
sewardj | a0645d5 | 2014-06-28 22:11:16 +0000 | [diff] [blame] | 5332 | assign(qcV128, IRExpr_Get( OFFB_QCFLAG, Ity_V128 )); |
| 5333 | IRTemp qc64 = newTemp(Ity_I64); |
| 5334 | assign(qc64, binop(Iop_Or64, unop(Iop_V128HIto64, mkexpr(qcV128)), |
| 5335 | unop(Iop_V128to64, mkexpr(qcV128)))); |
| 5336 | IRExpr* res = binop(Iop_Shl64, |
| 5337 | unop(Iop_1Uto64, |
| 5338 | binop(Iop_CmpNE64, mkexpr(qc64), mkU64(0))), |
| 5339 | mkU8(27)); |
| 5340 | putIReg64orZR(tt, res); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5341 | DIP("mrs %s, fpsr\n", nameIReg64orZR(tt)); |
| 5342 | } |
| 5343 | return True; |
| 5344 | } |
sewardj | 6eb5ef8 | 2014-07-14 20:39:23 +0000 | [diff] [blame] | 5345 | /* ---- Cases for NZCV ---- |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5346 | D51B42 000 Rt MSR nzcv, rT |
| 5347 | D53B42 000 Rt MRS rT, nzcv |
sewardj | a0645d5 | 2014-06-28 22:11:16 +0000 | [diff] [blame] | 5348 | The only parts of NZCV that actually exist are bits 31:28, which |
| 5349 | are the N Z C and V bits themselves. Hence the flags thunk provides |
| 5350 | all the state we need. |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5351 | */ |
| 5352 | if ( (INSN(31,0) & 0xFFFFFFE0) == 0xD51B4200 /*MSR*/ |
| 5353 | || (INSN(31,0) & 0xFFFFFFE0) == 0xD53B4200 /*MRS*/) { |
| 5354 | Bool toSys = INSN(21,21) == 0; |
| 5355 | UInt tt = INSN(4,0); |
| 5356 | if (toSys) { |
| 5357 | IRTemp t = newTemp(Ity_I64); |
| 5358 | assign(t, binop(Iop_And64, getIReg64orZR(tt), mkU64(0xF0000000ULL))); |
| 5359 | setFlags_COPY(t); |
| 5360 | DIP("msr %s, nzcv\n", nameIReg32orZR(tt)); |
| 5361 | } else { |
| 5362 | IRTemp res = newTemp(Ity_I64); |
| 5363 | assign(res, mk_arm64g_calculate_flags_nzcv()); |
| 5364 | putIReg32orZR(tt, unop(Iop_64to32, mkexpr(res))); |
| 5365 | DIP("mrs %s, nzcv\n", nameIReg64orZR(tt)); |
| 5366 | } |
| 5367 | return True; |
| 5368 | } |
sewardj | 6eb5ef8 | 2014-07-14 20:39:23 +0000 | [diff] [blame] | 5369 | /* ---- Cases for DCZID_EL0 ---- |
sewardj | d512d10 | 2014-02-21 14:49:44 +0000 | [diff] [blame] | 5370 | Don't support arbitrary reads and writes to this register. Just |
| 5371 | return the value 16, which indicates that the DC ZVA instruction |
| 5372 | is not permitted, so we don't have to emulate it. |
| 5373 | D5 3B 00 111 Rt MRS rT, dczid_el0 |
| 5374 | */ |
| 5375 | if ((INSN(31,0) & 0xFFFFFFE0) == 0xD53B00E0) { |
| 5376 | UInt tt = INSN(4,0); |
| 5377 | putIReg64orZR(tt, mkU64(1<<4)); |
| 5378 | DIP("mrs %s, dczid_el0 (FAKED)\n", nameIReg64orZR(tt)); |
| 5379 | return True; |
| 5380 | } |
sewardj | 6eb5ef8 | 2014-07-14 20:39:23 +0000 | [diff] [blame] | 5381 | /* ---- Cases for CTR_EL0 ---- |
sewardj | 6590299 | 2014-05-03 21:20:56 +0000 | [diff] [blame] | 5382 | We just handle reads, and make up a value from the D and I line |
| 5383 | sizes in the VexArchInfo we are given, and patch in the following |
| 5384 | fields that the Foundation model gives ("natively"): |
| 5385 | CWG = 0b0100, ERG = 0b0100, L1Ip = 0b11 |
| 5386 | D5 3B 00 001 Rt MRS rT, dczid_el0 |
| 5387 | */ |
| 5388 | if ((INSN(31,0) & 0xFFFFFFE0) == 0xD53B0020) { |
| 5389 | UInt tt = INSN(4,0); |
| 5390 | /* Need to generate a value from dMinLine_lg2_szB and |
| 5391 | dMinLine_lg2_szB. The value in the register is in 32-bit |
| 5392 | units, so need to subtract 2 from the values in the |
| 5393 | VexArchInfo. We can assume that the values here are valid -- |
| 5394 | disInstr_ARM64 checks them -- so there's no need to deal with |
| 5395 | out-of-range cases. */ |
| 5396 | vassert(archinfo->arm64_dMinLine_lg2_szB >= 2 |
| 5397 | && archinfo->arm64_dMinLine_lg2_szB <= 17 |
| 5398 | && archinfo->arm64_iMinLine_lg2_szB >= 2 |
| 5399 | && archinfo->arm64_iMinLine_lg2_szB <= 17); |
| 5400 | UInt val |
| 5401 | = 0x8440c000 | ((0xF & (archinfo->arm64_dMinLine_lg2_szB - 2)) << 16) |
| 5402 | | ((0xF & (archinfo->arm64_iMinLine_lg2_szB - 2)) << 0); |
| 5403 | putIReg64orZR(tt, mkU64(val)); |
| 5404 | DIP("mrs %s, ctr_el0\n", nameIReg64orZR(tt)); |
| 5405 | return True; |
| 5406 | } |
sewardj | 6eb5ef8 | 2014-07-14 20:39:23 +0000 | [diff] [blame] | 5407 | /* ---- Cases for CNTVCT_EL0 ---- |
| 5408 | This is a timestamp counter of some sort. Support reads of it only |
| 5409 | by passing through to the host. |
| 5410 | D5 3B E0 010 Rt MRS Xt, cntvct_el0 |
| 5411 | */ |
| 5412 | if ((INSN(31,0) & 0xFFFFFFE0) == 0xD53BE040) { |
| 5413 | UInt tt = INSN(4,0); |
| 5414 | IRTemp val = newTemp(Ity_I64); |
| 5415 | IRExpr** args = mkIRExprVec_0(); |
| 5416 | IRDirty* d = unsafeIRDirty_1_N ( |
| 5417 | val, |
| 5418 | 0/*regparms*/, |
| 5419 | "arm64g_dirtyhelper_MRS_CNTVCT_EL0", |
| 5420 | &arm64g_dirtyhelper_MRS_CNTVCT_EL0, |
| 5421 | args |
| 5422 | ); |
| 5423 | /* execute the dirty call, dumping the result in val. */ |
| 5424 | stmt( IRStmt_Dirty(d) ); |
| 5425 | putIReg64orZR(tt, mkexpr(val)); |
| 5426 | DIP("mrs %s, cntvct_el0\n", nameIReg64orZR(tt)); |
| 5427 | return True; |
| 5428 | } |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5429 | |
sewardj | 6590299 | 2014-05-03 21:20:56 +0000 | [diff] [blame] | 5430 | /* ------------------ IC_IVAU ------------------ */ |
| 5431 | /* D5 0B 75 001 Rt ic ivau, rT |
| 5432 | */ |
| 5433 | if ((INSN(31,0) & 0xFFFFFFE0) == 0xD50B7520) { |
| 5434 | /* We will always be provided with a valid iMinLine value. */ |
| 5435 | vassert(archinfo->arm64_iMinLine_lg2_szB >= 2 |
| 5436 | && archinfo->arm64_iMinLine_lg2_szB <= 17); |
| 5437 | /* Round the requested address, in rT, down to the start of the |
| 5438 | containing block. */ |
| 5439 | UInt tt = INSN(4,0); |
| 5440 | ULong lineszB = 1ULL << archinfo->arm64_iMinLine_lg2_szB; |
| 5441 | IRTemp addr = newTemp(Ity_I64); |
| 5442 | assign( addr, binop( Iop_And64, |
| 5443 | getIReg64orZR(tt), |
| 5444 | mkU64(~(lineszB - 1))) ); |
| 5445 | /* Set the invalidation range, request exit-and-invalidate, with |
| 5446 | continuation at the next instruction. */ |
sewardj | 05f5e01 | 2014-05-04 10:52:11 +0000 | [diff] [blame] | 5447 | stmt(IRStmt_Put(OFFB_CMSTART, mkexpr(addr))); |
| 5448 | stmt(IRStmt_Put(OFFB_CMLEN, mkU64(lineszB))); |
sewardj | 6590299 | 2014-05-03 21:20:56 +0000 | [diff] [blame] | 5449 | /* be paranoid ... */ |
| 5450 | stmt( IRStmt_MBE(Imbe_Fence) ); |
| 5451 | putPC(mkU64( guest_PC_curr_instr + 4 )); |
| 5452 | dres->whatNext = Dis_StopHere; |
sewardj | 05f5e01 | 2014-05-04 10:52:11 +0000 | [diff] [blame] | 5453 | dres->jk_StopHere = Ijk_InvalICache; |
sewardj | 6590299 | 2014-05-03 21:20:56 +0000 | [diff] [blame] | 5454 | DIP("ic ivau, %s\n", nameIReg64orZR(tt)); |
| 5455 | return True; |
| 5456 | } |
| 5457 | |
| 5458 | /* ------------------ DC_CVAU ------------------ */ |
| 5459 | /* D5 0B 7B 001 Rt dc cvau, rT |
| 5460 | */ |
| 5461 | if ((INSN(31,0) & 0xFFFFFFE0) == 0xD50B7B20) { |
| 5462 | /* Exactly the same scheme as for IC IVAU, except we observe the |
sewardj | 05f5e01 | 2014-05-04 10:52:11 +0000 | [diff] [blame] | 5463 | dMinLine size, and request an Ijk_FlushDCache instead of |
| 5464 | Ijk_InvalICache. */ |
sewardj | 6590299 | 2014-05-03 21:20:56 +0000 | [diff] [blame] | 5465 | /* We will always be provided with a valid dMinLine value. */ |
| 5466 | vassert(archinfo->arm64_dMinLine_lg2_szB >= 2 |
| 5467 | && archinfo->arm64_dMinLine_lg2_szB <= 17); |
| 5468 | /* Round the requested address, in rT, down to the start of the |
| 5469 | containing block. */ |
| 5470 | UInt tt = INSN(4,0); |
| 5471 | ULong lineszB = 1ULL << archinfo->arm64_dMinLine_lg2_szB; |
| 5472 | IRTemp addr = newTemp(Ity_I64); |
| 5473 | assign( addr, binop( Iop_And64, |
| 5474 | getIReg64orZR(tt), |
| 5475 | mkU64(~(lineszB - 1))) ); |
| 5476 | /* Set the flush range, request exit-and-flush, with |
| 5477 | continuation at the next instruction. */ |
sewardj | 05f5e01 | 2014-05-04 10:52:11 +0000 | [diff] [blame] | 5478 | stmt(IRStmt_Put(OFFB_CMSTART, mkexpr(addr))); |
| 5479 | stmt(IRStmt_Put(OFFB_CMLEN, mkU64(lineszB))); |
sewardj | 6590299 | 2014-05-03 21:20:56 +0000 | [diff] [blame] | 5480 | /* be paranoid ... */ |
| 5481 | stmt( IRStmt_MBE(Imbe_Fence) ); |
| 5482 | putPC(mkU64( guest_PC_curr_instr + 4 )); |
| 5483 | dres->whatNext = Dis_StopHere; |
| 5484 | dres->jk_StopHere = Ijk_FlushDCache; |
| 5485 | DIP("dc cvau, %s\n", nameIReg64orZR(tt)); |
| 5486 | return True; |
| 5487 | } |
| 5488 | |
| 5489 | /* ------------------ ISB, DMB, DSB ------------------ */ |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5490 | if (INSN(31,0) == 0xD5033FDF) { |
sewardj | d512d10 | 2014-02-21 14:49:44 +0000 | [diff] [blame] | 5491 | stmt(IRStmt_MBE(Imbe_Fence)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5492 | DIP("isb\n"); |
| 5493 | return True; |
| 5494 | } |
| 5495 | if (INSN(31,0) == 0xD5033BBF) { |
sewardj | d512d10 | 2014-02-21 14:49:44 +0000 | [diff] [blame] | 5496 | stmt(IRStmt_MBE(Imbe_Fence)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5497 | DIP("dmb ish\n"); |
| 5498 | return True; |
| 5499 | } |
sewardj | ab102bd | 2014-06-04 11:44:45 +0000 | [diff] [blame] | 5500 | if (INSN(31,0) == 0xD5033ABF) { |
| 5501 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 5502 | DIP("dmb ishst\n"); |
| 5503 | return True; |
| 5504 | } |
sewardj | 6590299 | 2014-05-03 21:20:56 +0000 | [diff] [blame] | 5505 | if (INSN(31,0) == 0xD5033B9F) { |
| 5506 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 5507 | DIP("dsb ish\n"); |
| 5508 | return True; |
| 5509 | } |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5510 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 5511 | /* -------------------- NOP -------------------- */ |
| 5512 | if (INSN(31,0) == 0xD503201F) { |
| 5513 | DIP("nop\n"); |
| 5514 | return True; |
| 5515 | } |
| 5516 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5517 | //fail: |
| 5518 | vex_printf("ARM64 front end: branch_etc\n"); |
| 5519 | return False; |
| 5520 | # undef INSN |
| 5521 | } |
| 5522 | |
| 5523 | |
| 5524 | /*------------------------------------------------------------*/ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5525 | /*--- SIMD and FP instructions: helper functions ---*/ |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5526 | /*------------------------------------------------------------*/ |
| 5527 | |
sewardj | d96daf6 | 2014-06-15 08:17:35 +0000 | [diff] [blame] | 5528 | /* Some constructors for interleave/deinterleave expressions. */ |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 5529 | |
sewardj | d96daf6 | 2014-06-15 08:17:35 +0000 | [diff] [blame] | 5530 | static IRExpr* mk_CatEvenLanes64x2 ( IRTemp a10, IRTemp b10 ) { |
| 5531 | // returns a0 b0 |
| 5532 | return binop(Iop_InterleaveLO64x2, mkexpr(a10), mkexpr(b10)); |
| 5533 | } |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 5534 | |
sewardj | d96daf6 | 2014-06-15 08:17:35 +0000 | [diff] [blame] | 5535 | static IRExpr* mk_CatOddLanes64x2 ( IRTemp a10, IRTemp b10 ) { |
| 5536 | // returns a1 b1 |
| 5537 | return binop(Iop_InterleaveHI64x2, mkexpr(a10), mkexpr(b10)); |
| 5538 | } |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 5539 | |
sewardj | d96daf6 | 2014-06-15 08:17:35 +0000 | [diff] [blame] | 5540 | static IRExpr* mk_CatEvenLanes32x4 ( IRTemp a3210, IRTemp b3210 ) { |
| 5541 | // returns a2 a0 b2 b0 |
| 5542 | return binop(Iop_CatEvenLanes32x4, mkexpr(a3210), mkexpr(b3210)); |
| 5543 | } |
| 5544 | |
| 5545 | static IRExpr* mk_CatOddLanes32x4 ( IRTemp a3210, IRTemp b3210 ) { |
| 5546 | // returns a3 a1 b3 b1 |
| 5547 | return binop(Iop_CatOddLanes32x4, mkexpr(a3210), mkexpr(b3210)); |
| 5548 | } |
| 5549 | |
| 5550 | static IRExpr* mk_InterleaveLO32x4 ( IRTemp a3210, IRTemp b3210 ) { |
| 5551 | // returns a1 b1 a0 b0 |
| 5552 | return binop(Iop_InterleaveLO32x4, mkexpr(a3210), mkexpr(b3210)); |
| 5553 | } |
| 5554 | |
| 5555 | static IRExpr* mk_InterleaveHI32x4 ( IRTemp a3210, IRTemp b3210 ) { |
| 5556 | // returns a3 b3 a2 b2 |
| 5557 | return binop(Iop_InterleaveHI32x4, mkexpr(a3210), mkexpr(b3210)); |
| 5558 | } |
| 5559 | |
| 5560 | static IRExpr* mk_CatEvenLanes16x8 ( IRTemp a76543210, IRTemp b76543210 ) { |
| 5561 | // returns a6 a4 a2 a0 b6 b4 b2 b0 |
| 5562 | return binop(Iop_CatEvenLanes16x8, mkexpr(a76543210), mkexpr(b76543210)); |
| 5563 | } |
| 5564 | |
| 5565 | static IRExpr* mk_CatOddLanes16x8 ( IRTemp a76543210, IRTemp b76543210 ) { |
| 5566 | // returns a7 a5 a3 a1 b7 b5 b3 b1 |
| 5567 | return binop(Iop_CatOddLanes16x8, mkexpr(a76543210), mkexpr(b76543210)); |
| 5568 | } |
| 5569 | |
| 5570 | static IRExpr* mk_InterleaveLO16x8 ( IRTemp a76543210, IRTemp b76543210 ) { |
| 5571 | // returns a3 b3 a2 b2 a1 b1 a0 b0 |
| 5572 | return binop(Iop_InterleaveLO16x8, mkexpr(a76543210), mkexpr(b76543210)); |
| 5573 | } |
| 5574 | |
| 5575 | static IRExpr* mk_InterleaveHI16x8 ( IRTemp a76543210, IRTemp b76543210 ) { |
| 5576 | // returns a7 b7 a6 b6 a5 b5 a4 b4 |
| 5577 | return binop(Iop_InterleaveHI16x8, mkexpr(a76543210), mkexpr(b76543210)); |
| 5578 | } |
| 5579 | |
| 5580 | static IRExpr* mk_CatEvenLanes8x16 ( IRTemp aFEDCBA9876543210, |
| 5581 | IRTemp bFEDCBA9876543210 ) { |
| 5582 | // returns aE aC aA a8 a6 a4 a2 a0 bE bC bA b8 b6 b4 b2 b0 |
| 5583 | return binop(Iop_CatEvenLanes8x16, mkexpr(aFEDCBA9876543210), |
| 5584 | mkexpr(bFEDCBA9876543210)); |
| 5585 | } |
| 5586 | |
| 5587 | static IRExpr* mk_CatOddLanes8x16 ( IRTemp aFEDCBA9876543210, |
| 5588 | IRTemp bFEDCBA9876543210 ) { |
| 5589 | // returns aF aD aB a9 a7 a5 a3 a1 bF bD bB b9 b7 b5 b3 b1 |
| 5590 | return binop(Iop_CatOddLanes8x16, mkexpr(aFEDCBA9876543210), |
| 5591 | mkexpr(bFEDCBA9876543210)); |
| 5592 | } |
| 5593 | |
| 5594 | static IRExpr* mk_InterleaveLO8x16 ( IRTemp aFEDCBA9876543210, |
| 5595 | IRTemp bFEDCBA9876543210 ) { |
| 5596 | // returns a7 b7 a6 b6 a5 b5 a4 b4 a3 b3 a2 b2 a1 b1 a0 b0 |
| 5597 | return binop(Iop_InterleaveLO8x16, mkexpr(aFEDCBA9876543210), |
| 5598 | mkexpr(bFEDCBA9876543210)); |
| 5599 | } |
| 5600 | |
| 5601 | static IRExpr* mk_InterleaveHI8x16 ( IRTemp aFEDCBA9876543210, |
| 5602 | IRTemp bFEDCBA9876543210 ) { |
| 5603 | // returns aF bF aE bE aD bD aC bC aB bB aA bA a9 b9 a8 b8 |
| 5604 | return binop(Iop_InterleaveHI8x16, mkexpr(aFEDCBA9876543210), |
| 5605 | mkexpr(bFEDCBA9876543210)); |
| 5606 | } |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5607 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5608 | /* Generate N copies of |bit| in the bottom of a ULong. */ |
| 5609 | static ULong Replicate ( ULong bit, Int N ) |
| 5610 | { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5611 | vassert(bit <= 1 && N >= 1 && N < 64); |
| 5612 | if (bit == 0) { |
| 5613 | return 0; |
| 5614 | } else { |
| 5615 | /* Careful. This won't work for N == 64. */ |
| 5616 | return (1ULL << N) - 1; |
| 5617 | } |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5618 | } |
| 5619 | |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5620 | static ULong Replicate32x2 ( ULong bits32 ) |
| 5621 | { |
| 5622 | vassert(0 == (bits32 & ~0xFFFFFFFFULL)); |
| 5623 | return (bits32 << 32) | bits32; |
| 5624 | } |
| 5625 | |
| 5626 | static ULong Replicate16x4 ( ULong bits16 ) |
| 5627 | { |
| 5628 | vassert(0 == (bits16 & ~0xFFFFULL)); |
| 5629 | return Replicate32x2((bits16 << 16) | bits16); |
| 5630 | } |
| 5631 | |
| 5632 | static ULong Replicate8x8 ( ULong bits8 ) |
| 5633 | { |
| 5634 | vassert(0 == (bits8 & ~0xFFULL)); |
| 5635 | return Replicate16x4((bits8 << 8) | bits8); |
| 5636 | } |
| 5637 | |
| 5638 | /* Expand the VFPExpandImm-style encoding in the bottom 8 bits of |
| 5639 | |imm8| to either a 32-bit value if N is 32 or a 64 bit value if N |
| 5640 | is 64. In the former case, the upper 32 bits of the returned value |
| 5641 | are guaranteed to be zero. */ |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5642 | static ULong VFPExpandImm ( ULong imm8, Int N ) |
| 5643 | { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5644 | vassert(imm8 <= 0xFF); |
| 5645 | vassert(N == 32 || N == 64); |
| 5646 | Int E = ((N == 32) ? 8 : 11) - 2; // The spec incorrectly omits the -2. |
| 5647 | Int F = N - E - 1; |
| 5648 | ULong imm8_6 = (imm8 >> 6) & 1; |
| 5649 | /* sign: 1 bit */ |
| 5650 | /* exp: E bits */ |
| 5651 | /* frac: F bits */ |
| 5652 | ULong sign = (imm8 >> 7) & 1; |
| 5653 | ULong exp = ((imm8_6 ^ 1) << (E-1)) | Replicate(imm8_6, E-1); |
| 5654 | ULong frac = ((imm8 & 63) << (F-6)) | Replicate(0, F-6); |
| 5655 | vassert(sign < (1ULL << 1)); |
| 5656 | vassert(exp < (1ULL << E)); |
| 5657 | vassert(frac < (1ULL << F)); |
| 5658 | vassert(1 + E + F == N); |
| 5659 | ULong res = (sign << (E+F)) | (exp << F) | frac; |
| 5660 | return res; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5661 | } |
| 5662 | |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5663 | /* Expand an AdvSIMDExpandImm-style encoding into a 64-bit value. |
| 5664 | This might fail, as indicated by the returned Bool. Page 2530 of |
| 5665 | the manual. */ |
| 5666 | static Bool AdvSIMDExpandImm ( /*OUT*/ULong* res, |
| 5667 | UInt op, UInt cmode, UInt imm8 ) |
| 5668 | { |
| 5669 | vassert(op <= 1); |
| 5670 | vassert(cmode <= 15); |
| 5671 | vassert(imm8 <= 255); |
| 5672 | |
| 5673 | *res = 0; /* will overwrite iff returning True */ |
| 5674 | |
| 5675 | ULong imm64 = 0; |
| 5676 | Bool testimm8 = False; |
| 5677 | |
| 5678 | switch (cmode >> 1) { |
| 5679 | case 0: |
| 5680 | testimm8 = False; imm64 = Replicate32x2(imm8); break; |
| 5681 | case 1: |
| 5682 | testimm8 = True; imm64 = Replicate32x2(imm8 << 8); break; |
| 5683 | case 2: |
| 5684 | testimm8 = True; imm64 = Replicate32x2(imm8 << 16); break; |
| 5685 | case 3: |
| 5686 | testimm8 = True; imm64 = Replicate32x2(imm8 << 24); break; |
| 5687 | case 4: |
| 5688 | testimm8 = False; imm64 = Replicate16x4(imm8); break; |
| 5689 | case 5: |
| 5690 | testimm8 = True; imm64 = Replicate16x4(imm8 << 8); break; |
| 5691 | case 6: |
| 5692 | testimm8 = True; |
| 5693 | if ((cmode & 1) == 0) |
| 5694 | imm64 = Replicate32x2((imm8 << 8) | 0xFF); |
| 5695 | else |
| 5696 | imm64 = Replicate32x2((imm8 << 16) | 0xFFFF); |
| 5697 | break; |
| 5698 | case 7: |
| 5699 | testimm8 = False; |
| 5700 | if ((cmode & 1) == 0 && op == 0) |
| 5701 | imm64 = Replicate8x8(imm8); |
| 5702 | if ((cmode & 1) == 0 && op == 1) { |
| 5703 | imm64 = 0; imm64 |= (imm8 & 0x80) ? 0xFF : 0x00; |
| 5704 | imm64 <<= 8; imm64 |= (imm8 & 0x40) ? 0xFF : 0x00; |
| 5705 | imm64 <<= 8; imm64 |= (imm8 & 0x20) ? 0xFF : 0x00; |
| 5706 | imm64 <<= 8; imm64 |= (imm8 & 0x10) ? 0xFF : 0x00; |
| 5707 | imm64 <<= 8; imm64 |= (imm8 & 0x08) ? 0xFF : 0x00; |
| 5708 | imm64 <<= 8; imm64 |= (imm8 & 0x04) ? 0xFF : 0x00; |
| 5709 | imm64 <<= 8; imm64 |= (imm8 & 0x02) ? 0xFF : 0x00; |
| 5710 | imm64 <<= 8; imm64 |= (imm8 & 0x01) ? 0xFF : 0x00; |
| 5711 | } |
| 5712 | if ((cmode & 1) == 1 && op == 0) { |
| 5713 | ULong imm8_7 = (imm8 >> 7) & 1; |
| 5714 | ULong imm8_6 = (imm8 >> 6) & 1; |
| 5715 | ULong imm8_50 = imm8 & 63; |
| 5716 | ULong imm32 = (imm8_7 << (1 + 5 + 6 + 19)) |
| 5717 | | ((imm8_6 ^ 1) << (5 + 6 + 19)) |
| 5718 | | (Replicate(imm8_6, 5) << (6 + 19)) |
| 5719 | | (imm8_50 << 19); |
| 5720 | imm64 = Replicate32x2(imm32); |
| 5721 | } |
| 5722 | if ((cmode & 1) == 1 && op == 1) { |
| 5723 | // imm64 = imm8<7>:NOT(imm8<6>) |
| 5724 | // :Replicate(imm8<6>,8):imm8<5:0>:Zeros(48); |
| 5725 | ULong imm8_7 = (imm8 >> 7) & 1; |
| 5726 | ULong imm8_6 = (imm8 >> 6) & 1; |
| 5727 | ULong imm8_50 = imm8 & 63; |
| 5728 | imm64 = (imm8_7 << 63) | ((imm8_6 ^ 1) << 62) |
| 5729 | | (Replicate(imm8_6, 8) << 54) |
| 5730 | | (imm8_50 << 48); |
| 5731 | } |
| 5732 | break; |
| 5733 | default: |
| 5734 | vassert(0); |
| 5735 | } |
| 5736 | |
| 5737 | if (testimm8 && imm8 == 0) |
| 5738 | return False; |
| 5739 | |
| 5740 | *res = imm64; |
| 5741 | return True; |
| 5742 | } |
| 5743 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5744 | /* Help a bit for decoding laneage for vector operations that can be |
| 5745 | of the form 4x32, 2x64 or 2x32-and-zero-upper-half, as encoded by Q |
| 5746 | and SZ bits, typically for vector floating point. */ |
| 5747 | static Bool getLaneInfo_Q_SZ ( /*OUT*/IRType* tyI, /*OUT*/IRType* tyF, |
| 5748 | /*OUT*/UInt* nLanes, /*OUT*/Bool* zeroUpper, |
| 5749 | /*OUT*/const HChar** arrSpec, |
| 5750 | Bool bitQ, Bool bitSZ ) |
| 5751 | { |
| 5752 | vassert(bitQ == True || bitQ == False); |
| 5753 | vassert(bitSZ == True || bitSZ == False); |
| 5754 | if (bitQ && bitSZ) { // 2x64 |
| 5755 | if (tyI) *tyI = Ity_I64; |
| 5756 | if (tyF) *tyF = Ity_F64; |
| 5757 | if (nLanes) *nLanes = 2; |
| 5758 | if (zeroUpper) *zeroUpper = False; |
| 5759 | if (arrSpec) *arrSpec = "2d"; |
| 5760 | return True; |
| 5761 | } |
| 5762 | if (bitQ && !bitSZ) { // 4x32 |
| 5763 | if (tyI) *tyI = Ity_I32; |
| 5764 | if (tyF) *tyF = Ity_F32; |
| 5765 | if (nLanes) *nLanes = 4; |
| 5766 | if (zeroUpper) *zeroUpper = False; |
| 5767 | if (arrSpec) *arrSpec = "4s"; |
| 5768 | return True; |
| 5769 | } |
| 5770 | if (!bitQ && !bitSZ) { // 2x32 |
| 5771 | if (tyI) *tyI = Ity_I32; |
| 5772 | if (tyF) *tyF = Ity_F32; |
| 5773 | if (nLanes) *nLanes = 2; |
| 5774 | if (zeroUpper) *zeroUpper = True; |
| 5775 | if (arrSpec) *arrSpec = "2s"; |
| 5776 | return True; |
| 5777 | } |
| 5778 | // Else impliedly 1x64, which isn't allowed. |
| 5779 | return False; |
| 5780 | } |
| 5781 | |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 5782 | /* Helper for decoding laneage for shift-style vector operations |
| 5783 | that involve an immediate shift amount. */ |
| 5784 | static Bool getLaneInfo_IMMH_IMMB ( /*OUT*/UInt* shift, /*OUT*/UInt* szBlg2, |
| 5785 | UInt immh, UInt immb ) |
| 5786 | { |
| 5787 | vassert(immh < (1<<4)); |
| 5788 | vassert(immb < (1<<3)); |
| 5789 | UInt immhb = (immh << 3) | immb; |
| 5790 | if (immh & 8) { |
| 5791 | if (shift) *shift = 128 - immhb; |
| 5792 | if (szBlg2) *szBlg2 = 3; |
| 5793 | return True; |
| 5794 | } |
| 5795 | if (immh & 4) { |
| 5796 | if (shift) *shift = 64 - immhb; |
| 5797 | if (szBlg2) *szBlg2 = 2; |
| 5798 | return True; |
| 5799 | } |
| 5800 | if (immh & 2) { |
| 5801 | if (shift) *shift = 32 - immhb; |
| 5802 | if (szBlg2) *szBlg2 = 1; |
| 5803 | return True; |
| 5804 | } |
| 5805 | if (immh & 1) { |
| 5806 | if (shift) *shift = 16 - immhb; |
| 5807 | if (szBlg2) *szBlg2 = 0; |
| 5808 | return True; |
| 5809 | } |
| 5810 | return False; |
| 5811 | } |
| 5812 | |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5813 | /* Generate IR to fold all lanes of the V128 value in 'src' as |
| 5814 | characterised by the operator 'op', and return the result in the |
| 5815 | bottom bits of a V128, with all other bits set to zero. */ |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 5816 | static IRTemp math_FOLDV ( IRTemp src, IROp op ) |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5817 | { |
| 5818 | /* The basic idea is to use repeated applications of Iop_CatEven* |
| 5819 | and Iop_CatOdd* operators to 'src' so as to clone each lane into |
| 5820 | a complete vector. Then fold all those vectors with 'op' and |
| 5821 | zero out all but the least significant lane. */ |
| 5822 | switch (op) { |
| 5823 | case Iop_Min8Sx16: case Iop_Min8Ux16: |
sewardj | b9aff1e | 2014-06-15 21:55:33 +0000 | [diff] [blame] | 5824 | case Iop_Max8Sx16: case Iop_Max8Ux16: case Iop_Add8x16: { |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5825 | /* NB: temp naming here is misleading -- the naming is for 8 |
| 5826 | lanes of 16 bit, whereas what is being operated on is 16 |
| 5827 | lanes of 8 bits. */ |
| 5828 | IRTemp x76543210 = src; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5829 | IRTemp x76547654 = newTempV128(); |
| 5830 | IRTemp x32103210 = newTempV128(); |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5831 | assign(x76547654, mk_CatOddLanes64x2 (x76543210, x76543210)); |
| 5832 | assign(x32103210, mk_CatEvenLanes64x2(x76543210, x76543210)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5833 | IRTemp x76767676 = newTempV128(); |
| 5834 | IRTemp x54545454 = newTempV128(); |
| 5835 | IRTemp x32323232 = newTempV128(); |
| 5836 | IRTemp x10101010 = newTempV128(); |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5837 | assign(x76767676, mk_CatOddLanes32x4 (x76547654, x76547654)); |
| 5838 | assign(x54545454, mk_CatEvenLanes32x4(x76547654, x76547654)); |
| 5839 | assign(x32323232, mk_CatOddLanes32x4 (x32103210, x32103210)); |
| 5840 | assign(x10101010, mk_CatEvenLanes32x4(x32103210, x32103210)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5841 | IRTemp x77777777 = newTempV128(); |
| 5842 | IRTemp x66666666 = newTempV128(); |
| 5843 | IRTemp x55555555 = newTempV128(); |
| 5844 | IRTemp x44444444 = newTempV128(); |
| 5845 | IRTemp x33333333 = newTempV128(); |
| 5846 | IRTemp x22222222 = newTempV128(); |
| 5847 | IRTemp x11111111 = newTempV128(); |
| 5848 | IRTemp x00000000 = newTempV128(); |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5849 | assign(x77777777, mk_CatOddLanes16x8 (x76767676, x76767676)); |
| 5850 | assign(x66666666, mk_CatEvenLanes16x8(x76767676, x76767676)); |
| 5851 | assign(x55555555, mk_CatOddLanes16x8 (x54545454, x54545454)); |
| 5852 | assign(x44444444, mk_CatEvenLanes16x8(x54545454, x54545454)); |
| 5853 | assign(x33333333, mk_CatOddLanes16x8 (x32323232, x32323232)); |
| 5854 | assign(x22222222, mk_CatEvenLanes16x8(x32323232, x32323232)); |
| 5855 | assign(x11111111, mk_CatOddLanes16x8 (x10101010, x10101010)); |
| 5856 | assign(x00000000, mk_CatEvenLanes16x8(x10101010, x10101010)); |
| 5857 | /* Naming not misleading after here. */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5858 | IRTemp xAllF = newTempV128(); |
| 5859 | IRTemp xAllE = newTempV128(); |
| 5860 | IRTemp xAllD = newTempV128(); |
| 5861 | IRTemp xAllC = newTempV128(); |
| 5862 | IRTemp xAllB = newTempV128(); |
| 5863 | IRTemp xAllA = newTempV128(); |
| 5864 | IRTemp xAll9 = newTempV128(); |
| 5865 | IRTemp xAll8 = newTempV128(); |
| 5866 | IRTemp xAll7 = newTempV128(); |
| 5867 | IRTemp xAll6 = newTempV128(); |
| 5868 | IRTemp xAll5 = newTempV128(); |
| 5869 | IRTemp xAll4 = newTempV128(); |
| 5870 | IRTemp xAll3 = newTempV128(); |
| 5871 | IRTemp xAll2 = newTempV128(); |
| 5872 | IRTemp xAll1 = newTempV128(); |
| 5873 | IRTemp xAll0 = newTempV128(); |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5874 | assign(xAllF, mk_CatOddLanes8x16 (x77777777, x77777777)); |
| 5875 | assign(xAllE, mk_CatEvenLanes8x16(x77777777, x77777777)); |
| 5876 | assign(xAllD, mk_CatOddLanes8x16 (x66666666, x66666666)); |
| 5877 | assign(xAllC, mk_CatEvenLanes8x16(x66666666, x66666666)); |
| 5878 | assign(xAllB, mk_CatOddLanes8x16 (x55555555, x55555555)); |
| 5879 | assign(xAllA, mk_CatEvenLanes8x16(x55555555, x55555555)); |
| 5880 | assign(xAll9, mk_CatOddLanes8x16 (x44444444, x44444444)); |
| 5881 | assign(xAll8, mk_CatEvenLanes8x16(x44444444, x44444444)); |
| 5882 | assign(xAll7, mk_CatOddLanes8x16 (x33333333, x33333333)); |
| 5883 | assign(xAll6, mk_CatEvenLanes8x16(x33333333, x33333333)); |
| 5884 | assign(xAll5, mk_CatOddLanes8x16 (x22222222, x22222222)); |
| 5885 | assign(xAll4, mk_CatEvenLanes8x16(x22222222, x22222222)); |
| 5886 | assign(xAll3, mk_CatOddLanes8x16 (x11111111, x11111111)); |
| 5887 | assign(xAll2, mk_CatEvenLanes8x16(x11111111, x11111111)); |
| 5888 | assign(xAll1, mk_CatOddLanes8x16 (x00000000, x00000000)); |
| 5889 | assign(xAll0, mk_CatEvenLanes8x16(x00000000, x00000000)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5890 | IRTemp maxFE = newTempV128(); |
| 5891 | IRTemp maxDC = newTempV128(); |
| 5892 | IRTemp maxBA = newTempV128(); |
| 5893 | IRTemp max98 = newTempV128(); |
| 5894 | IRTemp max76 = newTempV128(); |
| 5895 | IRTemp max54 = newTempV128(); |
| 5896 | IRTemp max32 = newTempV128(); |
| 5897 | IRTemp max10 = newTempV128(); |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5898 | assign(maxFE, binop(op, mkexpr(xAllF), mkexpr(xAllE))); |
| 5899 | assign(maxDC, binop(op, mkexpr(xAllD), mkexpr(xAllC))); |
| 5900 | assign(maxBA, binop(op, mkexpr(xAllB), mkexpr(xAllA))); |
| 5901 | assign(max98, binop(op, mkexpr(xAll9), mkexpr(xAll8))); |
| 5902 | assign(max76, binop(op, mkexpr(xAll7), mkexpr(xAll6))); |
| 5903 | assign(max54, binop(op, mkexpr(xAll5), mkexpr(xAll4))); |
| 5904 | assign(max32, binop(op, mkexpr(xAll3), mkexpr(xAll2))); |
| 5905 | assign(max10, binop(op, mkexpr(xAll1), mkexpr(xAll0))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5906 | IRTemp maxFEDC = newTempV128(); |
| 5907 | IRTemp maxBA98 = newTempV128(); |
| 5908 | IRTemp max7654 = newTempV128(); |
| 5909 | IRTemp max3210 = newTempV128(); |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5910 | assign(maxFEDC, binop(op, mkexpr(maxFE), mkexpr(maxDC))); |
| 5911 | assign(maxBA98, binop(op, mkexpr(maxBA), mkexpr(max98))); |
| 5912 | assign(max7654, binop(op, mkexpr(max76), mkexpr(max54))); |
| 5913 | assign(max3210, binop(op, mkexpr(max32), mkexpr(max10))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5914 | IRTemp maxFEDCBA98 = newTempV128(); |
| 5915 | IRTemp max76543210 = newTempV128(); |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5916 | assign(maxFEDCBA98, binop(op, mkexpr(maxFEDC), mkexpr(maxBA98))); |
| 5917 | assign(max76543210, binop(op, mkexpr(max7654), mkexpr(max3210))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5918 | IRTemp maxAllLanes = newTempV128(); |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5919 | assign(maxAllLanes, binop(op, mkexpr(maxFEDCBA98), |
| 5920 | mkexpr(max76543210))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5921 | IRTemp res = newTempV128(); |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5922 | assign(res, unop(Iop_ZeroHI120ofV128, mkexpr(maxAllLanes))); |
| 5923 | return res; |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5924 | } |
| 5925 | case Iop_Min16Sx8: case Iop_Min16Ux8: |
sewardj | b9aff1e | 2014-06-15 21:55:33 +0000 | [diff] [blame] | 5926 | case Iop_Max16Sx8: case Iop_Max16Ux8: case Iop_Add16x8: { |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5927 | IRTemp x76543210 = src; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5928 | IRTemp x76547654 = newTempV128(); |
| 5929 | IRTemp x32103210 = newTempV128(); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5930 | assign(x76547654, mk_CatOddLanes64x2 (x76543210, x76543210)); |
| 5931 | assign(x32103210, mk_CatEvenLanes64x2(x76543210, x76543210)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5932 | IRTemp x76767676 = newTempV128(); |
| 5933 | IRTemp x54545454 = newTempV128(); |
| 5934 | IRTemp x32323232 = newTempV128(); |
| 5935 | IRTemp x10101010 = newTempV128(); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5936 | assign(x76767676, mk_CatOddLanes32x4 (x76547654, x76547654)); |
| 5937 | assign(x54545454, mk_CatEvenLanes32x4(x76547654, x76547654)); |
| 5938 | assign(x32323232, mk_CatOddLanes32x4 (x32103210, x32103210)); |
| 5939 | assign(x10101010, mk_CatEvenLanes32x4(x32103210, x32103210)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5940 | IRTemp x77777777 = newTempV128(); |
| 5941 | IRTemp x66666666 = newTempV128(); |
| 5942 | IRTemp x55555555 = newTempV128(); |
| 5943 | IRTemp x44444444 = newTempV128(); |
| 5944 | IRTemp x33333333 = newTempV128(); |
| 5945 | IRTemp x22222222 = newTempV128(); |
| 5946 | IRTemp x11111111 = newTempV128(); |
| 5947 | IRTemp x00000000 = newTempV128(); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5948 | assign(x77777777, mk_CatOddLanes16x8 (x76767676, x76767676)); |
| 5949 | assign(x66666666, mk_CatEvenLanes16x8(x76767676, x76767676)); |
| 5950 | assign(x55555555, mk_CatOddLanes16x8 (x54545454, x54545454)); |
| 5951 | assign(x44444444, mk_CatEvenLanes16x8(x54545454, x54545454)); |
| 5952 | assign(x33333333, mk_CatOddLanes16x8 (x32323232, x32323232)); |
| 5953 | assign(x22222222, mk_CatEvenLanes16x8(x32323232, x32323232)); |
| 5954 | assign(x11111111, mk_CatOddLanes16x8 (x10101010, x10101010)); |
| 5955 | assign(x00000000, mk_CatEvenLanes16x8(x10101010, x10101010)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5956 | IRTemp max76 = newTempV128(); |
| 5957 | IRTemp max54 = newTempV128(); |
| 5958 | IRTemp max32 = newTempV128(); |
| 5959 | IRTemp max10 = newTempV128(); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5960 | assign(max76, binop(op, mkexpr(x77777777), mkexpr(x66666666))); |
| 5961 | assign(max54, binop(op, mkexpr(x55555555), mkexpr(x44444444))); |
| 5962 | assign(max32, binop(op, mkexpr(x33333333), mkexpr(x22222222))); |
| 5963 | assign(max10, binop(op, mkexpr(x11111111), mkexpr(x00000000))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5964 | IRTemp max7654 = newTempV128(); |
| 5965 | IRTemp max3210 = newTempV128(); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5966 | assign(max7654, binop(op, mkexpr(max76), mkexpr(max54))); |
| 5967 | assign(max3210, binop(op, mkexpr(max32), mkexpr(max10))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5968 | IRTemp max76543210 = newTempV128(); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5969 | assign(max76543210, binop(op, mkexpr(max7654), mkexpr(max3210))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5970 | IRTemp res = newTempV128(); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5971 | assign(res, unop(Iop_ZeroHI112ofV128, mkexpr(max76543210))); |
| 5972 | return res; |
| 5973 | } |
| 5974 | case Iop_Min32Sx4: case Iop_Min32Ux4: |
sewardj | b9aff1e | 2014-06-15 21:55:33 +0000 | [diff] [blame] | 5975 | case Iop_Max32Sx4: case Iop_Max32Ux4: case Iop_Add32x4: { |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5976 | IRTemp x3210 = src; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5977 | IRTemp x3232 = newTempV128(); |
| 5978 | IRTemp x1010 = newTempV128(); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5979 | assign(x3232, mk_CatOddLanes64x2 (x3210, x3210)); |
| 5980 | assign(x1010, mk_CatEvenLanes64x2(x3210, x3210)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5981 | IRTemp x3333 = newTempV128(); |
| 5982 | IRTemp x2222 = newTempV128(); |
| 5983 | IRTemp x1111 = newTempV128(); |
| 5984 | IRTemp x0000 = newTempV128(); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5985 | assign(x3333, mk_CatOddLanes32x4 (x3232, x3232)); |
| 5986 | assign(x2222, mk_CatEvenLanes32x4(x3232, x3232)); |
| 5987 | assign(x1111, mk_CatOddLanes32x4 (x1010, x1010)); |
| 5988 | assign(x0000, mk_CatEvenLanes32x4(x1010, x1010)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5989 | IRTemp max32 = newTempV128(); |
| 5990 | IRTemp max10 = newTempV128(); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5991 | assign(max32, binop(op, mkexpr(x3333), mkexpr(x2222))); |
| 5992 | assign(max10, binop(op, mkexpr(x1111), mkexpr(x0000))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5993 | IRTemp max3210 = newTempV128(); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5994 | assign(max3210, binop(op, mkexpr(max32), mkexpr(max10))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 5995 | IRTemp res = newTempV128(); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5996 | assign(res, unop(Iop_ZeroHI96ofV128, mkexpr(max3210))); |
| 5997 | return res; |
| 5998 | } |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 5999 | case Iop_Add64x2: { |
| 6000 | IRTemp x10 = src; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6001 | IRTemp x00 = newTempV128(); |
| 6002 | IRTemp x11 = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 6003 | assign(x11, binop(Iop_InterleaveHI64x2, mkexpr(x10), mkexpr(x10))); |
| 6004 | assign(x00, binop(Iop_InterleaveLO64x2, mkexpr(x10), mkexpr(x10))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6005 | IRTemp max10 = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 6006 | assign(max10, binop(op, mkexpr(x11), mkexpr(x00))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6007 | IRTemp res = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 6008 | assign(res, unop(Iop_ZeroHI64ofV128, mkexpr(max10))); |
| 6009 | return res; |
| 6010 | } |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 6011 | default: |
| 6012 | vassert(0); |
| 6013 | } |
| 6014 | } |
| 6015 | |
| 6016 | |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6017 | /* Generate IR for TBL and TBX. This deals with the 128 bit case |
| 6018 | only. */ |
| 6019 | static IRTemp math_TBL_TBX ( IRTemp tab[4], UInt len, IRTemp src, |
| 6020 | IRTemp oor_values ) |
| 6021 | { |
| 6022 | vassert(len >= 0 && len <= 3); |
| 6023 | |
| 6024 | /* Generate some useful constants as concisely as possible. */ |
| 6025 | IRTemp half15 = newTemp(Ity_I64); |
| 6026 | assign(half15, mkU64(0x0F0F0F0F0F0F0F0FULL)); |
| 6027 | IRTemp half16 = newTemp(Ity_I64); |
| 6028 | assign(half16, mkU64(0x1010101010101010ULL)); |
| 6029 | |
| 6030 | /* A zero vector */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6031 | IRTemp allZero = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6032 | assign(allZero, mkV128(0x0000)); |
| 6033 | /* A vector containing 15 in each 8-bit lane */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6034 | IRTemp all15 = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6035 | assign(all15, binop(Iop_64HLtoV128, mkexpr(half15), mkexpr(half15))); |
| 6036 | /* A vector containing 16 in each 8-bit lane */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6037 | IRTemp all16 = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6038 | assign(all16, binop(Iop_64HLtoV128, mkexpr(half16), mkexpr(half16))); |
| 6039 | /* A vector containing 32 in each 8-bit lane */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6040 | IRTemp all32 = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6041 | assign(all32, binop(Iop_Add8x16, mkexpr(all16), mkexpr(all16))); |
| 6042 | /* A vector containing 48 in each 8-bit lane */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6043 | IRTemp all48 = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6044 | assign(all48, binop(Iop_Add8x16, mkexpr(all16), mkexpr(all32))); |
| 6045 | /* A vector containing 64 in each 8-bit lane */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6046 | IRTemp all64 = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6047 | assign(all64, binop(Iop_Add8x16, mkexpr(all32), mkexpr(all32))); |
| 6048 | |
| 6049 | /* Group the 16/32/48/64 vectors so as to be indexable. */ |
| 6050 | IRTemp allXX[4] = { all16, all32, all48, all64 }; |
| 6051 | |
| 6052 | /* Compute the result for each table vector, with zeroes in places |
| 6053 | where the index values are out of range, and OR them into the |
| 6054 | running vector. */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6055 | IRTemp running_result = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6056 | assign(running_result, mkV128(0)); |
| 6057 | |
| 6058 | UInt tabent; |
| 6059 | for (tabent = 0; tabent <= len; tabent++) { |
| 6060 | vassert(tabent >= 0 && tabent < 4); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6061 | IRTemp bias = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6062 | assign(bias, |
| 6063 | mkexpr(tabent == 0 ? allZero : allXX[tabent-1])); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6064 | IRTemp biased_indices = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6065 | assign(biased_indices, |
| 6066 | binop(Iop_Sub8x16, mkexpr(src), mkexpr(bias))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6067 | IRTemp valid_mask = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6068 | assign(valid_mask, |
| 6069 | binop(Iop_CmpGT8Ux16, mkexpr(all16), mkexpr(biased_indices))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6070 | IRTemp safe_biased_indices = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6071 | assign(safe_biased_indices, |
| 6072 | binop(Iop_AndV128, mkexpr(biased_indices), mkexpr(all15))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6073 | IRTemp results_or_junk = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6074 | assign(results_or_junk, |
| 6075 | binop(Iop_Perm8x16, mkexpr(tab[tabent]), |
| 6076 | mkexpr(safe_biased_indices))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6077 | IRTemp results_or_zero = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6078 | assign(results_or_zero, |
| 6079 | binop(Iop_AndV128, mkexpr(results_or_junk), mkexpr(valid_mask))); |
| 6080 | /* And OR that into the running result. */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6081 | IRTemp tmp = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6082 | assign(tmp, binop(Iop_OrV128, mkexpr(results_or_zero), |
| 6083 | mkexpr(running_result))); |
| 6084 | running_result = tmp; |
| 6085 | } |
| 6086 | |
| 6087 | /* So now running_result holds the overall result where the indices |
| 6088 | are in range, and zero in out-of-range lanes. Now we need to |
| 6089 | compute an overall validity mask and use this to copy in the |
| 6090 | lanes in the oor_values for out of range indices. This is |
| 6091 | unnecessary for TBL but will get folded out by iropt, so we lean |
| 6092 | on that and generate the same code for TBL and TBX here. */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6093 | IRTemp overall_valid_mask = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6094 | assign(overall_valid_mask, |
| 6095 | binop(Iop_CmpGT8Ux16, mkexpr(allXX[len]), mkexpr(src))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6096 | IRTemp result = newTempV128(); |
sewardj | 92d0ae3 | 2014-04-03 13:48:54 +0000 | [diff] [blame] | 6097 | assign(result, |
| 6098 | binop(Iop_OrV128, |
| 6099 | mkexpr(running_result), |
| 6100 | binop(Iop_AndV128, |
| 6101 | mkexpr(oor_values), |
| 6102 | unop(Iop_NotV128, mkexpr(overall_valid_mask))))); |
| 6103 | return result; |
| 6104 | } |
| 6105 | |
| 6106 | |
sewardj | 31b5a95 | 2014-06-26 07:41:14 +0000 | [diff] [blame] | 6107 | /* Let |argL| and |argR| be V128 values, and let |opI64x2toV128| be |
| 6108 | an op which takes two I64s and produces a V128. That is, a widening |
| 6109 | operator. Generate IR which applies |opI64x2toV128| to either the |
| 6110 | lower (if |is2| is False) or upper (if |is2| is True) halves of |
| 6111 | |argL| and |argR|, and return the value in a new IRTemp. |
| 6112 | */ |
| 6113 | static |
| 6114 | IRTemp math_BINARY_WIDENING_V128 ( Bool is2, IROp opI64x2toV128, |
| 6115 | IRExpr* argL, IRExpr* argR ) |
| 6116 | { |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6117 | IRTemp res = newTempV128(); |
sewardj | 31b5a95 | 2014-06-26 07:41:14 +0000 | [diff] [blame] | 6118 | IROp slice = is2 ? Iop_V128HIto64 : Iop_V128to64; |
| 6119 | assign(res, binop(opI64x2toV128, unop(slice, argL), |
| 6120 | unop(slice, argR))); |
| 6121 | return res; |
| 6122 | } |
| 6123 | |
| 6124 | |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 6125 | /* Generate signed/unsigned absolute difference vector IR. */ |
| 6126 | static |
| 6127 | IRTemp math_ABD ( Bool isU, UInt size, IRExpr* argLE, IRExpr* argRE ) |
| 6128 | { |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 6129 | vassert(size <= 3); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6130 | IRTemp argL = newTempV128(); |
| 6131 | IRTemp argR = newTempV128(); |
| 6132 | IRTemp msk = newTempV128(); |
| 6133 | IRTemp res = newTempV128(); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 6134 | assign(argL, argLE); |
| 6135 | assign(argR, argRE); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6136 | assign(msk, binop(isU ? mkVecCMPGTU(size) : mkVecCMPGTS(size), |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 6137 | mkexpr(argL), mkexpr(argR))); |
| 6138 | assign(res, |
| 6139 | binop(Iop_OrV128, |
| 6140 | binop(Iop_AndV128, |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6141 | binop(mkVecSUB(size), mkexpr(argL), mkexpr(argR)), |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 6142 | mkexpr(msk)), |
| 6143 | binop(Iop_AndV128, |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6144 | binop(mkVecSUB(size), mkexpr(argR), mkexpr(argL)), |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 6145 | unop(Iop_NotV128, mkexpr(msk))))); |
| 6146 | return res; |
| 6147 | } |
| 6148 | |
| 6149 | |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 6150 | /* Generate IR that takes a V128 and sign- or zero-widens |
| 6151 | either the lower or upper set of lanes to twice-as-wide, |
| 6152 | resulting in a new V128 value. */ |
| 6153 | static |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 6154 | IRTemp math_WIDEN_LO_OR_HI_LANES ( Bool zWiden, Bool fromUpperHalf, |
| 6155 | UInt sizeNarrow, IRExpr* srcE ) |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 6156 | { |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6157 | IRTemp src = newTempV128(); |
| 6158 | IRTemp res = newTempV128(); |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 6159 | assign(src, srcE); |
| 6160 | switch (sizeNarrow) { |
| 6161 | case X10: |
| 6162 | assign(res, |
| 6163 | binop(zWiden ? Iop_ShrN64x2 : Iop_SarN64x2, |
| 6164 | binop(fromUpperHalf ? Iop_InterleaveHI32x4 |
| 6165 | : Iop_InterleaveLO32x4, |
| 6166 | mkexpr(src), |
| 6167 | mkexpr(src)), |
| 6168 | mkU8(32))); |
| 6169 | break; |
| 6170 | case X01: |
| 6171 | assign(res, |
| 6172 | binop(zWiden ? Iop_ShrN32x4 : Iop_SarN32x4, |
| 6173 | binop(fromUpperHalf ? Iop_InterleaveHI16x8 |
| 6174 | : Iop_InterleaveLO16x8, |
| 6175 | mkexpr(src), |
| 6176 | mkexpr(src)), |
| 6177 | mkU8(16))); |
| 6178 | break; |
| 6179 | case X00: |
| 6180 | assign(res, |
| 6181 | binop(zWiden ? Iop_ShrN16x8 : Iop_SarN16x8, |
| 6182 | binop(fromUpperHalf ? Iop_InterleaveHI8x16 |
| 6183 | : Iop_InterleaveLO8x16, |
| 6184 | mkexpr(src), |
| 6185 | mkexpr(src)), |
| 6186 | mkU8(8))); |
| 6187 | break; |
| 6188 | default: |
| 6189 | vassert(0); |
| 6190 | } |
| 6191 | return res; |
| 6192 | } |
| 6193 | |
| 6194 | |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 6195 | /* Generate IR that takes a V128 and sign- or zero-widens |
| 6196 | either the even or odd lanes to twice-as-wide, |
| 6197 | resulting in a new V128 value. */ |
| 6198 | static |
| 6199 | IRTemp math_WIDEN_EVEN_OR_ODD_LANES ( Bool zWiden, Bool fromOdd, |
| 6200 | UInt sizeNarrow, IRExpr* srcE ) |
| 6201 | { |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6202 | IRTemp src = newTempV128(); |
| 6203 | IRTemp res = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 6204 | IROp opSAR = mkVecSARN(sizeNarrow+1); |
| 6205 | IROp opSHR = mkVecSHRN(sizeNarrow+1); |
| 6206 | IROp opSHL = mkVecSHLN(sizeNarrow+1); |
| 6207 | IROp opSxR = zWiden ? opSHR : opSAR; |
| 6208 | UInt amt = 0; |
| 6209 | switch (sizeNarrow) { |
| 6210 | case X10: amt = 32; break; |
| 6211 | case X01: amt = 16; break; |
| 6212 | case X00: amt = 8; break; |
| 6213 | default: vassert(0); |
| 6214 | } |
| 6215 | assign(src, srcE); |
| 6216 | if (fromOdd) { |
| 6217 | assign(res, binop(opSxR, mkexpr(src), mkU8(amt))); |
| 6218 | } else { |
| 6219 | assign(res, binop(opSxR, binop(opSHL, mkexpr(src), mkU8(amt)), |
| 6220 | mkU8(amt))); |
| 6221 | } |
| 6222 | return res; |
| 6223 | } |
| 6224 | |
| 6225 | |
| 6226 | /* Generate IR that takes two V128s and narrows (takes lower half) |
| 6227 | of each lane, producing a single V128 value. */ |
| 6228 | static |
| 6229 | IRTemp math_NARROW_LANES ( IRTemp argHi, IRTemp argLo, UInt sizeNarrow ) |
| 6230 | { |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6231 | IRTemp res = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 6232 | assign(res, binop(mkVecCATEVENLANES(sizeNarrow), |
| 6233 | mkexpr(argHi), mkexpr(argLo))); |
| 6234 | return res; |
| 6235 | } |
| 6236 | |
| 6237 | |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 6238 | /* Return a temp which holds the vector dup of the lane of width |
| 6239 | (1 << size) obtained from src[laneNo]. */ |
| 6240 | static |
| 6241 | IRTemp math_DUP_VEC_ELEM ( IRExpr* src, UInt size, UInt laneNo ) |
| 6242 | { |
| 6243 | vassert(size <= 3); |
| 6244 | /* Normalise |laneNo| so it is of the form |
| 6245 | x000 for D, xx00 for S, xxx0 for H, and xxxx for B. |
| 6246 | This puts the bits we want to inspect at constant offsets |
| 6247 | regardless of the value of |size|. |
| 6248 | */ |
| 6249 | UInt ix = laneNo << size; |
| 6250 | vassert(ix <= 15); |
| 6251 | IROp ops[4] = { Iop_INVALID, Iop_INVALID, Iop_INVALID, Iop_INVALID }; |
| 6252 | switch (size) { |
| 6253 | case 0: /* B */ |
| 6254 | ops[0] = (ix & 1) ? Iop_CatOddLanes8x16 : Iop_CatEvenLanes8x16; |
| 6255 | /* fallthrough */ |
| 6256 | case 1: /* H */ |
| 6257 | ops[1] = (ix & 2) ? Iop_CatOddLanes16x8 : Iop_CatEvenLanes16x8; |
| 6258 | /* fallthrough */ |
| 6259 | case 2: /* S */ |
| 6260 | ops[2] = (ix & 4) ? Iop_CatOddLanes32x4 : Iop_CatEvenLanes32x4; |
| 6261 | /* fallthrough */ |
| 6262 | case 3: /* D */ |
| 6263 | ops[3] = (ix & 8) ? Iop_InterleaveHI64x2 : Iop_InterleaveLO64x2; |
| 6264 | break; |
| 6265 | default: |
| 6266 | vassert(0); |
| 6267 | } |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6268 | IRTemp res = newTempV128(); |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 6269 | assign(res, src); |
| 6270 | Int i; |
| 6271 | for (i = 3; i >= 0; i--) { |
| 6272 | if (ops[i] == Iop_INVALID) |
| 6273 | break; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6274 | IRTemp tmp = newTempV128(); |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 6275 | assign(tmp, binop(ops[i], mkexpr(res), mkexpr(res))); |
| 6276 | res = tmp; |
| 6277 | } |
| 6278 | return res; |
| 6279 | } |
| 6280 | |
| 6281 | |
| 6282 | /* Let |srcV| be a V128 value, and let |imm5| be a lane-and-size |
| 6283 | selector encoded as shown below. Return a new V128 holding the |
| 6284 | selected lane from |srcV| dup'd out to V128, and also return the |
| 6285 | lane number, log2 of the lane size in bytes, and width-character via |
| 6286 | *laneNo, *laneSzLg2 and *laneCh respectively. It may be that imm5 |
| 6287 | is an invalid selector, in which case return |
| 6288 | IRTemp_INVALID, 0, 0 and '?' respectively. |
| 6289 | |
| 6290 | imm5 = xxxx1 signifies .b[xxxx] |
| 6291 | = xxx10 .h[xxx] |
| 6292 | = xx100 .s[xx] |
| 6293 | = x1000 .d[x] |
| 6294 | otherwise invalid |
| 6295 | */ |
| 6296 | static |
| 6297 | IRTemp handle_DUP_VEC_ELEM ( /*OUT*/UInt* laneNo, |
| 6298 | /*OUT*/UInt* laneSzLg2, /*OUT*/HChar* laneCh, |
| 6299 | IRExpr* srcV, UInt imm5 ) |
| 6300 | { |
| 6301 | *laneNo = 0; |
| 6302 | *laneSzLg2 = 0; |
| 6303 | *laneCh = '?'; |
| 6304 | |
| 6305 | if (imm5 & 1) { |
| 6306 | *laneNo = (imm5 >> 1) & 15; |
| 6307 | *laneSzLg2 = 0; |
| 6308 | *laneCh = 'b'; |
| 6309 | } |
| 6310 | else if (imm5 & 2) { |
| 6311 | *laneNo = (imm5 >> 2) & 7; |
| 6312 | *laneSzLg2 = 1; |
| 6313 | *laneCh = 'h'; |
| 6314 | } |
| 6315 | else if (imm5 & 4) { |
| 6316 | *laneNo = (imm5 >> 3) & 3; |
| 6317 | *laneSzLg2 = 2; |
| 6318 | *laneCh = 's'; |
| 6319 | } |
| 6320 | else if (imm5 & 8) { |
| 6321 | *laneNo = (imm5 >> 4) & 1; |
| 6322 | *laneSzLg2 = 3; |
| 6323 | *laneCh = 'd'; |
| 6324 | } |
| 6325 | else { |
| 6326 | /* invalid */ |
| 6327 | return IRTemp_INVALID; |
| 6328 | } |
| 6329 | |
| 6330 | return math_DUP_VEC_ELEM(srcV, *laneSzLg2, *laneNo); |
| 6331 | } |
| 6332 | |
| 6333 | |
| 6334 | /* Clone |imm| to every lane of a V128, with lane size log2 of |size|. */ |
| 6335 | static |
| 6336 | IRTemp math_VEC_DUP_IMM ( UInt size, ULong imm ) |
| 6337 | { |
| 6338 | IRType ty = Ity_INVALID; |
| 6339 | IRTemp rcS = IRTemp_INVALID; |
| 6340 | switch (size) { |
| 6341 | case X01: |
| 6342 | vassert(imm <= 0xFFFFULL); |
| 6343 | ty = Ity_I16; |
| 6344 | rcS = newTemp(ty); assign(rcS, mkU16( (UShort)imm )); |
| 6345 | break; |
| 6346 | case X10: |
| 6347 | vassert(imm <= 0xFFFFFFFFULL); |
| 6348 | ty = Ity_I32; |
| 6349 | rcS = newTemp(ty); assign(rcS, mkU32( (UInt)imm )); |
| 6350 | break; |
| 6351 | case X11: |
| 6352 | ty = Ity_I64; |
| 6353 | rcS = newTemp(ty); assign(rcS, mkU64(imm)); break; |
| 6354 | default: |
| 6355 | vassert(0); |
| 6356 | } |
| 6357 | IRTemp rcV = math_DUP_TO_V128(rcS, ty); |
| 6358 | return rcV; |
| 6359 | } |
| 6360 | |
| 6361 | |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 6362 | /* Let |new64| be a V128 in which only the lower 64 bits are interesting, |
| 6363 | and the upper can contain any value -- it is ignored. If |is2| is False, |
| 6364 | generate IR to put |new64| in the lower half of vector reg |dd| and zero |
| 6365 | the upper half. If |is2| is True, generate IR to put |new64| in the upper |
| 6366 | half of vector reg |dd| and leave the lower half unchanged. This |
| 6367 | simulates the behaviour of the "foo/foo2" instructions in which the |
| 6368 | destination is half the width of sources, for example addhn/addhn2. |
| 6369 | */ |
| 6370 | static |
| 6371 | void putLO64andZUorPutHI64 ( Bool is2, UInt dd, IRTemp new64 ) |
| 6372 | { |
| 6373 | if (is2) { |
| 6374 | /* Get the old contents of Vdd, zero the upper half, and replace |
| 6375 | it with 'x'. */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6376 | IRTemp t_zero_oldLO = newTempV128(); |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 6377 | assign(t_zero_oldLO, unop(Iop_ZeroHI64ofV128, getQReg128(dd))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6378 | IRTemp t_newHI_zero = newTempV128(); |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 6379 | assign(t_newHI_zero, binop(Iop_InterleaveLO64x2, mkexpr(new64), |
| 6380 | mkV128(0x0000))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6381 | IRTemp res = newTempV128(); |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 6382 | assign(res, binop(Iop_OrV128, mkexpr(t_zero_oldLO), |
| 6383 | mkexpr(t_newHI_zero))); |
| 6384 | putQReg128(dd, mkexpr(res)); |
| 6385 | } else { |
| 6386 | /* This is simple. */ |
| 6387 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, mkexpr(new64))); |
| 6388 | } |
| 6389 | } |
| 6390 | |
| 6391 | |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6392 | /* Compute vector SQABS at lane size |size| for |srcE|, returning |
| 6393 | the q result in |*qabs| and the normal result in |*nabs|. */ |
| 6394 | static |
| 6395 | void math_SQABS ( /*OUT*/IRTemp* qabs, /*OUT*/IRTemp* nabs, |
| 6396 | IRExpr* srcE, UInt size ) |
| 6397 | { |
| 6398 | IRTemp src, mask, maskn, nsub, qsub; |
| 6399 | src = mask = maskn = nsub = qsub = IRTemp_INVALID; |
| 6400 | newTempsV128_7(&src, &mask, &maskn, &nsub, &qsub, nabs, qabs); |
| 6401 | assign(src, srcE); |
| 6402 | assign(mask, binop(mkVecCMPGTS(size), mkV128(0x0000), mkexpr(src))); |
| 6403 | assign(maskn, unop(Iop_NotV128, mkexpr(mask))); |
| 6404 | assign(nsub, binop(mkVecSUB(size), mkV128(0x0000), mkexpr(src))); |
| 6405 | assign(qsub, binop(mkVecQSUBS(size), mkV128(0x0000), mkexpr(src))); |
| 6406 | assign(*nabs, binop(Iop_OrV128, |
| 6407 | binop(Iop_AndV128, mkexpr(nsub), mkexpr(mask)), |
| 6408 | binop(Iop_AndV128, mkexpr(src), mkexpr(maskn)))); |
| 6409 | assign(*qabs, binop(Iop_OrV128, |
| 6410 | binop(Iop_AndV128, mkexpr(qsub), mkexpr(mask)), |
| 6411 | binop(Iop_AndV128, mkexpr(src), mkexpr(maskn)))); |
| 6412 | } |
| 6413 | |
| 6414 | |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 6415 | /* Compute vector SQNEG at lane size |size| for |srcE|, returning |
| 6416 | the q result in |*qneg| and the normal result in |*nneg|. */ |
| 6417 | static |
| 6418 | void math_SQNEG ( /*OUT*/IRTemp* qneg, /*OUT*/IRTemp* nneg, |
| 6419 | IRExpr* srcE, UInt size ) |
| 6420 | { |
| 6421 | IRTemp src = IRTemp_INVALID; |
| 6422 | newTempsV128_3(&src, nneg, qneg); |
| 6423 | assign(src, srcE); |
| 6424 | assign(*nneg, binop(mkVecSUB(size), mkV128(0x0000), mkexpr(src))); |
| 6425 | assign(*qneg, binop(mkVecQSUBS(size), mkV128(0x0000), mkexpr(src))); |
| 6426 | } |
| 6427 | |
| 6428 | |
sewardj | ecedd98 | 2014-08-11 14:02:47 +0000 | [diff] [blame] | 6429 | /* Zero all except the least significant lane of |srcE|, where |size| |
| 6430 | indicates the lane size in the usual way. */ |
sewardj | 257e99f | 2014-08-03 12:45:19 +0000 | [diff] [blame] | 6431 | static IRTemp math_ZERO_ALL_EXCEPT_LOWEST_LANE ( UInt size, IRExpr* srcE ) |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6432 | { |
| 6433 | vassert(size < 4); |
| 6434 | IRTemp t = newTempV128(); |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 6435 | assign(t, unop(mkVecZEROHIxxOFV128(size), srcE)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6436 | return t; |
| 6437 | } |
| 6438 | |
| 6439 | |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 6440 | /* Generate IR to compute vector widening MULL from either the lower |
| 6441 | (is2==False) or upper (is2==True) halves of vecN and vecM. The |
| 6442 | widening multiplies are unsigned when isU==True and signed when |
| 6443 | isU==False. |size| is the narrow lane size indication. Optionally, |
| 6444 | the product may be added to or subtracted from vecD, at the wide lane |
| 6445 | size. This happens when |mas| is 'a' (add) or 's' (sub). When |mas| |
| 6446 | is 'm' (only multiply) then the accumulate part does not happen, and |
| 6447 | |vecD| is expected to == IRTemp_INVALID. |
| 6448 | |
| 6449 | Only size==0 (h_b_b), size==1 (s_h_h) and size==2 (d_s_s) variants |
| 6450 | are allowed. The result is returned in a new IRTemp, which is |
| 6451 | returned in *res. */ |
| 6452 | static |
| 6453 | void math_MULL_ACC ( /*OUT*/IRTemp* res, |
| 6454 | Bool is2, Bool isU, UInt size, HChar mas, |
| 6455 | IRTemp vecN, IRTemp vecM, IRTemp vecD ) |
| 6456 | { |
| 6457 | vassert(res && *res == IRTemp_INVALID); |
| 6458 | vassert(size <= 2); |
| 6459 | vassert(mas == 'm' || mas == 'a' || mas == 's'); |
| 6460 | if (mas == 'm') vassert(vecD == IRTemp_INVALID); |
| 6461 | IROp mulOp = isU ? mkVecMULLU(size) : mkVecMULLS(size); |
| 6462 | IROp accOp = (mas == 'a') ? mkVecADD(size+1) |
| 6463 | : (mas == 's' ? mkVecSUB(size+1) |
| 6464 | : Iop_INVALID); |
| 6465 | IRTemp mul = math_BINARY_WIDENING_V128(is2, mulOp, |
| 6466 | mkexpr(vecN), mkexpr(vecM)); |
| 6467 | *res = newTempV128(); |
| 6468 | assign(*res, mas == 'm' ? mkexpr(mul) |
| 6469 | : binop(accOp, mkexpr(vecD), mkexpr(mul))); |
| 6470 | } |
| 6471 | |
| 6472 | |
| 6473 | /* Same as math_MULL_ACC, except the multiply is signed widening, |
| 6474 | the multiplied value is then doubled, before being added to or |
| 6475 | subtracted from the accumulated value. And everything is |
| 6476 | saturated. In all cases, saturation residuals are returned |
| 6477 | via (sat1q, sat1n), and in the accumulate cases, |
| 6478 | via (sat2q, sat2n) too. All results are returned in new temporaries. |
| 6479 | In the no-accumulate case, *sat2q and *sat2n are never instantiated, |
| 6480 | so the caller can tell this has happened. */ |
| 6481 | static |
| 6482 | void math_SQDMULL_ACC ( /*OUT*/IRTemp* res, |
| 6483 | /*OUT*/IRTemp* sat1q, /*OUT*/IRTemp* sat1n, |
| 6484 | /*OUT*/IRTemp* sat2q, /*OUT*/IRTemp* sat2n, |
| 6485 | Bool is2, UInt size, HChar mas, |
| 6486 | IRTemp vecN, IRTemp vecM, IRTemp vecD ) |
| 6487 | { |
| 6488 | vassert(size <= 2); |
| 6489 | vassert(mas == 'm' || mas == 'a' || mas == 's'); |
| 6490 | /* Compute |
| 6491 | sat1q = vecN.D[is2] *sq vecM.d[is2] *q 2 |
| 6492 | sat1n = vecN.D[is2] *s vecM.d[is2] * 2 |
| 6493 | IOW take either the low or high halves of vecN and vecM, signed widen, |
| 6494 | multiply, double that, and signedly saturate. Also compute the same |
| 6495 | but without saturation. |
| 6496 | */ |
| 6497 | vassert(sat2q && *sat2q == IRTemp_INVALID); |
| 6498 | vassert(sat2n && *sat2n == IRTemp_INVALID); |
| 6499 | newTempsV128_3(sat1q, sat1n, res); |
| 6500 | IRTemp tq = math_BINARY_WIDENING_V128(is2, mkVecQDMULLS(size), |
| 6501 | mkexpr(vecN), mkexpr(vecM)); |
| 6502 | IRTemp tn = math_BINARY_WIDENING_V128(is2, mkVecMULLS(size), |
| 6503 | mkexpr(vecN), mkexpr(vecM)); |
| 6504 | assign(*sat1q, mkexpr(tq)); |
| 6505 | assign(*sat1n, binop(mkVecADD(size+1), mkexpr(tn), mkexpr(tn))); |
| 6506 | |
| 6507 | /* If there is no accumulation, the final result is sat1q, |
| 6508 | and there's no assignment to sat2q or sat2n. */ |
| 6509 | if (mas == 'm') { |
| 6510 | assign(*res, mkexpr(*sat1q)); |
| 6511 | return; |
| 6512 | } |
| 6513 | |
| 6514 | /* Compute |
| 6515 | sat2q = vecD +sq/-sq sat1q |
| 6516 | sat2n = vecD +/- sat1n |
| 6517 | result = sat2q |
| 6518 | */ |
| 6519 | newTempsV128_2(sat2q, sat2n); |
| 6520 | assign(*sat2q, binop(mas == 'a' ? mkVecQADDS(size+1) : mkVecQSUBS(size+1), |
| 6521 | mkexpr(vecD), mkexpr(*sat1q))); |
| 6522 | assign(*sat2n, binop(mas == 'a' ? mkVecADD(size+1) : mkVecSUB(size+1), |
| 6523 | mkexpr(vecD), mkexpr(*sat1n))); |
| 6524 | assign(*res, mkexpr(*sat2q)); |
| 6525 | } |
| 6526 | |
| 6527 | |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 6528 | /* Generate IR for widening signed vector multiplies. The operands |
| 6529 | have their lane width signedly widened, and they are then multiplied |
| 6530 | at the wider width, returning results in two new IRTemps. */ |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 6531 | static |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 6532 | void math_MULLS ( /*OUT*/IRTemp* resHI, /*OUT*/IRTemp* resLO, |
| 6533 | UInt sizeNarrow, IRTemp argL, IRTemp argR ) |
| 6534 | { |
| 6535 | vassert(sizeNarrow <= 2); |
| 6536 | newTempsV128_2(resHI, resLO); |
| 6537 | IRTemp argLhi = newTemp(Ity_I64); |
| 6538 | IRTemp argLlo = newTemp(Ity_I64); |
| 6539 | IRTemp argRhi = newTemp(Ity_I64); |
| 6540 | IRTemp argRlo = newTemp(Ity_I64); |
| 6541 | assign(argLhi, unop(Iop_V128HIto64, mkexpr(argL))); |
| 6542 | assign(argLlo, unop(Iop_V128to64, mkexpr(argL))); |
| 6543 | assign(argRhi, unop(Iop_V128HIto64, mkexpr(argR))); |
| 6544 | assign(argRlo, unop(Iop_V128to64, mkexpr(argR))); |
| 6545 | IROp opMulls = mkVecMULLS(sizeNarrow); |
| 6546 | assign(*resHI, binop(opMulls, mkexpr(argLhi), mkexpr(argRhi))); |
| 6547 | assign(*resLO, binop(opMulls, mkexpr(argLlo), mkexpr(argRlo))); |
| 6548 | } |
| 6549 | |
| 6550 | |
sewardj | 257e99f | 2014-08-03 12:45:19 +0000 | [diff] [blame] | 6551 | /* Generate IR for SQDMULH and SQRDMULH: signedly wideningly multiply, |
| 6552 | double that, possibly add a rounding constant (R variants), and take |
| 6553 | the high half. */ |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 6554 | static |
| 6555 | void math_SQDMULH ( /*OUT*/IRTemp* res, |
| 6556 | /*OUT*/IRTemp* sat1q, /*OUT*/IRTemp* sat1n, |
| 6557 | Bool isR, UInt size, IRTemp vN, IRTemp vM ) |
| 6558 | { |
| 6559 | vassert(size == X01 || size == X10); /* s or h only */ |
| 6560 | |
| 6561 | newTempsV128_3(res, sat1q, sat1n); |
| 6562 | |
| 6563 | IRTemp mullsHI = IRTemp_INVALID, mullsLO = IRTemp_INVALID; |
| 6564 | math_MULLS(&mullsHI, &mullsLO, size, vN, vM); |
| 6565 | |
| 6566 | IRTemp addWide = mkVecADD(size+1); |
| 6567 | |
| 6568 | if (isR) { |
| 6569 | assign(*sat1q, binop(mkVecQRDMULHIS(size), mkexpr(vN), mkexpr(vM))); |
| 6570 | |
| 6571 | Int rcShift = size == X01 ? 15 : 31; |
| 6572 | IRTemp roundConst = math_VEC_DUP_IMM(size+1, 1ULL << rcShift); |
| 6573 | assign(*sat1n, |
| 6574 | binop(mkVecCATODDLANES(size), |
| 6575 | binop(addWide, |
| 6576 | binop(addWide, mkexpr(mullsHI), mkexpr(mullsHI)), |
| 6577 | mkexpr(roundConst)), |
| 6578 | binop(addWide, |
| 6579 | binop(addWide, mkexpr(mullsLO), mkexpr(mullsLO)), |
| 6580 | mkexpr(roundConst)))); |
| 6581 | } else { |
| 6582 | assign(*sat1q, binop(mkVecQDMULHIS(size), mkexpr(vN), mkexpr(vM))); |
| 6583 | |
| 6584 | assign(*sat1n, |
| 6585 | binop(mkVecCATODDLANES(size), |
| 6586 | binop(addWide, mkexpr(mullsHI), mkexpr(mullsHI)), |
| 6587 | binop(addWide, mkexpr(mullsLO), mkexpr(mullsLO)))); |
| 6588 | } |
| 6589 | |
| 6590 | assign(*res, mkexpr(*sat1q)); |
| 6591 | } |
| 6592 | |
| 6593 | |
sewardj | a97dddf | 2014-08-14 22:26:52 +0000 | [diff] [blame^] | 6594 | /* Generate IR for SQSHL, UQSHL, SQSHLU by imm. Put the result in |
| 6595 | a new temp in *res, and the Q difference pair in new temps in |
| 6596 | *qDiff1 and *qDiff2 respectively. |nm| denotes which of the |
| 6597 | three operations it is. */ |
| 6598 | static |
| 6599 | void math_QSHL_IMM ( /*OUT*/IRTemp* res, |
| 6600 | /*OUT*/IRTemp* qDiff1, /*OUT*/IRTemp* qDiff2, |
| 6601 | IRTemp src, UInt size, UInt shift, const HChar* nm ) |
| 6602 | { |
| 6603 | vassert(size <= 3); |
| 6604 | UInt laneBits = 8 << size; |
| 6605 | vassert(shift < laneBits); |
| 6606 | newTempsV128_3(res, qDiff1, qDiff2); |
| 6607 | IRTemp z128 = newTempV128(); |
| 6608 | assign(z128, mkV128(0x0000)); |
| 6609 | |
| 6610 | /* UQSHL */ |
| 6611 | if (vex_streq(nm, "uqshl")) { |
| 6612 | IROp qop = mkVecQSHLNSATU2U(size); |
| 6613 | assign(*res, binop(qop, mkexpr(src), mkU8(shift))); |
| 6614 | if (shift == 0) { |
| 6615 | /* No shift means no saturation. */ |
| 6616 | assign(*qDiff1, mkexpr(z128)); |
| 6617 | assign(*qDiff2, mkexpr(z128)); |
| 6618 | } else { |
| 6619 | /* Saturation has occurred if any of the shifted-out bits are |
| 6620 | nonzero. We get the shifted-out bits by right-shifting the |
| 6621 | original value. */ |
| 6622 | UInt rshift = laneBits - shift; |
| 6623 | vassert(rshift >= 1 && rshift < laneBits); |
| 6624 | assign(*qDiff1, binop(mkVecSHRN(size), mkexpr(src), mkU8(rshift))); |
| 6625 | assign(*qDiff2, mkexpr(z128)); |
| 6626 | } |
| 6627 | return; |
| 6628 | } |
| 6629 | |
| 6630 | /* SQSHL */ |
| 6631 | if (vex_streq(nm, "sqshl")) { |
| 6632 | IROp qop = mkVecQSHLNSATS2S(size); |
| 6633 | assign(*res, binop(qop, mkexpr(src), mkU8(shift))); |
| 6634 | if (shift == 0) { |
| 6635 | /* No shift means no saturation. */ |
| 6636 | assign(*qDiff1, mkexpr(z128)); |
| 6637 | assign(*qDiff2, mkexpr(z128)); |
| 6638 | } else { |
| 6639 | /* Saturation has occurred if any of the shifted-out bits are |
| 6640 | different from the top bit of the original value. */ |
| 6641 | UInt rshift = laneBits - 1 - shift; |
| 6642 | vassert(rshift >= 0 && rshift < laneBits-1); |
| 6643 | /* qDiff1 is the shifted out bits, and the top bit of the original |
| 6644 | value, preceded by zeroes. */ |
| 6645 | assign(*qDiff1, binop(mkVecSHRN(size), mkexpr(src), mkU8(rshift))); |
| 6646 | /* qDiff2 is the top bit of the original value, cloned the |
| 6647 | correct number of times. */ |
| 6648 | assign(*qDiff2, binop(mkVecSHRN(size), |
| 6649 | binop(mkVecSARN(size), mkexpr(src), |
| 6650 | mkU8(laneBits-1)), |
| 6651 | mkU8(rshift))); |
| 6652 | /* This also succeeds in comparing the top bit of the original |
| 6653 | value to itself, which is a bit stupid, but not wrong. */ |
| 6654 | } |
| 6655 | return; |
| 6656 | } |
| 6657 | |
| 6658 | /* SQSHLU */ |
| 6659 | if (vex_streq(nm, "sqshlu")) { |
| 6660 | IROp qop = mkVecQSHLNSATS2U(size); |
| 6661 | assign(*res, binop(qop, mkexpr(src), mkU8(shift))); |
| 6662 | /* This is different from the other two cases, in that |
| 6663 | saturation can occur even if there is no shift. */ |
| 6664 | /* Saturation has occurred if any of the shifted-out bits, or |
| 6665 | the top bit of the original value, are nonzero. */ |
| 6666 | UInt rshift = laneBits - 1 - shift; |
| 6667 | vassert(rshift >= 0 && rshift < laneBits); |
| 6668 | /* qDiff1 is the shifted out bits, and the top bit of the original |
| 6669 | value, preceded by zeroes. */ |
| 6670 | assign(*qDiff1, binop(mkVecSHRN(size), mkexpr(src), mkU8(rshift))); |
| 6671 | assign(*qDiff2, mkexpr(z128)); |
| 6672 | return; |
| 6673 | } |
| 6674 | |
| 6675 | vassert(0); |
| 6676 | } |
| 6677 | |
| 6678 | |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 6679 | /* QCFLAG tracks the SIMD sticky saturation status. Update the status |
| 6680 | thusly: if, after application of |opZHI| to both |qres| and |nres|, |
| 6681 | they have the same value, leave QCFLAG unchanged. Otherwise, set it |
| 6682 | (implicitly) to 1. |opZHI| may only be one of the Iop_ZeroHIxxofV128 |
| 6683 | operators, or Iop_INVALID, in which case |qres| and |nres| are used |
| 6684 | unmodified. The presence |opZHI| means this function can be used to |
| 6685 | generate QCFLAG update code for both scalar and vector SIMD operations. |
| 6686 | */ |
| 6687 | static |
| 6688 | void updateQCFLAGwithDifferenceZHI ( IRTemp qres, IRTemp nres, IROp opZHI ) |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 6689 | { |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6690 | IRTemp diff = newTempV128(); |
| 6691 | IRTemp oldQCFLAG = newTempV128(); |
| 6692 | IRTemp newQCFLAG = newTempV128(); |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 6693 | if (opZHI == Iop_INVALID) { |
| 6694 | assign(diff, binop(Iop_XorV128, mkexpr(qres), mkexpr(nres))); |
| 6695 | } else { |
sewardj | 257e99f | 2014-08-03 12:45:19 +0000 | [diff] [blame] | 6696 | vassert(opZHI == Iop_ZeroHI64ofV128 |
| 6697 | || opZHI == Iop_ZeroHI96ofV128 || opZHI == Iop_ZeroHI112ofV128); |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 6698 | assign(diff, unop(opZHI, binop(Iop_XorV128, mkexpr(qres), mkexpr(nres)))); |
| 6699 | } |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 6700 | assign(oldQCFLAG, IRExpr_Get(OFFB_QCFLAG, Ity_V128)); |
| 6701 | assign(newQCFLAG, binop(Iop_OrV128, mkexpr(oldQCFLAG), mkexpr(diff))); |
| 6702 | stmt(IRStmt_Put(OFFB_QCFLAG, mkexpr(newQCFLAG))); |
| 6703 | } |
| 6704 | |
| 6705 | |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 6706 | /* A variant of updateQCFLAGwithDifferenceZHI in which |qres| and |nres| |
| 6707 | are used unmodified, hence suitable for QCFLAG updates for whole-vector |
| 6708 | operations. */ |
| 6709 | static |
| 6710 | void updateQCFLAGwithDifference ( IRTemp qres, IRTemp nres ) |
| 6711 | { |
| 6712 | updateQCFLAGwithDifferenceZHI(qres, nres, Iop_INVALID); |
| 6713 | } |
| 6714 | |
| 6715 | |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6716 | /*------------------------------------------------------------*/ |
| 6717 | /*--- SIMD and FP instructions ---*/ |
| 6718 | /*------------------------------------------------------------*/ |
| 6719 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6720 | static |
| 6721 | Bool dis_AdvSIMD_EXT(/*MB_OUT*/DisResult* dres, UInt insn) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 6722 | { |
sewardj | ab33a7a | 2014-06-19 22:20:47 +0000 | [diff] [blame] | 6723 | /* 31 29 23 21 20 15 14 10 9 4 |
| 6724 | 0 q 101110 op2 0 m 0 imm4 0 n d |
| 6725 | Decode fields: op2 |
| 6726 | */ |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 6727 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
sewardj | ab33a7a | 2014-06-19 22:20:47 +0000 | [diff] [blame] | 6728 | if (INSN(31,31) != 0 |
| 6729 | || INSN(29,24) != BITS6(1,0,1,1,1,0) |
| 6730 | || INSN(21,21) != 0 || INSN(15,15) != 0 || INSN(10,10) != 0) { |
| 6731 | return False; |
| 6732 | } |
| 6733 | UInt bitQ = INSN(30,30); |
| 6734 | UInt op2 = INSN(23,22); |
| 6735 | UInt mm = INSN(20,16); |
| 6736 | UInt imm4 = INSN(14,11); |
| 6737 | UInt nn = INSN(9,5); |
| 6738 | UInt dd = INSN(4,0); |
| 6739 | |
| 6740 | if (op2 == BITS2(0,0)) { |
| 6741 | /* -------- 00: EXT 16b_16b_16b, 8b_8b_8b -------- */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6742 | IRTemp sHi = newTempV128(); |
| 6743 | IRTemp sLo = newTempV128(); |
| 6744 | IRTemp res = newTempV128(); |
sewardj | ab33a7a | 2014-06-19 22:20:47 +0000 | [diff] [blame] | 6745 | assign(sHi, getQReg128(mm)); |
| 6746 | assign(sLo, getQReg128(nn)); |
| 6747 | if (bitQ == 1) { |
| 6748 | if (imm4 == 0) { |
| 6749 | assign(res, mkexpr(sLo)); |
| 6750 | } else { |
| 6751 | vassert(imm4 <= 15); |
| 6752 | assign(res, |
| 6753 | binop(Iop_OrV128, |
| 6754 | binop(Iop_ShlV128, mkexpr(sHi), mkU8(8 * (16-imm4))), |
| 6755 | binop(Iop_ShrV128, mkexpr(sLo), mkU8(8 * imm4)))); |
| 6756 | } |
| 6757 | putQReg128(dd, mkexpr(res)); |
| 6758 | DIP("ext v%u.16b, v%u.16b, v%u.16b, #%u\n", dd, nn, mm, imm4); |
| 6759 | } else { |
| 6760 | if (imm4 >= 8) return False; |
| 6761 | if (imm4 == 0) { |
| 6762 | assign(res, mkexpr(sLo)); |
| 6763 | } else { |
| 6764 | assign(res, |
| 6765 | binop(Iop_ShrV128, |
| 6766 | binop(Iop_InterleaveLO64x2, mkexpr(sHi), mkexpr(sLo)), |
| 6767 | mkU8(8 * imm4))); |
| 6768 | } |
| 6769 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, mkexpr(res))); |
| 6770 | DIP("ext v%u.8b, v%u.8b, v%u.8b, #%u\n", dd, nn, mm, imm4); |
| 6771 | } |
| 6772 | return True; |
| 6773 | } |
| 6774 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6775 | return False; |
| 6776 | # undef INSN |
| 6777 | } |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 6778 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 6779 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6780 | static |
| 6781 | Bool dis_AdvSIMD_TBL_TBX(/*MB_OUT*/DisResult* dres, UInt insn) |
| 6782 | { |
| 6783 | /* 31 29 23 21 20 15 14 12 11 9 4 |
| 6784 | 0 q 001110 op2 0 m 0 len op 00 n d |
| 6785 | Decode fields: op2,len,op |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 6786 | */ |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6787 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 6788 | if (INSN(31,31) != 0 |
| 6789 | || INSN(29,24) != BITS6(0,0,1,1,1,0) |
| 6790 | || INSN(21,21) != 0 |
| 6791 | || INSN(15,15) != 0 |
| 6792 | || INSN(11,10) != BITS2(0,0)) { |
| 6793 | return False; |
| 6794 | } |
| 6795 | UInt bitQ = INSN(30,30); |
| 6796 | UInt op2 = INSN(23,22); |
| 6797 | UInt mm = INSN(20,16); |
| 6798 | UInt len = INSN(14,13); |
| 6799 | UInt bitOP = INSN(12,12); |
| 6800 | UInt nn = INSN(9,5); |
| 6801 | UInt dd = INSN(4,0); |
| 6802 | |
| 6803 | if (op2 == X00) { |
| 6804 | /* -------- 00,xx,0 TBL, xx register table -------- */ |
| 6805 | /* -------- 00,xx,1 TBX, xx register table -------- */ |
| 6806 | /* 31 28 20 15 14 12 9 4 |
| 6807 | 0q0 01110 000 m 0 len 000 n d TBL Vd.Ta, {Vn .. V(n+len)%32}, Vm.Ta |
| 6808 | 0q0 01110 000 m 0 len 100 n d TBX Vd.Ta, {Vn .. V(n+len)%32}, Vm.Ta |
| 6809 | where Ta = 16b(q=1) or 8b(q=0) |
| 6810 | */ |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6811 | Bool isTBX = bitOP == 1; |
| 6812 | /* The out-of-range values to use. */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6813 | IRTemp oor_values = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6814 | assign(oor_values, isTBX ? getQReg128(dd) : mkV128(0)); |
| 6815 | /* src value */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6816 | IRTemp src = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6817 | assign(src, getQReg128(mm)); |
| 6818 | /* The table values */ |
| 6819 | IRTemp tab[4]; |
| 6820 | UInt i; |
| 6821 | for (i = 0; i <= len; i++) { |
| 6822 | vassert(i < 4); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6823 | tab[i] = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6824 | assign(tab[i], getQReg128((nn + i) % 32)); |
| 6825 | } |
| 6826 | IRTemp res = math_TBL_TBX(tab, len, src, oor_values); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 6827 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
| 6828 | const HChar* Ta = bitQ ==1 ? "16b" : "8b"; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6829 | const HChar* nm = isTBX ? "tbx" : "tbl"; |
| 6830 | DIP("%s %s.%s, {v%d.16b .. v%d.16b}, %s.%s\n", |
| 6831 | nm, nameQReg128(dd), Ta, nn, (nn + len) % 32, nameQReg128(mm), Ta); |
| 6832 | return True; |
| 6833 | } |
| 6834 | |
| 6835 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 6836 | return False; |
| 6837 | # undef INSN |
| 6838 | } |
| 6839 | |
| 6840 | |
| 6841 | static |
| 6842 | Bool dis_AdvSIMD_ZIP_UZP_TRN(/*MB_OUT*/DisResult* dres, UInt insn) |
| 6843 | { |
| 6844 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 6845 | return False; |
| 6846 | # undef INSN |
| 6847 | } |
| 6848 | |
| 6849 | |
| 6850 | static |
| 6851 | Bool dis_AdvSIMD_across_lanes(/*MB_OUT*/DisResult* dres, UInt insn) |
| 6852 | { |
| 6853 | /* 31 28 23 21 16 11 9 4 |
| 6854 | 0 q u 01110 size 11000 opcode 10 n d |
| 6855 | Decode fields: u,size,opcode |
| 6856 | */ |
| 6857 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 6858 | if (INSN(31,31) != 0 |
| 6859 | || INSN(28,24) != BITS5(0,1,1,1,0) |
| 6860 | || INSN(21,17) != BITS5(1,1,0,0,0) || INSN(11,10) != BITS2(1,0)) { |
| 6861 | return False; |
| 6862 | } |
| 6863 | UInt bitQ = INSN(30,30); |
| 6864 | UInt bitU = INSN(29,29); |
| 6865 | UInt size = INSN(23,22); |
| 6866 | UInt opcode = INSN(16,12); |
| 6867 | UInt nn = INSN(9,5); |
| 6868 | UInt dd = INSN(4,0); |
| 6869 | |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 6870 | if (opcode == BITS5(0,0,0,1,1)) { |
| 6871 | /* -------- 0,xx,00011 SADDLV -------- */ |
| 6872 | /* -------- 1,xx,00011 UADDLV -------- */ |
| 6873 | /* size is the narrow size */ |
| 6874 | if (size == X11 || (size == X10 && bitQ == 0)) return False; |
| 6875 | Bool isU = bitU == 1; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6876 | IRTemp src = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 6877 | assign(src, getQReg128(nn)); |
| 6878 | /* The basic plan is to widen the lower half, and if Q = 1, |
| 6879 | the upper half too. Add them together (if Q = 1), and in |
| 6880 | either case fold with add at twice the lane width. |
| 6881 | */ |
| 6882 | IRExpr* widened |
| 6883 | = mkexpr(math_WIDEN_LO_OR_HI_LANES( |
| 6884 | isU, False/*!fromUpperHalf*/, size, mkexpr(src))); |
| 6885 | if (bitQ == 1) { |
| 6886 | widened |
| 6887 | = binop(mkVecADD(size+1), |
| 6888 | widened, |
| 6889 | mkexpr(math_WIDEN_LO_OR_HI_LANES( |
| 6890 | isU, True/*fromUpperHalf*/, size, mkexpr(src))) |
| 6891 | ); |
| 6892 | } |
| 6893 | /* Now fold. */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6894 | IRTemp tWi = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 6895 | assign(tWi, widened); |
| 6896 | IRTemp res = math_FOLDV(tWi, mkVecADD(size+1)); |
| 6897 | putQReg128(dd, mkexpr(res)); |
| 6898 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 6899 | const HChar ch = "bhsd"[size]; |
| 6900 | DIP("%s %s.%c, %s.%s\n", isU ? "uaddlv" : "saddlv", |
| 6901 | nameQReg128(dd), ch, nameQReg128(nn), arr); |
| 6902 | return True; |
| 6903 | } |
| 6904 | |
sewardj | b9aff1e | 2014-06-15 21:55:33 +0000 | [diff] [blame] | 6905 | UInt ix = 0; |
| 6906 | /**/ if (opcode == BITS5(0,1,0,1,0)) { ix = bitU == 0 ? 1 : 2; } |
| 6907 | else if (opcode == BITS5(1,1,0,1,0)) { ix = bitU == 0 ? 3 : 4; } |
| 6908 | else if (opcode == BITS5(1,1,0,1,1) && bitU == 0) { ix = 5; } |
| 6909 | /**/ |
| 6910 | if (ix != 0) { |
| 6911 | /* -------- 0,xx,01010: SMAXV -------- (1) */ |
| 6912 | /* -------- 1,xx,01010: UMAXV -------- (2) */ |
| 6913 | /* -------- 0,xx,11010: SMINV -------- (3) */ |
| 6914 | /* -------- 1,xx,11010: UMINV -------- (4) */ |
| 6915 | /* -------- 0,xx,11011: ADDV -------- (5) */ |
| 6916 | vassert(ix >= 1 && ix <= 5); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6917 | if (size == X11) return False; // 1d,2d cases not allowed |
| 6918 | if (size == X10 && bitQ == 0) return False; // 2s case not allowed |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6919 | const IROp opMAXS[3] |
| 6920 | = { Iop_Max8Sx16, Iop_Max16Sx8, Iop_Max32Sx4 }; |
| 6921 | const IROp opMAXU[3] |
| 6922 | = { Iop_Max8Ux16, Iop_Max16Ux8, Iop_Max32Ux4 }; |
sewardj | b9aff1e | 2014-06-15 21:55:33 +0000 | [diff] [blame] | 6923 | const IROp opMINS[3] |
| 6924 | = { Iop_Min8Sx16, Iop_Min16Sx8, Iop_Min32Sx4 }; |
| 6925 | const IROp opMINU[3] |
| 6926 | = { Iop_Min8Ux16, Iop_Min16Ux8, Iop_Min32Ux4 }; |
| 6927 | const IROp opADD[3] |
| 6928 | = { Iop_Add8x16, Iop_Add16x8, Iop_Add32x4 }; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6929 | vassert(size < 3); |
sewardj | b9aff1e | 2014-06-15 21:55:33 +0000 | [diff] [blame] | 6930 | IROp op = Iop_INVALID; |
| 6931 | const HChar* nm = NULL; |
| 6932 | switch (ix) { |
| 6933 | case 1: op = opMAXS[size]; nm = "smaxv"; break; |
| 6934 | case 2: op = opMAXU[size]; nm = "umaxv"; break; |
| 6935 | case 3: op = opMINS[size]; nm = "sminv"; break; |
| 6936 | case 4: op = opMINU[size]; nm = "uminv"; break; |
| 6937 | case 5: op = opADD[size]; nm = "addv"; break; |
| 6938 | default: vassert(0); |
| 6939 | } |
| 6940 | vassert(op != Iop_INVALID && nm != NULL); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6941 | IRTemp tN1 = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6942 | assign(tN1, getQReg128(nn)); |
| 6943 | /* If Q == 0, we're just folding lanes in the lower half of |
| 6944 | the value. In which case, copy the lower half of the |
| 6945 | source into the upper half, so we can then treat it the |
sewardj | b9aff1e | 2014-06-15 21:55:33 +0000 | [diff] [blame] | 6946 | same as the full width case. Except for the addition case, |
| 6947 | in which we have to zero out the upper half. */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 6948 | IRTemp tN2 = newTempV128(); |
sewardj | b9aff1e | 2014-06-15 21:55:33 +0000 | [diff] [blame] | 6949 | assign(tN2, bitQ == 0 |
| 6950 | ? (ix == 5 ? unop(Iop_ZeroHI64ofV128, mkexpr(tN1)) |
| 6951 | : mk_CatEvenLanes64x2(tN1,tN1)) |
| 6952 | : mkexpr(tN1)); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 6953 | IRTemp res = math_FOLDV(tN2, op); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6954 | if (res == IRTemp_INVALID) |
| 6955 | return False; /* means math_MINMAXV |
| 6956 | doesn't handle this case yet */ |
| 6957 | putQReg128(dd, mkexpr(res)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 6958 | const IRType tys[3] = { Ity_I8, Ity_I16, Ity_I32 }; |
| 6959 | IRType laneTy = tys[size]; |
| 6960 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 6961 | DIP("%s %s, %s.%s\n", nm, |
| 6962 | nameQRegLO(dd, laneTy), nameQReg128(nn), arr); |
| 6963 | return True; |
| 6964 | } |
| 6965 | |
| 6966 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 6967 | return False; |
| 6968 | # undef INSN |
| 6969 | } |
| 6970 | |
| 6971 | |
| 6972 | static |
| 6973 | Bool dis_AdvSIMD_copy(/*MB_OUT*/DisResult* dres, UInt insn) |
| 6974 | { |
| 6975 | /* 31 28 20 15 14 10 9 4 |
| 6976 | 0 q op 01110000 imm5 0 imm4 1 n d |
| 6977 | Decode fields: q,op,imm4 |
| 6978 | */ |
| 6979 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 6980 | if (INSN(31,31) != 0 |
| 6981 | || INSN(28,21) != BITS8(0,1,1,1,0,0,0,0) |
| 6982 | || INSN(15,15) != 0 || INSN(10,10) != 1) { |
| 6983 | return False; |
| 6984 | } |
| 6985 | UInt bitQ = INSN(30,30); |
| 6986 | UInt bitOP = INSN(29,29); |
| 6987 | UInt imm5 = INSN(20,16); |
| 6988 | UInt imm4 = INSN(14,11); |
| 6989 | UInt nn = INSN(9,5); |
| 6990 | UInt dd = INSN(4,0); |
| 6991 | |
| 6992 | /* -------- x,0,0000: DUP (element, vector) -------- */ |
| 6993 | /* 31 28 20 15 9 4 |
| 6994 | 0q0 01110000 imm5 000001 n d DUP Vd.T, Vn.Ts[index] |
| 6995 | */ |
| 6996 | if (bitOP == 0 && imm4 == BITS4(0,0,0,0)) { |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 6997 | UInt laneNo = 0; |
| 6998 | UInt laneSzLg2 = 0; |
| 6999 | HChar laneCh = '?'; |
| 7000 | IRTemp res = handle_DUP_VEC_ELEM(&laneNo, &laneSzLg2, &laneCh, |
| 7001 | getQReg128(nn), imm5); |
| 7002 | if (res == IRTemp_INVALID) |
| 7003 | return False; |
| 7004 | if (bitQ == 0 && laneSzLg2 == X11) |
| 7005 | return False; /* .1d case */ |
| 7006 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
| 7007 | const HChar* arT = nameArr_Q_SZ(bitQ, laneSzLg2); |
| 7008 | DIP("dup %s.%s, %s.%c[%u]\n", |
| 7009 | nameQReg128(dd), arT, nameQReg128(nn), laneCh, laneNo); |
| 7010 | return True; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7011 | } |
| 7012 | |
| 7013 | /* -------- x,0,0001: DUP (general, vector) -------- */ |
| 7014 | /* 31 28 20 15 9 4 |
| 7015 | 0q0 01110000 imm5 0 0001 1 n d DUP Vd.T, Rn |
| 7016 | Q=0 writes 64, Q=1 writes 128 |
| 7017 | imm5: xxxx1 8B(q=0) or 16b(q=1), R=W |
| 7018 | xxx10 4H(q=0) or 8H(q=1), R=W |
| 7019 | xx100 2S(q=0) or 4S(q=1), R=W |
| 7020 | x1000 Invalid(q=0) or 2D(q=1), R=X |
| 7021 | x0000 Invalid(q=0) or Invalid(q=1) |
| 7022 | Require op=0, imm4=0001 |
| 7023 | */ |
| 7024 | if (bitOP == 0 && imm4 == BITS4(0,0,0,1)) { |
| 7025 | Bool isQ = bitQ == 1; |
| 7026 | IRTemp w0 = newTemp(Ity_I64); |
| 7027 | const HChar* arT = "??"; |
| 7028 | IRType laneTy = Ity_INVALID; |
| 7029 | if (imm5 & 1) { |
| 7030 | arT = isQ ? "16b" : "8b"; |
| 7031 | laneTy = Ity_I8; |
| 7032 | assign(w0, unop(Iop_8Uto64, unop(Iop_64to8, getIReg64orZR(nn)))); |
| 7033 | } |
| 7034 | else if (imm5 & 2) { |
| 7035 | arT = isQ ? "8h" : "4h"; |
| 7036 | laneTy = Ity_I16; |
| 7037 | assign(w0, unop(Iop_16Uto64, unop(Iop_64to16, getIReg64orZR(nn)))); |
| 7038 | } |
| 7039 | else if (imm5 & 4) { |
| 7040 | arT = isQ ? "4s" : "2s"; |
| 7041 | laneTy = Ity_I32; |
| 7042 | assign(w0, unop(Iop_32Uto64, unop(Iop_64to32, getIReg64orZR(nn)))); |
| 7043 | } |
| 7044 | else if ((imm5 & 8) && isQ) { |
| 7045 | arT = "2d"; |
| 7046 | laneTy = Ity_I64; |
| 7047 | assign(w0, getIReg64orZR(nn)); |
| 7048 | } |
| 7049 | else { |
| 7050 | /* invalid; leave laneTy unchanged. */ |
| 7051 | } |
| 7052 | /* */ |
| 7053 | if (laneTy != Ity_INVALID) { |
| 7054 | IRTemp w1 = math_DUP_TO_64(w0, laneTy); |
| 7055 | putQReg128(dd, binop(Iop_64HLtoV128, |
| 7056 | isQ ? mkexpr(w1) : mkU64(0), mkexpr(w1))); |
| 7057 | DIP("dup %s.%s, %s\n", |
| 7058 | nameQReg128(dd), arT, nameIRegOrZR(laneTy == Ity_I64, nn)); |
| 7059 | return True; |
| 7060 | } |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 7061 | /* invalid */ |
| 7062 | return False; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7063 | } |
| 7064 | |
| 7065 | /* -------- 1,0,0011: INS (general) -------- */ |
| 7066 | /* 31 28 20 15 9 4 |
| 7067 | 010 01110000 imm5 000111 n d INS Vd.Ts[ix], Rn |
| 7068 | where Ts,ix = case imm5 of xxxx1 -> B, xxxx |
| 7069 | xxx10 -> H, xxx |
| 7070 | xx100 -> S, xx |
| 7071 | x1000 -> D, x |
| 7072 | */ |
| 7073 | if (bitQ == 1 && bitOP == 0 && imm4 == BITS4(0,0,1,1)) { |
| 7074 | HChar ts = '?'; |
| 7075 | UInt laneNo = 16; |
| 7076 | IRExpr* src = NULL; |
| 7077 | if (imm5 & 1) { |
| 7078 | src = unop(Iop_64to8, getIReg64orZR(nn)); |
| 7079 | laneNo = (imm5 >> 1) & 15; |
| 7080 | ts = 'b'; |
| 7081 | } |
| 7082 | else if (imm5 & 2) { |
| 7083 | src = unop(Iop_64to16, getIReg64orZR(nn)); |
| 7084 | laneNo = (imm5 >> 2) & 7; |
| 7085 | ts = 'h'; |
| 7086 | } |
| 7087 | else if (imm5 & 4) { |
| 7088 | src = unop(Iop_64to32, getIReg64orZR(nn)); |
| 7089 | laneNo = (imm5 >> 3) & 3; |
| 7090 | ts = 's'; |
| 7091 | } |
| 7092 | else if (imm5 & 8) { |
| 7093 | src = getIReg64orZR(nn); |
| 7094 | laneNo = (imm5 >> 4) & 1; |
| 7095 | ts = 'd'; |
| 7096 | } |
| 7097 | /* */ |
| 7098 | if (src) { |
| 7099 | vassert(laneNo < 16); |
| 7100 | putQRegLane(dd, laneNo, src); |
| 7101 | DIP("ins %s.%c[%u], %s\n", |
| 7102 | nameQReg128(dd), ts, laneNo, nameIReg64orZR(nn)); |
| 7103 | return True; |
| 7104 | } |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 7105 | /* invalid */ |
| 7106 | return False; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7107 | } |
| 7108 | |
| 7109 | /* -------- x,0,0101: SMOV -------- */ |
| 7110 | /* -------- x,0,0111: UMOV -------- */ |
| 7111 | /* 31 28 20 15 9 4 |
| 7112 | 0q0 01110 000 imm5 001111 n d UMOV Xd/Wd, Vn.Ts[index] |
| 7113 | 0q0 01110 000 imm5 001011 n d SMOV Xd/Wd, Vn.Ts[index] |
| 7114 | dest is Xd when q==1, Wd when q==0 |
| 7115 | UMOV: |
| 7116 | Ts,index,ops = case q:imm5 of |
| 7117 | 0:xxxx1 -> B, xxxx, 8Uto64 |
| 7118 | 1:xxxx1 -> invalid |
| 7119 | 0:xxx10 -> H, xxx, 16Uto64 |
| 7120 | 1:xxx10 -> invalid |
| 7121 | 0:xx100 -> S, xx, 32Uto64 |
| 7122 | 1:xx100 -> invalid |
| 7123 | 1:x1000 -> D, x, copy64 |
| 7124 | other -> invalid |
| 7125 | SMOV: |
| 7126 | Ts,index,ops = case q:imm5 of |
| 7127 | 0:xxxx1 -> B, xxxx, (32Uto64 . 8Sto32) |
| 7128 | 1:xxxx1 -> B, xxxx, 8Sto64 |
| 7129 | 0:xxx10 -> H, xxx, (32Uto64 . 16Sto32) |
| 7130 | 1:xxx10 -> H, xxx, 16Sto64 |
| 7131 | 0:xx100 -> invalid |
| 7132 | 1:xx100 -> S, xx, 32Sto64 |
| 7133 | 1:x1000 -> invalid |
| 7134 | other -> invalid |
| 7135 | */ |
| 7136 | if (bitOP == 0 && (imm4 == BITS4(0,1,0,1) || imm4 == BITS4(0,1,1,1))) { |
| 7137 | Bool isU = (imm4 & 2) == 2; |
| 7138 | const HChar* arTs = "??"; |
| 7139 | UInt laneNo = 16; /* invalid */ |
| 7140 | // Setting 'res' to non-NULL determines valid/invalid |
| 7141 | IRExpr* res = NULL; |
| 7142 | if (!bitQ && (imm5 & 1)) { // 0:xxxx1 |
| 7143 | laneNo = (imm5 >> 1) & 15; |
| 7144 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I8); |
| 7145 | res = isU ? unop(Iop_8Uto64, lane) |
| 7146 | : unop(Iop_32Uto64, unop(Iop_8Sto32, lane)); |
| 7147 | arTs = "b"; |
| 7148 | } |
| 7149 | else if (bitQ && (imm5 & 1)) { // 1:xxxx1 |
| 7150 | laneNo = (imm5 >> 1) & 15; |
| 7151 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I8); |
| 7152 | res = isU ? NULL |
| 7153 | : unop(Iop_8Sto64, lane); |
| 7154 | arTs = "b"; |
| 7155 | } |
| 7156 | else if (!bitQ && (imm5 & 2)) { // 0:xxx10 |
| 7157 | laneNo = (imm5 >> 2) & 7; |
| 7158 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I16); |
| 7159 | res = isU ? unop(Iop_16Uto64, lane) |
| 7160 | : unop(Iop_32Uto64, unop(Iop_16Sto32, lane)); |
| 7161 | arTs = "h"; |
| 7162 | } |
| 7163 | else if (bitQ && (imm5 & 2)) { // 1:xxx10 |
| 7164 | laneNo = (imm5 >> 2) & 7; |
| 7165 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I16); |
| 7166 | res = isU ? NULL |
| 7167 | : unop(Iop_16Sto64, lane); |
| 7168 | arTs = "h"; |
| 7169 | } |
| 7170 | else if (!bitQ && (imm5 & 4)) { // 0:xx100 |
| 7171 | laneNo = (imm5 >> 3) & 3; |
| 7172 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I32); |
| 7173 | res = isU ? unop(Iop_32Uto64, lane) |
| 7174 | : NULL; |
| 7175 | arTs = "s"; |
| 7176 | } |
| 7177 | else if (bitQ && (imm5 & 4)) { // 1:xxx10 |
| 7178 | laneNo = (imm5 >> 3) & 3; |
| 7179 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I32); |
| 7180 | res = isU ? NULL |
| 7181 | : unop(Iop_32Sto64, lane); |
| 7182 | arTs = "s"; |
| 7183 | } |
| 7184 | else if (bitQ && (imm5 & 8)) { // 1:x1000 |
| 7185 | laneNo = (imm5 >> 4) & 1; |
| 7186 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I64); |
| 7187 | res = isU ? lane |
| 7188 | : NULL; |
| 7189 | arTs = "d"; |
| 7190 | } |
| 7191 | /* */ |
| 7192 | if (res) { |
| 7193 | vassert(laneNo < 16); |
| 7194 | putIReg64orZR(dd, res); |
| 7195 | DIP("%cmov %s, %s.%s[%u]\n", isU ? 'u' : 's', |
| 7196 | nameIRegOrZR(bitQ == 1, dd), |
| 7197 | nameQReg128(nn), arTs, laneNo); |
| 7198 | return True; |
| 7199 | } |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 7200 | /* invalid */ |
| 7201 | return False; |
| 7202 | } |
| 7203 | |
| 7204 | /* -------- 1,1,xxxx: INS (element) -------- */ |
| 7205 | /* 31 28 20 14 9 4 |
| 7206 | 011 01110000 imm5 0 imm4 n d INS Vd.Ts[ix1], Vn.Ts[ix2] |
| 7207 | where Ts,ix1,ix2 |
| 7208 | = case imm5 of xxxx1 -> B, xxxx, imm4[3:0] |
| 7209 | xxx10 -> H, xxx, imm4[3:1] |
| 7210 | xx100 -> S, xx, imm4[3:2] |
| 7211 | x1000 -> D, x, imm4[3:3] |
| 7212 | */ |
| 7213 | if (bitQ == 1 && bitOP == 1) { |
| 7214 | HChar ts = '?'; |
| 7215 | IRType ity = Ity_INVALID; |
| 7216 | UInt ix1 = 16; |
| 7217 | UInt ix2 = 16; |
| 7218 | if (imm5 & 1) { |
| 7219 | ts = 'b'; |
| 7220 | ity = Ity_I8; |
| 7221 | ix1 = (imm5 >> 1) & 15; |
| 7222 | ix2 = (imm4 >> 0) & 15; |
| 7223 | } |
| 7224 | else if (imm5 & 2) { |
| 7225 | ts = 'h'; |
| 7226 | ity = Ity_I16; |
| 7227 | ix1 = (imm5 >> 2) & 7; |
| 7228 | ix2 = (imm4 >> 1) & 7; |
| 7229 | } |
| 7230 | else if (imm5 & 4) { |
| 7231 | ts = 's'; |
| 7232 | ity = Ity_I32; |
| 7233 | ix1 = (imm5 >> 3) & 3; |
| 7234 | ix2 = (imm4 >> 2) & 3; |
| 7235 | } |
| 7236 | else if (imm5 & 8) { |
| 7237 | ts = 'd'; |
| 7238 | ity = Ity_I64; |
| 7239 | ix1 = (imm5 >> 4) & 1; |
| 7240 | ix2 = (imm4 >> 3) & 1; |
| 7241 | } |
| 7242 | /* */ |
| 7243 | if (ity != Ity_INVALID) { |
| 7244 | vassert(ix1 < 16); |
| 7245 | vassert(ix2 < 16); |
| 7246 | putQRegLane(dd, ix1, getQRegLane(nn, ix2, ity)); |
| 7247 | DIP("ins %s.%c[%u], %s.%c[%u]\n", |
| 7248 | nameQReg128(dd), ts, ix1, nameQReg128(nn), ts, ix2); |
| 7249 | return True; |
| 7250 | } |
| 7251 | /* invalid */ |
| 7252 | return False; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7253 | } |
| 7254 | |
| 7255 | return False; |
| 7256 | # undef INSN |
| 7257 | } |
| 7258 | |
| 7259 | |
| 7260 | static |
| 7261 | Bool dis_AdvSIMD_modified_immediate(/*MB_OUT*/DisResult* dres, UInt insn) |
| 7262 | { |
| 7263 | /* 31 28 18 15 11 9 4 |
| 7264 | 0q op 01111 00000 abc cmode 01 defgh d |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7265 | Decode fields: q,op,cmode |
| 7266 | Bit 11 is really "o2", but it is always zero. |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7267 | */ |
| 7268 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 7269 | if (INSN(31,31) != 0 |
| 7270 | || INSN(28,19) != BITS10(0,1,1,1,1,0,0,0,0,0) |
| 7271 | || INSN(11,10) != BITS2(0,1)) { |
| 7272 | return False; |
| 7273 | } |
| 7274 | UInt bitQ = INSN(30,30); |
| 7275 | UInt bitOP = INSN(29,29); |
| 7276 | UInt cmode = INSN(15,12); |
| 7277 | UInt abcdefgh = (INSN(18,16) << 5) | INSN(9,5); |
| 7278 | UInt dd = INSN(4,0); |
| 7279 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7280 | ULong imm64lo = 0; |
| 7281 | UInt op_cmode = (bitOP << 4) | cmode; |
| 7282 | Bool ok = False; |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7283 | Bool isORR = False; |
| 7284 | Bool isBIC = False; |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 7285 | Bool isMOV = False; |
| 7286 | Bool isMVN = False; |
| 7287 | Bool isFMOV = False; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7288 | switch (op_cmode) { |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7289 | /* -------- x,0,0000 MOVI 32-bit shifted imm -------- */ |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7290 | /* -------- x,0,0010 MOVI 32-bit shifted imm -------- */ |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 7291 | /* -------- x,0,0100 MOVI 32-bit shifted imm -------- */ |
| 7292 | /* -------- x,0,0110 MOVI 32-bit shifted imm -------- */ |
| 7293 | case BITS5(0,0,0,0,0): case BITS5(0,0,0,1,0): |
| 7294 | case BITS5(0,0,1,0,0): case BITS5(0,0,1,1,0): // 0:0xx0 |
| 7295 | ok = True; isMOV = True; break; |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7296 | |
| 7297 | /* -------- x,0,0001 ORR (vector, immediate) 32-bit -------- */ |
| 7298 | /* -------- x,0,0011 ORR (vector, immediate) 32-bit -------- */ |
| 7299 | /* -------- x,0,0101 ORR (vector, immediate) 32-bit -------- */ |
| 7300 | /* -------- x,0,0111 ORR (vector, immediate) 32-bit -------- */ |
| 7301 | case BITS5(0,0,0,0,1): case BITS5(0,0,0,1,1): |
| 7302 | case BITS5(0,0,1,0,1): case BITS5(0,0,1,1,1): // 0:0xx1 |
| 7303 | ok = True; isORR = True; break; |
| 7304 | |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 7305 | /* -------- x,0,1000 MOVI 16-bit shifted imm -------- */ |
| 7306 | /* -------- x,0,1010 MOVI 16-bit shifted imm -------- */ |
| 7307 | case BITS5(0,1,0,0,0): case BITS5(0,1,0,1,0): // 0:10x0 |
| 7308 | ok = True; isMOV = True; break; |
| 7309 | |
| 7310 | /* -------- x,0,1001 ORR (vector, immediate) 16-bit -------- */ |
| 7311 | /* -------- x,0,1011 ORR (vector, immediate) 16-bit -------- */ |
| 7312 | case BITS5(0,1,0,0,1): case BITS5(0,1,0,1,1): // 0:10x1 |
| 7313 | ok = True; isORR = True; break; |
| 7314 | |
| 7315 | /* -------- x,0,1100 MOVI 32-bit shifting ones -------- */ |
| 7316 | /* -------- x,0,1101 MOVI 32-bit shifting ones -------- */ |
| 7317 | case BITS5(0,1,1,0,0): case BITS5(0,1,1,0,1): // 0:110x |
| 7318 | ok = True; isMOV = True; break; |
| 7319 | |
| 7320 | /* -------- x,0,1110 MOVI 8-bit -------- */ |
| 7321 | case BITS5(0,1,1,1,0): |
| 7322 | ok = True; isMOV = True; break; |
| 7323 | |
| 7324 | /* FMOV (vector, immediate, single precision) */ |
| 7325 | |
| 7326 | /* -------- x,1,0000 MVNI 32-bit shifted imm -------- */ |
| 7327 | /* -------- x,1,0010 MVNI 32-bit shifted imm -------- */ |
| 7328 | /* -------- x,1,0100 MVNI 32-bit shifted imm -------- */ |
| 7329 | /* -------- x,1,0110 MVNI 32-bit shifted imm -------- */ |
| 7330 | case BITS5(1,0,0,0,0): case BITS5(1,0,0,1,0): |
| 7331 | case BITS5(1,0,1,0,0): case BITS5(1,0,1,1,0): // 1:0xx0 |
| 7332 | ok = True; isMVN = True; break; |
| 7333 | |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7334 | /* -------- x,1,0001 BIC (vector, immediate) 32-bit -------- */ |
| 7335 | /* -------- x,1,0011 BIC (vector, immediate) 32-bit -------- */ |
| 7336 | /* -------- x,1,0101 BIC (vector, immediate) 32-bit -------- */ |
| 7337 | /* -------- x,1,0111 BIC (vector, immediate) 32-bit -------- */ |
| 7338 | case BITS5(1,0,0,0,1): case BITS5(1,0,0,1,1): |
| 7339 | case BITS5(1,0,1,0,1): case BITS5(1,0,1,1,1): // 1:0xx1 |
| 7340 | ok = True; isBIC = True; break; |
| 7341 | |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 7342 | /* -------- x,1,1000 MVNI 16-bit shifted imm -------- */ |
| 7343 | /* -------- x,1,1010 MVNI 16-bit shifted imm -------- */ |
| 7344 | case BITS5(1,1,0,0,0): case BITS5(1,1,0,1,0): // 1:10x0 |
| 7345 | ok = True; isMVN = True; break; |
| 7346 | |
| 7347 | /* -------- x,1,1001 BIC (vector, immediate) 16-bit -------- */ |
| 7348 | /* -------- x,1,1011 BIC (vector, immediate) 16-bit -------- */ |
| 7349 | case BITS5(1,1,0,0,1): case BITS5(1,1,0,1,1): // 1:10x1 |
| 7350 | ok = True; isBIC = True; break; |
| 7351 | |
| 7352 | /* -------- x,1,1100 MVNI 32-bit shifting ones -------- */ |
| 7353 | /* -------- x,1,1101 MVNI 32-bit shifting ones -------- */ |
| 7354 | case BITS5(1,1,1,0,0): case BITS5(1,1,1,0,1): // 1:110x |
| 7355 | ok = True; isMVN = True; break; |
| 7356 | |
| 7357 | /* -------- 0,1,1110 MOVI 64-bit scalar -------- */ |
| 7358 | /* -------- 1,1,1110 MOVI 64-bit vector -------- */ |
| 7359 | case BITS5(1,1,1,1,0): |
| 7360 | ok = True; isMOV = True; break; |
| 7361 | |
| 7362 | /* -------- 1,1,1111 FMOV (vector, immediate) -------- */ |
| 7363 | case BITS5(1,1,1,1,1): // 1:1111 |
| 7364 | ok = bitQ == 1; isFMOV = True; break; |
| 7365 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7366 | default: |
| 7367 | break; |
| 7368 | } |
| 7369 | if (ok) { |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 7370 | vassert(1 == (isMOV ? 1 : 0) + (isMVN ? 1 : 0) |
| 7371 | + (isORR ? 1 : 0) + (isBIC ? 1 : 0) + (isFMOV ? 1 : 0)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7372 | ok = AdvSIMDExpandImm(&imm64lo, bitOP, cmode, abcdefgh); |
| 7373 | } |
| 7374 | if (ok) { |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7375 | if (isORR || isBIC) { |
| 7376 | ULong inv |
| 7377 | = isORR ? 0ULL : ~0ULL; |
| 7378 | IRExpr* immV128 |
| 7379 | = binop(Iop_64HLtoV128, mkU64(inv ^ imm64lo), mkU64(inv ^ imm64lo)); |
| 7380 | IRExpr* res |
| 7381 | = binop(isORR ? Iop_OrV128 : Iop_AndV128, getQReg128(dd), immV128); |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7382 | const HChar* nm = isORR ? "orr" : "bic"; |
| 7383 | if (bitQ == 0) { |
| 7384 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, res)); |
| 7385 | DIP("%s %s.1d, %016llx\n", nm, nameQReg128(dd), imm64lo); |
| 7386 | } else { |
| 7387 | putQReg128(dd, res); |
| 7388 | DIP("%s %s.2d, #0x%016llx'%016llx\n", nm, |
| 7389 | nameQReg128(dd), imm64lo, imm64lo); |
| 7390 | } |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 7391 | } |
| 7392 | else if (isMOV || isMVN || isFMOV) { |
| 7393 | if (isMVN) imm64lo = ~imm64lo; |
| 7394 | ULong imm64hi = bitQ == 0 ? 0 : imm64lo; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 7395 | IRExpr* immV128 = binop(Iop_64HLtoV128, mkU64(imm64hi), |
| 7396 | mkU64(imm64lo)); |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7397 | putQReg128(dd, immV128); |
| 7398 | DIP("mov %s, #0x%016llx'%016llx\n", nameQReg128(dd), imm64hi, imm64lo); |
| 7399 | } |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7400 | return True; |
| 7401 | } |
| 7402 | /* else fall through */ |
| 7403 | |
| 7404 | return False; |
| 7405 | # undef INSN |
| 7406 | } |
| 7407 | |
| 7408 | |
| 7409 | static |
| 7410 | Bool dis_AdvSIMD_scalar_copy(/*MB_OUT*/DisResult* dres, UInt insn) |
| 7411 | { |
sewardj | ab33a7a | 2014-06-19 22:20:47 +0000 | [diff] [blame] | 7412 | /* 31 28 20 15 14 10 9 4 |
| 7413 | 01 op 11110000 imm5 0 imm4 1 n d |
| 7414 | Decode fields: op,imm4 |
| 7415 | */ |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7416 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
sewardj | ab33a7a | 2014-06-19 22:20:47 +0000 | [diff] [blame] | 7417 | if (INSN(31,30) != BITS2(0,1) |
| 7418 | || INSN(28,21) != BITS8(1,1,1,1,0,0,0,0) |
| 7419 | || INSN(15,15) != 0 || INSN(10,10) != 1) { |
| 7420 | return False; |
| 7421 | } |
| 7422 | UInt bitOP = INSN(29,29); |
| 7423 | UInt imm5 = INSN(20,16); |
| 7424 | UInt imm4 = INSN(14,11); |
| 7425 | UInt nn = INSN(9,5); |
| 7426 | UInt dd = INSN(4,0); |
| 7427 | |
| 7428 | if (bitOP == 0 && imm4 == BITS4(0,0,0,0)) { |
| 7429 | /* -------- 0,0000 DUP (element, scalar) -------- */ |
| 7430 | IRTemp w0 = newTemp(Ity_I64); |
| 7431 | const HChar* arTs = "??"; |
| 7432 | IRType laneTy = Ity_INVALID; |
| 7433 | UInt laneNo = 16; /* invalid */ |
| 7434 | if (imm5 & 1) { |
| 7435 | arTs = "b"; |
| 7436 | laneNo = (imm5 >> 1) & 15; |
| 7437 | laneTy = Ity_I8; |
| 7438 | assign(w0, unop(Iop_8Uto64, getQRegLane(nn, laneNo, laneTy))); |
| 7439 | } |
| 7440 | else if (imm5 & 2) { |
| 7441 | arTs = "h"; |
| 7442 | laneNo = (imm5 >> 2) & 7; |
| 7443 | laneTy = Ity_I16; |
| 7444 | assign(w0, unop(Iop_16Uto64, getQRegLane(nn, laneNo, laneTy))); |
| 7445 | } |
| 7446 | else if (imm5 & 4) { |
| 7447 | arTs = "s"; |
| 7448 | laneNo = (imm5 >> 3) & 3; |
| 7449 | laneTy = Ity_I32; |
| 7450 | assign(w0, unop(Iop_32Uto64, getQRegLane(nn, laneNo, laneTy))); |
| 7451 | } |
| 7452 | else if (imm5 & 8) { |
| 7453 | arTs = "d"; |
| 7454 | laneNo = (imm5 >> 4) & 1; |
| 7455 | laneTy = Ity_I64; |
| 7456 | assign(w0, getQRegLane(nn, laneNo, laneTy)); |
| 7457 | } |
| 7458 | else { |
| 7459 | /* invalid; leave laneTy unchanged. */ |
| 7460 | } |
| 7461 | /* */ |
| 7462 | if (laneTy != Ity_INVALID) { |
| 7463 | vassert(laneNo < 16); |
| 7464 | putQReg128(dd, binop(Iop_64HLtoV128, mkU64(0), mkexpr(w0))); |
| 7465 | DIP("dup %s, %s.%s[%u]\n", |
| 7466 | nameQRegLO(dd, laneTy), nameQReg128(nn), arTs, laneNo); |
| 7467 | return True; |
| 7468 | } |
| 7469 | /* else fall through */ |
| 7470 | } |
| 7471 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7472 | return False; |
| 7473 | # undef INSN |
| 7474 | } |
| 7475 | |
sewardj | fc83d2c | 2014-06-12 10:15:46 +0000 | [diff] [blame] | 7476 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7477 | static |
| 7478 | Bool dis_AdvSIMD_scalar_pairwise(/*MB_OUT*/DisResult* dres, UInt insn) |
| 7479 | { |
sewardj | b9aff1e | 2014-06-15 21:55:33 +0000 | [diff] [blame] | 7480 | /* 31 28 23 21 16 11 9 4 |
| 7481 | 01 u 11110 sz 11000 opcode 10 n d |
| 7482 | Decode fields: u,sz,opcode |
| 7483 | */ |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7484 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
sewardj | b9aff1e | 2014-06-15 21:55:33 +0000 | [diff] [blame] | 7485 | if (INSN(31,30) != BITS2(0,1) |
| 7486 | || INSN(28,24) != BITS5(1,1,1,1,0) |
| 7487 | || INSN(21,17) != BITS5(1,1,0,0,0) |
| 7488 | || INSN(11,10) != BITS2(1,0)) { |
| 7489 | return False; |
| 7490 | } |
| 7491 | UInt bitU = INSN(29,29); |
| 7492 | UInt sz = INSN(23,22); |
| 7493 | UInt opcode = INSN(16,12); |
| 7494 | UInt nn = INSN(9,5); |
| 7495 | UInt dd = INSN(4,0); |
| 7496 | |
| 7497 | if (bitU == 0 && sz == X11 && opcode == BITS5(1,1,0,1,1)) { |
| 7498 | /* -------- 0,11,11011 ADDP d_2d -------- */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 7499 | IRTemp xy = newTempV128(); |
| 7500 | IRTemp xx = newTempV128(); |
sewardj | b9aff1e | 2014-06-15 21:55:33 +0000 | [diff] [blame] | 7501 | assign(xy, getQReg128(nn)); |
| 7502 | assign(xx, binop(Iop_InterleaveHI64x2, mkexpr(xy), mkexpr(xy))); |
| 7503 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, |
| 7504 | binop(Iop_Add64x2, mkexpr(xy), mkexpr(xx)))); |
| 7505 | DIP("addp d%u, %s.2d\n", dd, nameQReg128(nn)); |
| 7506 | return True; |
| 7507 | } |
| 7508 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7509 | return False; |
| 7510 | # undef INSN |
| 7511 | } |
| 7512 | |
sewardj | fc83d2c | 2014-06-12 10:15:46 +0000 | [diff] [blame] | 7513 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7514 | static |
| 7515 | Bool dis_AdvSIMD_scalar_shift_by_imm(/*MB_OUT*/DisResult* dres, UInt insn) |
| 7516 | { |
| 7517 | /* 31 28 22 18 15 10 9 4 |
| 7518 | 01 u 111110 immh immb opcode 1 n d |
| 7519 | Decode fields: u,immh,opcode |
| 7520 | */ |
| 7521 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 7522 | if (INSN(31,30) != BITS2(0,1) |
| 7523 | || INSN(28,23) != BITS6(1,1,1,1,1,0) || INSN(10,10) != 1) { |
| 7524 | return False; |
| 7525 | } |
| 7526 | UInt bitU = INSN(29,29); |
| 7527 | UInt immh = INSN(22,19); |
| 7528 | UInt immb = INSN(18,16); |
| 7529 | UInt opcode = INSN(15,11); |
| 7530 | UInt nn = INSN(9,5); |
| 7531 | UInt dd = INSN(4,0); |
| 7532 | UInt immhb = (immh << 3) | immb; |
| 7533 | |
sewardj | fc83d2c | 2014-06-12 10:15:46 +0000 | [diff] [blame] | 7534 | if (bitU == 1 && (immh & 8) == 8 && opcode == BITS5(0,0,0,0,0)) { |
| 7535 | /* -------- 1,1xxx,00000 SHR d_d_#imm -------- */ |
| 7536 | UInt sh = 128 - immhb; |
| 7537 | vassert(sh >= 1 && sh <= 64); |
| 7538 | /* Don't generate an out of range IR shift */ |
| 7539 | putQReg128(dd, sh == 64 |
| 7540 | ? mkV128(0x0000) |
| 7541 | : unop(Iop_ZeroHI64ofV128, |
| 7542 | binop(Iop_ShrN64x2, getQReg128(nn), mkU8(sh)))); |
| 7543 | DIP("shr d%u, d%u, #%u\n", dd, nn, sh); |
| 7544 | return True; |
| 7545 | } |
| 7546 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7547 | if (bitU == 0 && (immh & 8) == 8 && opcode == BITS5(0,1,0,1,0)) { |
| 7548 | /* -------- 0,1xxx,01010 SHL d_d_#imm -------- */ |
| 7549 | UInt sh = immhb - 64; |
| 7550 | vassert(sh >= 0 && sh < 64); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 7551 | putQReg128(dd, |
| 7552 | unop(Iop_ZeroHI64ofV128, |
| 7553 | sh == 0 ? getQReg128(nn) |
| 7554 | : binop(Iop_ShlN64x2, getQReg128(nn), mkU8(sh)))); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7555 | DIP("shl d%u, d%u, #%u\n", dd, nn, sh); |
| 7556 | return True; |
| 7557 | } |
| 7558 | |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 7559 | if (bitU == 1 && (immh & 8) == 8 && opcode == BITS5(0,1,0,0,0)) { |
| 7560 | /* -------- 1,1xxx,01000 SRI d_d_#imm -------- */ |
| 7561 | UInt sh = 128 - immhb; |
| 7562 | vassert(sh >= 1 && sh <= 64); |
| 7563 | if (sh == 64) { |
| 7564 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, getQReg128(dd))); |
| 7565 | } else { |
| 7566 | /* sh is in range 1 .. 63 */ |
| 7567 | ULong nmask = (ULong)(((Long)0x8000000000000000ULL) >> (sh-1)); |
| 7568 | IRExpr* nmaskV = binop(Iop_64HLtoV128, mkU64(nmask), mkU64(nmask)); |
| 7569 | IRTemp res = newTempV128(); |
| 7570 | assign(res, binop(Iop_OrV128, |
| 7571 | binop(Iop_AndV128, getQReg128(dd), nmaskV), |
| 7572 | binop(Iop_ShrN64x2, getQReg128(nn), mkU8(sh)))); |
| 7573 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, mkexpr(res))); |
| 7574 | } |
| 7575 | DIP("sri d%u, d%u, #%u\n", dd, nn, sh); |
| 7576 | return True; |
| 7577 | } |
| 7578 | |
| 7579 | if (bitU == 1 && (immh & 8) == 8 && opcode == BITS5(0,1,0,1,0)) { |
| 7580 | /* -------- 1,1xxx,01010 SLI d_d_#imm -------- */ |
| 7581 | UInt sh = immhb - 64; |
| 7582 | vassert(sh >= 0 && sh < 64); |
| 7583 | if (sh == 0) { |
| 7584 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, getQReg128(nn))); |
| 7585 | } else { |
| 7586 | /* sh is in range 1 .. 63 */ |
| 7587 | ULong nmask = (1ULL << sh) - 1; |
| 7588 | IRExpr* nmaskV = binop(Iop_64HLtoV128, mkU64(nmask), mkU64(nmask)); |
| 7589 | IRTemp res = newTempV128(); |
| 7590 | assign(res, binop(Iop_OrV128, |
| 7591 | binop(Iop_AndV128, getQReg128(dd), nmaskV), |
| 7592 | binop(Iop_ShlN64x2, getQReg128(nn), mkU8(sh)))); |
| 7593 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, mkexpr(res))); |
| 7594 | } |
| 7595 | DIP("sli d%u, d%u, #%u\n", dd, nn, sh); |
| 7596 | return True; |
| 7597 | } |
| 7598 | |
sewardj | e741d16 | 2014-08-13 13:10:47 +0000 | [diff] [blame] | 7599 | if (opcode == BITS5(1,0,0,1,0) || opcode == BITS5(1,0,0,1,1) |
| 7600 | || (bitU == 1 |
| 7601 | && (opcode == BITS5(1,0,0,0,0) || opcode == BITS5(1,0,0,0,1)))) { |
| 7602 | /* -------- 0,10010 SQSHRN #imm -------- */ |
| 7603 | /* -------- 1,10010 UQSHRN #imm -------- */ |
| 7604 | /* -------- 0,10011 SQRSHRN #imm -------- */ |
| 7605 | /* -------- 1,10011 UQRSHRN #imm -------- */ |
| 7606 | /* -------- 1,10000 SQSHRUN #imm -------- */ |
| 7607 | /* -------- 1,10001 SQRSHRUN #imm -------- */ |
| 7608 | UInt size = 0; |
| 7609 | UInt shift = 0; |
| 7610 | Bool ok = getLaneInfo_IMMH_IMMB(&shift, &size, immh, immb); |
| 7611 | if (!ok || size == X11) return False; |
| 7612 | vassert(size >= X00 && size <= X10); |
| 7613 | vassert(shift >= 1 && shift <= (8 << size)); |
| 7614 | const HChar* nm = "??"; |
| 7615 | IROp op = Iop_INVALID; |
| 7616 | /* Decide on the name and the operation. */ |
| 7617 | /**/ if (bitU == 0 && opcode == BITS5(1,0,0,1,0)) { |
| 7618 | nm = "sqshrn"; op = mkVecQANDqsarNNARROWSS(size); |
| 7619 | } |
| 7620 | else if (bitU == 1 && opcode == BITS5(1,0,0,1,0)) { |
| 7621 | nm = "uqshrn"; op = mkVecQANDqshrNNARROWUU(size); |
| 7622 | } |
| 7623 | else if (bitU == 0 && opcode == BITS5(1,0,0,1,1)) { |
| 7624 | nm = "sqrshrn"; op = mkVecQANDqrsarNNARROWSS(size); |
| 7625 | } |
| 7626 | else if (bitU == 1 && opcode == BITS5(1,0,0,1,1)) { |
| 7627 | nm = "uqrshrn"; op = mkVecQANDqrshrNNARROWUU(size); |
| 7628 | } |
| 7629 | else if (bitU == 1 && opcode == BITS5(1,0,0,0,0)) { |
| 7630 | nm = "sqshrun"; op = mkVecQANDqsarNNARROWSU(size); |
| 7631 | } |
| 7632 | else if (bitU == 1 && opcode == BITS5(1,0,0,0,1)) { |
| 7633 | nm = "sqrshrun"; op = mkVecQANDqrsarNNARROWSU(size); |
| 7634 | } |
| 7635 | else vassert(0); |
| 7636 | /* Compute the result (Q, shifted value) pair. */ |
| 7637 | IRTemp src128 = math_ZERO_ALL_EXCEPT_LOWEST_LANE(size+1, getQReg128(nn)); |
| 7638 | IRTemp pair = newTempV128(); |
| 7639 | assign(pair, binop(op, mkexpr(src128), mkU8(shift))); |
| 7640 | /* Update the result reg */ |
| 7641 | IRTemp res64in128 = newTempV128(); |
| 7642 | assign(res64in128, unop(Iop_ZeroHI64ofV128, mkexpr(pair))); |
| 7643 | putQReg128(dd, mkexpr(res64in128)); |
| 7644 | /* Update the Q flag. */ |
| 7645 | IRTemp q64q64 = newTempV128(); |
| 7646 | assign(q64q64, binop(Iop_InterleaveHI64x2, mkexpr(pair), mkexpr(pair))); |
| 7647 | IRTemp z128 = newTempV128(); |
| 7648 | assign(z128, mkV128(0x0000)); |
| 7649 | updateQCFLAGwithDifference(q64q64, z128); |
| 7650 | /* */ |
| 7651 | const HChar arrNarrow = "bhsd"[size]; |
| 7652 | const HChar arrWide = "bhsd"[size+1]; |
| 7653 | DIP("%s %c%u, %c%u, #%u\n", nm, arrNarrow, dd, arrWide, nn, shift); |
| 7654 | return True; |
| 7655 | } |
| 7656 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7657 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 7658 | return False; |
| 7659 | # undef INSN |
| 7660 | } |
| 7661 | |
sewardj | fc83d2c | 2014-06-12 10:15:46 +0000 | [diff] [blame] | 7662 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7663 | static |
| 7664 | Bool dis_AdvSIMD_scalar_three_different(/*MB_OUT*/DisResult* dres, UInt insn) |
| 7665 | { |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 7666 | /* 31 29 28 23 21 20 15 11 9 4 |
| 7667 | 01 U 11110 size 1 m opcode 00 n d |
| 7668 | Decode fields: u,opcode |
| 7669 | */ |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7670 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 7671 | if (INSN(31,30) != BITS2(0,1) |
| 7672 | || INSN(28,24) != BITS5(1,1,1,1,0) |
| 7673 | || INSN(21,21) != 1 |
| 7674 | || INSN(11,10) != BITS2(0,0)) { |
| 7675 | return False; |
| 7676 | } |
| 7677 | UInt bitU = INSN(29,29); |
| 7678 | UInt size = INSN(23,22); |
| 7679 | UInt mm = INSN(20,16); |
| 7680 | UInt opcode = INSN(15,12); |
| 7681 | UInt nn = INSN(9,5); |
| 7682 | UInt dd = INSN(4,0); |
| 7683 | vassert(size < 4); |
| 7684 | |
| 7685 | if (bitU == 0 |
| 7686 | && (opcode == BITS4(1,1,0,1) |
| 7687 | || opcode == BITS4(1,0,0,1) || opcode == BITS4(1,0,1,1))) { |
| 7688 | /* -------- 0,1101 SQDMULL -------- */ // 0 (ks) |
| 7689 | /* -------- 0,1001 SQDMLAL -------- */ // 1 |
| 7690 | /* -------- 0,1011 SQDMLSL -------- */ // 2 |
| 7691 | /* Widens, and size refers to the narrowed lanes. */ |
| 7692 | UInt ks = 3; |
| 7693 | switch (opcode) { |
| 7694 | case BITS4(1,1,0,1): ks = 0; break; |
| 7695 | case BITS4(1,0,0,1): ks = 1; break; |
| 7696 | case BITS4(1,0,1,1): ks = 2; break; |
| 7697 | default: vassert(0); |
| 7698 | } |
| 7699 | vassert(ks >= 0 && ks <= 2); |
| 7700 | if (size == X00 || size == X11) return False; |
| 7701 | vassert(size <= 2); |
| 7702 | IRTemp vecN, vecM, vecD, res, sat1q, sat1n, sat2q, sat2n; |
| 7703 | vecN = vecM = vecD = res = sat1q = sat1n = sat2q = sat2n = IRTemp_INVALID; |
| 7704 | newTempsV128_3(&vecN, &vecM, &vecD); |
| 7705 | assign(vecN, getQReg128(nn)); |
| 7706 | assign(vecM, getQReg128(mm)); |
| 7707 | assign(vecD, getQReg128(dd)); |
| 7708 | math_SQDMULL_ACC(&res, &sat1q, &sat1n, &sat2q, &sat2n, |
| 7709 | False/*!is2*/, size, "mas"[ks], |
| 7710 | vecN, vecM, ks == 0 ? IRTemp_INVALID : vecD); |
| 7711 | IROp opZHI = mkVecZEROHIxxOFV128(size+1); |
| 7712 | putQReg128(dd, unop(opZHI, mkexpr(res))); |
| 7713 | vassert(sat1q != IRTemp_INVALID && sat1n != IRTemp_INVALID); |
| 7714 | updateQCFLAGwithDifferenceZHI(sat1q, sat1n, opZHI); |
| 7715 | if (sat2q != IRTemp_INVALID || sat2n != IRTemp_INVALID) { |
| 7716 | updateQCFLAGwithDifferenceZHI(sat2q, sat2n, opZHI); |
| 7717 | } |
| 7718 | const HChar* nm = ks == 0 ? "sqdmull" |
| 7719 | : (ks == 1 ? "sqdmlal" : "sqdmlsl"); |
| 7720 | const HChar arrNarrow = "bhsd"[size]; |
| 7721 | const HChar arrWide = "bhsd"[size+1]; |
| 7722 | DIP("%s %c%d, %c%d, %c%d\n", |
| 7723 | nm, arrWide, dd, arrNarrow, nn, arrNarrow, mm); |
| 7724 | return True; |
| 7725 | } |
| 7726 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7727 | return False; |
| 7728 | # undef INSN |
| 7729 | } |
| 7730 | |
| 7731 | |
| 7732 | static |
| 7733 | Bool dis_AdvSIMD_scalar_three_same(/*MB_OUT*/DisResult* dres, UInt insn) |
| 7734 | { |
| 7735 | /* 31 29 28 23 21 20 15 10 9 4 |
| 7736 | 01 U 11110 size 1 m opcode 1 n d |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 7737 | Decode fields: u,size,opcode |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7738 | */ |
| 7739 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 7740 | if (INSN(31,30) != BITS2(0,1) |
| 7741 | || INSN(28,24) != BITS5(1,1,1,1,0) |
| 7742 | || INSN(21,21) != 1 |
| 7743 | || INSN(10,10) != 1) { |
| 7744 | return False; |
| 7745 | } |
| 7746 | UInt bitU = INSN(29,29); |
| 7747 | UInt size = INSN(23,22); |
| 7748 | UInt mm = INSN(20,16); |
| 7749 | UInt opcode = INSN(15,11); |
| 7750 | UInt nn = INSN(9,5); |
| 7751 | UInt dd = INSN(4,0); |
| 7752 | vassert(size < 4); |
| 7753 | |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 7754 | if (opcode == BITS5(0,0,0,0,1) || opcode == BITS5(0,0,1,0,1)) { |
| 7755 | /* -------- 0,xx,00001 SQADD std4_std4_std4 -------- */ |
| 7756 | /* -------- 1,xx,00001 UQADD std4_std4_std4 -------- */ |
| 7757 | /* -------- 0,xx,00101 SQSUB std4_std4_std4 -------- */ |
| 7758 | /* -------- 1,xx,00101 UQSUB std4_std4_std4 -------- */ |
| 7759 | Bool isADD = opcode == BITS5(0,0,0,0,1); |
| 7760 | Bool isU = bitU == 1; |
| 7761 | IROp qop = Iop_INVALID; |
| 7762 | IROp nop = Iop_INVALID; |
| 7763 | if (isADD) { |
| 7764 | qop = isU ? mkVecQADDU(size) : mkVecQADDS(size); |
| 7765 | nop = mkVecADD(size); |
| 7766 | } else { |
| 7767 | qop = isU ? mkVecQSUBU(size) : mkVecQSUBS(size); |
| 7768 | nop = mkVecSUB(size); |
| 7769 | } |
| 7770 | IRTemp argL = newTempV128(); |
| 7771 | IRTemp argR = newTempV128(); |
| 7772 | IRTemp qres = newTempV128(); |
| 7773 | IRTemp nres = newTempV128(); |
| 7774 | assign(argL, getQReg128(nn)); |
| 7775 | assign(argR, getQReg128(mm)); |
| 7776 | assign(qres, mkexpr(math_ZERO_ALL_EXCEPT_LOWEST_LANE( |
sewardj | 257e99f | 2014-08-03 12:45:19 +0000 | [diff] [blame] | 7777 | size, binop(qop, mkexpr(argL), mkexpr(argR))))); |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 7778 | assign(nres, mkexpr(math_ZERO_ALL_EXCEPT_LOWEST_LANE( |
sewardj | 257e99f | 2014-08-03 12:45:19 +0000 | [diff] [blame] | 7779 | size, binop(nop, mkexpr(argL), mkexpr(argR))))); |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 7780 | putQReg128(dd, mkexpr(qres)); |
| 7781 | updateQCFLAGwithDifference(qres, nres); |
| 7782 | const HChar* nm = isADD ? (isU ? "uqadd" : "sqadd") |
| 7783 | : (isU ? "uqsub" : "sqsub"); |
| 7784 | const HChar arr = "bhsd"[size]; |
sewardj | 1297218 | 2014-08-04 08:09:47 +0000 | [diff] [blame] | 7785 | DIP("%s %c%u, %c%u, %c%u\n", nm, arr, dd, arr, nn, arr, mm); |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 7786 | return True; |
| 7787 | } |
| 7788 | |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7789 | if (size == X11 && opcode == BITS5(0,0,1,1,0)) { |
| 7790 | /* -------- 0,11,00110 CMGT d_d_d -------- */ // >s |
| 7791 | /* -------- 1,11,00110 CMHI d_d_d -------- */ // >u |
| 7792 | Bool isGT = bitU == 0; |
| 7793 | IRExpr* argL = getQReg128(nn); |
| 7794 | IRExpr* argR = getQReg128(mm); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 7795 | IRTemp res = newTempV128(); |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7796 | assign(res, |
| 7797 | isGT ? binop(Iop_CmpGT64Sx2, argL, argR) |
| 7798 | : binop(Iop_CmpGT64Ux2, argL, argR)); |
| 7799 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, mkexpr(res))); |
| 7800 | DIP("%s %s, %s, %s\n",isGT ? "cmgt" : "cmhi", |
| 7801 | nameQRegLO(dd, Ity_I64), |
| 7802 | nameQRegLO(nn, Ity_I64), nameQRegLO(mm, Ity_I64)); |
| 7803 | return True; |
| 7804 | } |
| 7805 | |
| 7806 | if (size == X11 && opcode == BITS5(0,0,1,1,1)) { |
| 7807 | /* -------- 0,11,00111 CMGE d_d_d -------- */ // >=s |
| 7808 | /* -------- 1,11,00111 CMHS d_d_d -------- */ // >=u |
| 7809 | Bool isGE = bitU == 0; |
| 7810 | IRExpr* argL = getQReg128(nn); |
| 7811 | IRExpr* argR = getQReg128(mm); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 7812 | IRTemp res = newTempV128(); |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7813 | assign(res, |
| 7814 | isGE ? unop(Iop_NotV128, binop(Iop_CmpGT64Sx2, argR, argL)) |
| 7815 | : unop(Iop_NotV128, binop(Iop_CmpGT64Ux2, argR, argL))); |
| 7816 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, mkexpr(res))); |
| 7817 | DIP("%s %s, %s, %s\n", isGE ? "cmge" : "cmhs", |
| 7818 | nameQRegLO(dd, Ity_I64), |
| 7819 | nameQRegLO(nn, Ity_I64), nameQRegLO(mm, Ity_I64)); |
| 7820 | return True; |
| 7821 | } |
| 7822 | |
sewardj | 1297218 | 2014-08-04 08:09:47 +0000 | [diff] [blame] | 7823 | if (opcode == BITS5(0,1,0,0,1) || opcode == BITS5(0,1,0,1,1)) { |
| 7824 | /* -------- 0,xx,01001 SQSHL std4_std4_std4 -------- */ |
| 7825 | /* -------- 0,xx,01011 SQRSHL std4_std4_std4 -------- */ |
| 7826 | /* -------- 1,xx,01001 UQSHL std4_std4_std4 -------- */ |
| 7827 | /* -------- 1,xx,01011 UQRSHL std4_std4_std4 -------- */ |
| 7828 | Bool isU = bitU == 1; |
| 7829 | Bool isR = opcode == BITS5(0,1,0,1,1); |
| 7830 | IROp op = isR ? (isU ? mkVecQANDUQRSH(size) : mkVecQANDSQRSH(size)) |
| 7831 | : (isU ? mkVecQANDUQSH(size) : mkVecQANDSQSH(size)); |
| 7832 | /* This is a bit tricky. Since we're only interested in the lowest |
| 7833 | lane of the result, we zero out all the rest in the operands, so |
| 7834 | as to ensure that other lanes don't pollute the returned Q value. |
| 7835 | This works because it means, for the lanes we don't care about, we |
| 7836 | are shifting zero by zero, which can never saturate. */ |
| 7837 | IRTemp res256 = newTemp(Ity_V256); |
| 7838 | IRTemp resSH = newTempV128(); |
| 7839 | IRTemp resQ = newTempV128(); |
| 7840 | IRTemp zero = newTempV128(); |
| 7841 | assign( |
| 7842 | res256, |
| 7843 | binop(op, |
| 7844 | mkexpr(math_ZERO_ALL_EXCEPT_LOWEST_LANE(size, getQReg128(nn))), |
| 7845 | mkexpr(math_ZERO_ALL_EXCEPT_LOWEST_LANE(size, getQReg128(mm))))); |
| 7846 | assign(resSH, unop(Iop_V256toV128_0, mkexpr(res256))); |
| 7847 | assign(resQ, unop(Iop_V256toV128_1, mkexpr(res256))); |
| 7848 | assign(zero, mkV128(0x0000)); |
| 7849 | putQReg128(dd, mkexpr(resSH)); |
| 7850 | updateQCFLAGwithDifference(resQ, zero); |
| 7851 | const HChar* nm = isR ? (isU ? "uqrshl" : "sqrshl") |
| 7852 | : (isU ? "uqshl" : "sqshl"); |
| 7853 | const HChar arr = "bhsd"[size]; |
| 7854 | DIP("%s %c%u, %c%u, %c%u\n", nm, arr, dd, arr, nn, arr, mm); |
| 7855 | return True; |
| 7856 | } |
| 7857 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7858 | if (size == X11 && opcode == BITS5(1,0,0,0,0)) { |
| 7859 | /* -------- 0,11,10000 ADD d_d_d -------- */ |
| 7860 | /* -------- 1,11,10000 SUB d_d_d -------- */ |
| 7861 | Bool isSUB = bitU == 1; |
| 7862 | IRTemp res = newTemp(Ity_I64); |
| 7863 | assign(res, binop(isSUB ? Iop_Sub64 : Iop_Add64, |
| 7864 | getQRegLane(nn, 0, Ity_I64), |
| 7865 | getQRegLane(mm, 0, Ity_I64))); |
| 7866 | putQRegLane(dd, 0, mkexpr(res)); |
| 7867 | putQRegLane(dd, 1, mkU64(0)); |
| 7868 | DIP("%s %s, %s, %s\n", isSUB ? "sub" : "add", |
| 7869 | nameQRegLO(dd, Ity_I64), |
| 7870 | nameQRegLO(nn, Ity_I64), nameQRegLO(mm, Ity_I64)); |
| 7871 | return True; |
| 7872 | } |
| 7873 | |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7874 | if (size == X11 && opcode == BITS5(1,0,0,0,1)) { |
| 7875 | /* -------- 0,11,10001 CMTST d_d_d -------- */ // &, != 0 |
| 7876 | /* -------- 1,11,10001 CMEQ d_d_d -------- */ // == |
| 7877 | Bool isEQ = bitU == 1; |
| 7878 | IRExpr* argL = getQReg128(nn); |
| 7879 | IRExpr* argR = getQReg128(mm); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 7880 | IRTemp res = newTempV128(); |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7881 | assign(res, |
| 7882 | isEQ ? binop(Iop_CmpEQ64x2, argL, argR) |
| 7883 | : unop(Iop_NotV128, binop(Iop_CmpEQ64x2, |
| 7884 | binop(Iop_AndV128, argL, argR), |
| 7885 | mkV128(0x0000)))); |
| 7886 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, mkexpr(res))); |
| 7887 | DIP("%s %s, %s, %s\n", isEQ ? "cmeq" : "cmtst", |
| 7888 | nameQRegLO(dd, Ity_I64), |
| 7889 | nameQRegLO(nn, Ity_I64), nameQRegLO(mm, Ity_I64)); |
| 7890 | return True; |
| 7891 | } |
| 7892 | |
sewardj | 257e99f | 2014-08-03 12:45:19 +0000 | [diff] [blame] | 7893 | if (opcode == BITS5(1,0,1,1,0)) { |
| 7894 | /* -------- 0,xx,10110 SQDMULH s and h variants only -------- */ |
| 7895 | /* -------- 1,xx,10110 SQRDMULH s and h variants only -------- */ |
| 7896 | if (size == X00 || size == X11) return False; |
| 7897 | Bool isR = bitU == 1; |
| 7898 | IRTemp res, sat1q, sat1n, vN, vM; |
| 7899 | res = sat1q = sat1n = vN = vM = IRTemp_INVALID; |
| 7900 | newTempsV128_2(&vN, &vM); |
| 7901 | assign(vN, getQReg128(nn)); |
| 7902 | assign(vM, getQReg128(mm)); |
| 7903 | math_SQDMULH(&res, &sat1q, &sat1n, isR, size, vN, vM); |
| 7904 | putQReg128(dd, |
| 7905 | mkexpr(math_ZERO_ALL_EXCEPT_LOWEST_LANE(size, mkexpr(res)))); |
| 7906 | updateQCFLAGwithDifference( |
| 7907 | math_ZERO_ALL_EXCEPT_LOWEST_LANE(size, mkexpr(sat1q)), |
| 7908 | math_ZERO_ALL_EXCEPT_LOWEST_LANE(size, mkexpr(sat1n))); |
| 7909 | const HChar arr = "bhsd"[size]; |
| 7910 | const HChar* nm = isR ? "sqrdmulh" : "sqdmulh"; |
| 7911 | DIP("%s %c%d, %c%d, %c%d\n", nm, arr, dd, arr, nn, arr, mm); |
| 7912 | return True; |
| 7913 | } |
| 7914 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7915 | if (bitU == 1 && size >= X10 && opcode == BITS5(1,1,0,1,0)) { |
| 7916 | /* -------- 1,1x,11010 FABD d_d_d, s_s_s -------- */ |
| 7917 | IRType ity = size == X11 ? Ity_F64 : Ity_F32; |
| 7918 | IRTemp res = newTemp(ity); |
| 7919 | assign(res, unop(mkABSF(ity), |
| 7920 | triop(mkSUBF(ity), |
| 7921 | mkexpr(mk_get_IR_rounding_mode()), |
| 7922 | getQRegLO(nn,ity), getQRegLO(mm,ity)))); |
| 7923 | putQReg128(dd, mkV128(0x0000)); |
| 7924 | putQRegLO(dd, mkexpr(res)); |
| 7925 | DIP("fabd %s, %s, %s\n", |
| 7926 | nameQRegLO(dd, ity), nameQRegLO(nn, ity), nameQRegLO(mm, ity)); |
| 7927 | return True; |
| 7928 | } |
| 7929 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7930 | return False; |
| 7931 | # undef INSN |
| 7932 | } |
| 7933 | |
| 7934 | |
| 7935 | static |
| 7936 | Bool dis_AdvSIMD_scalar_two_reg_misc(/*MB_OUT*/DisResult* dres, UInt insn) |
| 7937 | { |
| 7938 | /* 31 29 28 23 21 16 11 9 4 |
| 7939 | 01 U 11110 size 10000 opcode 10 n d |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 7940 | Decode fields: u,size,opcode |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 7941 | */ |
| 7942 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 7943 | if (INSN(31,30) != BITS2(0,1) |
| 7944 | || INSN(28,24) != BITS5(1,1,1,1,0) |
| 7945 | || INSN(21,17) != BITS5(1,0,0,0,0) |
| 7946 | || INSN(11,10) != BITS2(1,0)) { |
| 7947 | return False; |
| 7948 | } |
| 7949 | UInt bitU = INSN(29,29); |
| 7950 | UInt size = INSN(23,22); |
| 7951 | UInt opcode = INSN(16,12); |
| 7952 | UInt nn = INSN(9,5); |
| 7953 | UInt dd = INSN(4,0); |
| 7954 | vassert(size < 4); |
| 7955 | |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 7956 | if (opcode == BITS5(0,0,1,1,1)) { |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 7957 | /* -------- 0,xx,00111 SQABS std4_std4 -------- */ |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 7958 | /* -------- 1,xx,00111 SQNEG std4_std4 -------- */ |
| 7959 | Bool isNEG = bitU == 1; |
| 7960 | IRTemp qresFW = IRTemp_INVALID, nresFW = IRTemp_INVALID; |
| 7961 | (isNEG ? math_SQNEG : math_SQABS)( &qresFW, &nresFW, |
| 7962 | getQReg128(nn), size ); |
sewardj | 257e99f | 2014-08-03 12:45:19 +0000 | [diff] [blame] | 7963 | IRTemp qres = math_ZERO_ALL_EXCEPT_LOWEST_LANE(size, mkexpr(qresFW)); |
| 7964 | IRTemp nres = math_ZERO_ALL_EXCEPT_LOWEST_LANE(size, mkexpr(nresFW)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 7965 | putQReg128(dd, mkexpr(qres)); |
| 7966 | updateQCFLAGwithDifference(qres, nres); |
| 7967 | const HChar arr = "bhsd"[size]; |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 7968 | DIP("%s %c%u, %c%u\n", isNEG ? "sqneg" : "sqabs", arr, dd, arr, nn); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 7969 | return True; |
| 7970 | } |
| 7971 | |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7972 | if (size == X11 && opcode == BITS5(0,1,0,0,0)) { |
| 7973 | /* -------- 0,11,01000: CMGT d_d_#0 -------- */ // >s 0 |
| 7974 | /* -------- 1,11,01000: CMGE d_d_#0 -------- */ // >=s 0 |
| 7975 | Bool isGT = bitU == 0; |
| 7976 | IRExpr* argL = getQReg128(nn); |
| 7977 | IRExpr* argR = mkV128(0x0000); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 7978 | IRTemp res = newTempV128(); |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7979 | assign(res, isGT ? binop(Iop_CmpGT64Sx2, argL, argR) |
| 7980 | : unop(Iop_NotV128, binop(Iop_CmpGT64Sx2, argR, argL))); |
| 7981 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, mkexpr(res))); |
| 7982 | DIP("cm%s d%u, d%u, #0\n", isGT ? "gt" : "ge", dd, nn); |
| 7983 | return True; |
| 7984 | } |
| 7985 | |
| 7986 | if (size == X11 && opcode == BITS5(0,1,0,0,1)) { |
| 7987 | /* -------- 0,11,01001: CMEQ d_d_#0 -------- */ // == 0 |
| 7988 | /* -------- 1,11,01001: CMLE d_d_#0 -------- */ // <=s 0 |
| 7989 | Bool isEQ = bitU == 0; |
| 7990 | IRExpr* argL = getQReg128(nn); |
| 7991 | IRExpr* argR = mkV128(0x0000); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 7992 | IRTemp res = newTempV128(); |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 7993 | assign(res, isEQ ? binop(Iop_CmpEQ64x2, argL, argR) |
| 7994 | : unop(Iop_NotV128, |
| 7995 | binop(Iop_CmpGT64Sx2, argL, argR))); |
| 7996 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, mkexpr(res))); |
| 7997 | DIP("cm%s d%u, d%u, #0\n", isEQ ? "eq" : "le", dd, nn); |
| 7998 | return True; |
| 7999 | } |
| 8000 | |
| 8001 | if (bitU == 0 && size == X11 && opcode == BITS5(0,1,0,1,0)) { |
| 8002 | /* -------- 0,11,01010: CMLT d_d_#0 -------- */ // <s 0 |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8003 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 8004 | binop(Iop_CmpGT64Sx2, mkV128(0x0000), |
| 8005 | getQReg128(nn)))); |
| 8006 | DIP("cm%s d%u, d%u, #0\n", "lt", dd, nn); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8007 | return True; |
| 8008 | } |
| 8009 | |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 8010 | if (bitU == 0 && size == X11 && opcode == BITS5(0,1,0,1,1)) { |
| 8011 | /* -------- 0,11,01011 ABS d_d -------- */ |
| 8012 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, |
| 8013 | unop(Iop_Abs64x2, getQReg128(nn)))); |
| 8014 | DIP("abs d%u, d%u\n", dd, nn); |
| 8015 | return True; |
| 8016 | } |
| 8017 | |
| 8018 | if (bitU == 1 && size == X11 && opcode == BITS5(0,1,0,1,1)) { |
| 8019 | /* -------- 1,11,01011 NEG d_d -------- */ |
| 8020 | putQReg128(dd, unop(Iop_ZeroHI64ofV128, |
| 8021 | binop(Iop_Sub64x2, mkV128(0x0000), getQReg128(nn)))); |
| 8022 | DIP("neg d%u, d%u\n", dd, nn); |
| 8023 | return True; |
| 8024 | } |
| 8025 | |
sewardj | ecedd98 | 2014-08-11 14:02:47 +0000 | [diff] [blame] | 8026 | if (opcode == BITS5(1,0,1,0,0) |
| 8027 | || (bitU == 1 && opcode == BITS5(1,0,0,1,0))) { |
| 8028 | /* -------- 0,xx,10100: SQXTN -------- */ |
| 8029 | /* -------- 1,xx,10100: UQXTN -------- */ |
| 8030 | /* -------- 1,xx,10010: SQXTUN -------- */ |
| 8031 | if (size == X11) return False; |
| 8032 | vassert(size < 3); |
| 8033 | IROp opN = Iop_INVALID; |
| 8034 | Bool zWiden = True; |
| 8035 | const HChar* nm = "??"; |
| 8036 | /**/ if (bitU == 0 && opcode == BITS5(1,0,1,0,0)) { |
| 8037 | opN = mkVecQNARROWUNSS(size); nm = "sqxtn"; zWiden = False; |
| 8038 | } |
| 8039 | else if (bitU == 1 && opcode == BITS5(1,0,1,0,0)) { |
| 8040 | opN = mkVecQNARROWUNUU(size); nm = "uqxtn"; |
| 8041 | } |
| 8042 | else if (bitU == 1 && opcode == BITS5(1,0,0,1,0)) { |
| 8043 | opN = mkVecQNARROWUNSU(size); nm = "sqxtun"; |
| 8044 | } |
| 8045 | else vassert(0); |
| 8046 | IRTemp src = math_ZERO_ALL_EXCEPT_LOWEST_LANE( |
| 8047 | size+1, getQReg128(nn)); |
| 8048 | IRTemp resN = math_ZERO_ALL_EXCEPT_LOWEST_LANE( |
| 8049 | size, unop(Iop_64UtoV128, unop(opN, mkexpr(src)))); |
| 8050 | putQReg128(dd, mkexpr(resN)); |
| 8051 | /* This widens zero lanes to zero, and compares it against zero, so all |
| 8052 | of the non-participating lanes make no contribution to the |
| 8053 | Q flag state. */ |
| 8054 | IRTemp resW = math_WIDEN_LO_OR_HI_LANES(zWiden, False/*!fromUpperHalf*/, |
| 8055 | size, mkexpr(resN)); |
| 8056 | updateQCFLAGwithDifference(src, resW); |
| 8057 | const HChar arrNarrow = "bhsd"[size]; |
| 8058 | const HChar arrWide = "bhsd"[size+1]; |
| 8059 | DIP("%s %c%u, %c%u\n", nm, arrNarrow, dd, arrWide, nn); |
| 8060 | return True; |
| 8061 | } |
| 8062 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8063 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 8064 | return False; |
| 8065 | # undef INSN |
| 8066 | } |
| 8067 | |
sewardj | fc83d2c | 2014-06-12 10:15:46 +0000 | [diff] [blame] | 8068 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8069 | static |
| 8070 | Bool dis_AdvSIMD_scalar_x_indexed_element(/*MB_OUT*/DisResult* dres, UInt insn) |
| 8071 | { |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 8072 | /* 31 28 23 21 20 19 15 11 9 4 |
| 8073 | 01 U 11111 size L M m opcode H 0 n d |
| 8074 | Decode fields are: u,size,opcode |
| 8075 | M is really part of the mm register number. Individual |
| 8076 | cases need to inspect L and H though. |
| 8077 | */ |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8078 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 8079 | if (INSN(31,30) != BITS2(0,1) |
| 8080 | || INSN(28,24) != BITS5(1,1,1,1,1) || INSN(10,10) !=0) { |
| 8081 | return False; |
| 8082 | } |
| 8083 | UInt bitU = INSN(29,29); |
| 8084 | UInt size = INSN(23,22); |
| 8085 | UInt bitL = INSN(21,21); |
| 8086 | UInt bitM = INSN(20,20); |
| 8087 | UInt mmLO4 = INSN(19,16); |
| 8088 | UInt opcode = INSN(15,12); |
| 8089 | UInt bitH = INSN(11,11); |
| 8090 | UInt nn = INSN(9,5); |
| 8091 | UInt dd = INSN(4,0); |
| 8092 | vassert(size < 4); |
| 8093 | vassert(bitH < 2 && bitM < 2 && bitL < 2); |
| 8094 | |
| 8095 | if (bitU == 0 |
| 8096 | && (opcode == BITS4(1,0,1,1) |
| 8097 | || opcode == BITS4(0,0,1,1) || opcode == BITS4(0,1,1,1))) { |
| 8098 | /* -------- 0,xx,1011 SQDMULL s/h variants only -------- */ // 0 (ks) |
| 8099 | /* -------- 0,xx,0011 SQDMLAL s/h variants only -------- */ // 1 |
| 8100 | /* -------- 0,xx,0111 SQDMLSL s/h variants only -------- */ // 2 |
| 8101 | /* Widens, and size refers to the narrowed lanes. */ |
| 8102 | UInt ks = 3; |
| 8103 | switch (opcode) { |
| 8104 | case BITS4(1,0,1,1): ks = 0; break; |
| 8105 | case BITS4(0,0,1,1): ks = 1; break; |
| 8106 | case BITS4(0,1,1,1): ks = 2; break; |
| 8107 | default: vassert(0); |
| 8108 | } |
| 8109 | vassert(ks >= 0 && ks <= 2); |
| 8110 | UInt mm = 32; // invalid |
| 8111 | UInt ix = 16; // invalid |
| 8112 | switch (size) { |
| 8113 | case X00: |
| 8114 | return False; // h_b_b[] case is not allowed |
| 8115 | case X01: |
| 8116 | mm = mmLO4; ix = (bitH << 2) | (bitL << 1) | (bitM << 0); break; |
| 8117 | case X10: |
| 8118 | mm = (bitM << 4) | mmLO4; ix = (bitH << 1) | (bitL << 0); break; |
| 8119 | case X11: |
| 8120 | return False; // q_d_d[] case is not allowed |
| 8121 | default: |
| 8122 | vassert(0); |
| 8123 | } |
| 8124 | vassert(mm < 32 && ix < 16); |
| 8125 | IRTemp vecN, vecD, res, sat1q, sat1n, sat2q, sat2n; |
| 8126 | vecN = vecD = res = sat1q = sat1n = sat2q = sat2n = IRTemp_INVALID; |
| 8127 | newTempsV128_2(&vecN, &vecD); |
| 8128 | assign(vecN, getQReg128(nn)); |
| 8129 | IRTemp vecM = math_DUP_VEC_ELEM(getQReg128(mm), size, ix); |
| 8130 | assign(vecD, getQReg128(dd)); |
| 8131 | math_SQDMULL_ACC(&res, &sat1q, &sat1n, &sat2q, &sat2n, |
| 8132 | False/*!is2*/, size, "mas"[ks], |
| 8133 | vecN, vecM, ks == 0 ? IRTemp_INVALID : vecD); |
| 8134 | IROp opZHI = mkVecZEROHIxxOFV128(size+1); |
| 8135 | putQReg128(dd, unop(opZHI, mkexpr(res))); |
| 8136 | vassert(sat1q != IRTemp_INVALID && sat1n != IRTemp_INVALID); |
| 8137 | updateQCFLAGwithDifferenceZHI(sat1q, sat1n, opZHI); |
| 8138 | if (sat2q != IRTemp_INVALID || sat2n != IRTemp_INVALID) { |
| 8139 | updateQCFLAGwithDifferenceZHI(sat2q, sat2n, opZHI); |
| 8140 | } |
| 8141 | const HChar* nm = ks == 0 ? "sqmull" |
| 8142 | : (ks == 1 ? "sqdmlal" : "sqdmlsl"); |
| 8143 | const HChar arrNarrow = "bhsd"[size]; |
| 8144 | const HChar arrWide = "bhsd"[size+1]; |
| 8145 | DIP("%s %c%d, %c%d, v%d.%c[%u]\n", |
| 8146 | nm, arrWide, dd, arrNarrow, nn, dd, arrNarrow, ix); |
| 8147 | return True; |
| 8148 | } |
| 8149 | |
sewardj | 257e99f | 2014-08-03 12:45:19 +0000 | [diff] [blame] | 8150 | if (opcode == BITS4(1,1,0,0) || opcode == BITS4(1,1,0,1)) { |
| 8151 | /* -------- 0,xx,1100 SQDMULH s and h variants only -------- */ |
| 8152 | /* -------- 0,xx,1101 SQRDMULH s and h variants only -------- */ |
| 8153 | UInt mm = 32; // invalid |
| 8154 | UInt ix = 16; // invalid |
| 8155 | switch (size) { |
| 8156 | case X00: |
| 8157 | return False; // b case is not allowed |
| 8158 | case X01: |
| 8159 | mm = mmLO4; ix = (bitH << 2) | (bitL << 1) | (bitM << 0); break; |
| 8160 | case X10: |
| 8161 | mm = (bitM << 4) | mmLO4; ix = (bitH << 1) | (bitL << 0); break; |
| 8162 | case X11: |
| 8163 | return False; // q case is not allowed |
| 8164 | default: |
| 8165 | vassert(0); |
| 8166 | } |
| 8167 | vassert(mm < 32 && ix < 16); |
| 8168 | Bool isR = opcode == BITS4(1,1,0,1); |
| 8169 | IRTemp res, sat1q, sat1n, vN, vM; |
| 8170 | res = sat1q = sat1n = vN = vM = IRTemp_INVALID; |
| 8171 | vN = newTempV128(); |
| 8172 | assign(vN, getQReg128(nn)); |
| 8173 | vM = math_DUP_VEC_ELEM(getQReg128(mm), size, ix); |
| 8174 | math_SQDMULH(&res, &sat1q, &sat1n, isR, size, vN, vM); |
| 8175 | IROp opZHI = mkVecZEROHIxxOFV128(size); |
| 8176 | putQReg128(dd, unop(opZHI, mkexpr(res))); |
| 8177 | updateQCFLAGwithDifferenceZHI(sat1q, sat1n, opZHI); |
| 8178 | const HChar* nm = isR ? "sqrdmulh" : "sqdmulh"; |
| 8179 | HChar ch = size == X01 ? 'h' : 's'; |
| 8180 | DIP("%s %c%d, %c%d, v%d.%c[%u]\n", nm, ch, dd, ch, nn, ch, dd, ix); |
| 8181 | return True; |
| 8182 | } |
| 8183 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8184 | return False; |
| 8185 | # undef INSN |
| 8186 | } |
| 8187 | |
sewardj | fc83d2c | 2014-06-12 10:15:46 +0000 | [diff] [blame] | 8188 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8189 | static |
| 8190 | Bool dis_AdvSIMD_shift_by_immediate(/*MB_OUT*/DisResult* dres, UInt insn) |
| 8191 | { |
| 8192 | /* 31 28 22 18 15 10 9 4 |
| 8193 | 0 q u 011110 immh immb opcode 1 n d |
| 8194 | Decode fields: u,opcode |
| 8195 | */ |
| 8196 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 8197 | if (INSN(31,31) != 0 |
| 8198 | || INSN(28,23) != BITS6(0,1,1,1,1,0) || INSN(10,10) != 1) { |
| 8199 | return False; |
| 8200 | } |
| 8201 | UInt bitQ = INSN(30,30); |
| 8202 | UInt bitU = INSN(29,29); |
| 8203 | UInt immh = INSN(22,19); |
| 8204 | UInt immb = INSN(18,16); |
| 8205 | UInt opcode = INSN(15,11); |
| 8206 | UInt nn = INSN(9,5); |
| 8207 | UInt dd = INSN(4,0); |
| 8208 | |
| 8209 | if (opcode == BITS5(0,0,0,0,0)) { |
| 8210 | /* -------- 0,00000 SSHR std7_std7_#imm -------- */ |
| 8211 | /* -------- 1,00000 USHR std7_std7_#imm -------- */ |
| 8212 | /* laneTy, shift = case immh:immb of |
| 8213 | 0001:xxx -> B, SHR:8-xxx |
| 8214 | 001x:xxx -> H, SHR:16-xxxx |
| 8215 | 01xx:xxx -> S, SHR:32-xxxxx |
| 8216 | 1xxx:xxx -> D, SHR:64-xxxxxx |
| 8217 | other -> invalid |
| 8218 | */ |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8219 | UInt size = 0; |
| 8220 | UInt shift = 0; |
| 8221 | Bool isQ = bitQ == 1; |
| 8222 | Bool isU = bitU == 1; |
| 8223 | Bool ok = getLaneInfo_IMMH_IMMB(&shift, &size, immh, immb); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8224 | if (!ok || (bitQ == 0 && size == X11)) return False; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8225 | vassert(size >= 0 && size <= 3); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8226 | UInt lanebits = 8 << size; |
| 8227 | vassert(shift >= 1 && shift <= lanebits); |
| 8228 | IROp op = isU ? mkVecSHRN(size) : mkVecSARN(size); |
| 8229 | IRExpr* src = getQReg128(nn); |
| 8230 | IRTemp res = newTempV128(); |
| 8231 | if (shift == lanebits && isU) { |
| 8232 | assign(res, mkV128(0x0000)); |
| 8233 | } else { |
| 8234 | UInt nudge = 0; |
| 8235 | if (shift == lanebits) { |
| 8236 | vassert(!isU); |
| 8237 | nudge = 1; |
| 8238 | } |
| 8239 | assign(res, binop(op, src, mkU8(shift - nudge))); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8240 | } |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8241 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
| 8242 | HChar laneCh = "bhsd"[size]; |
| 8243 | UInt nLanes = (isQ ? 128 : 64) / lanebits; |
| 8244 | const HChar* nm = isU ? "ushr" : "sshr"; |
| 8245 | DIP("%s %s.%u%c, %s.%u%c, #%u\n", nm, |
| 8246 | nameQReg128(dd), nLanes, laneCh, |
| 8247 | nameQReg128(nn), nLanes, laneCh, shift); |
| 8248 | return True; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8249 | } |
| 8250 | |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8251 | if (bitU == 1 && opcode == BITS5(0,1,0,0,0)) { |
| 8252 | /* -------- 1,01000 SRI std7_std7_#imm -------- */ |
| 8253 | /* laneTy, shift = case immh:immb of |
| 8254 | 0001:xxx -> B, SHR:8-xxx |
| 8255 | 001x:xxx -> H, SHR:16-xxxx |
| 8256 | 01xx:xxx -> S, SHR:32-xxxxx |
| 8257 | 1xxx:xxx -> D, SHR:64-xxxxxx |
| 8258 | other -> invalid |
| 8259 | */ |
| 8260 | UInt size = 0; |
| 8261 | UInt shift = 0; |
| 8262 | Bool isQ = bitQ == 1; |
| 8263 | Bool ok = getLaneInfo_IMMH_IMMB(&shift, &size, immh, immb); |
| 8264 | if (!ok || (bitQ == 0 && size == X11)) return False; |
| 8265 | vassert(size >= 0 && size <= 3); |
| 8266 | UInt lanebits = 8 << size; |
| 8267 | vassert(shift >= 1 && shift <= lanebits); |
| 8268 | IRExpr* src = getQReg128(nn); |
| 8269 | IRTemp res = newTempV128(); |
| 8270 | if (shift == lanebits) { |
| 8271 | assign(res, getQReg128(dd)); |
| 8272 | } else { |
| 8273 | assign(res, binop(mkVecSHRN(size), src, mkU8(shift))); |
| 8274 | IRExpr* nmask = binop(mkVecSHLN(size), |
| 8275 | mkV128(0xFFFF), mkU8(lanebits - shift)); |
| 8276 | IRTemp tmp = newTempV128(); |
| 8277 | assign(tmp, binop(Iop_OrV128, |
| 8278 | mkexpr(res), |
| 8279 | binop(Iop_AndV128, getQReg128(dd), nmask))); |
| 8280 | res = tmp; |
| 8281 | } |
| 8282 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
| 8283 | HChar laneCh = "bhsd"[size]; |
| 8284 | UInt nLanes = (isQ ? 128 : 64) / lanebits; |
| 8285 | DIP("%s %s.%u%c, %s.%u%c, #%u\n", "sri", |
| 8286 | nameQReg128(dd), nLanes, laneCh, |
| 8287 | nameQReg128(nn), nLanes, laneCh, shift); |
| 8288 | return True; |
| 8289 | } |
| 8290 | |
| 8291 | if (opcode == BITS5(0,1,0,1,0)) { |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8292 | /* -------- 0,01010 SHL std7_std7_#imm -------- */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8293 | /* -------- 1,01010 SLI std7_std7_#imm -------- */ |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8294 | /* laneTy, shift = case immh:immb of |
| 8295 | 0001:xxx -> B, xxx |
| 8296 | 001x:xxx -> H, xxxx |
| 8297 | 01xx:xxx -> S, xxxxx |
| 8298 | 1xxx:xxx -> D, xxxxxx |
| 8299 | other -> invalid |
| 8300 | */ |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8301 | UInt size = 0; |
| 8302 | UInt shift = 0; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8303 | Bool isSLI = bitU == 1; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8304 | Bool isQ = bitQ == 1; |
| 8305 | Bool ok = getLaneInfo_IMMH_IMMB(&shift, &size, immh, immb); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8306 | if (!ok || (bitQ == 0 && size == X11)) return False; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8307 | vassert(size >= 0 && size <= 3); |
| 8308 | /* The shift encoding has opposite sign for the leftwards case. |
| 8309 | Adjust shift to compensate. */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8310 | UInt lanebits = 8 << size; |
| 8311 | shift = lanebits - shift; |
| 8312 | vassert(shift >= 0 && shift < lanebits); |
| 8313 | IROp op = mkVecSHLN(size); |
| 8314 | IRExpr* src = getQReg128(nn); |
| 8315 | IRTemp res = newTempV128(); |
| 8316 | if (shift == 0) { |
| 8317 | assign(res, src); |
| 8318 | } else { |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 8319 | assign(res, binop(op, src, mkU8(shift))); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8320 | if (isSLI) { |
| 8321 | IRExpr* nmask = binop(mkVecSHRN(size), |
| 8322 | mkV128(0xFFFF), mkU8(lanebits - shift)); |
| 8323 | IRTemp tmp = newTempV128(); |
| 8324 | assign(tmp, binop(Iop_OrV128, |
| 8325 | mkexpr(res), |
| 8326 | binop(Iop_AndV128, getQReg128(dd), nmask))); |
| 8327 | res = tmp; |
| 8328 | } |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8329 | } |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8330 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
| 8331 | HChar laneCh = "bhsd"[size]; |
| 8332 | UInt nLanes = (isQ ? 128 : 64) / lanebits; |
| 8333 | const HChar* nm = isSLI ? "sli" : "shl"; |
| 8334 | DIP("%s %s.%u%c, %s.%u%c, #%u\n", nm, |
| 8335 | nameQReg128(dd), nLanes, laneCh, |
| 8336 | nameQReg128(nn), nLanes, laneCh, shift); |
| 8337 | return True; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8338 | } |
| 8339 | |
sewardj | a97dddf | 2014-08-14 22:26:52 +0000 | [diff] [blame^] | 8340 | if (opcode == BITS5(0,1,1,1,0) |
| 8341 | || (bitU == 1 && opcode == BITS5(0,1,1,0,0))) { |
| 8342 | /* -------- 0,01110 SQSHL std7_std7_#imm -------- */ |
| 8343 | /* -------- 1,01110 UQSHL std7_std7_#imm -------- */ |
| 8344 | /* -------- 1,01100 SQSHLU std7_std7_#imm -------- */ |
| 8345 | UInt size = 0; |
| 8346 | UInt shift = 0; |
| 8347 | Bool isQ = bitQ == 1; |
| 8348 | Bool ok = getLaneInfo_IMMH_IMMB(&shift, &size, immh, immb); |
| 8349 | if (!ok || (bitQ == 0 && size == X11)) return False; |
| 8350 | vassert(size >= 0 && size <= 3); |
| 8351 | /* The shift encoding has opposite sign for the leftwards case. |
| 8352 | Adjust shift to compensate. */ |
| 8353 | UInt lanebits = 8 << size; |
| 8354 | shift = lanebits - shift; |
| 8355 | vassert(shift >= 0 && shift < lanebits); |
| 8356 | const HChar* nm = NULL; |
| 8357 | /**/ if (bitU == 0 && opcode == BITS5(0,1,1,1,0)) nm = "sqshl"; |
| 8358 | else if (bitU == 1 && opcode == BITS5(0,1,1,1,0)) nm = "uqshl"; |
| 8359 | else if (bitU == 1 && opcode == BITS5(0,1,1,0,0)) nm = "sqshlu"; |
| 8360 | else vassert(0); |
| 8361 | IRTemp qDiff1 = IRTemp_INVALID; |
| 8362 | IRTemp qDiff2 = IRTemp_INVALID; |
| 8363 | IRTemp res = IRTemp_INVALID; |
| 8364 | IRTemp src = newTempV128(); |
| 8365 | assign(src, getQReg128(nn)); |
| 8366 | math_QSHL_IMM(&res, &qDiff1, &qDiff2, src, size, shift, nm); |
| 8367 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
| 8368 | updateQCFLAGwithDifferenceZHI(qDiff1, qDiff2, |
| 8369 | isQ ? Iop_ZeroHI64ofV128 : Iop_INVALID); |
| 8370 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 8371 | DIP("%s %s.%s, %s.%s, #%u\n", nm, |
| 8372 | nameQReg128(dd), arr, nameQReg128(nn), arr, shift); |
| 8373 | return True; |
| 8374 | } |
| 8375 | |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 8376 | if (bitU == 0 |
| 8377 | && (opcode == BITS5(1,0,0,0,0) || opcode == BITS5(1,0,0,0,1))) { |
| 8378 | /* -------- 0,10000 SHRN{,2} #imm -------- */ |
| 8379 | /* -------- 0,10001 RSHRN{,2} #imm -------- */ |
| 8380 | /* Narrows, and size is the narrow size. */ |
| 8381 | UInt size = 0; |
| 8382 | UInt shift = 0; |
| 8383 | Bool is2 = bitQ == 1; |
| 8384 | Bool isR = opcode == BITS5(1,0,0,0,1); |
| 8385 | Bool ok = getLaneInfo_IMMH_IMMB(&shift, &size, immh, immb); |
| 8386 | if (!ok || size == X11) return False; |
| 8387 | vassert(shift >= 1); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8388 | IRTemp t1 = newTempV128(); |
| 8389 | IRTemp t2 = newTempV128(); |
| 8390 | IRTemp t3 = newTempV128(); |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 8391 | assign(t1, getQReg128(nn)); |
| 8392 | assign(t2, isR ? binop(mkVecADD(size+1), |
| 8393 | mkexpr(t1), |
| 8394 | mkexpr(math_VEC_DUP_IMM(size+1, 1ULL<<(shift-1)))) |
| 8395 | : mkexpr(t1)); |
| 8396 | assign(t3, binop(mkVecSHRN(size+1), mkexpr(t2), mkU8(shift))); |
| 8397 | IRTemp t4 = math_NARROW_LANES(t3, t3, size); |
| 8398 | putLO64andZUorPutHI64(is2, dd, t4); |
| 8399 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 8400 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
| 8401 | DIP("%s %s.%s, %s.%s, #%u\n", isR ? "rshrn" : "shrn", |
| 8402 | nameQReg128(dd), arrNarrow, nameQReg128(nn), arrWide, shift); |
| 8403 | return True; |
| 8404 | } |
| 8405 | |
sewardj | ecedd98 | 2014-08-11 14:02:47 +0000 | [diff] [blame] | 8406 | if (opcode == BITS5(1,0,0,1,0) || opcode == BITS5(1,0,0,1,1) |
| 8407 | || (bitU == 1 |
| 8408 | && (opcode == BITS5(1,0,0,0,0) || opcode == BITS5(1,0,0,0,1)))) { |
| 8409 | /* -------- 0,10010 SQSHRN{,2} #imm -------- */ |
| 8410 | /* -------- 1,10010 UQSHRN{,2} #imm -------- */ |
| 8411 | /* -------- 0,10011 SQRSHRN{,2} #imm -------- */ |
| 8412 | /* -------- 1,10011 UQRSHRN{,2} #imm -------- */ |
| 8413 | /* -------- 1,10000 SQSHRUN{,2} #imm -------- */ |
| 8414 | /* -------- 1,10001 SQRSHRUN{,2} #imm -------- */ |
| 8415 | UInt size = 0; |
| 8416 | UInt shift = 0; |
| 8417 | Bool is2 = bitQ == 1; |
| 8418 | Bool ok = getLaneInfo_IMMH_IMMB(&shift, &size, immh, immb); |
| 8419 | if (!ok || size == X11) return False; |
| 8420 | vassert(shift >= 1 && shift <= (8 << size)); |
| 8421 | const HChar* nm = "??"; |
| 8422 | IROp op = Iop_INVALID; |
| 8423 | /* Decide on the name and the operation. */ |
| 8424 | /**/ if (bitU == 0 && opcode == BITS5(1,0,0,1,0)) { |
| 8425 | nm = "sqshrn"; op = mkVecQANDqsarNNARROWSS(size); |
| 8426 | } |
| 8427 | else if (bitU == 1 && opcode == BITS5(1,0,0,1,0)) { |
| 8428 | nm = "uqshrn"; op = mkVecQANDqshrNNARROWUU(size); |
| 8429 | } |
| 8430 | else if (bitU == 0 && opcode == BITS5(1,0,0,1,1)) { |
| 8431 | nm = "sqrshrn"; op = mkVecQANDqrsarNNARROWSS(size); |
| 8432 | } |
| 8433 | else if (bitU == 1 && opcode == BITS5(1,0,0,1,1)) { |
| 8434 | nm = "uqrshrn"; op = mkVecQANDqrshrNNARROWUU(size); |
| 8435 | } |
| 8436 | else if (bitU == 1 && opcode == BITS5(1,0,0,0,0)) { |
| 8437 | nm = "sqshrun"; op = mkVecQANDqsarNNARROWSU(size); |
| 8438 | } |
| 8439 | else if (bitU == 1 && opcode == BITS5(1,0,0,0,1)) { |
| 8440 | nm = "sqrshrun"; op = mkVecQANDqrsarNNARROWSU(size); |
| 8441 | } |
| 8442 | else vassert(0); |
| 8443 | /* Compute the result (Q, shifted value) pair. */ |
| 8444 | IRTemp src128 = newTempV128(); |
| 8445 | assign(src128, getQReg128(nn)); |
| 8446 | IRTemp pair = newTempV128(); |
| 8447 | assign(pair, binop(op, mkexpr(src128), mkU8(shift))); |
| 8448 | /* Update the result reg */ |
| 8449 | IRTemp res64in128 = newTempV128(); |
| 8450 | assign(res64in128, unop(Iop_ZeroHI64ofV128, mkexpr(pair))); |
| 8451 | putLO64andZUorPutHI64(is2, dd, res64in128); |
| 8452 | /* Update the Q flag. */ |
| 8453 | IRTemp q64q64 = newTempV128(); |
| 8454 | assign(q64q64, binop(Iop_InterleaveHI64x2, mkexpr(pair), mkexpr(pair))); |
| 8455 | IRTemp z128 = newTempV128(); |
| 8456 | assign(z128, mkV128(0x0000)); |
| 8457 | updateQCFLAGwithDifference(q64q64, z128); |
| 8458 | /* */ |
| 8459 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 8460 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
| 8461 | DIP("%s %s.%s, %s.%s, #%u\n", nm, |
| 8462 | nameQReg128(dd), arrNarrow, nameQReg128(nn), arrWide, shift); |
| 8463 | return True; |
| 8464 | } |
| 8465 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8466 | if (opcode == BITS5(1,0,1,0,0)) { |
| 8467 | /* -------- 0,10100 SSHLL{,2} #imm -------- */ |
| 8468 | /* -------- 1,10100 USHLL{,2} #imm -------- */ |
| 8469 | /* 31 28 22 18 15 9 4 |
| 8470 | 0q0 011110 immh immb 101001 n d SSHLL Vd.Ta, Vn.Tb, #sh |
| 8471 | 0q1 011110 immh immb 101001 n d USHLL Vd.Ta, Vn.Tb, #sh |
| 8472 | where Ta,Tb,sh |
| 8473 | = case immh of 1xxx -> invalid |
| 8474 | 01xx -> 2d, 2s(q0)/4s(q1), immh:immb - 32 (0..31) |
| 8475 | 001x -> 4s, 4h(q0)/8h(q1), immh:immb - 16 (0..15) |
| 8476 | 0001 -> 8h, 8b(q0)/16b(q1), immh:immb - 8 (0..7) |
| 8477 | 0000 -> AdvSIMD modified immediate (???) |
| 8478 | */ |
| 8479 | Bool isQ = bitQ == 1; |
| 8480 | Bool isU = bitU == 1; |
| 8481 | UInt immhb = (immh << 3) | immb; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8482 | IRTemp src = newTempV128(); |
| 8483 | IRTemp zero = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8484 | IRExpr* res = NULL; |
| 8485 | UInt sh = 0; |
| 8486 | const HChar* ta = "??"; |
| 8487 | const HChar* tb = "??"; |
| 8488 | assign(src, getQReg128(nn)); |
| 8489 | assign(zero, mkV128(0x0000)); |
| 8490 | if (immh & 8) { |
| 8491 | /* invalid; don't assign to res */ |
| 8492 | } |
| 8493 | else if (immh & 4) { |
| 8494 | sh = immhb - 32; |
| 8495 | vassert(sh < 32); /* so 32-sh is 1..32 */ |
| 8496 | ta = "2d"; |
| 8497 | tb = isQ ? "4s" : "2s"; |
| 8498 | IRExpr* tmp = isQ ? mk_InterleaveHI32x4(src, zero) |
| 8499 | : mk_InterleaveLO32x4(src, zero); |
| 8500 | res = binop(isU ? Iop_ShrN64x2 : Iop_SarN64x2, tmp, mkU8(32-sh)); |
| 8501 | } |
| 8502 | else if (immh & 2) { |
| 8503 | sh = immhb - 16; |
| 8504 | vassert(sh < 16); /* so 16-sh is 1..16 */ |
| 8505 | ta = "4s"; |
| 8506 | tb = isQ ? "8h" : "4h"; |
| 8507 | IRExpr* tmp = isQ ? mk_InterleaveHI16x8(src, zero) |
| 8508 | : mk_InterleaveLO16x8(src, zero); |
| 8509 | res = binop(isU ? Iop_ShrN32x4 : Iop_SarN32x4, tmp, mkU8(16-sh)); |
| 8510 | } |
| 8511 | else if (immh & 1) { |
| 8512 | sh = immhb - 8; |
| 8513 | vassert(sh < 8); /* so 8-sh is 1..8 */ |
| 8514 | ta = "8h"; |
| 8515 | tb = isQ ? "16b" : "8b"; |
| 8516 | IRExpr* tmp = isQ ? mk_InterleaveHI8x16(src, zero) |
| 8517 | : mk_InterleaveLO8x16(src, zero); |
| 8518 | res = binop(isU ? Iop_ShrN16x8 : Iop_SarN16x8, tmp, mkU8(8-sh)); |
| 8519 | } else { |
| 8520 | vassert(immh == 0); |
| 8521 | /* invalid; don't assign to res */ |
| 8522 | } |
| 8523 | /* */ |
| 8524 | if (res) { |
| 8525 | putQReg128(dd, res); |
| 8526 | DIP("%cshll%s %s.%s, %s.%s, #%d\n", |
| 8527 | isU ? 'u' : 's', isQ ? "2" : "", |
| 8528 | nameQReg128(dd), ta, nameQReg128(nn), tb, sh); |
| 8529 | return True; |
| 8530 | } |
| 8531 | return False; |
| 8532 | } |
| 8533 | |
| 8534 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 8535 | return False; |
| 8536 | # undef INSN |
| 8537 | } |
| 8538 | |
sewardj | fc83d2c | 2014-06-12 10:15:46 +0000 | [diff] [blame] | 8539 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8540 | static |
| 8541 | Bool dis_AdvSIMD_three_different(/*MB_OUT*/DisResult* dres, UInt insn) |
| 8542 | { |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 8543 | /* 31 30 29 28 23 21 20 15 11 9 4 |
| 8544 | 0 Q U 01110 size 1 m opcode 00 n d |
| 8545 | Decode fields: u,opcode |
| 8546 | */ |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8547 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 8548 | if (INSN(31,31) != 0 |
| 8549 | || INSN(28,24) != BITS5(0,1,1,1,0) |
| 8550 | || INSN(21,21) != 1 |
| 8551 | || INSN(11,10) != BITS2(0,0)) { |
| 8552 | return False; |
| 8553 | } |
| 8554 | UInt bitQ = INSN(30,30); |
| 8555 | UInt bitU = INSN(29,29); |
| 8556 | UInt size = INSN(23,22); |
| 8557 | UInt mm = INSN(20,16); |
| 8558 | UInt opcode = INSN(15,12); |
| 8559 | UInt nn = INSN(9,5); |
| 8560 | UInt dd = INSN(4,0); |
| 8561 | vassert(size < 4); |
| 8562 | Bool is2 = bitQ == 1; |
| 8563 | |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8564 | if (opcode == BITS4(0,0,0,0) || opcode == BITS4(0,0,1,0)) { |
| 8565 | /* -------- 0,0000 SADDL{2} -------- */ |
| 8566 | /* -------- 1,0000 UADDL{2} -------- */ |
| 8567 | /* -------- 0,0010 SSUBL{2} -------- */ |
| 8568 | /* -------- 1,0010 USUBL{2} -------- */ |
| 8569 | /* Widens, and size refers to the narrowed lanes. */ |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8570 | if (size == X11) return False; |
| 8571 | vassert(size <= 2); |
| 8572 | Bool isU = bitU == 1; |
| 8573 | Bool isADD = opcode == BITS4(0,0,0,0); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 8574 | IRTemp argL = math_WIDEN_LO_OR_HI_LANES(isU, is2, size, getQReg128(nn)); |
| 8575 | IRTemp argR = math_WIDEN_LO_OR_HI_LANES(isU, is2, size, getQReg128(mm)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8576 | IRTemp res = newTempV128(); |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 8577 | assign(res, binop(isADD ? mkVecADD(size+1) : mkVecSUB(size+1), |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8578 | mkexpr(argL), mkexpr(argR))); |
| 8579 | putQReg128(dd, mkexpr(res)); |
| 8580 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 8581 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
| 8582 | const HChar* nm = isADD ? (isU ? "uaddl" : "saddl") |
| 8583 | : (isU ? "usubl" : "ssubl"); |
| 8584 | DIP("%s%s %s.%s, %s.%s, %s.%s\n", nm, is2 ? "2" : "", |
| 8585 | nameQReg128(dd), arrWide, |
| 8586 | nameQReg128(nn), arrNarrow, nameQReg128(mm), arrNarrow); |
| 8587 | return True; |
| 8588 | } |
| 8589 | |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 8590 | if (opcode == BITS4(0,0,0,1) || opcode == BITS4(0,0,1,1)) { |
| 8591 | /* -------- 0,0001 SADDW{2} -------- */ |
| 8592 | /* -------- 1,0001 UADDW{2} -------- */ |
| 8593 | /* -------- 0,0011 SSUBW{2} -------- */ |
| 8594 | /* -------- 1,0011 USUBW{2} -------- */ |
| 8595 | /* Widens, and size refers to the narrowed lanes. */ |
| 8596 | if (size == X11) return False; |
| 8597 | vassert(size <= 2); |
| 8598 | Bool isU = bitU == 1; |
| 8599 | Bool isADD = opcode == BITS4(0,0,0,1); |
| 8600 | IRTemp argR = math_WIDEN_LO_OR_HI_LANES(isU, is2, size, getQReg128(mm)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8601 | IRTemp res = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 8602 | assign(res, binop(isADD ? mkVecADD(size+1) : mkVecSUB(size+1), |
| 8603 | getQReg128(nn), mkexpr(argR))); |
| 8604 | putQReg128(dd, mkexpr(res)); |
| 8605 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 8606 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
| 8607 | const HChar* nm = isADD ? (isU ? "uaddw" : "saddw") |
| 8608 | : (isU ? "usubw" : "ssubw"); |
| 8609 | DIP("%s%s %s.%s, %s.%s, %s.%s\n", nm, is2 ? "2" : "", |
| 8610 | nameQReg128(dd), arrWide, |
| 8611 | nameQReg128(nn), arrWide, nameQReg128(mm), arrNarrow); |
| 8612 | return True; |
| 8613 | } |
| 8614 | |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 8615 | if (opcode == BITS4(0,1,0,0) || opcode == BITS4(0,1,1,0)) { |
| 8616 | /* -------- 0,0100 ADDHN{2} -------- */ |
| 8617 | /* -------- 1,0100 RADDHN{2} -------- */ |
| 8618 | /* -------- 0,0110 SUBHN{2} -------- */ |
| 8619 | /* -------- 1,0110 RSUBHN{2} -------- */ |
| 8620 | /* Narrows, and size refers to the narrowed lanes. */ |
| 8621 | if (size == X11) return False; |
| 8622 | vassert(size <= 2); |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 8623 | const UInt shift[3] = { 8, 16, 32 }; |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 8624 | Bool isADD = opcode == BITS4(0,1,0,0); |
| 8625 | Bool isR = bitU == 1; |
| 8626 | /* Combined elements in wide lanes */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8627 | IRTemp wide = newTempV128(); |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 8628 | IRExpr* wideE = binop(isADD ? mkVecADD(size+1) : mkVecSUB(size+1), |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 8629 | getQReg128(nn), getQReg128(mm)); |
| 8630 | if (isR) { |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 8631 | wideE = binop(mkVecADD(size+1), |
| 8632 | wideE, |
| 8633 | mkexpr(math_VEC_DUP_IMM(size+1, |
| 8634 | 1ULL << (shift[size]-1)))); |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 8635 | } |
| 8636 | assign(wide, wideE); |
| 8637 | /* Top halves of elements, still in wide lanes */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8638 | IRTemp shrd = newTempV128(); |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 8639 | assign(shrd, binop(mkVecSHRN(size+1), mkexpr(wide), mkU8(shift[size]))); |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 8640 | /* Elements now compacted into lower 64 bits */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8641 | IRTemp new64 = newTempV128(); |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 8642 | assign(new64, binop(mkVecCATEVENLANES(size), mkexpr(shrd), mkexpr(shrd))); |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 8643 | putLO64andZUorPutHI64(is2, dd, new64); |
| 8644 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 8645 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
| 8646 | const HChar* nm = isADD ? (isR ? "raddhn" : "addhn") |
| 8647 | : (isR ? "rsubhn" : "subhn"); |
| 8648 | DIP("%s%s %s.%s, %s.%s, %s.%s\n", nm, is2 ? "2" : "", |
| 8649 | nameQReg128(dd), arrNarrow, |
| 8650 | nameQReg128(nn), arrWide, nameQReg128(mm), arrWide); |
| 8651 | return True; |
| 8652 | } |
| 8653 | |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8654 | if (opcode == BITS4(0,1,0,1) || opcode == BITS4(0,1,1,1)) { |
| 8655 | /* -------- 0,0101 SABAL{2} -------- */ |
| 8656 | /* -------- 1,0101 UABAL{2} -------- */ |
| 8657 | /* -------- 0,0111 SABDL{2} -------- */ |
| 8658 | /* -------- 1,0111 UABDL{2} -------- */ |
| 8659 | /* Widens, and size refers to the narrowed lanes. */ |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8660 | if (size == X11) return False; |
| 8661 | vassert(size <= 2); |
| 8662 | Bool isU = bitU == 1; |
| 8663 | Bool isACC = opcode == BITS4(0,1,0,1); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 8664 | IRTemp argL = math_WIDEN_LO_OR_HI_LANES(isU, is2, size, getQReg128(nn)); |
| 8665 | IRTemp argR = math_WIDEN_LO_OR_HI_LANES(isU, is2, size, getQReg128(mm)); |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8666 | IRTemp abd = math_ABD(isU, size+1, mkexpr(argL), mkexpr(argR)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8667 | IRTemp res = newTempV128(); |
| 8668 | assign(res, isACC ? binop(mkVecADD(size+1), mkexpr(abd), getQReg128(dd)) |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8669 | : mkexpr(abd)); |
| 8670 | putQReg128(dd, mkexpr(res)); |
| 8671 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 8672 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
| 8673 | const HChar* nm = isACC ? (isU ? "uabal" : "sabal") |
| 8674 | : (isU ? "uabdl" : "sabdl"); |
| 8675 | DIP("%s%s %s.%s, %s.%s, %s.%s\n", nm, is2 ? "2" : "", |
| 8676 | nameQReg128(dd), arrWide, |
| 8677 | nameQReg128(nn), arrNarrow, nameQReg128(mm), arrNarrow); |
| 8678 | return True; |
| 8679 | } |
| 8680 | |
| 8681 | if (opcode == BITS4(1,1,0,0) |
| 8682 | || opcode == BITS4(1,0,0,0) || opcode == BITS4(1,0,1,0)) { |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 8683 | /* -------- 0,1100 SMULL{2} -------- */ // 0 (ks) |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8684 | /* -------- 1,1100 UMULL{2} -------- */ // 0 |
| 8685 | /* -------- 0,1000 SMLAL{2} -------- */ // 1 |
| 8686 | /* -------- 1,1000 UMLAL{2} -------- */ // 1 |
| 8687 | /* -------- 0,1010 SMLSL{2} -------- */ // 2 |
| 8688 | /* -------- 1,1010 UMLSL{2} -------- */ // 2 |
| 8689 | /* Widens, and size refers to the narrowed lanes. */ |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 8690 | UInt ks = 3; |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8691 | switch (opcode) { |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 8692 | case BITS4(1,1,0,0): ks = 0; break; |
| 8693 | case BITS4(1,0,0,0): ks = 1; break; |
| 8694 | case BITS4(1,0,1,0): ks = 2; break; |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8695 | default: vassert(0); |
| 8696 | } |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 8697 | vassert(ks >= 0 && ks <= 2); |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8698 | if (size == X11) return False; |
| 8699 | vassert(size <= 2); |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 8700 | Bool isU = bitU == 1; |
| 8701 | IRTemp vecN = newTempV128(); |
| 8702 | IRTemp vecM = newTempV128(); |
| 8703 | IRTemp vecD = newTempV128(); |
| 8704 | assign(vecN, getQReg128(nn)); |
| 8705 | assign(vecM, getQReg128(mm)); |
| 8706 | assign(vecD, getQReg128(dd)); |
| 8707 | IRTemp res = IRTemp_INVALID; |
| 8708 | math_MULL_ACC(&res, is2, isU, size, "mas"[ks], |
| 8709 | vecN, vecM, ks == 0 ? IRTemp_INVALID : vecD); |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8710 | putQReg128(dd, mkexpr(res)); |
| 8711 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 8712 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 8713 | const HChar* nm = ks == 0 ? "mull" : (ks == 1 ? "mlal" : "mlsl"); |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8714 | DIP("%c%s%s %s.%s, %s.%s, %s.%s\n", isU ? 'u' : 's', nm, is2 ? "2" : "", |
| 8715 | nameQReg128(dd), arrWide, |
| 8716 | nameQReg128(nn), arrNarrow, nameQReg128(mm), arrNarrow); |
| 8717 | return True; |
| 8718 | } |
| 8719 | |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 8720 | if (bitU == 0 |
| 8721 | && (opcode == BITS4(1,1,0,1) |
| 8722 | || opcode == BITS4(1,0,0,1) || opcode == BITS4(1,0,1,1))) { |
| 8723 | /* -------- 0,1101 SQDMULL{2} -------- */ // 0 (ks) |
| 8724 | /* -------- 0,1001 SQDMLAL{2} -------- */ // 1 |
| 8725 | /* -------- 0,1011 SQDMLSL{2} -------- */ // 2 |
| 8726 | /* Widens, and size refers to the narrowed lanes. */ |
| 8727 | UInt ks = 3; |
| 8728 | switch (opcode) { |
| 8729 | case BITS4(1,1,0,1): ks = 0; break; |
| 8730 | case BITS4(1,0,0,1): ks = 1; break; |
| 8731 | case BITS4(1,0,1,1): ks = 2; break; |
| 8732 | default: vassert(0); |
| 8733 | } |
| 8734 | vassert(ks >= 0 && ks <= 2); |
| 8735 | if (size == X00 || size == X11) return False; |
| 8736 | vassert(size <= 2); |
| 8737 | IRTemp vecN, vecM, vecD, res, sat1q, sat1n, sat2q, sat2n; |
| 8738 | vecN = vecM = vecD = res = sat1q = sat1n = sat2q = sat2n = IRTemp_INVALID; |
| 8739 | newTempsV128_3(&vecN, &vecM, &vecD); |
| 8740 | assign(vecN, getQReg128(nn)); |
| 8741 | assign(vecM, getQReg128(mm)); |
| 8742 | assign(vecD, getQReg128(dd)); |
| 8743 | math_SQDMULL_ACC(&res, &sat1q, &sat1n, &sat2q, &sat2n, |
| 8744 | is2, size, "mas"[ks], |
| 8745 | vecN, vecM, ks == 0 ? IRTemp_INVALID : vecD); |
| 8746 | putQReg128(dd, mkexpr(res)); |
| 8747 | vassert(sat1q != IRTemp_INVALID && sat1n != IRTemp_INVALID); |
| 8748 | updateQCFLAGwithDifference(sat1q, sat1n); |
| 8749 | if (sat2q != IRTemp_INVALID || sat2n != IRTemp_INVALID) { |
| 8750 | updateQCFLAGwithDifference(sat2q, sat2n); |
| 8751 | } |
| 8752 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 8753 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
| 8754 | const HChar* nm = ks == 0 ? "sqdmull" |
| 8755 | : (ks == 1 ? "sqdmlal" : "sqdmlsl"); |
| 8756 | DIP("%s%s %s.%s, %s.%s, %s.%s\n", nm, is2 ? "2" : "", |
| 8757 | nameQReg128(dd), arrWide, |
| 8758 | nameQReg128(nn), arrNarrow, nameQReg128(mm), arrNarrow); |
| 8759 | return True; |
| 8760 | } |
| 8761 | |
sewardj | 31b5a95 | 2014-06-26 07:41:14 +0000 | [diff] [blame] | 8762 | if (bitU == 0 && opcode == BITS4(1,1,1,0)) { |
| 8763 | /* -------- 0,1110 PMULL{2} -------- */ |
sewardj | 6f312d0 | 2014-06-28 12:21:37 +0000 | [diff] [blame] | 8764 | /* Widens, and size refers to the narrowed lanes. */ |
sewardj | 31b5a95 | 2014-06-26 07:41:14 +0000 | [diff] [blame] | 8765 | if (size != X00) return False; |
| 8766 | IRTemp res |
| 8767 | = math_BINARY_WIDENING_V128(is2, Iop_PolynomialMull8x8, |
| 8768 | getQReg128(nn), getQReg128(mm)); |
| 8769 | putQReg128(dd, mkexpr(res)); |
| 8770 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 8771 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
| 8772 | DIP("%s%s %s.%s, %s.%s, %s.%s\n", "pmull", is2 ? "2" : "", |
| 8773 | nameQReg128(dd), arrNarrow, |
| 8774 | nameQReg128(nn), arrWide, nameQReg128(mm), arrWide); |
| 8775 | return True; |
| 8776 | } |
| 8777 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8778 | return False; |
| 8779 | # undef INSN |
| 8780 | } |
| 8781 | |
| 8782 | |
| 8783 | static |
| 8784 | Bool dis_AdvSIMD_three_same(/*MB_OUT*/DisResult* dres, UInt insn) |
| 8785 | { |
| 8786 | /* 31 30 29 28 23 21 20 15 10 9 4 |
| 8787 | 0 Q U 01110 size 1 m opcode 1 n d |
| 8788 | Decode fields: u,size,opcode |
| 8789 | */ |
| 8790 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 8791 | if (INSN(31,31) != 0 |
| 8792 | || INSN(28,24) != BITS5(0,1,1,1,0) |
| 8793 | || INSN(21,21) != 1 |
| 8794 | || INSN(10,10) != 1) { |
| 8795 | return False; |
| 8796 | } |
| 8797 | UInt bitQ = INSN(30,30); |
| 8798 | UInt bitU = INSN(29,29); |
| 8799 | UInt size = INSN(23,22); |
| 8800 | UInt mm = INSN(20,16); |
| 8801 | UInt opcode = INSN(15,11); |
| 8802 | UInt nn = INSN(9,5); |
| 8803 | UInt dd = INSN(4,0); |
| 8804 | vassert(size < 4); |
| 8805 | |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 8806 | if (opcode == BITS5(0,0,0,0,0) || opcode == BITS5(0,0,1,0,0)) { |
| 8807 | /* -------- 0,xx,00000 SHADD std6_std6_std6 -------- */ |
| 8808 | /* -------- 1,xx,00000 UHADD std6_std6_std6 -------- */ |
| 8809 | /* -------- 0,xx,00100 SHSUB std6_std6_std6 -------- */ |
| 8810 | /* -------- 1,xx,00100 UHSUB std6_std6_std6 -------- */ |
| 8811 | if (size == X11) return False; |
| 8812 | Bool isADD = opcode == BITS5(0,0,0,0,0); |
| 8813 | Bool isU = bitU == 1; |
| 8814 | /* Widen both args out, do the math, narrow to final result. */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8815 | IRTemp argL = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 8816 | IRTemp argLhi = IRTemp_INVALID; |
| 8817 | IRTemp argLlo = IRTemp_INVALID; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8818 | IRTemp argR = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 8819 | IRTemp argRhi = IRTemp_INVALID; |
| 8820 | IRTemp argRlo = IRTemp_INVALID; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8821 | IRTemp resHi = newTempV128(); |
| 8822 | IRTemp resLo = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 8823 | IRTemp res = IRTemp_INVALID; |
| 8824 | assign(argL, getQReg128(nn)); |
| 8825 | argLlo = math_WIDEN_LO_OR_HI_LANES(isU, False, size, mkexpr(argL)); |
| 8826 | argLhi = math_WIDEN_LO_OR_HI_LANES(isU, True, size, mkexpr(argL)); |
| 8827 | assign(argR, getQReg128(mm)); |
| 8828 | argRlo = math_WIDEN_LO_OR_HI_LANES(isU, False, size, mkexpr(argR)); |
| 8829 | argRhi = math_WIDEN_LO_OR_HI_LANES(isU, True, size, mkexpr(argR)); |
| 8830 | IROp opADDSUB = isADD ? mkVecADD(size+1) : mkVecSUB(size+1); |
| 8831 | IROp opSxR = isU ? mkVecSHRN(size+1) : mkVecSARN(size+1); |
| 8832 | assign(resHi, binop(opSxR, |
| 8833 | binop(opADDSUB, mkexpr(argLhi), mkexpr(argRhi)), |
| 8834 | mkU8(1))); |
| 8835 | assign(resLo, binop(opSxR, |
| 8836 | binop(opADDSUB, mkexpr(argLlo), mkexpr(argRlo)), |
| 8837 | mkU8(1))); |
| 8838 | res = math_NARROW_LANES ( resHi, resLo, size ); |
| 8839 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
| 8840 | const HChar* nm = isADD ? (isU ? "uhadd" : "shadd") |
| 8841 | : (isU ? "uhsub" : "shsub"); |
| 8842 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 8843 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 8844 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 8845 | return True; |
| 8846 | } |
| 8847 | |
| 8848 | if (opcode == BITS5(0,0,0,0,1) || opcode == BITS5(0,0,1,0,1)) { |
| 8849 | /* -------- 0,xx,00001 SQADD std7_std7_std7 -------- */ |
| 8850 | /* -------- 1,xx,00001 UQADD std7_std7_std7 -------- */ |
| 8851 | /* -------- 0,xx,00101 SQSUB std7_std7_std7 -------- */ |
| 8852 | /* -------- 1,xx,00101 UQSUB std7_std7_std7 -------- */ |
| 8853 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
| 8854 | Bool isADD = opcode == BITS5(0,0,0,0,1); |
| 8855 | Bool isU = bitU == 1; |
| 8856 | IROp qop = Iop_INVALID; |
| 8857 | IROp nop = Iop_INVALID; |
| 8858 | if (isADD) { |
| 8859 | qop = isU ? mkVecQADDU(size) : mkVecQADDS(size); |
| 8860 | nop = mkVecADD(size); |
| 8861 | } else { |
| 8862 | qop = isU ? mkVecQSUBU(size) : mkVecQSUBS(size); |
| 8863 | nop = mkVecSUB(size); |
| 8864 | } |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8865 | IRTemp argL = newTempV128(); |
| 8866 | IRTemp argR = newTempV128(); |
| 8867 | IRTemp qres = newTempV128(); |
| 8868 | IRTemp nres = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 8869 | assign(argL, getQReg128(nn)); |
| 8870 | assign(argR, getQReg128(mm)); |
| 8871 | assign(qres, math_MAYBE_ZERO_HI64_fromE( |
| 8872 | bitQ, binop(qop, mkexpr(argL), mkexpr(argR)))); |
| 8873 | assign(nres, math_MAYBE_ZERO_HI64_fromE( |
| 8874 | bitQ, binop(nop, mkexpr(argL), mkexpr(argR)))); |
| 8875 | putQReg128(dd, mkexpr(qres)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8876 | updateQCFLAGwithDifference(qres, nres); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 8877 | const HChar* nm = isADD ? (isU ? "uqadd" : "sqadd") |
| 8878 | : (isU ? "uqsub" : "sqsub"); |
| 8879 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 8880 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 8881 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 8882 | return True; |
| 8883 | } |
| 8884 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8885 | if (bitU == 0 && opcode == BITS5(0,0,0,1,1)) { |
| 8886 | /* -------- 0,00,00011 AND 16b_16b_16b, 8b_8b_8b -------- */ |
| 8887 | /* -------- 0,01,00011 BIC 16b_16b_16b, 8b_8b_8b -------- */ |
| 8888 | /* -------- 0,10,00011 ORR 16b_16b_16b, 8b_8b_8b -------- */ |
| 8889 | /* -------- 0,10,00011 ORN 16b_16b_16b, 8b_8b_8b -------- */ |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 8890 | Bool isORx = (size & 2) == 2; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8891 | Bool invert = (size & 1) == 1; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8892 | IRTemp res = newTempV128(); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 8893 | assign(res, binop(isORx ? Iop_OrV128 : Iop_AndV128, |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8894 | getQReg128(nn), |
| 8895 | invert ? unop(Iop_NotV128, getQReg128(mm)) |
| 8896 | : getQReg128(mm))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 8897 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8898 | const HChar* names[4] = { "and", "bic", "orr", "orn" }; |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 8899 | const HChar* ar = bitQ == 1 ? "16b" : "8b"; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8900 | DIP("%s %s.%s, %s.%s, %s.%s\n", names[INSN(23,22)], |
| 8901 | nameQReg128(dd), ar, nameQReg128(nn), ar, nameQReg128(mm), ar); |
| 8902 | return True; |
| 8903 | } |
| 8904 | |
| 8905 | if (bitU == 1 && opcode == BITS5(0,0,0,1,1)) { |
| 8906 | /* -------- 1,00,00011 EOR 16b_16b_16b, 8b_8b_8b -------- */ |
| 8907 | /* -------- 1,01,00011 BSL 16b_16b_16b, 8b_8b_8b -------- */ |
| 8908 | /* -------- 1,10,00011 BIT 16b_16b_16b, 8b_8b_8b -------- */ |
| 8909 | /* -------- 1,10,00011 BIF 16b_16b_16b, 8b_8b_8b -------- */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8910 | IRTemp argD = newTempV128(); |
| 8911 | IRTemp argN = newTempV128(); |
| 8912 | IRTemp argM = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8913 | assign(argD, getQReg128(dd)); |
| 8914 | assign(argN, getQReg128(nn)); |
| 8915 | assign(argM, getQReg128(mm)); |
| 8916 | const IROp opXOR = Iop_XorV128; |
| 8917 | const IROp opAND = Iop_AndV128; |
| 8918 | const IROp opNOT = Iop_NotV128; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8919 | IRTemp res = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8920 | switch (size) { |
| 8921 | case BITS2(0,0): /* EOR */ |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 8922 | assign(res, binop(opXOR, mkexpr(argM), mkexpr(argN))); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8923 | break; |
| 8924 | case BITS2(0,1): /* BSL */ |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 8925 | assign(res, binop(opXOR, mkexpr(argM), |
| 8926 | binop(opAND, |
| 8927 | binop(opXOR, mkexpr(argM), mkexpr(argN)), |
| 8928 | mkexpr(argD)))); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8929 | break; |
| 8930 | case BITS2(1,0): /* BIT */ |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 8931 | assign(res, binop(opXOR, mkexpr(argD), |
| 8932 | binop(opAND, |
| 8933 | binop(opXOR, mkexpr(argD), mkexpr(argN)), |
| 8934 | mkexpr(argM)))); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8935 | break; |
| 8936 | case BITS2(1,1): /* BIF */ |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 8937 | assign(res, binop(opXOR, mkexpr(argD), |
| 8938 | binop(opAND, |
| 8939 | binop(opXOR, mkexpr(argD), mkexpr(argN)), |
| 8940 | unop(opNOT, mkexpr(argM))))); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8941 | break; |
| 8942 | default: |
| 8943 | vassert(0); |
| 8944 | } |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 8945 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8946 | const HChar* nms[4] = { "eor", "bsl", "bit", "bif" }; |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 8947 | const HChar* arr = bitQ == 1 ? "16b" : "8b"; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8948 | DIP("%s %s.%s, %s.%s, %s.%s\n", nms[size], |
| 8949 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 8950 | return True; |
| 8951 | } |
| 8952 | |
| 8953 | if (opcode == BITS5(0,0,1,1,0)) { |
| 8954 | /* -------- 0,xx,00110 CMGT std7_std7_std7 -------- */ // >s |
| 8955 | /* -------- 1,xx,00110 CMHI std7_std7_std7 -------- */ // >u |
| 8956 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8957 | Bool isGT = bitU == 0; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8958 | IRExpr* argL = getQReg128(nn); |
| 8959 | IRExpr* argR = getQReg128(mm); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8960 | IRTemp res = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8961 | assign(res, |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8962 | isGT ? binop(mkVecCMPGTS(size), argL, argR) |
| 8963 | : binop(mkVecCMPGTU(size), argL, argR)); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 8964 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8965 | const HChar* nm = isGT ? "cmgt" : "cmhi"; |
| 8966 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 8967 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 8968 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 8969 | return True; |
| 8970 | } |
| 8971 | |
| 8972 | if (opcode == BITS5(0,0,1,1,1)) { |
| 8973 | /* -------- 0,xx,00111 CMGE std7_std7_std7 -------- */ // >=s |
| 8974 | /* -------- 1,xx,00111 CMHS std7_std7_std7 -------- */ // >=u |
| 8975 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8976 | Bool isGE = bitU == 0; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8977 | IRExpr* argL = getQReg128(nn); |
| 8978 | IRExpr* argR = getQReg128(mm); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8979 | IRTemp res = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8980 | assign(res, |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 8981 | isGE ? unop(Iop_NotV128, binop(mkVecCMPGTS(size), argR, argL)) |
| 8982 | : unop(Iop_NotV128, binop(mkVecCMPGTU(size), argR, argL))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 8983 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 8984 | const HChar* nm = isGE ? "cmge" : "cmhs"; |
| 8985 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 8986 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 8987 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 8988 | return True; |
| 8989 | } |
| 8990 | |
sewardj | 1297218 | 2014-08-04 08:09:47 +0000 | [diff] [blame] | 8991 | if (opcode == BITS5(0,1,0,0,1) || opcode == BITS5(0,1,0,1,1)) { |
| 8992 | /* -------- 0,xx,01001 SQSHL std7_std7_std7 -------- */ |
| 8993 | /* -------- 0,xx,01011 SQRSHL std7_std7_std7 -------- */ |
| 8994 | /* -------- 1,xx,01001 UQSHL std7_std7_std7 -------- */ |
| 8995 | /* -------- 1,xx,01011 UQRSHL std7_std7_std7 -------- */ |
| 8996 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
| 8997 | Bool isU = bitU == 1; |
| 8998 | Bool isR = opcode == BITS5(0,1,0,1,1); |
| 8999 | IROp op = isR ? (isU ? mkVecQANDUQRSH(size) : mkVecQANDSQRSH(size)) |
| 9000 | : (isU ? mkVecQANDUQSH(size) : mkVecQANDSQSH(size)); |
| 9001 | /* This is a bit tricky. If we're only interested in the lowest 64 bits |
| 9002 | of the result (viz, bitQ == 0), then we must adjust the operands to |
| 9003 | ensure that the upper part of the result, that we don't care about, |
| 9004 | doesn't pollute the returned Q value. To do this, zero out the upper |
| 9005 | operand halves beforehand. This works because it means, for the |
| 9006 | lanes we don't care about, we are shifting zero by zero, which can |
| 9007 | never saturate. */ |
| 9008 | IRTemp res256 = newTemp(Ity_V256); |
| 9009 | IRTemp resSH = newTempV128(); |
| 9010 | IRTemp resQ = newTempV128(); |
| 9011 | IRTemp zero = newTempV128(); |
| 9012 | assign(res256, binop(op, |
| 9013 | math_MAYBE_ZERO_HI64_fromE(bitQ, getQReg128(nn)), |
| 9014 | math_MAYBE_ZERO_HI64_fromE(bitQ, getQReg128(mm)))); |
| 9015 | assign(resSH, unop(Iop_V256toV128_0, mkexpr(res256))); |
| 9016 | assign(resQ, unop(Iop_V256toV128_1, mkexpr(res256))); |
| 9017 | assign(zero, mkV128(0x0000)); |
| 9018 | putQReg128(dd, mkexpr(resSH)); |
| 9019 | updateQCFLAGwithDifference(resQ, zero); |
| 9020 | const HChar* nm = isR ? (isU ? "uqrshl" : "sqrshl") |
| 9021 | : (isU ? "uqshl" : "sqshl"); |
| 9022 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9023 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 9024 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9025 | return True; |
| 9026 | } |
| 9027 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9028 | if (opcode == BITS5(0,1,1,0,0) || opcode == BITS5(0,1,1,0,1)) { |
| 9029 | /* -------- 0,xx,01100 SMAX std7_std7_std7 -------- */ |
| 9030 | /* -------- 1,xx,01100 UMAX std7_std7_std7 -------- */ |
| 9031 | /* -------- 0,xx,01101 SMIN std7_std7_std7 -------- */ |
| 9032 | /* -------- 1,xx,01101 UMIN std7_std7_std7 -------- */ |
| 9033 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
| 9034 | Bool isU = bitU == 1; |
| 9035 | Bool isMAX = (opcode & 1) == 0; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9036 | IROp op = isMAX ? (isU ? mkVecMAXU(size) : mkVecMAXS(size)) |
| 9037 | : (isU ? mkVecMINU(size) : mkVecMINS(size)); |
| 9038 | IRTemp t = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9039 | assign(t, binop(op, getQReg128(nn), getQReg128(mm))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9040 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, t)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9041 | const HChar* nm = isMAX ? (isU ? "umax" : "smax") |
| 9042 | : (isU ? "umin" : "smin"); |
| 9043 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9044 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 9045 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9046 | return True; |
| 9047 | } |
| 9048 | |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9049 | if (opcode == BITS5(0,1,1,1,0) || opcode == BITS5(0,1,1,1,1)) { |
| 9050 | /* -------- 0,xx,01110 SABD std6_std6_std6 -------- */ |
| 9051 | /* -------- 1,xx,01110 UABD std6_std6_std6 -------- */ |
| 9052 | /* -------- 0,xx,01111 SABA std6_std6_std6 -------- */ |
| 9053 | /* -------- 1,xx,01111 UABA std6_std6_std6 -------- */ |
| 9054 | if (size == X11) return False; // 1d/2d cases not allowed |
| 9055 | Bool isU = bitU == 1; |
| 9056 | Bool isACC = opcode == BITS5(0,1,1,1,1); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9057 | vassert(size <= 2); |
| 9058 | IRTemp t1 = math_ABD(isU, size, getQReg128(nn), getQReg128(mm)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9059 | IRTemp t2 = newTempV128(); |
| 9060 | assign(t2, isACC ? binop(mkVecADD(size), mkexpr(t1), getQReg128(dd)) |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9061 | : mkexpr(t1)); |
| 9062 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, t2)); |
| 9063 | const HChar* nm = isACC ? (isU ? "uaba" : "saba") |
| 9064 | : (isU ? "uabd" : "sabd"); |
| 9065 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9066 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 9067 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9068 | return True; |
| 9069 | } |
| 9070 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9071 | if (opcode == BITS5(1,0,0,0,0)) { |
| 9072 | /* -------- 0,xx,10000 ADD std7_std7_std7 -------- */ |
| 9073 | /* -------- 1,xx,10000 SUB std7_std7_std7 -------- */ |
| 9074 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9075 | Bool isSUB = bitU == 1; |
| 9076 | IROp op = isSUB ? mkVecSUB(size) : mkVecADD(size); |
| 9077 | IRTemp t = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9078 | assign(t, binop(op, getQReg128(nn), getQReg128(mm))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9079 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, t)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9080 | const HChar* nm = isSUB ? "sub" : "add"; |
| 9081 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9082 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 9083 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9084 | return True; |
| 9085 | } |
| 9086 | |
| 9087 | if (opcode == BITS5(1,0,0,0,1)) { |
| 9088 | /* -------- 0,xx,10001 CMTST std7_std7_std7 -------- */ // &, != 0 |
| 9089 | /* -------- 1,xx,10001 CMEQ std7_std7_std7 -------- */ // == |
| 9090 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9091 | Bool isEQ = bitU == 1; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9092 | IRExpr* argL = getQReg128(nn); |
| 9093 | IRExpr* argR = getQReg128(mm); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9094 | IRTemp res = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9095 | assign(res, |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9096 | isEQ ? binop(mkVecCMPEQ(size), argL, argR) |
| 9097 | : unop(Iop_NotV128, binop(mkVecCMPEQ(size), |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9098 | binop(Iop_AndV128, argL, argR), |
| 9099 | mkV128(0x0000)))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9100 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9101 | const HChar* nm = isEQ ? "cmeq" : "cmtst"; |
| 9102 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9103 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 9104 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9105 | return True; |
| 9106 | } |
| 9107 | |
| 9108 | if (opcode == BITS5(1,0,0,1,0)) { |
| 9109 | /* -------- 0,xx,10010 MLA std7_std7_std7 -------- */ |
| 9110 | /* -------- 1,xx,10010 MLS std7_std7_std7 -------- */ |
| 9111 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
| 9112 | Bool isMLS = bitU == 1; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9113 | IROp opMUL = mkVecMUL(size); |
| 9114 | IROp opADDSUB = isMLS ? mkVecSUB(size) : mkVecADD(size); |
| 9115 | IRTemp res = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9116 | if (opMUL != Iop_INVALID && opADDSUB != Iop_INVALID) { |
| 9117 | assign(res, binop(opADDSUB, |
| 9118 | getQReg128(dd), |
| 9119 | binop(opMUL, getQReg128(nn), getQReg128(mm)))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9120 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9121 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9122 | DIP("%s %s.%s, %s.%s, %s.%s\n", isMLS ? "mls" : "mla", |
| 9123 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9124 | return True; |
| 9125 | } |
| 9126 | return False; |
| 9127 | } |
| 9128 | |
| 9129 | if (opcode == BITS5(1,0,0,1,1)) { |
| 9130 | /* -------- 0,xx,10011 MUL std7_std7_std7 -------- */ |
| 9131 | /* -------- 1,xx,10011 PMUL 16b_16b_16b, 8b_8b_8b -------- */ |
| 9132 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
| 9133 | Bool isPMUL = bitU == 1; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9134 | const IROp opsPMUL[4] |
| 9135 | = { Iop_PolynomialMul8x16, Iop_INVALID, Iop_INVALID, Iop_INVALID }; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9136 | IROp opMUL = isPMUL ? opsPMUL[size] : mkVecMUL(size); |
| 9137 | IRTemp res = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9138 | if (opMUL != Iop_INVALID) { |
| 9139 | assign(res, binop(opMUL, getQReg128(nn), getQReg128(mm))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9140 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9141 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9142 | DIP("%s %s.%s, %s.%s, %s.%s\n", isPMUL ? "pmul" : "mul", |
| 9143 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9144 | return True; |
| 9145 | } |
| 9146 | return False; |
| 9147 | } |
| 9148 | |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 9149 | if (opcode == BITS5(1,0,1,0,0) || opcode == BITS5(1,0,1,0,1)) { |
| 9150 | /* -------- 0,xx,10100 SMAXP std6_std6_std6 -------- */ |
| 9151 | /* -------- 1,xx,10100 UMAXP std6_std6_std6 -------- */ |
| 9152 | /* -------- 0,xx,10101 SMINP std6_std6_std6 -------- */ |
| 9153 | /* -------- 1,xx,10101 UMINP std6_std6_std6 -------- */ |
| 9154 | if (size == X11) return False; |
| 9155 | Bool isU = bitU == 1; |
| 9156 | Bool isMAX = opcode == BITS5(1,0,1,0,0); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9157 | IRTemp vN = newTempV128(); |
| 9158 | IRTemp vM = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 9159 | IROp op = isMAX ? (isU ? mkVecMAXU(size) : mkVecMAXS(size)) |
| 9160 | : (isU ? mkVecMINU(size) : mkVecMINS(size)); |
| 9161 | assign(vN, getQReg128(nn)); |
| 9162 | assign(vM, getQReg128(mm)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9163 | IRTemp res128 = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 9164 | assign(res128, |
| 9165 | binop(op, |
| 9166 | binop(mkVecCATEVENLANES(size), mkexpr(vM), mkexpr(vN)), |
| 9167 | binop(mkVecCATODDLANES(size), mkexpr(vM), mkexpr(vN)))); |
| 9168 | /* In the half-width case, use CatEL32x4 to extract the half-width |
| 9169 | result from the full-width result. */ |
| 9170 | IRExpr* res |
| 9171 | = bitQ == 0 ? unop(Iop_ZeroHI64ofV128, |
| 9172 | binop(Iop_CatEvenLanes32x4, mkexpr(res128), |
| 9173 | mkexpr(res128))) |
| 9174 | : mkexpr(res128); |
| 9175 | putQReg128(dd, res); |
| 9176 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9177 | const HChar* nm = isMAX ? (isU ? "umaxp" : "smaxp") |
| 9178 | : (isU ? "uminp" : "sminp"); |
| 9179 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 9180 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9181 | return True; |
| 9182 | } |
| 9183 | |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 9184 | if (opcode == BITS5(1,0,1,1,0)) { |
| 9185 | /* -------- 0,xx,10110 SQDMULH s and h variants only -------- */ |
| 9186 | /* -------- 1,xx,10110 SQRDMULH s and h variants only -------- */ |
| 9187 | if (size == X00 || size == X11) return False; |
| 9188 | Bool isR = bitU == 1; |
| 9189 | IRTemp res, sat1q, sat1n, vN, vM; |
| 9190 | res = sat1q = sat1n = vN = vM = IRTemp_INVALID; |
| 9191 | newTempsV128_2(&vN, &vM); |
| 9192 | assign(vN, getQReg128(nn)); |
| 9193 | assign(vM, getQReg128(mm)); |
| 9194 | math_SQDMULH(&res, &sat1q, &sat1n, isR, size, vN, vM); |
| 9195 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
| 9196 | IROp opZHI = bitQ == 0 ? Iop_ZeroHI64ofV128 : Iop_INVALID; |
| 9197 | updateQCFLAGwithDifferenceZHI(sat1q, sat1n, opZHI); |
| 9198 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9199 | const HChar* nm = isR ? "sqrdmulh" : "sqdmulh"; |
| 9200 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 9201 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9202 | return True; |
| 9203 | } |
| 9204 | |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 9205 | if (bitU == 0 && opcode == BITS5(1,0,1,1,1)) { |
| 9206 | /* -------- 0,xx,10111 ADDP std7_std7_std7 -------- */ |
| 9207 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9208 | IRTemp vN = newTempV128(); |
| 9209 | IRTemp vM = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 9210 | assign(vN, getQReg128(nn)); |
| 9211 | assign(vM, getQReg128(mm)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9212 | IRTemp res128 = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 9213 | assign(res128, |
| 9214 | binop(mkVecADD(size), |
| 9215 | binop(mkVecCATEVENLANES(size), mkexpr(vM), mkexpr(vN)), |
| 9216 | binop(mkVecCATODDLANES(size), mkexpr(vM), mkexpr(vN)))); |
| 9217 | /* In the half-width case, use CatEL32x4 to extract the half-width |
| 9218 | result from the full-width result. */ |
| 9219 | IRExpr* res |
| 9220 | = bitQ == 0 ? unop(Iop_ZeroHI64ofV128, |
| 9221 | binop(Iop_CatEvenLanes32x4, mkexpr(res128), |
| 9222 | mkexpr(res128))) |
| 9223 | : mkexpr(res128); |
| 9224 | putQReg128(dd, res); |
| 9225 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9226 | DIP("addp %s.%s, %s.%s, %s.%s\n", |
| 9227 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9228 | return True; |
| 9229 | } |
| 9230 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9231 | if (bitU == 0 && opcode == BITS5(1,1,0,0,1)) { |
| 9232 | /* -------- 0,0x,11001 FMLA 2d_2d_2d, 4s_4s_4s, 2s_2s_2s -------- */ |
| 9233 | /* -------- 0,1x,11001 FMLS 2d_2d_2d, 4s_4s_4s, 2s_2s_2s -------- */ |
| 9234 | Bool isD = (size & 1) == 1; |
| 9235 | Bool isSUB = (size & 2) == 2; |
| 9236 | if (bitQ == 0 && isD) return False; // implied 1d case |
| 9237 | IROp opADD = isD ? Iop_Add64Fx2 : Iop_Add32Fx4; |
| 9238 | IROp opSUB = isD ? Iop_Sub64Fx2 : Iop_Sub32Fx4; |
| 9239 | IROp opMUL = isD ? Iop_Mul64Fx2 : Iop_Mul32Fx4; |
| 9240 | IRTemp rm = mk_get_IR_rounding_mode(); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9241 | IRTemp t1 = newTempV128(); |
| 9242 | IRTemp t2 = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9243 | // FIXME: double rounding; use FMA primops instead |
| 9244 | assign(t1, triop(opMUL, |
| 9245 | mkexpr(rm), getQReg128(nn), getQReg128(mm))); |
| 9246 | assign(t2, triop(isSUB ? opSUB : opADD, |
| 9247 | mkexpr(rm), getQReg128(dd), mkexpr(t1))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9248 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, t2)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9249 | const HChar* arr = bitQ == 0 ? "2s" : (isD ? "2d" : "4s"); |
| 9250 | DIP("%s %s.%s, %s.%s, %s.%s\n", isSUB ? "fmls" : "fmla", |
| 9251 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9252 | return True; |
| 9253 | } |
| 9254 | |
| 9255 | if (bitU == 0 && opcode == BITS5(1,1,0,1,0)) { |
| 9256 | /* -------- 0,0x,11010 FADD 2d_2d_2d, 4s_4s_4s, 2s_2s_2s -------- */ |
| 9257 | /* -------- 0,1x,11010 FSUB 2d_2d_2d, 4s_4s_4s, 2s_2s_2s -------- */ |
| 9258 | Bool isD = (size & 1) == 1; |
| 9259 | Bool isSUB = (size & 2) == 2; |
| 9260 | if (bitQ == 0 && isD) return False; // implied 1d case |
| 9261 | const IROp ops[4] |
| 9262 | = { Iop_Add32Fx4, Iop_Add64Fx2, Iop_Sub32Fx4, Iop_Sub64Fx2 }; |
| 9263 | IROp op = ops[size]; |
| 9264 | IRTemp rm = mk_get_IR_rounding_mode(); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9265 | IRTemp t1 = newTempV128(); |
| 9266 | IRTemp t2 = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9267 | assign(t1, triop(op, mkexpr(rm), getQReg128(nn), getQReg128(mm))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9268 | assign(t2, math_MAYBE_ZERO_HI64(bitQ, t1)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9269 | putQReg128(dd, mkexpr(t2)); |
| 9270 | const HChar* arr = bitQ == 0 ? "2s" : (isD ? "2d" : "4s"); |
| 9271 | DIP("%s %s.%s, %s.%s, %s.%s\n", isSUB ? "fsub" : "fadd", |
| 9272 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9273 | return True; |
| 9274 | } |
| 9275 | |
| 9276 | if (bitU == 1 && size >= X10 && opcode == BITS5(1,1,0,1,0)) { |
| 9277 | /* -------- 1,1x,11010 FABD 2d_2d_2d, 4s_4s_4s, 2s_2s_2s -------- */ |
| 9278 | Bool isD = (size & 1) == 1; |
| 9279 | if (bitQ == 0 && isD) return False; // implied 1d case |
| 9280 | IROp opSUB = isD ? Iop_Sub64Fx2 : Iop_Sub32Fx4; |
| 9281 | IROp opABS = isD ? Iop_Abs64Fx2 : Iop_Abs32Fx4; |
| 9282 | IRTemp rm = mk_get_IR_rounding_mode(); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9283 | IRTemp t1 = newTempV128(); |
| 9284 | IRTemp t2 = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9285 | // FIXME: use Abd primop instead? |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9286 | assign(t1, triop(opSUB, mkexpr(rm), getQReg128(nn), getQReg128(mm))); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9287 | assign(t2, unop(opABS, mkexpr(t1))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9288 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, t2)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9289 | const HChar* arr = bitQ == 0 ? "2s" : (isD ? "2d" : "4s"); |
| 9290 | DIP("fabd %s.%s, %s.%s, %s.%s\n", |
| 9291 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9292 | return True; |
| 9293 | } |
| 9294 | |
| 9295 | if (bitU == 1 && size <= X01 && opcode == BITS5(1,1,0,1,1)) { |
| 9296 | /* -------- 1,0x,11011 FMUL 2d_2d_2d, 4s_4s_4s, 2s_2s_2s -------- */ |
| 9297 | Bool isD = (size & 1) == 1; |
| 9298 | if (bitQ == 0 && isD) return False; // implied 1d case |
| 9299 | IRTemp rm = mk_get_IR_rounding_mode(); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9300 | IRTemp t1 = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9301 | assign(t1, triop(isD ? Iop_Mul64Fx2 : Iop_Mul32Fx4, |
| 9302 | mkexpr(rm), getQReg128(nn), getQReg128(mm))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9303 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, t1)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9304 | const HChar* arr = bitQ == 0 ? "2s" : (isD ? "2d" : "4s"); |
| 9305 | DIP("fmul %s.%s, %s.%s, %s.%s\n", |
| 9306 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9307 | return True; |
| 9308 | } |
| 9309 | |
| 9310 | if (size <= X01 && opcode == BITS5(1,1,1,0,0)) { |
| 9311 | /* -------- 0,0x,11100 FCMEQ 2d_2d_2d, 4s_4s_4s, 2s_2s_2s -------- */ |
| 9312 | /* -------- 1,0x,11100 FCMGE 2d_2d_2d, 4s_4s_4s, 2s_2s_2s -------- */ |
| 9313 | Bool isD = (size & 1) == 1; |
| 9314 | if (bitQ == 0 && isD) return False; // implied 1d case |
| 9315 | Bool isGE = bitU == 1; |
| 9316 | IROp opCMP = isGE ? (isD ? Iop_CmpLE64Fx2 : Iop_CmpLE32Fx4) |
| 9317 | : (isD ? Iop_CmpEQ64Fx2 : Iop_CmpEQ32Fx4); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9318 | IRTemp t1 = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9319 | assign(t1, isGE ? binop(opCMP, getQReg128(mm), getQReg128(nn)) // swapd |
| 9320 | : binop(opCMP, getQReg128(nn), getQReg128(mm))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9321 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, t1)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9322 | const HChar* arr = bitQ == 0 ? "2s" : (isD ? "2d" : "4s"); |
| 9323 | DIP("%s %s.%s, %s.%s, %s.%s\n", isGE ? "fcmge" : "fcmeq", |
| 9324 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9325 | return True; |
| 9326 | } |
| 9327 | |
| 9328 | if (bitU == 1 && size >= X10 && opcode == BITS5(1,1,1,0,0)) { |
| 9329 | /* -------- 1,1x,11100 FCMGT 2d_2d_2d, 4s_4s_4s, 2s_2s_2s -------- */ |
| 9330 | Bool isD = (size & 1) == 1; |
| 9331 | if (bitQ == 0 && isD) return False; // implied 1d case |
| 9332 | IROp opCMP = isD ? Iop_CmpLT64Fx2 : Iop_CmpLT32Fx4; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9333 | IRTemp t1 = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9334 | assign(t1, binop(opCMP, getQReg128(mm), getQReg128(nn))); // swapd |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9335 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, t1)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9336 | const HChar* arr = bitQ == 0 ? "2s" : (isD ? "2d" : "4s"); |
| 9337 | DIP("%s %s.%s, %s.%s, %s.%s\n", "fcmgt", |
| 9338 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9339 | return True; |
| 9340 | } |
| 9341 | |
| 9342 | if (bitU == 1 && opcode == BITS5(1,1,1,0,1)) { |
| 9343 | /* -------- 1,0x,11101 FACGE 2d_2d_2d, 4s_4s_4s, 2s_2s_2s -------- */ |
| 9344 | /* -------- 1,1x,11101 FACGT 2d_2d_2d, 4s_4s_4s, 2s_2s_2s -------- */ |
| 9345 | Bool isD = (size & 1) == 1; |
| 9346 | Bool isGT = (size & 2) == 2; |
| 9347 | if (bitQ == 0 && isD) return False; // implied 1d case |
| 9348 | IROp opCMP = isGT ? (isD ? Iop_CmpLT64Fx2 : Iop_CmpLT32Fx4) |
| 9349 | : (isD ? Iop_CmpLE64Fx2 : Iop_CmpLE32Fx4); |
| 9350 | IROp opABS = isD ? Iop_Abs64Fx2 : Iop_Abs32Fx4; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9351 | IRTemp t1 = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9352 | assign(t1, binop(opCMP, unop(opABS, getQReg128(mm)), |
| 9353 | unop(opABS, getQReg128(nn)))); // swapd |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9354 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, t1)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9355 | const HChar* arr = bitQ == 0 ? "2s" : (isD ? "2d" : "4s"); |
| 9356 | DIP("%s %s.%s, %s.%s, %s.%s\n", isGT ? "facgt" : "facge", |
| 9357 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9358 | return True; |
| 9359 | } |
| 9360 | |
| 9361 | if (bitU == 1 && size <= X01 && opcode == BITS5(1,1,1,1,1)) { |
| 9362 | /* -------- 1,0x,11111 FDIV 2d_2d_2d, 4s_4s_4s, 2s_2s_2s -------- */ |
| 9363 | Bool isD = (size & 1) == 1; |
| 9364 | if (bitQ == 0 && isD) return False; // implied 1d case |
| 9365 | vassert(size <= 1); |
| 9366 | const IROp ops[2] = { Iop_Div32Fx4, Iop_Div64Fx2 }; |
| 9367 | IROp op = ops[size]; |
| 9368 | IRTemp rm = mk_get_IR_rounding_mode(); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9369 | IRTemp t1 = newTempV128(); |
| 9370 | IRTemp t2 = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9371 | assign(t1, triop(op, mkexpr(rm), getQReg128(nn), getQReg128(mm))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9372 | assign(t2, math_MAYBE_ZERO_HI64(bitQ, t1)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9373 | putQReg128(dd, mkexpr(t2)); |
| 9374 | const HChar* arr = bitQ == 0 ? "2s" : (isD ? "2d" : "4s"); |
| 9375 | DIP("%s %s.%s, %s.%s, %s.%s\n", "fdiv", |
| 9376 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 9377 | return True; |
| 9378 | } |
| 9379 | |
| 9380 | return False; |
| 9381 | # undef INSN |
| 9382 | } |
| 9383 | |
| 9384 | |
| 9385 | static |
| 9386 | Bool dis_AdvSIMD_two_reg_misc(/*MB_OUT*/DisResult* dres, UInt insn) |
| 9387 | { |
| 9388 | /* 31 30 29 28 23 21 16 11 9 4 |
| 9389 | 0 Q U 01110 size 10000 opcode 10 n d |
| 9390 | Decode fields: U,size,opcode |
| 9391 | */ |
| 9392 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 9393 | if (INSN(31,31) != 0 |
| 9394 | || INSN(28,24) != BITS5(0,1,1,1,0) |
| 9395 | || INSN(21,17) != BITS5(1,0,0,0,0) |
| 9396 | || INSN(11,10) != BITS2(1,0)) { |
| 9397 | return False; |
| 9398 | } |
| 9399 | UInt bitQ = INSN(30,30); |
| 9400 | UInt bitU = INSN(29,29); |
| 9401 | UInt size = INSN(23,22); |
| 9402 | UInt opcode = INSN(16,12); |
| 9403 | UInt nn = INSN(9,5); |
| 9404 | UInt dd = INSN(4,0); |
| 9405 | vassert(size < 4); |
| 9406 | |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9407 | if (bitU == 0 && size <= X10 && opcode == BITS5(0,0,0,0,0)) { |
| 9408 | /* -------- 0,00,00000: REV64 16b_16b, 8b_8b -------- */ |
| 9409 | /* -------- 0,01,00000: REV64 8h_8h, 4h_4h -------- */ |
| 9410 | /* -------- 0,10,00000: REV64 4s_4s, 2s_2s -------- */ |
| 9411 | const IROp iops[3] = { Iop_Reverse8sIn64_x2, |
| 9412 | Iop_Reverse16sIn64_x2, Iop_Reverse32sIn64_x2 }; |
| 9413 | vassert(size <= 2); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9414 | IRTemp res = newTempV128(); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9415 | assign(res, unop(iops[size], getQReg128(nn))); |
| 9416 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
| 9417 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9418 | DIP("%s %s.%s, %s.%s\n", "rev64", |
| 9419 | nameQReg128(dd), arr, nameQReg128(nn), arr); |
| 9420 | return True; |
| 9421 | } |
| 9422 | |
| 9423 | if (bitU == 1 && size <= X01 && opcode == BITS5(0,0,0,0,0)) { |
| 9424 | /* -------- 1,00,00000: REV32 16b_16b, 8b_8b -------- */ |
| 9425 | /* -------- 1,01,00000: REV32 8h_8h, 4h_4h -------- */ |
| 9426 | Bool isH = size == X01; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9427 | IRTemp res = newTempV128(); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9428 | IROp iop = isH ? Iop_Reverse16sIn32_x4 : Iop_Reverse8sIn32_x4; |
| 9429 | assign(res, unop(iop, getQReg128(nn))); |
| 9430 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
| 9431 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9432 | DIP("%s %s.%s, %s.%s\n", "rev32", |
| 9433 | nameQReg128(dd), arr, nameQReg128(nn), arr); |
| 9434 | return True; |
| 9435 | } |
| 9436 | |
sewardj | 715d162 | 2014-06-26 12:39:05 +0000 | [diff] [blame] | 9437 | if (bitU == 0 && size == X00 && opcode == BITS5(0,0,0,0,1)) { |
| 9438 | /* -------- 0,00,00001: REV16 16b_16b, 8b_8b -------- */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9439 | IRTemp res = newTempV128(); |
sewardj | 715d162 | 2014-06-26 12:39:05 +0000 | [diff] [blame] | 9440 | assign(res, unop(Iop_Reverse8sIn16_x8, getQReg128(nn))); |
| 9441 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9442 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
sewardj | 715d162 | 2014-06-26 12:39:05 +0000 | [diff] [blame] | 9443 | DIP("%s %s.%s, %s.%s\n", "rev16", |
| 9444 | nameQReg128(dd), arr, nameQReg128(nn), arr); |
| 9445 | return True; |
| 9446 | } |
| 9447 | |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 9448 | if (opcode == BITS5(0,0,0,1,0) || opcode == BITS5(0,0,1,1,0)) { |
| 9449 | /* -------- 0,xx,00010: SADDLP std6_std6 -------- */ |
| 9450 | /* -------- 1,xx,00010: UADDLP std6_std6 -------- */ |
| 9451 | /* -------- 0,xx,00110: SADALP std6_std6 -------- */ |
| 9452 | /* -------- 1,xx,00110: UADALP std6_std6 -------- */ |
| 9453 | /* Widens, and size refers to the narrow size. */ |
| 9454 | if (size == X11) return False; // no 1d or 2d cases |
| 9455 | Bool isU = bitU == 1; |
| 9456 | Bool isACC = opcode == BITS5(0,0,1,1,0); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9457 | IRTemp src = newTempV128(); |
| 9458 | IRTemp sum = newTempV128(); |
| 9459 | IRTemp res = newTempV128(); |
sewardj | a5a6b75 | 2014-06-30 07:33:56 +0000 | [diff] [blame] | 9460 | assign(src, getQReg128(nn)); |
| 9461 | assign(sum, |
| 9462 | binop(mkVecADD(size+1), |
| 9463 | mkexpr(math_WIDEN_EVEN_OR_ODD_LANES( |
| 9464 | isU, True/*fromOdd*/, size, mkexpr(src))), |
| 9465 | mkexpr(math_WIDEN_EVEN_OR_ODD_LANES( |
| 9466 | isU, False/*!fromOdd*/, size, mkexpr(src))))); |
| 9467 | assign(res, isACC ? binop(mkVecADD(size+1), mkexpr(sum), getQReg128(dd)) |
| 9468 | : mkexpr(sum)); |
| 9469 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
| 9470 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 9471 | const HChar* arrWide = nameArr_Q_SZ(bitQ, size+1); |
| 9472 | DIP("%s %s.%s, %s.%s\n", isACC ? (isU ? "uadalp" : "sadalp") |
| 9473 | : (isU ? "uaddlp" : "saddlp"), |
| 9474 | nameQReg128(dd), arrWide, nameQReg128(nn), arrNarrow); |
| 9475 | return True; |
| 9476 | } |
| 9477 | |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 9478 | if (opcode == BITS5(0,0,1,0,0)) { |
| 9479 | /* -------- 0,xx,00100: CLS std6_std6 -------- */ |
| 9480 | /* -------- 1,xx,00100: CLZ std6_std6 -------- */ |
| 9481 | if (size == X11) return False; // no 1d or 2d cases |
sewardj | a8c7b0f | 2014-06-26 08:18:08 +0000 | [diff] [blame] | 9482 | const IROp opsCLS[3] = { Iop_Cls8x16, Iop_Cls16x8, Iop_Cls32x4 }; |
| 9483 | const IROp opsCLZ[3] = { Iop_Clz8x16, Iop_Clz16x8, Iop_Clz32x4 }; |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 9484 | Bool isCLZ = bitU == 1; |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9485 | IRTemp res = newTempV128(); |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 9486 | vassert(size <= 2); |
| 9487 | assign(res, unop(isCLZ ? opsCLZ[size] : opsCLS[size], getQReg128(nn))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9488 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 9489 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9490 | DIP("%s %s.%s, %s.%s\n", isCLZ ? "clz" : "cls", |
| 9491 | nameQReg128(dd), arr, nameQReg128(nn), arr); |
| 9492 | return True; |
| 9493 | } |
| 9494 | |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 9495 | if (size == X00 && opcode == BITS5(0,0,1,0,1)) { |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 9496 | /* -------- 0,00,00101: CNT 16b_16b, 8b_8b -------- */ |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 9497 | /* -------- 1,00,00101: NOT 16b_16b, 8b_8b -------- */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9498 | IRTemp res = newTempV128(); |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 9499 | assign(res, unop(bitU == 0 ? Iop_Cnt8x16 : Iop_NotV128, getQReg128(nn))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9500 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | 715d162 | 2014-06-26 12:39:05 +0000 | [diff] [blame] | 9501 | const HChar* arr = nameArr_Q_SZ(bitQ, 0); |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 9502 | DIP("%s %s.%s, %s.%s\n", bitU == 0 ? "cnt" : "not", |
sewardj | 2b6fd5e | 2014-06-19 14:21:37 +0000 | [diff] [blame] | 9503 | nameQReg128(dd), arr, nameQReg128(nn), arr); |
| 9504 | return True; |
| 9505 | } |
| 9506 | |
sewardj | 715d162 | 2014-06-26 12:39:05 +0000 | [diff] [blame] | 9507 | if (bitU == 1 && size == X01 && opcode == BITS5(0,0,1,0,1)) { |
| 9508 | /* -------- 1,01,00101 RBIT 16b_16b, 8b_8b -------- */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9509 | IRTemp res = newTempV128(); |
sewardj | 715d162 | 2014-06-26 12:39:05 +0000 | [diff] [blame] | 9510 | assign(res, unop(Iop_Reverse1sIn8_x16, getQReg128(nn))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9511 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | 715d162 | 2014-06-26 12:39:05 +0000 | [diff] [blame] | 9512 | const HChar* arr = nameArr_Q_SZ(bitQ, 0); |
| 9513 | DIP("%s %s.%s, %s.%s\n", "rbit", |
| 9514 | nameQReg128(dd), arr, nameQReg128(nn), arr); |
| 9515 | return True; |
| 9516 | } |
| 9517 | |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 9518 | if (opcode == BITS5(0,0,1,1,1)) { |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9519 | /* -------- 0,xx,00111 SQABS std7_std7 -------- */ |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 9520 | /* -------- 1,xx,00111 SQNEG std7_std7 -------- */ |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9521 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 9522 | Bool isNEG = bitU == 1; |
| 9523 | IRTemp qresFW = IRTemp_INVALID, nresFW = IRTemp_INVALID; |
| 9524 | (isNEG ? math_SQNEG : math_SQABS)( &qresFW, &nresFW, |
| 9525 | getQReg128(nn), size ); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9526 | IRTemp qres = newTempV128(), nres = newTempV128(); |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 9527 | assign(qres, math_MAYBE_ZERO_HI64(bitQ, qresFW)); |
| 9528 | assign(nres, math_MAYBE_ZERO_HI64(bitQ, nresFW)); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9529 | putQReg128(dd, mkexpr(qres)); |
| 9530 | updateQCFLAGwithDifference(qres, nres); |
| 9531 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 9532 | DIP("%s %s.%s, %s.%s\n", isNEG ? "sqneg" : "sqabs", |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9533 | nameQReg128(dd), arr, nameQReg128(nn), arr); |
| 9534 | return True; |
| 9535 | } |
| 9536 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9537 | if (opcode == BITS5(0,1,0,0,0)) { |
| 9538 | /* -------- 0,xx,01000: CMGT std7_std7_#0 -------- */ // >s 0 |
| 9539 | /* -------- 1,xx,01000: CMGE std7_std7_#0 -------- */ // >=s 0 |
| 9540 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9541 | Bool isGT = bitU == 0; |
| 9542 | IRExpr* argL = getQReg128(nn); |
| 9543 | IRExpr* argR = mkV128(0x0000); |
| 9544 | IRTemp res = newTempV128(); |
| 9545 | IROp opGTS = mkVecCMPGTS(size); |
| 9546 | assign(res, isGT ? binop(opGTS, argL, argR) |
| 9547 | : unop(Iop_NotV128, binop(opGTS, argR, argL))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9548 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9549 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9550 | DIP("cm%s %s.%s, %s.%s, #0\n", isGT ? "gt" : "ge", |
| 9551 | nameQReg128(dd), arr, nameQReg128(nn), arr); |
| 9552 | return True; |
| 9553 | } |
| 9554 | |
| 9555 | if (opcode == BITS5(0,1,0,0,1)) { |
| 9556 | /* -------- 0,xx,01001: CMEQ std7_std7_#0 -------- */ // == 0 |
| 9557 | /* -------- 1,xx,01001: CMLE std7_std7_#0 -------- */ // <=s 0 |
| 9558 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9559 | Bool isEQ = bitU == 0; |
| 9560 | IRExpr* argL = getQReg128(nn); |
| 9561 | IRExpr* argR = mkV128(0x0000); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9562 | IRTemp res = newTempV128(); |
| 9563 | assign(res, isEQ ? binop(mkVecCMPEQ(size), argL, argR) |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9564 | : unop(Iop_NotV128, |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9565 | binop(mkVecCMPGTS(size), argL, argR))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9566 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9567 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9568 | DIP("cm%s %s.%s, %s.%s, #0\n", isEQ ? "eq" : "le", |
| 9569 | nameQReg128(dd), arr, nameQReg128(nn), arr); |
| 9570 | return True; |
| 9571 | } |
| 9572 | |
| 9573 | if (bitU == 0 && opcode == BITS5(0,1,0,1,0)) { |
| 9574 | /* -------- 0,xx,01010: CMLT std7_std7_#0 -------- */ // <s 0 |
| 9575 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9576 | IRExpr* argL = getQReg128(nn); |
| 9577 | IRExpr* argR = mkV128(0x0000); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9578 | IRTemp res = newTempV128(); |
| 9579 | assign(res, binop(mkVecCMPGTS(size), argR, argL)); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9580 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9581 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9582 | DIP("cm%s %s.%s, %s.%s, #0\n", "lt", |
| 9583 | nameQReg128(dd), arr, nameQReg128(nn), arr); |
| 9584 | return True; |
| 9585 | } |
| 9586 | |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 9587 | if (bitU == 0 && opcode == BITS5(0,1,0,1,1)) { |
| 9588 | /* -------- 0,xx,01011: ABS std7_std7 -------- */ |
| 9589 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9590 | IRTemp res = newTempV128(); |
| 9591 | assign(res, unop(mkVecABS(size), getQReg128(nn))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9592 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | 25523c4 | 2014-06-15 19:36:29 +0000 | [diff] [blame] | 9593 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9594 | DIP("abs %s.%s, %s.%s\n", nameQReg128(dd), arr, nameQReg128(nn), arr); |
| 9595 | return True; |
| 9596 | } |
| 9597 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9598 | if (bitU == 1 && opcode == BITS5(0,1,0,1,1)) { |
| 9599 | /* -------- 1,xx,01011: NEG std7_std7 -------- */ |
| 9600 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9601 | IRTemp res = newTempV128(); |
| 9602 | assign(res, binop(mkVecSUB(size), mkV128(0x0000), getQReg128(nn))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9603 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9604 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9605 | DIP("neg %s.%s, %s.%s\n", nameQReg128(dd), arr, nameQReg128(nn), arr); |
| 9606 | return True; |
| 9607 | } |
| 9608 | |
| 9609 | if (size >= X10 && opcode == BITS5(0,1,1,1,1)) { |
| 9610 | /* -------- 0,1x,01111: FABS 2d_2d, 4s_4s, 2s_2s -------- */ |
| 9611 | /* -------- 1,1x,01111: FNEG 2d_2d, 4s_4s, 2s_2s -------- */ |
| 9612 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
| 9613 | Bool isFNEG = bitU == 1; |
| 9614 | IROp op = isFNEG ? (size == X10 ? Iop_Neg32Fx4 : Iop_Neg64Fx2) |
| 9615 | : (size == X10 ? Iop_Abs32Fx4 : Iop_Abs64Fx2); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9616 | IRTemp res = newTempV128(); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9617 | assign(res, unop(op, getQReg128(nn))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9618 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9619 | const HChar* arr = bitQ == 0 ? "2s" : (size == X11 ? "2d" : "4s"); |
| 9620 | DIP("%s %s.%s, %s.%s\n", isFNEG ? "fneg" : "fabs", |
| 9621 | nameQReg128(dd), arr, nameQReg128(nn), arr); |
| 9622 | return True; |
| 9623 | } |
| 9624 | |
| 9625 | if (bitU == 0 && opcode == BITS5(1,0,0,1,0)) { |
| 9626 | /* -------- 0,xx,10010: XTN{,2} -------- */ |
sewardj | ecedd98 | 2014-08-11 14:02:47 +0000 | [diff] [blame] | 9627 | if (size == X11) return False; |
| 9628 | vassert(size < 3); |
| 9629 | Bool is2 = bitQ == 1; |
| 9630 | IROp opN = mkVecNARROWUN(size); |
| 9631 | IRTemp resN = newTempV128(); |
| 9632 | assign(resN, unop(Iop_64UtoV128, unop(opN, getQReg128(nn)))); |
| 9633 | putLO64andZUorPutHI64(is2, dd, resN); |
| 9634 | const HChar* nm = "xtn"; |
| 9635 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 9636 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
| 9637 | DIP("%s%s %s.%s, %s.%s\n", is2 ? "2" : "", nm, |
| 9638 | nameQReg128(dd), arrNarrow, nameQReg128(nn), arrWide); |
| 9639 | return True; |
| 9640 | } |
| 9641 | |
| 9642 | if (opcode == BITS5(1,0,1,0,0) |
| 9643 | || (bitU == 1 && opcode == BITS5(1,0,0,1,0))) { |
| 9644 | /* -------- 0,xx,10100: SQXTN{,2} -------- */ |
| 9645 | /* -------- 1,xx,10100: UQXTN{,2} -------- */ |
| 9646 | /* -------- 1,xx,10010: SQXTUN{,2} -------- */ |
| 9647 | if (size == X11) return False; |
| 9648 | vassert(size < 3); |
| 9649 | Bool is2 = bitQ == 1; |
| 9650 | IROp opN = Iop_INVALID; |
| 9651 | Bool zWiden = True; |
| 9652 | const HChar* nm = "??"; |
| 9653 | /**/ if (bitU == 0 && opcode == BITS5(1,0,1,0,0)) { |
| 9654 | opN = mkVecQNARROWUNSS(size); nm = "sqxtn"; zWiden = False; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9655 | } |
sewardj | ecedd98 | 2014-08-11 14:02:47 +0000 | [diff] [blame] | 9656 | else if (bitU == 1 && opcode == BITS5(1,0,1,0,0)) { |
| 9657 | opN = mkVecQNARROWUNUU(size); nm = "uqxtn"; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9658 | } |
sewardj | ecedd98 | 2014-08-11 14:02:47 +0000 | [diff] [blame] | 9659 | else if (bitU == 1 && opcode == BITS5(1,0,0,1,0)) { |
| 9660 | opN = mkVecQNARROWUNSU(size); nm = "sqxtun"; |
| 9661 | } |
| 9662 | else vassert(0); |
| 9663 | IRTemp src = newTempV128(); |
| 9664 | assign(src, getQReg128(nn)); |
| 9665 | IRTemp resN = newTempV128(); |
| 9666 | assign(resN, unop(Iop_64UtoV128, unop(opN, mkexpr(src)))); |
| 9667 | putLO64andZUorPutHI64(is2, dd, resN); |
| 9668 | IRTemp resW = math_WIDEN_LO_OR_HI_LANES(zWiden, False/*!fromUpperHalf*/, |
| 9669 | size, mkexpr(resN)); |
| 9670 | updateQCFLAGwithDifference(src, resW); |
| 9671 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 9672 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
| 9673 | DIP("%s%s %s.%s, %s.%s\n", is2 ? "2" : "", nm, |
| 9674 | nameQReg128(dd), arrNarrow, nameQReg128(nn), arrWide); |
| 9675 | return True; |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9676 | } |
| 9677 | |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 9678 | if (bitU == 1 && opcode == BITS5(1,0,0,1,1)) { |
| 9679 | /* -------- 1,xx,10011 SHLL{2} #lane-width -------- */ |
| 9680 | /* Widens, and size is the narrow size. */ |
| 9681 | if (size == X11) return False; |
| 9682 | Bool is2 = bitQ == 1; |
| 9683 | IROp opINT = is2 ? mkVecINTERLEAVEHI(size) : mkVecINTERLEAVELO(size); |
| 9684 | IROp opSHL = mkVecSHLN(size+1); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9685 | IRTemp src = newTempV128(); |
| 9686 | IRTemp res = newTempV128(); |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 9687 | assign(src, getQReg128(nn)); |
| 9688 | assign(res, binop(opSHL, binop(opINT, mkexpr(src), mkexpr(src)), |
| 9689 | mkU8(8 << size))); |
| 9690 | putQReg128(dd, mkexpr(res)); |
| 9691 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 9692 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
| 9693 | DIP("shll%s %s.%s, %s.%s, #%u\n", is2 ? "2" : "", |
| 9694 | nameQReg128(dd), arrWide, nameQReg128(nn), arrNarrow, 8 << size); |
| 9695 | return True; |
| 9696 | } |
| 9697 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9698 | if (bitU == 0 && size == X01 && opcode == BITS5(1,0,1,1,0)) { |
| 9699 | /* -------- 0,01,10110: FCVTN 2s/4s_2d -------- */ |
| 9700 | IRTemp rm = mk_get_IR_rounding_mode(); |
| 9701 | IRExpr* srcLo = getQRegLane(nn, 0, Ity_F64); |
| 9702 | IRExpr* srcHi = getQRegLane(nn, 1, Ity_F64); |
| 9703 | putQRegLane(dd, 2 * bitQ + 0, binop(Iop_F64toF32, mkexpr(rm), srcLo)); |
| 9704 | putQRegLane(dd, 2 * bitQ + 1, binop(Iop_F64toF32, mkexpr(rm), srcHi)); |
| 9705 | if (bitQ == 0) { |
| 9706 | putQRegLane(dd, 1, mkU64(0)); |
| 9707 | } |
| 9708 | DIP("fcvtn%s %s.%s, %s.2d\n", bitQ ? "2" : "", |
| 9709 | nameQReg128(dd), bitQ ? "4s" : "2s", nameQReg128(nn)); |
| 9710 | return True; |
| 9711 | } |
| 9712 | |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 9713 | if (size <= X01 && opcode == BITS5(1,1,1,0,1)) { |
| 9714 | /* -------- 0,0x,11101: SCVTF -------- */ |
| 9715 | /* -------- 1,0x,11101: UCVTF -------- */ |
| 9716 | /* 31 28 22 21 15 9 4 |
| 9717 | 0q0 01110 0 sz 1 00001 110110 n d SCVTF Vd, Vn |
| 9718 | 0q1 01110 0 sz 1 00001 110110 n d UCVTF Vd, Vn |
| 9719 | with laneage: |
| 9720 | case sz:Q of 00 -> 2S, zero upper, 01 -> 4S, 10 -> illegal, 11 -> 2D |
| 9721 | */ |
| 9722 | Bool isQ = bitQ == 1; |
| 9723 | Bool isU = bitU == 1; |
| 9724 | Bool isF64 = (size & 1) == 1; |
| 9725 | if (isQ || !isF64) { |
| 9726 | IRType tyF = Ity_INVALID, tyI = Ity_INVALID; |
| 9727 | UInt nLanes = 0; |
| 9728 | Bool zeroHI = False; |
| 9729 | const HChar* arrSpec = NULL; |
| 9730 | Bool ok = getLaneInfo_Q_SZ(&tyI, &tyF, &nLanes, &zeroHI, &arrSpec, |
| 9731 | isQ, isF64 ); |
| 9732 | IROp iop = isU ? (isF64 ? Iop_I64UtoF64 : Iop_I32UtoF32) |
| 9733 | : (isF64 ? Iop_I64StoF64 : Iop_I32StoF32); |
| 9734 | IRTemp rm = mk_get_IR_rounding_mode(); |
| 9735 | UInt i; |
| 9736 | vassert(ok); /* the 'if' above should ensure this */ |
| 9737 | for (i = 0; i < nLanes; i++) { |
| 9738 | putQRegLane(dd, i, |
| 9739 | binop(iop, mkexpr(rm), getQRegLane(nn, i, tyI))); |
| 9740 | } |
| 9741 | if (zeroHI) { |
| 9742 | putQRegLane(dd, 1, mkU64(0)); |
| 9743 | } |
| 9744 | DIP("%ccvtf %s.%s, %s.%s\n", isU ? 'u' : 's', |
| 9745 | nameQReg128(dd), arrSpec, nameQReg128(nn), arrSpec); |
| 9746 | return True; |
| 9747 | } |
| 9748 | /* else fall through */ |
| 9749 | } |
| 9750 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9751 | return False; |
| 9752 | # undef INSN |
| 9753 | } |
| 9754 | |
sewardj | fc83d2c | 2014-06-12 10:15:46 +0000 | [diff] [blame] | 9755 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9756 | static |
| 9757 | Bool dis_AdvSIMD_vector_x_indexed_elem(/*MB_OUT*/DisResult* dres, UInt insn) |
| 9758 | { |
sewardj | 85fbb02 | 2014-06-12 13:16:01 +0000 | [diff] [blame] | 9759 | /* 31 28 23 21 20 19 15 11 9 4 |
| 9760 | 0 Q U 01111 size L M m opcode H 0 n d |
| 9761 | Decode fields are: u,size,opcode |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 9762 | M is really part of the mm register number. Individual |
| 9763 | cases need to inspect L and H though. |
sewardj | 85fbb02 | 2014-06-12 13:16:01 +0000 | [diff] [blame] | 9764 | */ |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 9765 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
sewardj | 85fbb02 | 2014-06-12 13:16:01 +0000 | [diff] [blame] | 9766 | if (INSN(31,31) != 0 |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9767 | || INSN(28,24) != BITS5(0,1,1,1,1) || INSN(10,10) !=0) { |
sewardj | 85fbb02 | 2014-06-12 13:16:01 +0000 | [diff] [blame] | 9768 | return False; |
| 9769 | } |
| 9770 | UInt bitQ = INSN(30,30); |
| 9771 | UInt bitU = INSN(29,29); |
| 9772 | UInt size = INSN(23,22); |
| 9773 | UInt bitL = INSN(21,21); |
| 9774 | UInt bitM = INSN(20,20); |
| 9775 | UInt mmLO4 = INSN(19,16); |
| 9776 | UInt opcode = INSN(15,12); |
| 9777 | UInt bitH = INSN(11,11); |
| 9778 | UInt nn = INSN(9,5); |
| 9779 | UInt dd = INSN(4,0); |
sewardj | 85fbb02 | 2014-06-12 13:16:01 +0000 | [diff] [blame] | 9780 | vassert(size < 4); |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 9781 | vassert(bitH < 2 && bitM < 2 && bitL < 2); |
sewardj | 85fbb02 | 2014-06-12 13:16:01 +0000 | [diff] [blame] | 9782 | |
| 9783 | if (bitU == 0 && size >= X10 && opcode == BITS4(1,0,0,1)) { |
| 9784 | /* -------- 0,1x,1001 FMUL 2d_2d_d[], 4s_4s_s[], 2s_2s_s[] -------- */ |
| 9785 | if (bitQ == 0 && size == X11) return False; // implied 1d case |
| 9786 | Bool isD = (size & 1) == 1; |
| 9787 | UInt index; |
| 9788 | if (!isD) index = (bitH << 1) | bitL; |
| 9789 | else if (isD && bitL == 0) index = bitH; |
| 9790 | else return False; // sz:L == x11 => unallocated encoding |
| 9791 | vassert(index < (isD ? 2 : 4)); |
| 9792 | IRType ity = isD ? Ity_F64 : Ity_F32; |
| 9793 | IRTemp elem = newTemp(ity); |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 9794 | UInt mm = (bitM << 4) | mmLO4; |
sewardj | 85fbb02 | 2014-06-12 13:16:01 +0000 | [diff] [blame] | 9795 | assign(elem, getQRegLane(mm, index, ity)); |
| 9796 | IRTemp dupd = math_DUP_TO_V128(elem, ity); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9797 | IRTemp res = newTempV128(); |
sewardj | 85fbb02 | 2014-06-12 13:16:01 +0000 | [diff] [blame] | 9798 | assign(res, triop(isD ? Iop_Mul64Fx2 : Iop_Mul32Fx4, |
| 9799 | mkexpr(mk_get_IR_rounding_mode()), |
| 9800 | getQReg128(nn), mkexpr(dupd))); |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9801 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | 85fbb02 | 2014-06-12 13:16:01 +0000 | [diff] [blame] | 9802 | const HChar* arr = bitQ == 0 ? "2s" : (isD ? "2d" : "4s"); |
| 9803 | DIP("fmul %s.%s, %s.%s, %s.%c[%u]\n", nameQReg128(dd), arr, |
| 9804 | nameQReg128(nn), arr, nameQReg128(mm), isD ? 'd' : 's', index); |
| 9805 | return True; |
| 9806 | } |
| 9807 | |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 9808 | if ((bitU == 1 && (opcode == BITS4(0,0,0,0) || opcode == BITS4(0,1,0,0))) |
| 9809 | || (bitU == 0 && opcode == BITS4(1,0,0,0))) { |
| 9810 | /* -------- 1,xx,0000 MLA s/h variants only -------- */ |
| 9811 | /* -------- 1,xx,0100 MLS s/h variants only -------- */ |
| 9812 | /* -------- 0,xx,1000 MUL s/h variants only -------- */ |
| 9813 | Bool isMLA = opcode == BITS4(0,0,0,0); |
| 9814 | Bool isMLS = opcode == BITS4(0,1,0,0); |
| 9815 | UInt mm = 32; // invalid |
| 9816 | UInt ix = 16; // invalid |
| 9817 | switch (size) { |
| 9818 | case X00: |
| 9819 | return False; // b case is not allowed |
| 9820 | case X01: |
| 9821 | mm = mmLO4; ix = (bitH << 2) | (bitL << 1) | (bitM << 0); break; |
| 9822 | case X10: |
| 9823 | mm = (bitM << 4) | mmLO4; ix = (bitH << 1) | (bitL << 0); break; |
| 9824 | case X11: |
| 9825 | return False; // d case is not allowed |
| 9826 | default: |
| 9827 | vassert(0); |
| 9828 | } |
| 9829 | vassert(mm < 32 && ix < 16); |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 9830 | IROp opMUL = mkVecMUL(size); |
| 9831 | IROp opADD = mkVecADD(size); |
| 9832 | IROp opSUB = mkVecSUB(size); |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 9833 | HChar ch = size == X01 ? 'h' : 's'; |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 9834 | IRTemp vecM = math_DUP_VEC_ELEM(getQReg128(mm), size, ix); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9835 | IRTemp vecD = newTempV128(); |
| 9836 | IRTemp vecN = newTempV128(); |
| 9837 | IRTemp res = newTempV128(); |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 9838 | assign(vecD, getQReg128(dd)); |
| 9839 | assign(vecN, getQReg128(nn)); |
| 9840 | IRExpr* prod = binop(opMUL, mkexpr(vecN), mkexpr(vecM)); |
| 9841 | if (isMLA || isMLS) { |
| 9842 | assign(res, binop(isMLA ? opADD : opSUB, mkexpr(vecD), prod)); |
| 9843 | } else { |
| 9844 | assign(res, prod); |
| 9845 | } |
sewardj | df9d6d5 | 2014-06-27 10:43:22 +0000 | [diff] [blame] | 9846 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
sewardj | 787a67f | 2014-06-23 09:09:41 +0000 | [diff] [blame] | 9847 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9848 | DIP("%s %s.%s, %s.%s, %s.%c[%u]\n", isMLA ? "mla" |
| 9849 | : (isMLS ? "mls" : "mul"), |
| 9850 | nameQReg128(dd), arr, |
| 9851 | nameQReg128(nn), arr, nameQReg128(dd), ch, ix); |
| 9852 | return True; |
| 9853 | } |
| 9854 | |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 9855 | if (opcode == BITS4(1,0,1,0) |
| 9856 | || opcode == BITS4(0,0,1,0) || opcode == BITS4(0,1,1,0)) { |
| 9857 | /* -------- 0,xx,1010 SMULL s/h variants only -------- */ // 0 (ks) |
| 9858 | /* -------- 1,xx,1010 UMULL s/h variants only -------- */ // 0 |
| 9859 | /* -------- 0,xx,0010 SMLAL s/h variants only -------- */ // 1 |
| 9860 | /* -------- 1,xx,0010 UMLAL s/h variants only -------- */ // 1 |
| 9861 | /* -------- 0,xx,0110 SMLSL s/h variants only -------- */ // 2 |
| 9862 | /* -------- 1,xx,0110 SMLSL s/h variants only -------- */ // 2 |
| 9863 | /* Widens, and size refers to the narrowed lanes. */ |
| 9864 | UInt ks = 3; |
| 9865 | switch (opcode) { |
| 9866 | case BITS4(1,0,1,0): ks = 0; break; |
| 9867 | case BITS4(0,0,1,0): ks = 1; break; |
| 9868 | case BITS4(0,1,1,0): ks = 2; break; |
| 9869 | default: vassert(0); |
| 9870 | } |
| 9871 | vassert(ks >= 0 && ks <= 2); |
| 9872 | Bool isU = bitU == 1; |
| 9873 | Bool is2 = bitQ == 1; |
| 9874 | UInt mm = 32; // invalid |
| 9875 | UInt ix = 16; // invalid |
| 9876 | switch (size) { |
| 9877 | case X00: |
| 9878 | return False; // h_b_b[] case is not allowed |
| 9879 | case X01: |
| 9880 | mm = mmLO4; ix = (bitH << 2) | (bitL << 1) | (bitM << 0); break; |
| 9881 | case X10: |
| 9882 | mm = (bitM << 4) | mmLO4; ix = (bitH << 1) | (bitL << 0); break; |
| 9883 | case X11: |
| 9884 | return False; // q_d_d[] case is not allowed |
| 9885 | default: |
| 9886 | vassert(0); |
| 9887 | } |
| 9888 | vassert(mm < 32 && ix < 16); |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 9889 | IRTemp vecN = newTempV128(); |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 9890 | IRTemp vecM = math_DUP_VEC_ELEM(getQReg128(mm), size, ix); |
sewardj | 8e91fd4 | 2014-07-11 12:05:47 +0000 | [diff] [blame] | 9891 | IRTemp vecD = newTempV128(); |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 9892 | assign(vecN, getQReg128(nn)); |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 9893 | assign(vecD, getQReg128(dd)); |
| 9894 | IRTemp res = IRTemp_INVALID; |
| 9895 | math_MULL_ACC(&res, is2, isU, size, "mas"[ks], |
| 9896 | vecN, vecM, ks == 0 ? IRTemp_INVALID : vecD); |
sewardj | 487559e | 2014-07-10 14:22:45 +0000 | [diff] [blame] | 9897 | putQReg128(dd, mkexpr(res)); |
| 9898 | const HChar* nm = ks == 0 ? "mull" : (ks == 1 ? "mlal" : "mlsl"); |
| 9899 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 9900 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
| 9901 | HChar ch = size == X01 ? 'h' : 's'; |
| 9902 | DIP("%c%s%s %s.%s, %s.%s, %s.%c[%u]\n", |
| 9903 | isU ? 'u' : 's', nm, is2 ? "2" : "", |
| 9904 | nameQReg128(dd), arrWide, |
| 9905 | nameQReg128(nn), arrNarrow, nameQReg128(dd), ch, ix); |
| 9906 | return True; |
| 9907 | } |
| 9908 | |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 9909 | if (bitU == 0 |
| 9910 | && (opcode == BITS4(1,0,1,1) |
| 9911 | || opcode == BITS4(0,0,1,1) || opcode == BITS4(0,1,1,1))) { |
| 9912 | /* -------- 0,xx,1011 SQDMULL s/h variants only -------- */ // 0 (ks) |
| 9913 | /* -------- 0,xx,0011 SQDMLAL s/h variants only -------- */ // 1 |
| 9914 | /* -------- 0,xx,0111 SQDMLSL s/h variants only -------- */ // 2 |
| 9915 | /* Widens, and size refers to the narrowed lanes. */ |
| 9916 | UInt ks = 3; |
| 9917 | switch (opcode) { |
| 9918 | case BITS4(1,0,1,1): ks = 0; break; |
| 9919 | case BITS4(0,0,1,1): ks = 1; break; |
| 9920 | case BITS4(0,1,1,1): ks = 2; break; |
| 9921 | default: vassert(0); |
| 9922 | } |
| 9923 | vassert(ks >= 0 && ks <= 2); |
| 9924 | Bool is2 = bitQ == 1; |
| 9925 | UInt mm = 32; // invalid |
| 9926 | UInt ix = 16; // invalid |
| 9927 | switch (size) { |
| 9928 | case X00: |
| 9929 | return False; // h_b_b[] case is not allowed |
| 9930 | case X01: |
| 9931 | mm = mmLO4; ix = (bitH << 2) | (bitL << 1) | (bitM << 0); break; |
| 9932 | case X10: |
| 9933 | mm = (bitM << 4) | mmLO4; ix = (bitH << 1) | (bitL << 0); break; |
| 9934 | case X11: |
| 9935 | return False; // q_d_d[] case is not allowed |
| 9936 | default: |
| 9937 | vassert(0); |
| 9938 | } |
| 9939 | vassert(mm < 32 && ix < 16); |
| 9940 | IRTemp vecN, vecD, res, sat1q, sat1n, sat2q, sat2n; |
| 9941 | vecN = vecD = res = sat1q = sat1n = sat2q = sat2n = IRTemp_INVALID; |
| 9942 | newTempsV128_2(&vecN, &vecD); |
| 9943 | assign(vecN, getQReg128(nn)); |
| 9944 | IRTemp vecM = math_DUP_VEC_ELEM(getQReg128(mm), size, ix); |
| 9945 | assign(vecD, getQReg128(dd)); |
| 9946 | math_SQDMULL_ACC(&res, &sat1q, &sat1n, &sat2q, &sat2n, |
| 9947 | is2, size, "mas"[ks], |
| 9948 | vecN, vecM, ks == 0 ? IRTemp_INVALID : vecD); |
| 9949 | putQReg128(dd, mkexpr(res)); |
| 9950 | vassert(sat1q != IRTemp_INVALID && sat1n != IRTemp_INVALID); |
| 9951 | updateQCFLAGwithDifference(sat1q, sat1n); |
| 9952 | if (sat2q != IRTemp_INVALID || sat2n != IRTemp_INVALID) { |
| 9953 | updateQCFLAGwithDifference(sat2q, sat2n); |
| 9954 | } |
sewardj | 54ffa1d | 2014-07-22 09:27:49 +0000 | [diff] [blame] | 9955 | const HChar* nm = ks == 0 ? "sqdmull" |
sewardj | 51d012a | 2014-07-21 09:19:50 +0000 | [diff] [blame] | 9956 | : (ks == 1 ? "sqdmlal" : "sqdmlsl"); |
| 9957 | const HChar* arrNarrow = nameArr_Q_SZ(bitQ, size); |
| 9958 | const HChar* arrWide = nameArr_Q_SZ(1, size+1); |
| 9959 | HChar ch = size == X01 ? 'h' : 's'; |
| 9960 | DIP("%s%s %s.%s, %s.%s, %s.%c[%u]\n", |
| 9961 | nm, is2 ? "2" : "", |
| 9962 | nameQReg128(dd), arrWide, |
| 9963 | nameQReg128(nn), arrNarrow, nameQReg128(dd), ch, ix); |
| 9964 | return True; |
| 9965 | } |
| 9966 | |
sewardj | 257e99f | 2014-08-03 12:45:19 +0000 | [diff] [blame] | 9967 | if (opcode == BITS4(1,1,0,0) || opcode == BITS4(1,1,0,1)) { |
| 9968 | /* -------- 0,xx,1100 SQDMULH s and h variants only -------- */ |
| 9969 | /* -------- 0,xx,1101 SQRDMULH s and h variants only -------- */ |
| 9970 | UInt mm = 32; // invalid |
| 9971 | UInt ix = 16; // invalid |
| 9972 | switch (size) { |
| 9973 | case X00: |
| 9974 | return False; // b case is not allowed |
| 9975 | case X01: |
| 9976 | mm = mmLO4; ix = (bitH << 2) | (bitL << 1) | (bitM << 0); break; |
| 9977 | case X10: |
| 9978 | mm = (bitM << 4) | mmLO4; ix = (bitH << 1) | (bitL << 0); break; |
| 9979 | case X11: |
| 9980 | return False; // q case is not allowed |
| 9981 | default: |
| 9982 | vassert(0); |
| 9983 | } |
| 9984 | vassert(mm < 32 && ix < 16); |
| 9985 | Bool isR = opcode == BITS4(1,1,0,1); |
| 9986 | IRTemp res, sat1q, sat1n, vN, vM; |
| 9987 | res = sat1q = sat1n = vN = vM = IRTemp_INVALID; |
| 9988 | vN = newTempV128(); |
| 9989 | assign(vN, getQReg128(nn)); |
| 9990 | vM = math_DUP_VEC_ELEM(getQReg128(mm), size, ix); |
| 9991 | math_SQDMULH(&res, &sat1q, &sat1n, isR, size, vN, vM); |
| 9992 | putQReg128(dd, math_MAYBE_ZERO_HI64(bitQ, res)); |
| 9993 | IROp opZHI = bitQ == 0 ? Iop_ZeroHI64ofV128 : Iop_INVALID; |
| 9994 | updateQCFLAGwithDifferenceZHI(sat1q, sat1n, opZHI); |
| 9995 | const HChar* nm = isR ? "sqrdmulh" : "sqdmulh"; |
| 9996 | const HChar* arr = nameArr_Q_SZ(bitQ, size); |
| 9997 | HChar ch = size == X01 ? 'h' : 's'; |
| 9998 | DIP("%s %s.%s, %s.%s, %s.%c[%u]\n", nm, |
| 9999 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(dd), ch, ix); |
| 10000 | return True; |
| 10001 | } |
| 10002 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10003 | return False; |
| 10004 | # undef INSN |
| 10005 | } |
| 10006 | |
sewardj | fc83d2c | 2014-06-12 10:15:46 +0000 | [diff] [blame] | 10007 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10008 | static |
| 10009 | Bool dis_AdvSIMD_crypto_aes(/*MB_OUT*/DisResult* dres, UInt insn) |
| 10010 | { |
| 10011 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 10012 | return False; |
| 10013 | # undef INSN |
| 10014 | } |
| 10015 | |
sewardj | fc83d2c | 2014-06-12 10:15:46 +0000 | [diff] [blame] | 10016 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10017 | static |
| 10018 | Bool dis_AdvSIMD_crypto_three_reg_sha(/*MB_OUT*/DisResult* dres, UInt insn) |
| 10019 | { |
| 10020 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 10021 | return False; |
| 10022 | # undef INSN |
| 10023 | } |
| 10024 | |
sewardj | fc83d2c | 2014-06-12 10:15:46 +0000 | [diff] [blame] | 10025 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10026 | static |
| 10027 | Bool dis_AdvSIMD_crypto_two_reg_sha(/*MB_OUT*/DisResult* dres, UInt insn) |
| 10028 | { |
| 10029 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 10030 | return False; |
| 10031 | # undef INSN |
| 10032 | } |
| 10033 | |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10034 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10035 | static |
| 10036 | Bool dis_AdvSIMD_fp_compare(/*MB_OUT*/DisResult* dres, UInt insn) |
| 10037 | { |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10038 | /* 31 28 23 21 20 15 13 9 4 |
| 10039 | 000 11110 ty 1 m op 1000 n opcode2 |
| 10040 | The first 3 bits are really "M 0 S", but M and S are always zero. |
| 10041 | Decode fields are: ty,op,opcode2 |
| 10042 | */ |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10043 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10044 | if (INSN(31,24) != BITS8(0,0,0,1,1,1,1,0) |
| 10045 | || INSN(21,21) != 1 || INSN(13,10) != BITS4(1,0,0,0)) { |
| 10046 | return False; |
| 10047 | } |
| 10048 | UInt ty = INSN(23,22); |
| 10049 | UInt mm = INSN(20,16); |
| 10050 | UInt op = INSN(15,14); |
| 10051 | UInt nn = INSN(9,5); |
| 10052 | UInt opcode2 = INSN(4,0); |
| 10053 | vassert(ty < 4); |
| 10054 | |
| 10055 | if (ty <= X01 && op == X00 |
| 10056 | && (opcode2 & BITS5(0,0,1,1,1)) == BITS5(0,0,0,0,0)) { |
| 10057 | /* -------- 0x,00,00000 FCMP d_d, s_s -------- */ |
| 10058 | /* -------- 0x,00,01000 FCMP d_#0, s_#0 -------- */ |
| 10059 | /* -------- 0x,00,10000 FCMPE d_d, s_s -------- */ |
| 10060 | /* -------- 0x,00,11000 FCMPE d_#0, s_#0 -------- */ |
| 10061 | /* 31 23 20 15 9 4 |
| 10062 | 000 11110 01 1 m 00 1000 n 10 000 FCMPE Dn, Dm |
| 10063 | 000 11110 01 1 00000 00 1000 n 11 000 FCMPE Dn, #0.0 |
| 10064 | 000 11110 01 1 m 00 1000 n 00 000 FCMP Dn, Dm |
| 10065 | 000 11110 01 1 00000 00 1000 n 01 000 FCMP Dn, #0.0 |
| 10066 | |
| 10067 | 000 11110 00 1 m 00 1000 n 10 000 FCMPE Sn, Sm |
| 10068 | 000 11110 00 1 00000 00 1000 n 11 000 FCMPE Sn, #0.0 |
| 10069 | 000 11110 00 1 m 00 1000 n 00 000 FCMP Sn, Sm |
| 10070 | 000 11110 00 1 00000 00 1000 n 01 000 FCMP Sn, #0.0 |
| 10071 | |
| 10072 | FCMPE generates Invalid Operation exn if either arg is any kind |
| 10073 | of NaN. FCMP generates Invalid Operation exn if either arg is a |
| 10074 | signalling NaN. We ignore this detail here and produce the same |
| 10075 | IR for both. |
| 10076 | */ |
| 10077 | Bool isD = (ty & 1) == 1; |
| 10078 | Bool isCMPE = (opcode2 & 16) == 16; |
| 10079 | Bool cmpZero = (opcode2 & 8) == 8; |
| 10080 | IRType ity = isD ? Ity_F64 : Ity_F32; |
| 10081 | Bool valid = True; |
| 10082 | if (cmpZero && mm != 0) valid = False; |
| 10083 | if (valid) { |
| 10084 | IRTemp argL = newTemp(ity); |
| 10085 | IRTemp argR = newTemp(ity); |
| 10086 | IRTemp irRes = newTemp(Ity_I32); |
| 10087 | assign(argL, getQRegLO(nn, ity)); |
| 10088 | assign(argR, |
| 10089 | cmpZero |
| 10090 | ? (IRExpr_Const(isD ? IRConst_F64i(0) : IRConst_F32i(0))) |
| 10091 | : getQRegLO(mm, ity)); |
| 10092 | assign(irRes, binop(isD ? Iop_CmpF64 : Iop_CmpF32, |
| 10093 | mkexpr(argL), mkexpr(argR))); |
| 10094 | IRTemp nzcv = mk_convert_IRCmpF64Result_to_NZCV(irRes); |
| 10095 | IRTemp nzcv_28x0 = newTemp(Ity_I64); |
| 10096 | assign(nzcv_28x0, binop(Iop_Shl64, mkexpr(nzcv), mkU8(28))); |
| 10097 | setFlags_COPY(nzcv_28x0); |
| 10098 | DIP("fcmp%s %s, %s\n", isCMPE ? "e" : "", nameQRegLO(nn, ity), |
| 10099 | cmpZero ? "#0.0" : nameQRegLO(mm, ity)); |
| 10100 | return True; |
| 10101 | } |
| 10102 | return False; |
| 10103 | } |
| 10104 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10105 | return False; |
| 10106 | # undef INSN |
| 10107 | } |
| 10108 | |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10109 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10110 | static |
| 10111 | Bool dis_AdvSIMD_fp_conditional_compare(/*MB_OUT*/DisResult* dres, UInt insn) |
| 10112 | { |
| 10113 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 10114 | return False; |
| 10115 | # undef INSN |
| 10116 | } |
| 10117 | |
sewardj | fc83d2c | 2014-06-12 10:15:46 +0000 | [diff] [blame] | 10118 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10119 | static |
| 10120 | Bool dis_AdvSIMD_fp_conditional_select(/*MB_OUT*/DisResult* dres, UInt insn) |
| 10121 | { |
| 10122 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 10123 | return False; |
| 10124 | # undef INSN |
| 10125 | } |
| 10126 | |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10127 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10128 | static |
| 10129 | Bool dis_AdvSIMD_fp_data_proc_1_source(/*MB_OUT*/DisResult* dres, UInt insn) |
| 10130 | { |
| 10131 | /* 31 28 23 21 20 14 9 4 |
| 10132 | 000 11110 ty 1 opcode 10000 n d |
| 10133 | The first 3 bits are really "M 0 S", but M and S are always zero. |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10134 | Decode fields: ty,opcode |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10135 | */ |
| 10136 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 10137 | if (INSN(31,24) != BITS8(0,0,0,1,1,1,1,0) |
| 10138 | || INSN(21,21) != 1 || INSN(14,10) != BITS5(1,0,0,0,0)) { |
| 10139 | return False; |
| 10140 | } |
| 10141 | UInt ty = INSN(23,22); |
| 10142 | UInt opcode = INSN(20,15); |
| 10143 | UInt nn = INSN(9,5); |
| 10144 | UInt dd = INSN(4,0); |
| 10145 | |
| 10146 | if (ty <= X01 && opcode <= BITS6(0,0,0,0,1,1)) { |
| 10147 | /* -------- 0x,000000: FMOV d_d, s_s -------- */ |
| 10148 | /* -------- 0x,000001: FABS d_d, s_s -------- */ |
| 10149 | /* -------- 0x,000010: FNEG d_d, s_s -------- */ |
| 10150 | /* -------- 0x,000011: FSQRT d_d, s_s -------- */ |
| 10151 | IRType ity = ty == X01 ? Ity_F64 : Ity_F32; |
| 10152 | IRTemp src = newTemp(ity); |
| 10153 | IRTemp res = newTemp(ity); |
| 10154 | const HChar* nm = "??"; |
| 10155 | assign(src, getQRegLO(nn, ity)); |
| 10156 | switch (opcode) { |
| 10157 | case BITS6(0,0,0,0,0,0): |
| 10158 | nm = "fmov"; assign(res, mkexpr(src)); break; |
| 10159 | case BITS6(0,0,0,0,0,1): |
| 10160 | nm = "fabs"; assign(res, unop(mkABSF(ity), mkexpr(src))); break; |
| 10161 | case BITS6(0,0,0,0,1,0): |
| 10162 | nm = "fabs"; assign(res, unop(mkNEGF(ity), mkexpr(src))); break; |
| 10163 | case BITS6(0,0,0,0,1,1): |
| 10164 | nm = "fsqrt"; |
| 10165 | assign(res, binop(mkSQRTF(ity), |
| 10166 | mkexpr(mk_get_IR_rounding_mode()), |
| 10167 | mkexpr(src))); break; |
| 10168 | default: |
| 10169 | vassert(0); |
| 10170 | } |
| 10171 | putQReg128(dd, mkV128(0x0000)); |
| 10172 | putQRegLO(dd, mkexpr(res)); |
| 10173 | DIP("%s %s, %s\n", nm, nameQRegLO(dd, ity), nameQRegLO(nn, ity)); |
| 10174 | return True; |
| 10175 | } |
| 10176 | |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10177 | if ( (ty == X11 && (opcode == BITS6(0,0,0,1,0,0) |
| 10178 | || opcode == BITS6(0,0,0,1,0,1))) |
| 10179 | || (ty == X00 && (opcode == BITS6(0,0,0,1,1,1) |
| 10180 | || opcode == BITS6(0,0,0,1,0,1))) |
| 10181 | || (ty == X01 && (opcode == BITS6(0,0,0,1,1,1) |
| 10182 | || opcode == BITS6(0,0,0,1,0,0)))) { |
| 10183 | /* -------- 11,000100: FCVT s_h -------- */ |
| 10184 | /* -------- 11,000101: FCVT d_h -------- */ |
| 10185 | /* -------- 00,000111: FCVT h_s -------- */ |
| 10186 | /* -------- 00,000101: FCVT d_s -------- */ |
| 10187 | /* -------- 01,000111: FCVT h_d -------- */ |
| 10188 | /* -------- 01,000100: FCVT s_d -------- */ |
| 10189 | /* 31 23 21 16 14 9 4 |
| 10190 | 000 11110 11 10001 00 10000 n d FCVT Sd, Hn (unimp) |
| 10191 | --------- 11 ----- 01 --------- FCVT Dd, Hn (unimp) |
| 10192 | --------- 00 ----- 11 --------- FCVT Hd, Sn (unimp) |
| 10193 | --------- 00 ----- 01 --------- FCVT Dd, Sn |
| 10194 | --------- 01 ----- 11 --------- FCVT Hd, Dn (unimp) |
| 10195 | --------- 01 ----- 00 --------- FCVT Sd, Dn |
| 10196 | Rounding, when dst is smaller than src, is per the FPCR. |
| 10197 | */ |
| 10198 | UInt b2322 = ty; |
| 10199 | UInt b1615 = opcode & BITS2(1,1); |
| 10200 | if (b2322 == BITS2(0,0) && b1615 == BITS2(0,1)) { |
| 10201 | /* Convert S to D */ |
| 10202 | IRTemp res = newTemp(Ity_F64); |
| 10203 | assign(res, unop(Iop_F32toF64, getQRegLO(nn, Ity_F32))); |
| 10204 | putQReg128(dd, mkV128(0x0000)); |
| 10205 | putQRegLO(dd, mkexpr(res)); |
| 10206 | DIP("fcvt %s, %s\n", |
| 10207 | nameQRegLO(dd, Ity_F64), nameQRegLO(nn, Ity_F32)); |
| 10208 | return True; |
| 10209 | } |
| 10210 | if (b2322 == BITS2(0,1) && b1615 == BITS2(0,0)) { |
| 10211 | /* Convert D to S */ |
| 10212 | IRTemp res = newTemp(Ity_F32); |
| 10213 | assign(res, binop(Iop_F64toF32, mkexpr(mk_get_IR_rounding_mode()), |
| 10214 | getQRegLO(nn, Ity_F64))); |
| 10215 | putQReg128(dd, mkV128(0x0000)); |
| 10216 | putQRegLO(dd, mkexpr(res)); |
| 10217 | DIP("fcvt %s, %s\n", |
| 10218 | nameQRegLO(dd, Ity_F32), nameQRegLO(nn, Ity_F64)); |
| 10219 | return True; |
| 10220 | } |
| 10221 | /* else unhandled */ |
| 10222 | return False; |
| 10223 | } |
| 10224 | |
| 10225 | if (ty <= X01 |
| 10226 | && opcode >= BITS6(0,0,1,0,0,0) && opcode <= BITS6(0,0,1,1,1,1) |
| 10227 | && opcode != BITS6(0,0,1,1,0,1)) { |
| 10228 | /* -------- 0x,001000 FRINTN d_d, s_s -------- */ |
| 10229 | /* -------- 0x,001001 FRINTP d_d, s_s -------- */ |
| 10230 | /* -------- 0x,001010 FRINTM d_d, s_s -------- */ |
| 10231 | /* -------- 0x,001011 FRINTZ d_d, s_s -------- */ |
| 10232 | /* -------- 0x,001100 FRINTA d_d, s_s -------- */ |
| 10233 | /* -------- 0x,001110 FRINTX d_d, s_s -------- */ |
| 10234 | /* -------- 0x,001111 FRINTI d_d, s_s -------- */ |
| 10235 | /* 31 23 21 17 14 9 4 |
| 10236 | 000 11110 0x 1001 111 10000 n d FRINTI Fd, Fm (round per FPCR) |
| 10237 | rm |
| 10238 | x==0 => S-registers, x==1 => D-registers |
| 10239 | rm (17:15) encodings: |
| 10240 | 111 per FPCR (FRINTI) |
| 10241 | 001 +inf (FRINTP) |
| 10242 | 010 -inf (FRINTM) |
| 10243 | 011 zero (FRINTZ) |
| 10244 | 000 tieeven |
| 10245 | 100 tieaway (FRINTA) -- !! FIXME KLUDGED !! |
| 10246 | 110 per FPCR + "exact = TRUE" |
| 10247 | 101 unallocated |
| 10248 | */ |
| 10249 | Bool isD = (ty & 1) == 1; |
| 10250 | UInt rm = opcode & BITS6(0,0,0,1,1,1); |
| 10251 | IRType ity = isD ? Ity_F64 : Ity_F32; |
| 10252 | IRExpr* irrmE = NULL; |
| 10253 | UChar ch = '?'; |
| 10254 | switch (rm) { |
| 10255 | case BITS3(0,1,1): ch = 'z'; irrmE = mkU32(Irrm_ZERO); break; |
| 10256 | case BITS3(0,1,0): ch = 'm'; irrmE = mkU32(Irrm_NegINF); break; |
| 10257 | case BITS3(0,0,1): ch = 'p'; irrmE = mkU32(Irrm_PosINF); break; |
| 10258 | // The following is a kludge. Should be: Irrm_NEAREST_TIE_AWAY_0 |
| 10259 | case BITS3(1,0,0): ch = 'a'; irrmE = mkU32(Irrm_NEAREST); break; |
| 10260 | default: break; |
| 10261 | } |
| 10262 | if (irrmE) { |
| 10263 | IRTemp src = newTemp(ity); |
| 10264 | IRTemp dst = newTemp(ity); |
| 10265 | assign(src, getQRegLO(nn, ity)); |
| 10266 | assign(dst, binop(isD ? Iop_RoundF64toInt : Iop_RoundF32toInt, |
| 10267 | irrmE, mkexpr(src))); |
| 10268 | putQReg128(dd, mkV128(0x0000)); |
| 10269 | putQRegLO(dd, mkexpr(dst)); |
| 10270 | DIP("frint%c %s, %s\n", |
| 10271 | ch, nameQRegLO(dd, ity), nameQRegLO(nn, ity)); |
| 10272 | return True; |
| 10273 | } |
| 10274 | return False; |
| 10275 | } |
| 10276 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10277 | return False; |
| 10278 | # undef INSN |
| 10279 | } |
| 10280 | |
| 10281 | |
| 10282 | static |
| 10283 | Bool dis_AdvSIMD_fp_data_proc_2_source(/*MB_OUT*/DisResult* dres, UInt insn) |
| 10284 | { |
| 10285 | /* 31 28 23 21 20 15 11 9 4 |
| 10286 | 000 11110 ty 1 m opcode 10 n d |
| 10287 | The first 3 bits are really "M 0 S", but M and S are always zero. |
| 10288 | */ |
| 10289 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 10290 | if (INSN(31,24) != BITS8(0,0,0,1,1,1,1,0) |
| 10291 | || INSN(21,21) != 1 || INSN(11,10) != BITS2(1,0)) { |
| 10292 | return False; |
| 10293 | } |
| 10294 | UInt ty = INSN(23,22); |
| 10295 | UInt mm = INSN(20,16); |
| 10296 | UInt opcode = INSN(15,12); |
| 10297 | UInt nn = INSN(9,5); |
| 10298 | UInt dd = INSN(4,0); |
| 10299 | |
| 10300 | if (ty <= X01 && opcode <= BITS4(0,0,1,1)) { |
| 10301 | /* ------- 0x,0000: FMUL d_d, s_s ------- */ |
| 10302 | /* ------- 0x,0001: FDIV d_d, s_s ------- */ |
| 10303 | /* ------- 0x,0010: FADD d_d, s_s ------- */ |
| 10304 | /* ------- 0x,0011: FSUB d_d, s_s ------- */ |
| 10305 | IRType ity = ty == X00 ? Ity_F32 : Ity_F64; |
| 10306 | IROp iop = Iop_INVALID; |
| 10307 | const HChar* nm = "???"; |
| 10308 | switch (opcode) { |
| 10309 | case BITS4(0,0,0,0): nm = "fmul"; iop = mkMULF(ity); break; |
| 10310 | case BITS4(0,0,0,1): nm = "fdiv"; iop = mkDIVF(ity); break; |
| 10311 | case BITS4(0,0,1,0): nm = "fadd"; iop = mkADDF(ity); break; |
| 10312 | case BITS4(0,0,1,1): nm = "fsub"; iop = mkSUBF(ity); break; |
| 10313 | default: vassert(0); |
| 10314 | } |
| 10315 | IRExpr* resE = triop(iop, mkexpr(mk_get_IR_rounding_mode()), |
| 10316 | getQRegLO(nn, ity), getQRegLO(mm, ity)); |
| 10317 | IRTemp res = newTemp(ity); |
| 10318 | assign(res, resE); |
| 10319 | putQReg128(dd, mkV128(0)); |
| 10320 | putQRegLO(dd, mkexpr(res)); |
| 10321 | DIP("%s %s, %s, %s\n", |
| 10322 | nm, nameQRegLO(dd, ity), nameQRegLO(nn, ity), nameQRegLO(mm, ity)); |
| 10323 | return True; |
| 10324 | } |
| 10325 | |
| 10326 | if (ty <= X01 && opcode == BITS4(1,0,0,0)) { |
| 10327 | /* ------- 0x,1000: FNMUL d_d, s_s ------- */ |
| 10328 | IRType ity = ty == X00 ? Ity_F32 : Ity_F64; |
| 10329 | IROp iop = mkMULF(ity); |
| 10330 | IROp iopn = mkNEGF(ity); |
| 10331 | const HChar* nm = "fnmul"; |
| 10332 | IRExpr* resE = unop(iopn, |
| 10333 | triop(iop, mkexpr(mk_get_IR_rounding_mode()), |
| 10334 | getQRegLO(nn, ity), getQRegLO(mm, ity))); |
| 10335 | IRTemp res = newTemp(ity); |
| 10336 | assign(res, resE); |
| 10337 | putQReg128(dd, mkV128(0)); |
| 10338 | putQRegLO(dd, mkexpr(res)); |
| 10339 | DIP("%s %s, %s, %s\n", |
| 10340 | nm, nameQRegLO(dd, ity), nameQRegLO(nn, ity), nameQRegLO(mm, ity)); |
| 10341 | return True; |
| 10342 | } |
| 10343 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10344 | return False; |
| 10345 | # undef INSN |
| 10346 | } |
| 10347 | |
| 10348 | |
| 10349 | static |
| 10350 | Bool dis_AdvSIMD_fp_data_proc_3_source(/*MB_OUT*/DisResult* dres, UInt insn) |
| 10351 | { |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10352 | /* 31 28 23 21 20 15 14 9 4 |
| 10353 | 000 11111 ty o1 m o0 a n d |
| 10354 | The first 3 bits are really "M 0 S", but M and S are always zero. |
| 10355 | Decode fields: ty,o1,o0 |
| 10356 | */ |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10357 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10358 | if (INSN(31,24) != BITS8(0,0,0,1,1,1,1,1)) { |
| 10359 | return False; |
| 10360 | } |
| 10361 | UInt ty = INSN(23,22); |
| 10362 | UInt bitO1 = INSN(21,21); |
| 10363 | UInt mm = INSN(20,16); |
| 10364 | UInt bitO0 = INSN(15,15); |
| 10365 | UInt aa = INSN(14,10); |
| 10366 | UInt nn = INSN(9,5); |
| 10367 | UInt dd = INSN(4,0); |
| 10368 | vassert(ty < 4); |
| 10369 | |
| 10370 | if (ty <= X01) { |
| 10371 | /* -------- 0x,0,0 FMADD d_d_d_d, s_s_s_s -------- */ |
| 10372 | /* -------- 0x,0,1 FMSUB d_d_d_d, s_s_s_s -------- */ |
| 10373 | /* -------- 0x,1,0 FNMADD d_d_d_d, s_s_s_s -------- */ |
| 10374 | /* -------- 0x,1,1 FNMSUB d_d_d_d, s_s_s_s -------- */ |
| 10375 | /* -------------------- F{N}M{ADD,SUB} -------------------- */ |
| 10376 | /* 31 22 20 15 14 9 4 ix |
| 10377 | 000 11111 0 sz 0 m 0 a n d 0 FMADD Fd,Fn,Fm,Fa |
| 10378 | 000 11111 0 sz 0 m 1 a n d 1 FMSUB Fd,Fn,Fm,Fa |
| 10379 | 000 11111 0 sz 1 m 0 a n d 2 FNMADD Fd,Fn,Fm,Fa |
| 10380 | 000 11111 0 sz 1 m 1 a n d 3 FNMSUB Fd,Fn,Fm,Fa |
| 10381 | where Fx=Dx when sz=1, Fx=Sx when sz=0 |
| 10382 | |
| 10383 | -----SPEC------ ----IMPL---- |
| 10384 | fmadd a + n * m a + n * m |
| 10385 | fmsub a + (-n) * m a - n * m |
| 10386 | fnmadd (-a) + (-n) * m -(a + n * m) |
| 10387 | fnmsub (-a) + n * m -(a - n * m) |
| 10388 | */ |
| 10389 | Bool isD = (ty & 1) == 1; |
| 10390 | UInt ix = (bitO1 << 1) | bitO0; |
| 10391 | IRType ity = isD ? Ity_F64 : Ity_F32; |
| 10392 | IROp opADD = mkADDF(ity); |
| 10393 | IROp opSUB = mkSUBF(ity); |
| 10394 | IROp opMUL = mkMULF(ity); |
| 10395 | IROp opNEG = mkNEGF(ity); |
| 10396 | IRTemp res = newTemp(ity); |
| 10397 | IRExpr* eA = getQRegLO(aa, ity); |
| 10398 | IRExpr* eN = getQRegLO(nn, ity); |
| 10399 | IRExpr* eM = getQRegLO(mm, ity); |
| 10400 | IRExpr* rm = mkexpr(mk_get_IR_rounding_mode()); |
| 10401 | IRExpr* eNxM = triop(opMUL, rm, eN, eM); |
| 10402 | switch (ix) { |
| 10403 | case 0: assign(res, triop(opADD, rm, eA, eNxM)); break; |
| 10404 | case 1: assign(res, triop(opSUB, rm, eA, eNxM)); break; |
| 10405 | case 2: assign(res, unop(opNEG, triop(opADD, rm, eA, eNxM))); break; |
| 10406 | case 3: assign(res, unop(opNEG, triop(opSUB, rm, eA, eNxM))); break; |
| 10407 | default: vassert(0); |
| 10408 | } |
| 10409 | putQReg128(dd, mkV128(0x0000)); |
| 10410 | putQRegLO(dd, mkexpr(res)); |
| 10411 | const HChar* names[4] = { "fmadd", "fmsub", "fnmadd", "fnmsub" }; |
| 10412 | DIP("%s %s, %s, %s, %s\n", |
| 10413 | names[ix], nameQRegLO(dd, ity), nameQRegLO(nn, ity), |
| 10414 | nameQRegLO(mm, ity), nameQRegLO(aa, ity)); |
| 10415 | return True; |
| 10416 | } |
| 10417 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10418 | return False; |
| 10419 | # undef INSN |
| 10420 | } |
| 10421 | |
| 10422 | |
| 10423 | static |
| 10424 | Bool dis_AdvSIMD_fp_immediate(/*MB_OUT*/DisResult* dres, UInt insn) |
| 10425 | { |
| 10426 | /* 31 28 23 21 20 12 9 4 |
| 10427 | 000 11110 ty 1 imm8 100 imm5 d |
| 10428 | The first 3 bits are really "M 0 S", but M and S are always zero. |
| 10429 | */ |
| 10430 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 10431 | if (INSN(31,24) != BITS8(0,0,0,1,1,1,1,0) |
| 10432 | || INSN(21,21) != 1 || INSN(12,10) != BITS3(1,0,0)) { |
| 10433 | return False; |
| 10434 | } |
| 10435 | UInt ty = INSN(23,22); |
| 10436 | UInt imm8 = INSN(20,13); |
| 10437 | UInt imm5 = INSN(9,5); |
| 10438 | UInt dd = INSN(4,0); |
| 10439 | |
| 10440 | /* ------- 00,00000: FMOV s_imm ------- */ |
| 10441 | /* ------- 01,00000: FMOV d_imm ------- */ |
| 10442 | if (ty <= X01 && imm5 == BITS5(0,0,0,0,0)) { |
| 10443 | Bool isD = (ty & 1) == 1; |
| 10444 | ULong imm = VFPExpandImm(imm8, isD ? 64 : 32); |
| 10445 | if (!isD) { |
| 10446 | vassert(0 == (imm & 0xFFFFFFFF00000000ULL)); |
| 10447 | } |
| 10448 | putQReg128(dd, mkV128(0)); |
| 10449 | putQRegLO(dd, isD ? mkU64(imm) : mkU32(imm & 0xFFFFFFFFULL)); |
| 10450 | DIP("fmov %s, #0x%llx\n", |
| 10451 | nameQRegLO(dd, isD ? Ity_F64 : Ity_F32), imm); |
| 10452 | return True; |
| 10453 | } |
| 10454 | |
| 10455 | return False; |
| 10456 | # undef INSN |
| 10457 | } |
| 10458 | |
| 10459 | |
| 10460 | static |
| 10461 | Bool dis_AdvSIMD_fp_to_fixedp_conv(/*MB_OUT*/DisResult* dres, UInt insn) |
| 10462 | { |
| 10463 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 10464 | return False; |
| 10465 | # undef INSN |
| 10466 | } |
| 10467 | |
| 10468 | |
| 10469 | static |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10470 | Bool dis_AdvSIMD_fp_to_from_int_conv(/*MB_OUT*/DisResult* dres, UInt insn) |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10471 | { |
| 10472 | /* 31 30 29 28 23 21 20 18 15 9 4 |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10473 | sf 0 0 11110 type 1 rmode opcode 000000 n d |
| 10474 | The first 3 bits are really "sf 0 S", but S is always zero. |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10475 | */ |
| 10476 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10477 | if (INSN(30,29) != BITS2(0,0) |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10478 | || INSN(28,24) != BITS5(1,1,1,1,0) |
| 10479 | || INSN(21,21) != 1 |
| 10480 | || INSN(15,10) != BITS6(0,0,0,0,0,0)) { |
| 10481 | return False; |
| 10482 | } |
| 10483 | UInt bitSF = INSN(31,31); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10484 | UInt ty = INSN(23,22); // type |
| 10485 | UInt rm = INSN(20,19); // rmode |
| 10486 | UInt op = INSN(18,16); // opcode |
| 10487 | UInt nn = INSN(9,5); |
| 10488 | UInt dd = INSN(4,0); |
| 10489 | |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10490 | // op = 000, 001 |
| 10491 | /* -------- FCVT{N,P,M,Z}{S,U} (scalar, integer) -------- */ |
| 10492 | /* 30 23 20 18 15 9 4 |
| 10493 | sf 00 11110 0x 1 00 000 000000 n d FCVTNS Rd, Fn (round to |
| 10494 | sf 00 11110 0x 1 00 001 000000 n d FCVTNU Rd, Fn nearest) |
| 10495 | ---------------- 01 -------------- FCVTP-------- (round to +inf) |
| 10496 | ---------------- 10 -------------- FCVTM-------- (round to -inf) |
| 10497 | ---------------- 11 -------------- FCVTZ-------- (round to zero) |
| 10498 | |
| 10499 | Rd is Xd when sf==1, Wd when sf==0 |
| 10500 | Fn is Dn when x==1, Sn when x==0 |
| 10501 | 20:19 carry the rounding mode, using the same encoding as FPCR |
| 10502 | */ |
| 10503 | if (ty <= X01 && (op == BITS3(0,0,0) || op == BITS3(0,0,1))) { |
| 10504 | Bool isI64 = bitSF == 1; |
| 10505 | Bool isF64 = (ty & 1) == 1; |
| 10506 | Bool isU = (op & 1) == 1; |
| 10507 | /* Decide on the IR rounding mode to use. */ |
| 10508 | IRRoundingMode irrm = 8; /*impossible*/ |
| 10509 | HChar ch = '?'; |
| 10510 | switch (rm) { |
| 10511 | case BITS2(0,0): ch = 'n'; irrm = Irrm_NEAREST; break; |
| 10512 | case BITS2(0,1): ch = 'p'; irrm = Irrm_PosINF; break; |
| 10513 | case BITS2(1,0): ch = 'm'; irrm = Irrm_NegINF; break; |
| 10514 | case BITS2(1,1): ch = 'z'; irrm = Irrm_ZERO; break; |
| 10515 | default: vassert(0); |
| 10516 | } |
| 10517 | vassert(irrm != 8); |
| 10518 | /* Decide on the conversion primop, based on the source size, |
| 10519 | dest size and signedness (8 possibilities). Case coding: |
| 10520 | F32 ->s I32 0 |
| 10521 | F32 ->u I32 1 |
| 10522 | F32 ->s I64 2 |
| 10523 | F32 ->u I64 3 |
| 10524 | F64 ->s I32 4 |
| 10525 | F64 ->u I32 5 |
| 10526 | F64 ->s I64 6 |
| 10527 | F64 ->u I64 7 |
| 10528 | */ |
| 10529 | UInt ix = (isF64 ? 4 : 0) | (isI64 ? 2 : 0) | (isU ? 1 : 0); |
| 10530 | vassert(ix < 8); |
| 10531 | const IROp iops[8] |
| 10532 | = { Iop_F32toI32S, Iop_F32toI32U, Iop_F32toI64S, Iop_F32toI64U, |
| 10533 | Iop_F64toI32S, Iop_F64toI32U, Iop_F64toI64S, Iop_F64toI64U }; |
| 10534 | IROp iop = iops[ix]; |
| 10535 | // A bit of ATCery: bounce all cases we haven't seen an example of. |
| 10536 | if (/* F32toI32S */ |
| 10537 | (iop == Iop_F32toI32S && irrm == Irrm_ZERO) /* FCVTZS Wd,Sn */ |
| 10538 | || (iop == Iop_F32toI32S && irrm == Irrm_NegINF) /* FCVTMS Wd,Sn */ |
| 10539 | || (iop == Iop_F32toI32S && irrm == Irrm_PosINF) /* FCVTPS Wd,Sn */ |
| 10540 | /* F32toI32U */ |
| 10541 | || (iop == Iop_F32toI32U && irrm == Irrm_ZERO) /* FCVTZU Wd,Sn */ |
| 10542 | || (iop == Iop_F32toI32U && irrm == Irrm_NegINF) /* FCVTMU Wd,Sn */ |
| 10543 | /* F32toI64S */ |
| 10544 | || (iop == Iop_F32toI64S && irrm == Irrm_ZERO) /* FCVTZS Xd,Sn */ |
| 10545 | /* F32toI64U */ |
| 10546 | || (iop == Iop_F32toI64U && irrm == Irrm_ZERO) /* FCVTZU Xd,Sn */ |
| 10547 | /* F64toI32S */ |
| 10548 | || (iop == Iop_F64toI32S && irrm == Irrm_ZERO) /* FCVTZS Wd,Dn */ |
| 10549 | || (iop == Iop_F64toI32S && irrm == Irrm_NegINF) /* FCVTMS Wd,Dn */ |
| 10550 | || (iop == Iop_F64toI32S && irrm == Irrm_PosINF) /* FCVTPS Wd,Dn */ |
| 10551 | /* F64toI32U */ |
| 10552 | || (iop == Iop_F64toI32U && irrm == Irrm_ZERO) /* FCVTZU Wd,Dn */ |
| 10553 | || (iop == Iop_F64toI32U && irrm == Irrm_NegINF) /* FCVTMU Wd,Dn */ |
| 10554 | || (iop == Iop_F64toI32U && irrm == Irrm_PosINF) /* FCVTPU Wd,Dn */ |
| 10555 | /* F64toI64S */ |
| 10556 | || (iop == Iop_F64toI64S && irrm == Irrm_ZERO) /* FCVTZS Xd,Dn */ |
| 10557 | || (iop == Iop_F64toI64S && irrm == Irrm_NegINF) /* FCVTMS Xd,Dn */ |
| 10558 | || (iop == Iop_F64toI64S && irrm == Irrm_PosINF) /* FCVTPS Xd,Dn */ |
| 10559 | /* F64toI64U */ |
| 10560 | || (iop == Iop_F64toI64U && irrm == Irrm_ZERO) /* FCVTZU Xd,Dn */ |
| 10561 | || (iop == Iop_F64toI64U && irrm == Irrm_PosINF) /* FCVTPU Xd,Dn */ |
| 10562 | ) { |
| 10563 | /* validated */ |
| 10564 | } else { |
| 10565 | return False; |
| 10566 | } |
| 10567 | IRType srcTy = isF64 ? Ity_F64 : Ity_F32; |
| 10568 | IRType dstTy = isI64 ? Ity_I64 : Ity_I32; |
| 10569 | IRTemp src = newTemp(srcTy); |
| 10570 | IRTemp dst = newTemp(dstTy); |
| 10571 | assign(src, getQRegLO(nn, srcTy)); |
| 10572 | assign(dst, binop(iop, mkU32(irrm), mkexpr(src))); |
| 10573 | putIRegOrZR(isI64, dd, mkexpr(dst)); |
| 10574 | DIP("fcvt%c%c %s, %s\n", ch, isU ? 'u' : 's', |
| 10575 | nameIRegOrZR(isI64, dd), nameQRegLO(nn, srcTy)); |
| 10576 | return True; |
| 10577 | } |
| 10578 | |
| 10579 | // op = 010, 011 |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10580 | /* -------------- {S,U}CVTF (scalar, integer) -------------- */ |
| 10581 | /* (ix) sf S 28 ty rm op 15 9 4 |
| 10582 | 0 0 0 0 11110 00 1 00 010 000000 n d SCVTF Sd, Wn |
| 10583 | 1 0 0 0 11110 01 1 00 010 000000 n d SCVTF Dd, Wn |
| 10584 | 2 1 0 0 11110 00 1 00 010 000000 n d SCVTF Sd, Xn |
| 10585 | 3 1 0 0 11110 01 1 00 010 000000 n d SCVTF Dd, Xn |
| 10586 | |
| 10587 | 4 0 0 0 11110 00 1 00 011 000000 n d UCVTF Sd, Wn |
| 10588 | 5 0 0 0 11110 01 1 00 011 000000 n d UCVTF Dd, Wn |
| 10589 | 6 1 0 0 11110 00 1 00 011 000000 n d UCVTF Sd, Xn |
| 10590 | 7 1 0 0 11110 01 1 00 011 000000 n d UCVTF Dd, Xn |
| 10591 | |
| 10592 | These are signed/unsigned conversion from integer registers to |
| 10593 | FP registers, all 4 32/64-bit combinations, rounded per FPCR. |
| 10594 | */ |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10595 | if (ty <= X01 && rm == X00 && (op == BITS3(0,1,0) || op == BITS3(0,1,1))) { |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10596 | Bool isI64 = bitSF == 1; |
| 10597 | Bool isF64 = (ty & 1) == 1; |
| 10598 | Bool isU = (op & 1) == 1; |
| 10599 | UInt ix = (isU ? 4 : 0) | (isI64 ? 2 : 0) | (isF64 ? 1 : 0); |
| 10600 | const IROp ops[8] |
| 10601 | = { Iop_I32StoF32, Iop_I32StoF64, Iop_I64StoF32, Iop_I64StoF64, |
| 10602 | Iop_I32UtoF32, Iop_I32UtoF64, Iop_I64UtoF32, Iop_I64UtoF64 }; |
| 10603 | IRExpr* src = getIRegOrZR(isI64, nn); |
| 10604 | IRExpr* res = (isF64 && !isI64) |
| 10605 | ? unop(ops[ix], src) |
| 10606 | : binop(ops[ix], mkexpr(mk_get_IR_rounding_mode()), src); |
| 10607 | putQReg128(dd, mkV128(0)); |
| 10608 | putQRegLO(dd, res); |
| 10609 | DIP("%ccvtf %s, %s\n", |
| 10610 | isU ? 'u' : 's', nameQRegLO(dd, isF64 ? Ity_F64 : Ity_F32), |
| 10611 | nameIRegOrZR(isI64, nn)); |
| 10612 | return True; |
| 10613 | } |
| 10614 | |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10615 | // op = 110, 111 |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10616 | /* -------- FMOV (general) -------- */ |
| 10617 | /* case sf S ty rm op 15 9 4 |
| 10618 | (1) 0 0 0 11110 00 1 00 111 000000 n d FMOV Sd, Wn |
| 10619 | (2) 1 0 0 11110 01 1 00 111 000000 n d FMOV Dd, Xn |
| 10620 | (3) 1 0 0 11110 10 1 01 111 000000 n d FMOV Vd.D[1], Xn |
| 10621 | |
| 10622 | (4) 0 0 0 11110 00 1 00 110 000000 n d FMOV Wd, Sn |
| 10623 | (5) 1 0 0 11110 01 1 00 110 000000 n d FMOV Xd, Dn |
| 10624 | (6) 1 0 0 11110 10 1 01 110 000000 n d FMOV Xd, Vn.D[1] |
| 10625 | */ |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10626 | if (1) { |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10627 | UInt ix = 0; // case |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10628 | if (bitSF == 0) { |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10629 | if (ty == BITS2(0,0) && rm == BITS2(0,0) && op == BITS3(1,1,1)) |
| 10630 | ix = 1; |
| 10631 | else |
| 10632 | if (ty == BITS2(0,0) && rm == BITS2(0,0) && op == BITS3(1,1,0)) |
| 10633 | ix = 4; |
| 10634 | } else { |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10635 | vassert(bitSF == 1); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10636 | if (ty == BITS2(0,1) && rm == BITS2(0,0) && op == BITS3(1,1,1)) |
| 10637 | ix = 2; |
| 10638 | else |
| 10639 | if (ty == BITS2(0,1) && rm == BITS2(0,0) && op == BITS3(1,1,0)) |
| 10640 | ix = 5; |
| 10641 | else |
| 10642 | if (ty == BITS2(1,0) && rm == BITS2(0,1) && op == BITS3(1,1,1)) |
| 10643 | ix = 3; |
| 10644 | else |
| 10645 | if (ty == BITS2(1,0) && rm == BITS2(0,1) && op == BITS3(1,1,0)) |
| 10646 | ix = 6; |
| 10647 | } |
| 10648 | if (ix > 0) { |
| 10649 | switch (ix) { |
| 10650 | case 1: |
| 10651 | putQReg128(dd, mkV128(0)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 10652 | putQRegLO(dd, getIReg32orZR(nn)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10653 | DIP("fmov s%u, w%u\n", dd, nn); |
| 10654 | break; |
| 10655 | case 2: |
| 10656 | putQReg128(dd, mkV128(0)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 10657 | putQRegLO(dd, getIReg64orZR(nn)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10658 | DIP("fmov d%u, x%u\n", dd, nn); |
| 10659 | break; |
| 10660 | case 3: |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 10661 | putQRegHI64(dd, getIReg64orZR(nn)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10662 | DIP("fmov v%u.d[1], x%u\n", dd, nn); |
| 10663 | break; |
| 10664 | case 4: |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 10665 | putIReg32orZR(dd, getQRegLO(nn, Ity_I32)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10666 | DIP("fmov w%u, s%u\n", dd, nn); |
| 10667 | break; |
| 10668 | case 5: |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 10669 | putIReg64orZR(dd, getQRegLO(nn, Ity_I64)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10670 | DIP("fmov x%u, d%u\n", dd, nn); |
| 10671 | break; |
| 10672 | case 6: |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 10673 | putIReg64orZR(dd, getQRegHI64(nn)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10674 | DIP("fmov x%u, v%u.d[1]\n", dd, nn); |
| 10675 | break; |
| 10676 | default: |
| 10677 | vassert(0); |
| 10678 | } |
| 10679 | return True; |
| 10680 | } |
| 10681 | /* undecodable; fall through */ |
| 10682 | } |
| 10683 | |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10684 | return False; |
| 10685 | # undef INSN |
| 10686 | } |
| 10687 | |
| 10688 | |
| 10689 | static |
| 10690 | Bool dis_ARM64_simd_and_fp(/*MB_OUT*/DisResult* dres, UInt insn) |
| 10691 | { |
| 10692 | Bool ok; |
| 10693 | ok = dis_AdvSIMD_EXT(dres, insn); |
| 10694 | if (UNLIKELY(ok)) return True; |
| 10695 | ok = dis_AdvSIMD_TBL_TBX(dres, insn); |
| 10696 | if (UNLIKELY(ok)) return True; |
| 10697 | ok = dis_AdvSIMD_ZIP_UZP_TRN(dres, insn); |
| 10698 | if (UNLIKELY(ok)) return True; |
| 10699 | ok = dis_AdvSIMD_across_lanes(dres, insn); |
| 10700 | if (UNLIKELY(ok)) return True; |
| 10701 | ok = dis_AdvSIMD_copy(dres, insn); |
| 10702 | if (UNLIKELY(ok)) return True; |
| 10703 | ok = dis_AdvSIMD_modified_immediate(dres, insn); |
| 10704 | if (UNLIKELY(ok)) return True; |
| 10705 | ok = dis_AdvSIMD_scalar_copy(dres, insn); |
| 10706 | if (UNLIKELY(ok)) return True; |
| 10707 | ok = dis_AdvSIMD_scalar_pairwise(dres, insn); |
| 10708 | if (UNLIKELY(ok)) return True; |
| 10709 | ok = dis_AdvSIMD_scalar_shift_by_imm(dres, insn); |
| 10710 | if (UNLIKELY(ok)) return True; |
| 10711 | ok = dis_AdvSIMD_scalar_three_different(dres, insn); |
| 10712 | if (UNLIKELY(ok)) return True; |
| 10713 | ok = dis_AdvSIMD_scalar_three_same(dres, insn); |
| 10714 | if (UNLIKELY(ok)) return True; |
| 10715 | ok = dis_AdvSIMD_scalar_two_reg_misc(dres, insn); |
| 10716 | if (UNLIKELY(ok)) return True; |
| 10717 | ok = dis_AdvSIMD_scalar_x_indexed_element(dres, insn); |
| 10718 | if (UNLIKELY(ok)) return True; |
| 10719 | ok = dis_AdvSIMD_shift_by_immediate(dres, insn); |
| 10720 | if (UNLIKELY(ok)) return True; |
| 10721 | ok = dis_AdvSIMD_three_different(dres, insn); |
| 10722 | if (UNLIKELY(ok)) return True; |
| 10723 | ok = dis_AdvSIMD_three_same(dres, insn); |
| 10724 | if (UNLIKELY(ok)) return True; |
| 10725 | ok = dis_AdvSIMD_two_reg_misc(dres, insn); |
| 10726 | if (UNLIKELY(ok)) return True; |
| 10727 | ok = dis_AdvSIMD_vector_x_indexed_elem(dres, insn); |
| 10728 | if (UNLIKELY(ok)) return True; |
| 10729 | ok = dis_AdvSIMD_crypto_aes(dres, insn); |
| 10730 | if (UNLIKELY(ok)) return True; |
| 10731 | ok = dis_AdvSIMD_crypto_three_reg_sha(dres, insn); |
| 10732 | if (UNLIKELY(ok)) return True; |
| 10733 | ok = dis_AdvSIMD_crypto_two_reg_sha(dres, insn); |
| 10734 | if (UNLIKELY(ok)) return True; |
| 10735 | ok = dis_AdvSIMD_fp_compare(dres, insn); |
| 10736 | if (UNLIKELY(ok)) return True; |
| 10737 | ok = dis_AdvSIMD_fp_conditional_compare(dres, insn); |
| 10738 | if (UNLIKELY(ok)) return True; |
| 10739 | ok = dis_AdvSIMD_fp_conditional_select(dres, insn); |
| 10740 | if (UNLIKELY(ok)) return True; |
| 10741 | ok = dis_AdvSIMD_fp_data_proc_1_source(dres, insn); |
| 10742 | if (UNLIKELY(ok)) return True; |
| 10743 | ok = dis_AdvSIMD_fp_data_proc_2_source(dres, insn); |
| 10744 | if (UNLIKELY(ok)) return True; |
| 10745 | ok = dis_AdvSIMD_fp_data_proc_3_source(dres, insn); |
| 10746 | if (UNLIKELY(ok)) return True; |
| 10747 | ok = dis_AdvSIMD_fp_immediate(dres, insn); |
| 10748 | if (UNLIKELY(ok)) return True; |
| 10749 | ok = dis_AdvSIMD_fp_to_fixedp_conv(dres, insn); |
| 10750 | if (UNLIKELY(ok)) return True; |
sewardj | 5747c4a | 2014-06-11 20:57:23 +0000 | [diff] [blame] | 10751 | ok = dis_AdvSIMD_fp_to_from_int_conv(dres, insn); |
sewardj | df1628c | 2014-06-10 22:52:05 +0000 | [diff] [blame] | 10752 | if (UNLIKELY(ok)) return True; |
| 10753 | return False; |
| 10754 | } |
| 10755 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10756 | |
| 10757 | /*------------------------------------------------------------*/ |
| 10758 | /*--- Disassemble a single ARM64 instruction ---*/ |
| 10759 | /*------------------------------------------------------------*/ |
| 10760 | |
| 10761 | /* Disassemble a single ARM64 instruction into IR. The instruction |
| 10762 | has is located at |guest_instr| and has guest IP of |
| 10763 | |guest_PC_curr_instr|, which will have been set before the call |
| 10764 | here. Returns True iff the instruction was decoded, in which case |
| 10765 | *dres will be set accordingly, or False, in which case *dres should |
| 10766 | be ignored by the caller. */ |
| 10767 | |
| 10768 | static |
| 10769 | Bool disInstr_ARM64_WRK ( |
| 10770 | /*MB_OUT*/DisResult* dres, |
| 10771 | Bool (*resteerOkFn) ( /*opaque*/void*, Addr64 ), |
| 10772 | Bool resteerCisOk, |
| 10773 | void* callback_opaque, |
| 10774 | UChar* guest_instr, |
| 10775 | VexArchInfo* archinfo, |
| 10776 | VexAbiInfo* abiinfo |
| 10777 | ) |
| 10778 | { |
| 10779 | // A macro to fish bits out of 'insn'. |
| 10780 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 10781 | |
| 10782 | //ZZ DisResult dres; |
| 10783 | //ZZ UInt insn; |
| 10784 | //ZZ //Bool allow_VFP = False; |
| 10785 | //ZZ //UInt hwcaps = archinfo->hwcaps; |
| 10786 | //ZZ IRTemp condT; /* :: Ity_I32 */ |
| 10787 | //ZZ UInt summary; |
| 10788 | //ZZ HChar dis_buf[128]; // big enough to hold LDMIA etc text |
| 10789 | //ZZ |
| 10790 | //ZZ /* What insn variants are we supporting today? */ |
| 10791 | //ZZ //allow_VFP = (0 != (hwcaps & VEX_HWCAPS_ARM_VFP)); |
| 10792 | //ZZ // etc etc |
| 10793 | |
| 10794 | /* Set result defaults. */ |
| 10795 | dres->whatNext = Dis_Continue; |
| 10796 | dres->len = 4; |
| 10797 | dres->continueAt = 0; |
| 10798 | dres->jk_StopHere = Ijk_INVALID; |
| 10799 | |
| 10800 | /* At least this is simple on ARM64: insns are all 4 bytes long, and |
| 10801 | 4-aligned. So just fish the whole thing out of memory right now |
| 10802 | and have done. */ |
| 10803 | UInt insn = getUIntLittleEndianly( guest_instr ); |
| 10804 | |
| 10805 | if (0) vex_printf("insn: 0x%x\n", insn); |
| 10806 | |
| 10807 | DIP("\t(arm64) 0x%llx: ", (ULong)guest_PC_curr_instr); |
| 10808 | |
| 10809 | vassert(0 == (guest_PC_curr_instr & 3ULL)); |
| 10810 | |
| 10811 | /* ----------------------------------------------------------- */ |
| 10812 | |
| 10813 | /* Spot "Special" instructions (see comment at top of file). */ |
| 10814 | { |
| 10815 | UChar* code = (UChar*)guest_instr; |
| 10816 | /* Spot the 16-byte preamble: |
| 10817 | 93CC0D8C ror x12, x12, #3 |
| 10818 | 93CC358C ror x12, x12, #13 |
| 10819 | 93CCCD8C ror x12, x12, #51 |
| 10820 | 93CCF58C ror x12, x12, #61 |
| 10821 | */ |
| 10822 | UInt word1 = 0x93CC0D8C; |
| 10823 | UInt word2 = 0x93CC358C; |
| 10824 | UInt word3 = 0x93CCCD8C; |
| 10825 | UInt word4 = 0x93CCF58C; |
| 10826 | if (getUIntLittleEndianly(code+ 0) == word1 && |
| 10827 | getUIntLittleEndianly(code+ 4) == word2 && |
| 10828 | getUIntLittleEndianly(code+ 8) == word3 && |
| 10829 | getUIntLittleEndianly(code+12) == word4) { |
| 10830 | /* Got a "Special" instruction preamble. Which one is it? */ |
| 10831 | if (getUIntLittleEndianly(code+16) == 0xAA0A014A |
| 10832 | /* orr x10,x10,x10 */) { |
| 10833 | /* X3 = client_request ( X4 ) */ |
| 10834 | DIP("x3 = client_request ( x4 )\n"); |
| 10835 | putPC(mkU64( guest_PC_curr_instr + 20 )); |
| 10836 | dres->jk_StopHere = Ijk_ClientReq; |
| 10837 | dres->whatNext = Dis_StopHere; |
| 10838 | return True; |
| 10839 | } |
| 10840 | else |
| 10841 | if (getUIntLittleEndianly(code+16) == 0xAA0B016B |
| 10842 | /* orr x11,x11,x11 */) { |
| 10843 | /* X3 = guest_NRADDR */ |
| 10844 | DIP("x3 = guest_NRADDR\n"); |
| 10845 | dres->len = 20; |
| 10846 | putIReg64orZR(3, IRExpr_Get( OFFB_NRADDR, Ity_I64 )); |
| 10847 | return True; |
| 10848 | } |
| 10849 | else |
| 10850 | if (getUIntLittleEndianly(code+16) == 0xAA0C018C |
| 10851 | /* orr x12,x12,x12 */) { |
| 10852 | /* branch-and-link-to-noredir X8 */ |
| 10853 | DIP("branch-and-link-to-noredir x8\n"); |
| 10854 | putIReg64orZR(30, mkU64(guest_PC_curr_instr + 20)); |
| 10855 | putPC(getIReg64orZR(8)); |
| 10856 | dres->jk_StopHere = Ijk_NoRedir; |
| 10857 | dres->whatNext = Dis_StopHere; |
| 10858 | return True; |
| 10859 | } |
| 10860 | else |
| 10861 | if (getUIntLittleEndianly(code+16) == 0xAA090129 |
| 10862 | /* orr x9,x9,x9 */) { |
| 10863 | /* IR injection */ |
| 10864 | DIP("IR injection\n"); |
| 10865 | vex_inject_ir(irsb, Iend_LE); |
| 10866 | // Invalidate the current insn. The reason is that the IRop we're |
| 10867 | // injecting here can change. In which case the translation has to |
| 10868 | // be redone. For ease of handling, we simply invalidate all the |
| 10869 | // time. |
sewardj | 05f5e01 | 2014-05-04 10:52:11 +0000 | [diff] [blame] | 10870 | stmt(IRStmt_Put(OFFB_CMSTART, mkU64(guest_PC_curr_instr))); |
| 10871 | stmt(IRStmt_Put(OFFB_CMLEN, mkU64(20))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10872 | putPC(mkU64( guest_PC_curr_instr + 20 )); |
| 10873 | dres->whatNext = Dis_StopHere; |
sewardj | 05f5e01 | 2014-05-04 10:52:11 +0000 | [diff] [blame] | 10874 | dres->jk_StopHere = Ijk_InvalICache; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10875 | return True; |
| 10876 | } |
| 10877 | /* We don't know what it is. */ |
| 10878 | return False; |
| 10879 | /*NOTREACHED*/ |
| 10880 | } |
| 10881 | } |
| 10882 | |
| 10883 | /* ----------------------------------------------------------- */ |
| 10884 | |
| 10885 | /* Main ARM64 instruction decoder starts here. */ |
| 10886 | |
| 10887 | Bool ok = False; |
| 10888 | |
| 10889 | /* insn[28:25] determines the top-level grouping, so let's start |
| 10890 | off with that. |
| 10891 | |
| 10892 | For all of these dis_ARM64_ functions, we pass *dres with the |
| 10893 | normal default results "insn OK, 4 bytes long, keep decoding" so |
| 10894 | they don't need to change it. However, decodes of control-flow |
| 10895 | insns may cause *dres to change. |
| 10896 | */ |
| 10897 | switch (INSN(28,25)) { |
| 10898 | case BITS4(1,0,0,0): case BITS4(1,0,0,1): |
| 10899 | // Data processing - immediate |
| 10900 | ok = dis_ARM64_data_processing_immediate(dres, insn); |
| 10901 | break; |
| 10902 | case BITS4(1,0,1,0): case BITS4(1,0,1,1): |
| 10903 | // Branch, exception generation and system instructions |
sewardj | 6590299 | 2014-05-03 21:20:56 +0000 | [diff] [blame] | 10904 | ok = dis_ARM64_branch_etc(dres, insn, archinfo); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10905 | break; |
| 10906 | case BITS4(0,1,0,0): case BITS4(0,1,1,0): |
| 10907 | case BITS4(1,1,0,0): case BITS4(1,1,1,0): |
| 10908 | // Loads and stores |
| 10909 | ok = dis_ARM64_load_store(dres, insn); |
| 10910 | break; |
| 10911 | case BITS4(0,1,0,1): case BITS4(1,1,0,1): |
| 10912 | // Data processing - register |
| 10913 | ok = dis_ARM64_data_processing_register(dres, insn); |
| 10914 | break; |
| 10915 | case BITS4(0,1,1,1): case BITS4(1,1,1,1): |
| 10916 | // Data processing - SIMD and floating point |
| 10917 | ok = dis_ARM64_simd_and_fp(dres, insn); |
| 10918 | break; |
| 10919 | case BITS4(0,0,0,0): case BITS4(0,0,0,1): |
| 10920 | case BITS4(0,0,1,0): case BITS4(0,0,1,1): |
| 10921 | // UNALLOCATED |
| 10922 | break; |
| 10923 | default: |
| 10924 | vassert(0); /* Can't happen */ |
| 10925 | } |
| 10926 | |
| 10927 | /* If the next-level down decoders failed, make sure |dres| didn't |
| 10928 | get changed. */ |
| 10929 | if (!ok) { |
| 10930 | vassert(dres->whatNext == Dis_Continue); |
| 10931 | vassert(dres->len == 4); |
| 10932 | vassert(dres->continueAt == 0); |
| 10933 | vassert(dres->jk_StopHere == Ijk_INVALID); |
| 10934 | } |
| 10935 | |
| 10936 | return ok; |
| 10937 | |
| 10938 | # undef INSN |
| 10939 | } |
| 10940 | |
| 10941 | |
| 10942 | /*------------------------------------------------------------*/ |
| 10943 | /*--- Top-level fn ---*/ |
| 10944 | /*------------------------------------------------------------*/ |
| 10945 | |
| 10946 | /* Disassemble a single instruction into IR. The instruction |
| 10947 | is located in host memory at &guest_code[delta]. */ |
| 10948 | |
| 10949 | DisResult disInstr_ARM64 ( IRSB* irsb_IN, |
| 10950 | Bool (*resteerOkFn) ( void*, Addr64 ), |
| 10951 | Bool resteerCisOk, |
| 10952 | void* callback_opaque, |
| 10953 | UChar* guest_code_IN, |
| 10954 | Long delta_IN, |
| 10955 | Addr64 guest_IP, |
| 10956 | VexArch guest_arch, |
| 10957 | VexArchInfo* archinfo, |
| 10958 | VexAbiInfo* abiinfo, |
sewardj | 9b76916 | 2014-07-24 12:42:03 +0000 | [diff] [blame] | 10959 | VexEndness host_endness_IN, |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10960 | Bool sigill_diag_IN ) |
| 10961 | { |
| 10962 | DisResult dres; |
| 10963 | vex_bzero(&dres, sizeof(dres)); |
| 10964 | |
| 10965 | /* Set globals (see top of this file) */ |
| 10966 | vassert(guest_arch == VexArchARM64); |
| 10967 | |
| 10968 | irsb = irsb_IN; |
sewardj | 9b76916 | 2014-07-24 12:42:03 +0000 | [diff] [blame] | 10969 | host_endness = host_endness_IN; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10970 | guest_PC_curr_instr = (Addr64)guest_IP; |
| 10971 | |
sewardj | 6590299 | 2014-05-03 21:20:56 +0000 | [diff] [blame] | 10972 | /* Sanity checks */ |
| 10973 | /* (x::UInt - 2) <= 15 === x >= 2 && x <= 17 (I hope) */ |
| 10974 | vassert((archinfo->arm64_dMinLine_lg2_szB - 2) <= 15); |
| 10975 | vassert((archinfo->arm64_iMinLine_lg2_szB - 2) <= 15); |
| 10976 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10977 | /* Try to decode */ |
| 10978 | Bool ok = disInstr_ARM64_WRK( &dres, |
| 10979 | resteerOkFn, resteerCisOk, callback_opaque, |
| 10980 | (UChar*)&guest_code_IN[delta_IN], |
| 10981 | archinfo, abiinfo ); |
| 10982 | if (ok) { |
| 10983 | /* All decode successes end up here. */ |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame] | 10984 | vassert(dres.len == 4 || dres.len == 20); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 10985 | switch (dres.whatNext) { |
| 10986 | case Dis_Continue: |
| 10987 | putPC( mkU64(dres.len + guest_PC_curr_instr) ); |
| 10988 | break; |
| 10989 | case Dis_ResteerU: |
| 10990 | case Dis_ResteerC: |
| 10991 | putPC(mkU64(dres.continueAt)); |
| 10992 | break; |
| 10993 | case Dis_StopHere: |
| 10994 | break; |
| 10995 | default: |
| 10996 | vassert(0); |
| 10997 | } |
| 10998 | DIP("\n"); |
| 10999 | } else { |
| 11000 | /* All decode failures end up here. */ |
| 11001 | if (sigill_diag_IN) { |
| 11002 | Int i, j; |
| 11003 | UChar buf[64]; |
| 11004 | UInt insn |
| 11005 | = getUIntLittleEndianly( (UChar*)&guest_code_IN[delta_IN] ); |
| 11006 | vex_bzero(buf, sizeof(buf)); |
| 11007 | for (i = j = 0; i < 32; i++) { |
| 11008 | if (i > 0) { |
| 11009 | if ((i & 7) == 0) buf[j++] = ' '; |
| 11010 | else if ((i & 3) == 0) buf[j++] = '\''; |
| 11011 | } |
| 11012 | buf[j++] = (insn & (1<<(31-i))) ? '1' : '0'; |
| 11013 | } |
| 11014 | vex_printf("disInstr(arm64): unhandled instruction 0x%08x\n", insn); |
| 11015 | vex_printf("disInstr(arm64): %s\n", buf); |
| 11016 | } |
| 11017 | |
| 11018 | /* Tell the dispatcher that this insn cannot be decoded, and so |
| 11019 | has not been executed, and (is currently) the next to be |
| 11020 | executed. PC should be up-to-date since it is made so at the |
| 11021 | start of each insn, but nevertheless be paranoid and update |
| 11022 | it again right now. */ |
| 11023 | putPC( mkU64(guest_PC_curr_instr) ); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 11024 | dres.len = 0; |
philippe | 2faf591 | 2014-08-11 22:45:47 +0000 | [diff] [blame] | 11025 | dres.whatNext = Dis_StopHere; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 11026 | dres.jk_StopHere = Ijk_NoDecode; |
philippe | 2faf591 | 2014-08-11 22:45:47 +0000 | [diff] [blame] | 11027 | dres.continueAt = 0; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 11028 | } |
| 11029 | return dres; |
| 11030 | } |
| 11031 | |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 11032 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 11033 | /*--------------------------------------------------------------------*/ |
| 11034 | /*--- end guest_arm64_toIR.c ---*/ |
| 11035 | /*--------------------------------------------------------------------*/ |