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 | |
| 122 | /* CONST: is the host bigendian? We need to know this in order to do |
| 123 | sub-register accesses to the SIMD/FP registers correctly. */ |
| 124 | static Bool host_is_bigendian; |
| 125 | |
| 126 | /* CONST: The guest address for the instruction currently being |
| 127 | translated. */ |
| 128 | static Addr64 guest_PC_curr_instr; |
| 129 | |
| 130 | /* MOD: The IRSB* into which we're generating code. */ |
| 131 | static IRSB* irsb; |
| 132 | |
| 133 | |
| 134 | /*------------------------------------------------------------*/ |
| 135 | /*--- Debugging output ---*/ |
| 136 | /*------------------------------------------------------------*/ |
| 137 | |
| 138 | #define DIP(format, args...) \ |
| 139 | if (vex_traceflags & VEX_TRACE_FE) \ |
| 140 | vex_printf(format, ## args) |
| 141 | |
| 142 | #define DIS(buf, format, args...) \ |
| 143 | if (vex_traceflags & VEX_TRACE_FE) \ |
| 144 | vex_sprintf(buf, format, ## args) |
| 145 | |
| 146 | |
| 147 | /*------------------------------------------------------------*/ |
| 148 | /*--- Helper bits and pieces for deconstructing the ---*/ |
| 149 | /*--- arm insn stream. ---*/ |
| 150 | /*------------------------------------------------------------*/ |
| 151 | |
| 152 | /* Do a little-endian load of a 32-bit word, regardless of the |
| 153 | endianness of the underlying host. */ |
| 154 | static inline UInt getUIntLittleEndianly ( UChar* p ) |
| 155 | { |
| 156 | UInt w = 0; |
| 157 | w = (w << 8) | p[3]; |
| 158 | w = (w << 8) | p[2]; |
| 159 | w = (w << 8) | p[1]; |
| 160 | w = (w << 8) | p[0]; |
| 161 | return w; |
| 162 | } |
| 163 | |
| 164 | /* Sign extend a N-bit value up to 64 bits, by copying |
| 165 | bit N-1 into all higher positions. */ |
| 166 | static ULong sx_to_64 ( ULong x, UInt n ) |
| 167 | { |
| 168 | vassert(n > 1 && n < 64); |
| 169 | Long r = (Long)x; |
| 170 | r = (r << (64-n)) >> (64-n); |
| 171 | return (ULong)r; |
| 172 | } |
| 173 | |
| 174 | //ZZ /* Do a little-endian load of a 16-bit word, regardless of the |
| 175 | //ZZ endianness of the underlying host. */ |
| 176 | //ZZ static inline UShort getUShortLittleEndianly ( UChar* p ) |
| 177 | //ZZ { |
| 178 | //ZZ UShort w = 0; |
| 179 | //ZZ w = (w << 8) | p[1]; |
| 180 | //ZZ w = (w << 8) | p[0]; |
| 181 | //ZZ return w; |
| 182 | //ZZ } |
| 183 | //ZZ |
| 184 | //ZZ static UInt ROR32 ( UInt x, UInt sh ) { |
| 185 | //ZZ vassert(sh >= 0 && sh < 32); |
| 186 | //ZZ if (sh == 0) |
| 187 | //ZZ return x; |
| 188 | //ZZ else |
| 189 | //ZZ return (x << (32-sh)) | (x >> sh); |
| 190 | //ZZ } |
| 191 | //ZZ |
| 192 | //ZZ static Int popcount32 ( UInt x ) |
| 193 | //ZZ { |
| 194 | //ZZ Int res = 0, i; |
| 195 | //ZZ for (i = 0; i < 32; i++) { |
| 196 | //ZZ res += (x & 1); |
| 197 | //ZZ x >>= 1; |
| 198 | //ZZ } |
| 199 | //ZZ return res; |
| 200 | //ZZ } |
| 201 | //ZZ |
| 202 | //ZZ static UInt setbit32 ( UInt x, Int ix, UInt b ) |
| 203 | //ZZ { |
| 204 | //ZZ UInt mask = 1 << ix; |
| 205 | //ZZ x &= ~mask; |
| 206 | //ZZ x |= ((b << ix) & mask); |
| 207 | //ZZ return x; |
| 208 | //ZZ } |
| 209 | |
| 210 | #define BITS2(_b1,_b0) \ |
| 211 | (((_b1) << 1) | (_b0)) |
| 212 | |
| 213 | #define BITS3(_b2,_b1,_b0) \ |
| 214 | (((_b2) << 2) | ((_b1) << 1) | (_b0)) |
| 215 | |
| 216 | #define BITS4(_b3,_b2,_b1,_b0) \ |
| 217 | (((_b3) << 3) | ((_b2) << 2) | ((_b1) << 1) | (_b0)) |
| 218 | |
| 219 | #define BITS8(_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 220 | ((BITS4((_b7),(_b6),(_b5),(_b4)) << 4) \ |
| 221 | | BITS4((_b3),(_b2),(_b1),(_b0))) |
| 222 | |
| 223 | #define BITS5(_b4,_b3,_b2,_b1,_b0) \ |
| 224 | (BITS8(0,0,0,(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 225 | #define BITS6(_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 226 | (BITS8(0,0,(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 227 | #define BITS7(_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 228 | (BITS8(0,(_b6),(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 229 | |
| 230 | #define BITS9(_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 231 | (((_b8) << 8) \ |
| 232 | | BITS8((_b7),(_b6),(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 233 | |
| 234 | #define BITS10(_b9,_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 235 | (((_b9) << 9) | ((_b8) << 8) \ |
| 236 | | BITS8((_b7),(_b6),(_b5),(_b4),(_b3),(_b2),(_b1),(_b0))) |
| 237 | |
| 238 | #define BITS11(_b10,_b9,_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 239 | (((_b10) << 10) \ |
| 240 | | BITS10(_b9,_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0)) |
| 241 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 242 | #define BITS12(_b11, _b10,_b9,_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0) \ |
| 243 | (((_b11) << 11) \ |
| 244 | | BITS11(_b10,_b9,_b8,_b7,_b6,_b5,_b4,_b3,_b2,_b1,_b0)) |
| 245 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 246 | // produces _uint[_bMax:_bMin] |
| 247 | #define SLICE_UInt(_uint,_bMax,_bMin) \ |
| 248 | (( ((UInt)(_uint)) >> (_bMin)) \ |
| 249 | & (UInt)((1ULL << ((_bMax) - (_bMin) + 1)) - 1ULL)) |
| 250 | |
| 251 | |
| 252 | /*------------------------------------------------------------*/ |
| 253 | /*--- Helper bits and pieces for creating IR fragments. ---*/ |
| 254 | /*------------------------------------------------------------*/ |
| 255 | |
| 256 | static IRExpr* mkV128 ( UShort w ) |
| 257 | { |
| 258 | return IRExpr_Const(IRConst_V128(w)); |
| 259 | } |
| 260 | |
| 261 | static IRExpr* mkU64 ( ULong i ) |
| 262 | { |
| 263 | return IRExpr_Const(IRConst_U64(i)); |
| 264 | } |
| 265 | |
| 266 | static IRExpr* mkU32 ( UInt i ) |
| 267 | { |
| 268 | return IRExpr_Const(IRConst_U32(i)); |
| 269 | } |
| 270 | |
| 271 | static IRExpr* mkU8 ( UInt i ) |
| 272 | { |
| 273 | vassert(i < 256); |
| 274 | return IRExpr_Const(IRConst_U8( (UChar)i )); |
| 275 | } |
| 276 | |
| 277 | static IRExpr* mkexpr ( IRTemp tmp ) |
| 278 | { |
| 279 | return IRExpr_RdTmp(tmp); |
| 280 | } |
| 281 | |
| 282 | static IRExpr* unop ( IROp op, IRExpr* a ) |
| 283 | { |
| 284 | return IRExpr_Unop(op, a); |
| 285 | } |
| 286 | |
| 287 | static IRExpr* binop ( IROp op, IRExpr* a1, IRExpr* a2 ) |
| 288 | { |
| 289 | return IRExpr_Binop(op, a1, a2); |
| 290 | } |
| 291 | |
| 292 | static IRExpr* triop ( IROp op, IRExpr* a1, IRExpr* a2, IRExpr* a3 ) |
| 293 | { |
| 294 | return IRExpr_Triop(op, a1, a2, a3); |
| 295 | } |
| 296 | |
| 297 | static IRExpr* loadLE ( IRType ty, IRExpr* addr ) |
| 298 | { |
| 299 | return IRExpr_Load(Iend_LE, ty, addr); |
| 300 | } |
| 301 | |
| 302 | /* Add a statement to the list held by "irbb". */ |
| 303 | static void stmt ( IRStmt* st ) |
| 304 | { |
| 305 | addStmtToIRSB( irsb, st ); |
| 306 | } |
| 307 | |
| 308 | static void assign ( IRTemp dst, IRExpr* e ) |
| 309 | { |
| 310 | stmt( IRStmt_WrTmp(dst, e) ); |
| 311 | } |
| 312 | |
| 313 | static void storeLE ( IRExpr* addr, IRExpr* data ) |
| 314 | { |
| 315 | stmt( IRStmt_Store(Iend_LE, addr, data) ); |
| 316 | } |
| 317 | |
| 318 | //ZZ static void storeGuardedLE ( IRExpr* addr, IRExpr* data, IRTemp guardT ) |
| 319 | //ZZ { |
| 320 | //ZZ if (guardT == IRTemp_INVALID) { |
| 321 | //ZZ /* unconditional */ |
| 322 | //ZZ storeLE(addr, data); |
| 323 | //ZZ } else { |
| 324 | //ZZ stmt( IRStmt_StoreG(Iend_LE, addr, data, |
| 325 | //ZZ binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0))) ); |
| 326 | //ZZ } |
| 327 | //ZZ } |
| 328 | //ZZ |
| 329 | //ZZ static void loadGuardedLE ( IRTemp dst, IRLoadGOp cvt, |
| 330 | //ZZ IRExpr* addr, IRExpr* alt, |
| 331 | //ZZ IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
| 332 | //ZZ { |
| 333 | //ZZ if (guardT == IRTemp_INVALID) { |
| 334 | //ZZ /* unconditional */ |
| 335 | //ZZ IRExpr* loaded = NULL; |
| 336 | //ZZ switch (cvt) { |
| 337 | //ZZ case ILGop_Ident32: |
| 338 | //ZZ loaded = loadLE(Ity_I32, addr); break; |
| 339 | //ZZ case ILGop_8Uto32: |
| 340 | //ZZ loaded = unop(Iop_8Uto32, loadLE(Ity_I8, addr)); break; |
| 341 | //ZZ case ILGop_8Sto32: |
| 342 | //ZZ loaded = unop(Iop_8Sto32, loadLE(Ity_I8, addr)); break; |
| 343 | //ZZ case ILGop_16Uto32: |
| 344 | //ZZ loaded = unop(Iop_16Uto32, loadLE(Ity_I16, addr)); break; |
| 345 | //ZZ case ILGop_16Sto32: |
| 346 | //ZZ loaded = unop(Iop_16Sto32, loadLE(Ity_I16, addr)); break; |
| 347 | //ZZ default: |
| 348 | //ZZ vassert(0); |
| 349 | //ZZ } |
| 350 | //ZZ vassert(loaded != NULL); |
| 351 | //ZZ assign(dst, loaded); |
| 352 | //ZZ } else { |
| 353 | //ZZ /* Generate a guarded load into 'dst', but apply 'cvt' to the |
| 354 | //ZZ loaded data before putting the data in 'dst'. If the load |
| 355 | //ZZ does not take place, 'alt' is placed directly in 'dst'. */ |
| 356 | //ZZ stmt( IRStmt_LoadG(Iend_LE, cvt, dst, addr, alt, |
| 357 | //ZZ binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0))) ); |
| 358 | //ZZ } |
| 359 | //ZZ } |
| 360 | |
| 361 | /* Generate a new temporary of the given type. */ |
| 362 | static IRTemp newTemp ( IRType ty ) |
| 363 | { |
| 364 | vassert(isPlausibleIRType(ty)); |
| 365 | return newIRTemp( irsb->tyenv, ty ); |
| 366 | } |
| 367 | |
| 368 | //ZZ /* Produces a value in 0 .. 3, which is encoded as per the type |
| 369 | //ZZ IRRoundingMode. */ |
| 370 | //ZZ static IRExpr* /* :: Ity_I32 */ get_FAKE_roundingmode ( void ) |
| 371 | //ZZ { |
| 372 | //ZZ return mkU32(Irrm_NEAREST); |
| 373 | //ZZ } |
| 374 | //ZZ |
| 375 | //ZZ /* Generate an expression for SRC rotated right by ROT. */ |
| 376 | //ZZ static IRExpr* genROR32( IRTemp src, Int rot ) |
| 377 | //ZZ { |
| 378 | //ZZ vassert(rot >= 0 && rot < 32); |
| 379 | //ZZ if (rot == 0) |
| 380 | //ZZ return mkexpr(src); |
| 381 | //ZZ return |
| 382 | //ZZ binop(Iop_Or32, |
| 383 | //ZZ binop(Iop_Shl32, mkexpr(src), mkU8(32 - rot)), |
| 384 | //ZZ binop(Iop_Shr32, mkexpr(src), mkU8(rot))); |
| 385 | //ZZ } |
| 386 | //ZZ |
| 387 | //ZZ static IRExpr* mkU128 ( ULong i ) |
| 388 | //ZZ { |
| 389 | //ZZ return binop(Iop_64HLtoV128, mkU64(i), mkU64(i)); |
| 390 | //ZZ } |
| 391 | //ZZ |
| 392 | //ZZ /* Generate a 4-aligned version of the given expression if |
| 393 | //ZZ the given condition is true. Else return it unchanged. */ |
| 394 | //ZZ static IRExpr* align4if ( IRExpr* e, Bool b ) |
| 395 | //ZZ { |
| 396 | //ZZ if (b) |
| 397 | //ZZ return binop(Iop_And32, e, mkU32(~3)); |
| 398 | //ZZ else |
| 399 | //ZZ return e; |
| 400 | //ZZ } |
| 401 | |
| 402 | /* Other IR construction helpers. */ |
| 403 | static IROp mkAND ( IRType ty ) { |
| 404 | switch (ty) { |
| 405 | case Ity_I32: return Iop_And32; |
| 406 | case Ity_I64: return Iop_And64; |
| 407 | default: vpanic("mkAND"); |
| 408 | } |
| 409 | } |
| 410 | |
| 411 | static IROp mkOR ( IRType ty ) { |
| 412 | switch (ty) { |
| 413 | case Ity_I32: return Iop_Or32; |
| 414 | case Ity_I64: return Iop_Or64; |
| 415 | default: vpanic("mkOR"); |
| 416 | } |
| 417 | } |
| 418 | |
| 419 | static IROp mkXOR ( IRType ty ) { |
| 420 | switch (ty) { |
| 421 | case Ity_I32: return Iop_Xor32; |
| 422 | case Ity_I64: return Iop_Xor64; |
| 423 | default: vpanic("mkXOR"); |
| 424 | } |
| 425 | } |
| 426 | |
| 427 | static IROp mkSHL ( IRType ty ) { |
| 428 | switch (ty) { |
| 429 | case Ity_I32: return Iop_Shl32; |
| 430 | case Ity_I64: return Iop_Shl64; |
| 431 | default: vpanic("mkSHL"); |
| 432 | } |
| 433 | } |
| 434 | |
| 435 | static IROp mkSHR ( IRType ty ) { |
| 436 | switch (ty) { |
| 437 | case Ity_I32: return Iop_Shr32; |
| 438 | case Ity_I64: return Iop_Shr64; |
| 439 | default: vpanic("mkSHR"); |
| 440 | } |
| 441 | } |
| 442 | |
| 443 | static IROp mkSAR ( IRType ty ) { |
| 444 | switch (ty) { |
| 445 | case Ity_I32: return Iop_Sar32; |
| 446 | case Ity_I64: return Iop_Sar64; |
| 447 | default: vpanic("mkSAR"); |
| 448 | } |
| 449 | } |
| 450 | |
| 451 | static IROp mkNOT ( IRType ty ) { |
| 452 | switch (ty) { |
| 453 | case Ity_I32: return Iop_Not32; |
| 454 | case Ity_I64: return Iop_Not64; |
| 455 | default: vpanic("mkNOT"); |
| 456 | } |
| 457 | } |
| 458 | |
| 459 | static IROp mkADD ( IRType ty ) { |
| 460 | switch (ty) { |
| 461 | case Ity_I32: return Iop_Add32; |
| 462 | case Ity_I64: return Iop_Add64; |
| 463 | default: vpanic("mkADD"); |
| 464 | } |
| 465 | } |
| 466 | |
| 467 | static IROp mkSUB ( IRType ty ) { |
| 468 | switch (ty) { |
| 469 | case Ity_I32: return Iop_Sub32; |
| 470 | case Ity_I64: return Iop_Sub64; |
| 471 | default: vpanic("mkSUB"); |
| 472 | } |
| 473 | } |
| 474 | |
| 475 | static IROp mkADDF ( IRType ty ) { |
| 476 | switch (ty) { |
| 477 | case Ity_F32: return Iop_AddF32; |
| 478 | case Ity_F64: return Iop_AddF64; |
| 479 | default: vpanic("mkADDF"); |
| 480 | } |
| 481 | } |
| 482 | |
| 483 | static IROp mkSUBF ( IRType ty ) { |
| 484 | switch (ty) { |
| 485 | case Ity_F32: return Iop_SubF32; |
| 486 | case Ity_F64: return Iop_SubF64; |
| 487 | default: vpanic("mkSUBF"); |
| 488 | } |
| 489 | } |
| 490 | |
| 491 | static IROp mkMULF ( IRType ty ) { |
| 492 | switch (ty) { |
| 493 | case Ity_F32: return Iop_MulF32; |
| 494 | case Ity_F64: return Iop_MulF64; |
| 495 | default: vpanic("mkMULF"); |
| 496 | } |
| 497 | } |
| 498 | |
| 499 | static IROp mkDIVF ( IRType ty ) { |
| 500 | switch (ty) { |
| 501 | case Ity_F32: return Iop_DivF32; |
| 502 | case Ity_F64: return Iop_DivF64; |
| 503 | default: vpanic("mkMULF"); |
| 504 | } |
| 505 | } |
| 506 | |
| 507 | static IROp mkNEGF ( IRType ty ) { |
| 508 | switch (ty) { |
| 509 | case Ity_F32: return Iop_NegF32; |
| 510 | case Ity_F64: return Iop_NegF64; |
| 511 | default: vpanic("mkNEGF"); |
| 512 | } |
| 513 | } |
| 514 | |
| 515 | static IROp mkABSF ( IRType ty ) { |
| 516 | switch (ty) { |
| 517 | case Ity_F32: return Iop_AbsF32; |
| 518 | case Ity_F64: return Iop_AbsF64; |
| 519 | default: vpanic("mkNEGF"); |
| 520 | } |
| 521 | } |
| 522 | |
| 523 | static IROp mkSQRTF ( IRType ty ) { |
| 524 | switch (ty) { |
| 525 | case Ity_F32: return Iop_SqrtF32; |
| 526 | case Ity_F64: return Iop_SqrtF64; |
| 527 | default: vpanic("mkNEGF"); |
| 528 | } |
| 529 | } |
| 530 | |
| 531 | static IRExpr* mkU ( IRType ty, ULong imm ) { |
| 532 | switch (ty) { |
| 533 | case Ity_I32: return mkU32((UInt)(imm & 0xFFFFFFFFULL)); |
| 534 | case Ity_I64: return mkU64(imm); |
| 535 | default: vpanic("mkU"); |
| 536 | } |
| 537 | } |
| 538 | |
| 539 | /* Generate IR to create 'arg rotated right by imm', for sane values |
| 540 | of 'ty' and 'imm'. */ |
| 541 | static IRTemp mathROR ( IRType ty, IRTemp arg, UInt imm ) |
| 542 | { |
| 543 | UInt w = 0; |
| 544 | if (ty == Ity_I64) { |
| 545 | w = 64; |
| 546 | } else { |
| 547 | vassert(ty == Ity_I32); |
| 548 | w = 32; |
| 549 | } |
| 550 | vassert(w != 0); |
| 551 | vassert(imm < w); |
| 552 | if (imm == 0) { |
| 553 | return arg; |
| 554 | } |
| 555 | IRTemp res = newTemp(ty); |
| 556 | assign(res, binop(mkOR(ty), |
| 557 | binop(mkSHL(ty), mkexpr(arg), mkU8(w - imm)), |
| 558 | binop(mkSHR(ty), mkexpr(arg), mkU8(imm)) )); |
| 559 | return res; |
| 560 | } |
| 561 | |
| 562 | /* Generate IR to set the returned temp to either all-zeroes or |
| 563 | all ones, as a copy of arg<imm>. */ |
| 564 | static IRTemp mathREPLICATE ( IRType ty, IRTemp arg, UInt imm ) |
| 565 | { |
| 566 | UInt w = 0; |
| 567 | if (ty == Ity_I64) { |
| 568 | w = 64; |
| 569 | } else { |
| 570 | vassert(ty == Ity_I32); |
| 571 | w = 32; |
| 572 | } |
| 573 | vassert(w != 0); |
| 574 | vassert(imm < w); |
| 575 | IRTemp res = newTemp(ty); |
| 576 | assign(res, binop(mkSAR(ty), |
| 577 | binop(mkSHL(ty), mkexpr(arg), mkU8(w - 1 - imm)), |
| 578 | mkU8(w - 1))); |
| 579 | return res; |
| 580 | } |
| 581 | |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 582 | /* U-widen 8/16/32/64 bit int expr to 64. */ |
| 583 | static IRExpr* widenUto64 ( IRType srcTy, IRExpr* e ) |
| 584 | { |
| 585 | switch (srcTy) { |
| 586 | case Ity_I64: return e; |
| 587 | case Ity_I32: return unop(Iop_32Uto64, e); |
| 588 | case Ity_I16: return unop(Iop_16Uto64, e); |
| 589 | case Ity_I8: return unop(Iop_8Uto64, e); |
| 590 | default: vpanic("widenUto64(arm64)"); |
| 591 | } |
| 592 | } |
| 593 | |
| 594 | /* Narrow 64 bit int expr to 8/16/32/64. Clearly only some |
| 595 | of these combinations make sense. */ |
| 596 | static IRExpr* narrowFrom64 ( IRType dstTy, IRExpr* e ) |
| 597 | { |
| 598 | switch (dstTy) { |
| 599 | case Ity_I64: return e; |
| 600 | case Ity_I32: return unop(Iop_64to32, e); |
| 601 | case Ity_I16: return unop(Iop_64to16, e); |
| 602 | case Ity_I8: return unop(Iop_64to8, e); |
| 603 | default: vpanic("narrowFrom64(arm64)"); |
| 604 | } |
| 605 | } |
| 606 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 607 | |
| 608 | /*------------------------------------------------------------*/ |
| 609 | /*--- Helpers for accessing guest registers. ---*/ |
| 610 | /*------------------------------------------------------------*/ |
| 611 | |
| 612 | #define OFFB_X0 offsetof(VexGuestARM64State,guest_X0) |
| 613 | #define OFFB_X1 offsetof(VexGuestARM64State,guest_X1) |
| 614 | #define OFFB_X2 offsetof(VexGuestARM64State,guest_X2) |
| 615 | #define OFFB_X3 offsetof(VexGuestARM64State,guest_X3) |
| 616 | #define OFFB_X4 offsetof(VexGuestARM64State,guest_X4) |
| 617 | #define OFFB_X5 offsetof(VexGuestARM64State,guest_X5) |
| 618 | #define OFFB_X6 offsetof(VexGuestARM64State,guest_X6) |
| 619 | #define OFFB_X7 offsetof(VexGuestARM64State,guest_X7) |
| 620 | #define OFFB_X8 offsetof(VexGuestARM64State,guest_X8) |
| 621 | #define OFFB_X9 offsetof(VexGuestARM64State,guest_X9) |
| 622 | #define OFFB_X10 offsetof(VexGuestARM64State,guest_X10) |
| 623 | #define OFFB_X11 offsetof(VexGuestARM64State,guest_X11) |
| 624 | #define OFFB_X12 offsetof(VexGuestARM64State,guest_X12) |
| 625 | #define OFFB_X13 offsetof(VexGuestARM64State,guest_X13) |
| 626 | #define OFFB_X14 offsetof(VexGuestARM64State,guest_X14) |
| 627 | #define OFFB_X15 offsetof(VexGuestARM64State,guest_X15) |
| 628 | #define OFFB_X16 offsetof(VexGuestARM64State,guest_X16) |
| 629 | #define OFFB_X17 offsetof(VexGuestARM64State,guest_X17) |
| 630 | #define OFFB_X18 offsetof(VexGuestARM64State,guest_X18) |
| 631 | #define OFFB_X19 offsetof(VexGuestARM64State,guest_X19) |
| 632 | #define OFFB_X20 offsetof(VexGuestARM64State,guest_X20) |
| 633 | #define OFFB_X21 offsetof(VexGuestARM64State,guest_X21) |
| 634 | #define OFFB_X22 offsetof(VexGuestARM64State,guest_X22) |
| 635 | #define OFFB_X23 offsetof(VexGuestARM64State,guest_X23) |
| 636 | #define OFFB_X24 offsetof(VexGuestARM64State,guest_X24) |
| 637 | #define OFFB_X25 offsetof(VexGuestARM64State,guest_X25) |
| 638 | #define OFFB_X26 offsetof(VexGuestARM64State,guest_X26) |
| 639 | #define OFFB_X27 offsetof(VexGuestARM64State,guest_X27) |
| 640 | #define OFFB_X28 offsetof(VexGuestARM64State,guest_X28) |
| 641 | #define OFFB_X29 offsetof(VexGuestARM64State,guest_X29) |
| 642 | #define OFFB_X30 offsetof(VexGuestARM64State,guest_X30) |
| 643 | |
sewardj | 6068788 | 2014-01-15 10:25:21 +0000 | [diff] [blame] | 644 | #define OFFB_XSP offsetof(VexGuestARM64State,guest_XSP) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 645 | #define OFFB_PC offsetof(VexGuestARM64State,guest_PC) |
| 646 | |
| 647 | #define OFFB_CC_OP offsetof(VexGuestARM64State,guest_CC_OP) |
| 648 | #define OFFB_CC_DEP1 offsetof(VexGuestARM64State,guest_CC_DEP1) |
| 649 | #define OFFB_CC_DEP2 offsetof(VexGuestARM64State,guest_CC_DEP2) |
| 650 | #define OFFB_CC_NDEP offsetof(VexGuestARM64State,guest_CC_NDEP) |
| 651 | |
| 652 | #define OFFB_TPIDR_EL0 offsetof(VexGuestARM64State,guest_TPIDR_EL0) |
| 653 | #define OFFB_NRADDR offsetof(VexGuestARM64State,guest_NRADDR) |
| 654 | |
| 655 | #define OFFB_Q0 offsetof(VexGuestARM64State,guest_Q0) |
| 656 | #define OFFB_Q1 offsetof(VexGuestARM64State,guest_Q1) |
| 657 | #define OFFB_Q2 offsetof(VexGuestARM64State,guest_Q2) |
| 658 | #define OFFB_Q3 offsetof(VexGuestARM64State,guest_Q3) |
| 659 | #define OFFB_Q4 offsetof(VexGuestARM64State,guest_Q4) |
| 660 | #define OFFB_Q5 offsetof(VexGuestARM64State,guest_Q5) |
| 661 | #define OFFB_Q6 offsetof(VexGuestARM64State,guest_Q6) |
| 662 | #define OFFB_Q7 offsetof(VexGuestARM64State,guest_Q7) |
| 663 | #define OFFB_Q8 offsetof(VexGuestARM64State,guest_Q8) |
| 664 | #define OFFB_Q9 offsetof(VexGuestARM64State,guest_Q9) |
| 665 | #define OFFB_Q10 offsetof(VexGuestARM64State,guest_Q10) |
| 666 | #define OFFB_Q11 offsetof(VexGuestARM64State,guest_Q11) |
| 667 | #define OFFB_Q12 offsetof(VexGuestARM64State,guest_Q12) |
| 668 | #define OFFB_Q13 offsetof(VexGuestARM64State,guest_Q13) |
| 669 | #define OFFB_Q14 offsetof(VexGuestARM64State,guest_Q14) |
| 670 | #define OFFB_Q15 offsetof(VexGuestARM64State,guest_Q15) |
| 671 | #define OFFB_Q16 offsetof(VexGuestARM64State,guest_Q16) |
| 672 | #define OFFB_Q17 offsetof(VexGuestARM64State,guest_Q17) |
| 673 | #define OFFB_Q18 offsetof(VexGuestARM64State,guest_Q18) |
| 674 | #define OFFB_Q19 offsetof(VexGuestARM64State,guest_Q19) |
| 675 | #define OFFB_Q20 offsetof(VexGuestARM64State,guest_Q20) |
| 676 | #define OFFB_Q21 offsetof(VexGuestARM64State,guest_Q21) |
| 677 | #define OFFB_Q22 offsetof(VexGuestARM64State,guest_Q22) |
| 678 | #define OFFB_Q23 offsetof(VexGuestARM64State,guest_Q23) |
| 679 | #define OFFB_Q24 offsetof(VexGuestARM64State,guest_Q24) |
| 680 | #define OFFB_Q25 offsetof(VexGuestARM64State,guest_Q25) |
| 681 | #define OFFB_Q26 offsetof(VexGuestARM64State,guest_Q26) |
| 682 | #define OFFB_Q27 offsetof(VexGuestARM64State,guest_Q27) |
| 683 | #define OFFB_Q28 offsetof(VexGuestARM64State,guest_Q28) |
| 684 | #define OFFB_Q29 offsetof(VexGuestARM64State,guest_Q29) |
| 685 | #define OFFB_Q30 offsetof(VexGuestARM64State,guest_Q30) |
| 686 | #define OFFB_Q31 offsetof(VexGuestARM64State,guest_Q31) |
| 687 | |
| 688 | #define OFFB_FPCR offsetof(VexGuestARM64State,guest_FPCR) |
| 689 | #define OFFB_FPSR offsetof(VexGuestARM64State,guest_FPSR) |
| 690 | //ZZ #define OFFB_TPIDRURO offsetof(VexGuestARMState,guest_TPIDRURO) |
| 691 | //ZZ #define OFFB_ITSTATE offsetof(VexGuestARMState,guest_ITSTATE) |
| 692 | //ZZ #define OFFB_QFLAG32 offsetof(VexGuestARMState,guest_QFLAG32) |
| 693 | //ZZ #define OFFB_GEFLAG0 offsetof(VexGuestARMState,guest_GEFLAG0) |
| 694 | //ZZ #define OFFB_GEFLAG1 offsetof(VexGuestARMState,guest_GEFLAG1) |
| 695 | //ZZ #define OFFB_GEFLAG2 offsetof(VexGuestARMState,guest_GEFLAG2) |
| 696 | //ZZ #define OFFB_GEFLAG3 offsetof(VexGuestARMState,guest_GEFLAG3) |
| 697 | |
| 698 | #define OFFB_TISTART offsetof(VexGuestARM64State,guest_TISTART) |
| 699 | #define OFFB_TILEN offsetof(VexGuestARM64State,guest_TILEN) |
| 700 | |
| 701 | |
| 702 | /* ---------------- Integer registers ---------------- */ |
| 703 | |
| 704 | static Int offsetIReg64 ( UInt iregNo ) |
| 705 | { |
| 706 | /* Do we care about endianness here? We do if sub-parts of integer |
| 707 | registers are accessed. */ |
| 708 | switch (iregNo) { |
| 709 | case 0: return OFFB_X0; |
| 710 | case 1: return OFFB_X1; |
| 711 | case 2: return OFFB_X2; |
| 712 | case 3: return OFFB_X3; |
| 713 | case 4: return OFFB_X4; |
| 714 | case 5: return OFFB_X5; |
| 715 | case 6: return OFFB_X6; |
| 716 | case 7: return OFFB_X7; |
| 717 | case 8: return OFFB_X8; |
| 718 | case 9: return OFFB_X9; |
| 719 | case 10: return OFFB_X10; |
| 720 | case 11: return OFFB_X11; |
| 721 | case 12: return OFFB_X12; |
| 722 | case 13: return OFFB_X13; |
| 723 | case 14: return OFFB_X14; |
| 724 | case 15: return OFFB_X15; |
| 725 | case 16: return OFFB_X16; |
| 726 | case 17: return OFFB_X17; |
| 727 | case 18: return OFFB_X18; |
| 728 | case 19: return OFFB_X19; |
| 729 | case 20: return OFFB_X20; |
| 730 | case 21: return OFFB_X21; |
| 731 | case 22: return OFFB_X22; |
| 732 | case 23: return OFFB_X23; |
| 733 | case 24: return OFFB_X24; |
| 734 | case 25: return OFFB_X25; |
| 735 | case 26: return OFFB_X26; |
| 736 | case 27: return OFFB_X27; |
| 737 | case 28: return OFFB_X28; |
| 738 | case 29: return OFFB_X29; |
| 739 | case 30: return OFFB_X30; |
| 740 | /* but not 31 */ |
| 741 | default: vassert(0); |
| 742 | } |
| 743 | } |
| 744 | |
| 745 | static Int offsetIReg64orSP ( UInt iregNo ) |
| 746 | { |
sewardj | 6068788 | 2014-01-15 10:25:21 +0000 | [diff] [blame] | 747 | return iregNo == 31 ? OFFB_XSP : offsetIReg64(iregNo); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 748 | } |
| 749 | |
| 750 | static const HChar* nameIReg64orZR ( UInt iregNo ) |
| 751 | { |
| 752 | vassert(iregNo < 32); |
| 753 | static const HChar* names[32] |
| 754 | = { "x0", "x1", "x2", "x3", "x4", "x5", "x6", "x7", |
| 755 | "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", |
| 756 | "x16", "x17", "x18", "x19", "x20", "x21", "x22", "x23", |
| 757 | "x24", "x25", "x26", "x27", "x28", "x29", "x30", "xzr" }; |
| 758 | return names[iregNo]; |
| 759 | } |
| 760 | |
| 761 | static const HChar* nameIReg64orSP ( UInt iregNo ) |
| 762 | { |
| 763 | if (iregNo == 31) { |
| 764 | return "sp"; |
| 765 | } |
| 766 | vassert(iregNo < 31); |
| 767 | return nameIReg64orZR(iregNo); |
| 768 | } |
| 769 | |
| 770 | static IRExpr* getIReg64orSP ( UInt iregNo ) |
| 771 | { |
| 772 | vassert(iregNo < 32); |
| 773 | return IRExpr_Get( offsetIReg64orSP(iregNo), Ity_I64 ); |
| 774 | } |
| 775 | |
| 776 | static IRExpr* getIReg64orZR ( UInt iregNo ) |
| 777 | { |
| 778 | if (iregNo == 31) { |
| 779 | return mkU64(0); |
| 780 | } |
| 781 | vassert(iregNo < 31); |
| 782 | return IRExpr_Get( offsetIReg64orSP(iregNo), Ity_I64 ); |
| 783 | } |
| 784 | |
| 785 | static void putIReg64orSP ( UInt iregNo, IRExpr* e ) |
| 786 | { |
| 787 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I64); |
| 788 | stmt( IRStmt_Put(offsetIReg64orSP(iregNo), e) ); |
| 789 | } |
| 790 | |
| 791 | static void putIReg64orZR ( UInt iregNo, IRExpr* e ) |
| 792 | { |
| 793 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I64); |
| 794 | if (iregNo == 31) { |
| 795 | return; |
| 796 | } |
| 797 | vassert(iregNo < 31); |
| 798 | stmt( IRStmt_Put(offsetIReg64orSP(iregNo), e) ); |
| 799 | } |
| 800 | |
| 801 | static const HChar* nameIReg32orZR ( UInt iregNo ) |
| 802 | { |
| 803 | vassert(iregNo < 32); |
| 804 | static const HChar* names[32] |
| 805 | = { "w0", "w1", "w2", "w3", "w4", "w5", "w6", "w7", |
| 806 | "w8", "w9", "w10", "w11", "w12", "w13", "w14", "w15", |
| 807 | "w16", "w17", "w18", "w19", "w20", "w21", "w22", "w23", |
| 808 | "w24", "w25", "w26", "w27", "w28", "w29", "w30", "wzr" }; |
| 809 | return names[iregNo]; |
| 810 | } |
| 811 | |
| 812 | static const HChar* nameIReg32orSP ( UInt iregNo ) |
| 813 | { |
| 814 | if (iregNo == 31) { |
| 815 | return "wsp"; |
| 816 | } |
| 817 | vassert(iregNo < 31); |
| 818 | return nameIReg32orZR(iregNo); |
| 819 | } |
| 820 | |
| 821 | static IRExpr* getIReg32orSP ( UInt iregNo ) |
| 822 | { |
| 823 | vassert(iregNo < 32); |
| 824 | return unop(Iop_64to32, |
| 825 | IRExpr_Get( offsetIReg64orSP(iregNo), Ity_I64 )); |
| 826 | } |
| 827 | |
| 828 | static IRExpr* getIReg32orZR ( UInt iregNo ) |
| 829 | { |
| 830 | if (iregNo == 31) { |
| 831 | return mkU32(0); |
| 832 | } |
| 833 | vassert(iregNo < 31); |
| 834 | return unop(Iop_64to32, |
| 835 | IRExpr_Get( offsetIReg64orSP(iregNo), Ity_I64 )); |
| 836 | } |
| 837 | |
| 838 | static void putIReg32orSP ( UInt iregNo, IRExpr* e ) |
| 839 | { |
| 840 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I32); |
| 841 | stmt( IRStmt_Put(offsetIReg64orSP(iregNo), unop(Iop_32Uto64, e)) ); |
| 842 | } |
| 843 | |
| 844 | static void putIReg32orZR ( UInt iregNo, IRExpr* e ) |
| 845 | { |
| 846 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I32); |
| 847 | if (iregNo == 31) { |
| 848 | return; |
| 849 | } |
| 850 | vassert(iregNo < 31); |
| 851 | stmt( IRStmt_Put(offsetIReg64orSP(iregNo), unop(Iop_32Uto64, e)) ); |
| 852 | } |
| 853 | |
| 854 | static const HChar* nameIRegOrSP ( Bool is64, UInt iregNo ) |
| 855 | { |
| 856 | vassert(is64 == True || is64 == False); |
| 857 | return is64 ? nameIReg64orSP(iregNo) : nameIReg32orSP(iregNo); |
| 858 | } |
| 859 | |
| 860 | static const HChar* nameIRegOrZR ( Bool is64, UInt iregNo ) |
| 861 | { |
| 862 | vassert(is64 == True || is64 == False); |
| 863 | return is64 ? nameIReg64orZR(iregNo) : nameIReg32orZR(iregNo); |
| 864 | } |
| 865 | |
| 866 | static IRExpr* getIRegOrZR ( Bool is64, UInt iregNo ) |
| 867 | { |
| 868 | vassert(is64 == True || is64 == False); |
| 869 | return is64 ? getIReg64orZR(iregNo) : getIReg32orZR(iregNo); |
| 870 | } |
| 871 | |
| 872 | static void putIRegOrZR ( Bool is64, UInt iregNo, IRExpr* e ) |
| 873 | { |
| 874 | vassert(is64 == True || is64 == False); |
| 875 | if (is64) putIReg64orZR(iregNo, e); else putIReg32orZR(iregNo, e); |
| 876 | } |
| 877 | |
| 878 | static void putPC ( IRExpr* e ) |
| 879 | { |
| 880 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I64); |
| 881 | stmt( IRStmt_Put(OFFB_PC, e) ); |
| 882 | } |
| 883 | |
| 884 | |
| 885 | /* ---------------- Vector (Q) registers ---------------- */ |
| 886 | |
| 887 | static Int offsetQReg128 ( UInt qregNo ) |
| 888 | { |
| 889 | /* We don't care about endianness at this point. It only becomes |
| 890 | relevant when dealing with sections of these registers.*/ |
| 891 | switch (qregNo) { |
| 892 | case 0: return OFFB_Q0; |
| 893 | case 1: return OFFB_Q1; |
| 894 | case 2: return OFFB_Q2; |
| 895 | case 3: return OFFB_Q3; |
| 896 | case 4: return OFFB_Q4; |
| 897 | case 5: return OFFB_Q5; |
| 898 | case 6: return OFFB_Q6; |
| 899 | case 7: return OFFB_Q7; |
| 900 | case 8: return OFFB_Q8; |
| 901 | case 9: return OFFB_Q9; |
| 902 | case 10: return OFFB_Q10; |
| 903 | case 11: return OFFB_Q11; |
| 904 | case 12: return OFFB_Q12; |
| 905 | case 13: return OFFB_Q13; |
| 906 | case 14: return OFFB_Q14; |
| 907 | case 15: return OFFB_Q15; |
| 908 | case 16: return OFFB_Q16; |
| 909 | case 17: return OFFB_Q17; |
| 910 | case 18: return OFFB_Q18; |
| 911 | case 19: return OFFB_Q19; |
| 912 | case 20: return OFFB_Q20; |
| 913 | case 21: return OFFB_Q21; |
| 914 | case 22: return OFFB_Q22; |
| 915 | case 23: return OFFB_Q23; |
| 916 | case 24: return OFFB_Q24; |
| 917 | case 25: return OFFB_Q25; |
| 918 | case 26: return OFFB_Q26; |
| 919 | case 27: return OFFB_Q27; |
| 920 | case 28: return OFFB_Q28; |
| 921 | case 29: return OFFB_Q29; |
| 922 | case 30: return OFFB_Q30; |
| 923 | case 31: return OFFB_Q31; |
| 924 | default: vassert(0); |
| 925 | } |
| 926 | } |
| 927 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 928 | /* Write to a complete Qreg. */ |
| 929 | static void putQReg128 ( UInt qregNo, IRExpr* e ) |
| 930 | { |
| 931 | vassert(qregNo < 32); |
| 932 | vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_V128); |
| 933 | stmt( IRStmt_Put(offsetQReg128(qregNo), e) ); |
| 934 | } |
| 935 | |
| 936 | /* Read a complete Qreg. */ |
| 937 | static IRExpr* getQReg128 ( UInt qregNo ) |
| 938 | { |
| 939 | vassert(qregNo < 32); |
| 940 | return IRExpr_Get(offsetQReg128(qregNo), Ity_V128); |
| 941 | } |
| 942 | |
| 943 | /* Produce the IR type for some sub-part of a vector. For 32- and 64- |
| 944 | bit sub-parts we can choose either integer or float types, and |
| 945 | choose float on the basis that that is the common use case and so |
| 946 | will give least interference with Put-to-Get forwarding later |
| 947 | on. */ |
| 948 | static IRType preferredVectorSubTypeFromSize ( UInt szB ) |
| 949 | { |
| 950 | switch (szB) { |
| 951 | case 1: return Ity_I8; |
| 952 | case 2: return Ity_I16; |
| 953 | case 4: return Ity_I32; //Ity_F32; |
| 954 | case 8: return Ity_F64; |
| 955 | case 16: return Ity_V128; |
| 956 | default: vassert(0); |
| 957 | } |
| 958 | } |
| 959 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 960 | /* Find the offset of the laneNo'th lane of type laneTy in the given |
| 961 | Qreg. Since the host is little-endian, the least significant lane |
| 962 | has the lowest offset. */ |
| 963 | static Int offsetQRegLane ( UInt qregNo, IRType laneTy, UInt laneNo ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 964 | { |
| 965 | vassert(!host_is_bigendian); |
| 966 | Int base = offsetQReg128(qregNo); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 967 | /* Since the host is little-endian, the least significant lane |
| 968 | will be at the lowest address. */ |
| 969 | /* Restrict this to known types, so as to avoid silently accepting |
| 970 | stupid types. */ |
| 971 | UInt laneSzB = 0; |
| 972 | switch (laneTy) { |
| 973 | case Ity_F32: case Ity_I32: laneSzB = 4; break; |
| 974 | case Ity_F64: case Ity_I64: laneSzB = 8; break; |
| 975 | case Ity_V128: laneSzB = 16; break; |
| 976 | default: break; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 977 | } |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 978 | vassert(laneSzB > 0); |
| 979 | UInt minOff = laneNo * laneSzB; |
| 980 | UInt maxOff = minOff + laneSzB - 1; |
| 981 | vassert(maxOff < 16); |
| 982 | return base + minOff; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 983 | } |
| 984 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 985 | /* Put to the least significant lane of a Qreg. */ |
| 986 | static void putQRegLO ( UInt qregNo, IRExpr* e ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 987 | { |
| 988 | IRType ty = typeOfIRExpr(irsb->tyenv, e); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 989 | Int off = offsetQRegLane(qregNo, ty, 0); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 990 | switch (ty) { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 991 | case Ity_I8: case Ity_I16: case Ity_I32: case Ity_I64: |
| 992 | case Ity_F32: case Ity_F64: case Ity_V128: |
| 993 | break; |
| 994 | default: |
| 995 | vassert(0); // Other cases are probably invalid |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 996 | } |
| 997 | stmt(IRStmt_Put(off, e)); |
| 998 | } |
| 999 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1000 | /* Get from the least significant lane of a Qreg. */ |
| 1001 | static IRExpr* getQRegLO ( UInt qregNo, IRType ty ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1002 | { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1003 | Int off = offsetQRegLane(qregNo, ty, 0); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1004 | switch (ty) { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1005 | case Ity_I32: case Ity_I64: |
| 1006 | case Ity_F32: case Ity_F64: case Ity_V128: |
| 1007 | break; |
| 1008 | default: |
| 1009 | vassert(0); // Other cases are ATC |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1010 | } |
| 1011 | return IRExpr_Get(off, ty); |
| 1012 | } |
| 1013 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1014 | static const HChar* nameQRegLO ( UInt qregNo, IRType laneTy ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1015 | { |
| 1016 | static const HChar* namesQ[32] |
| 1017 | = { "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7", |
| 1018 | "q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15", |
| 1019 | "q16", "q17", "q18", "q19", "q20", "q21", "q22", "q23", |
| 1020 | "q24", "q25", "q26", "q27", "q28", "q29", "q30", "q31" }; |
| 1021 | static const HChar* namesD[32] |
| 1022 | = { "d0", "d1", "d2", "d3", "d4", "d5", "d6", "d7", |
| 1023 | "d8", "d9", "d10", "d11", "d12", "d13", "d14", "d15", |
| 1024 | "d16", "d17", "d18", "d19", "d20", "d21", "d22", "d23", |
| 1025 | "d24", "d25", "d26", "d27", "d28", "d29", "d30", "d31" }; |
| 1026 | static const HChar* namesS[32] |
| 1027 | = { "s0", "s1", "s2", "s3", "s4", "s5", "s6", "s7", |
| 1028 | "s8", "s9", "s10", "s11", "s12", "s13", "s14", "s15", |
| 1029 | "s16", "s17", "s18", "s19", "s20", "s21", "s22", "s23", |
| 1030 | "s24", "s25", "s26", "s27", "s28", "s29", "s30", "s31" }; |
| 1031 | static const HChar* namesH[32] |
| 1032 | = { "h0", "h1", "h2", "h3", "h4", "h5", "h6", "h7", |
| 1033 | "h8", "h9", "h10", "h11", "h12", "h13", "h14", "h15", |
| 1034 | "h16", "h17", "h18", "h19", "h20", "h21", "h22", "h23", |
| 1035 | "h24", "h25", "h26", "h27", "h28", "h29", "h30", "h31" }; |
| 1036 | static const HChar* namesB[32] |
| 1037 | = { "b0", "b1", "b2", "b3", "b4", "b5", "b6", "b7", |
| 1038 | "b8", "b9", "b10", "b11", "b12", "b13", "b14", "b15", |
| 1039 | "b16", "b17", "b18", "b19", "b20", "b21", "b22", "b23", |
| 1040 | "b24", "b25", "b26", "b27", "b28", "b29", "b30", "b31" }; |
| 1041 | vassert(qregNo < 32); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1042 | switch (sizeofIRType(laneTy)) { |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1043 | case 1: return namesB[qregNo]; |
| 1044 | case 2: return namesH[qregNo]; |
| 1045 | case 4: return namesS[qregNo]; |
| 1046 | case 8: return namesD[qregNo]; |
| 1047 | case 16: return namesQ[qregNo]; |
| 1048 | default: vassert(0); |
| 1049 | } |
| 1050 | /*NOTREACHED*/ |
| 1051 | } |
| 1052 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1053 | static const HChar* nameQReg128 ( UInt qregNo ) |
| 1054 | { |
| 1055 | return nameQRegLO(qregNo, Ity_V128); |
| 1056 | } |
| 1057 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1058 | /* Find the offset of the most significant half (8 bytes) of the given |
| 1059 | Qreg. This requires knowing the endianness of the host. */ |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1060 | static Int offsetQRegHI64 ( UInt qregNo ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1061 | { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1062 | return offsetQRegLane(qregNo, Ity_I64, 1); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1063 | } |
| 1064 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1065 | static IRExpr* getQRegHI64 ( UInt qregNo ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1066 | { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1067 | return IRExpr_Get(offsetQRegHI64(qregNo), Ity_I64); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1068 | } |
| 1069 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1070 | static void putQRegHI64 ( UInt qregNo, IRExpr* e ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1071 | { |
| 1072 | IRType ty = typeOfIRExpr(irsb->tyenv, e); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1073 | Int off = offsetQRegHI64(qregNo); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1074 | switch (ty) { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1075 | case Ity_I64: case Ity_F64: |
| 1076 | break; |
| 1077 | default: |
| 1078 | vassert(0); // Other cases are plain wrong |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1079 | } |
| 1080 | stmt(IRStmt_Put(off, e)); |
| 1081 | } |
| 1082 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1083 | /* Put to a specified lane of a Qreg. */ |
| 1084 | static void putQRegLane ( UInt qregNo, UInt laneNo, IRExpr* e ) |
| 1085 | { |
| 1086 | IRType laneTy = typeOfIRExpr(irsb->tyenv, e); |
| 1087 | Int off = offsetQRegLane(qregNo, laneTy, laneNo); |
| 1088 | switch (laneTy) { |
| 1089 | case Ity_F64: case Ity_I64: |
| 1090 | break; |
| 1091 | default: |
| 1092 | vassert(0); // Other cases are ATC |
| 1093 | } |
| 1094 | stmt(IRStmt_Put(off, e)); |
| 1095 | } |
| 1096 | |
| 1097 | /* Get from the least significant lane of a Qreg. */ |
| 1098 | static IRExpr* getQRegLane ( UInt qregNo, UInt laneNo, IRType laneTy ) |
| 1099 | { |
| 1100 | Int off = offsetQRegLane(qregNo, laneTy, laneNo); |
| 1101 | switch (laneTy) { |
| 1102 | case Ity_I64: case Ity_I32: |
| 1103 | break; |
| 1104 | default: |
| 1105 | vassert(0); // Other cases are ATC |
| 1106 | } |
| 1107 | return IRExpr_Get(off, laneTy); |
| 1108 | } |
| 1109 | |
| 1110 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1111 | //ZZ /* ---------------- Misc registers ---------------- */ |
| 1112 | //ZZ |
| 1113 | //ZZ static void putMiscReg32 ( UInt gsoffset, |
| 1114 | //ZZ IRExpr* e, /* :: Ity_I32 */ |
| 1115 | //ZZ IRTemp guardT /* :: Ity_I32, 0 or 1 */) |
| 1116 | //ZZ { |
| 1117 | //ZZ switch (gsoffset) { |
| 1118 | //ZZ case OFFB_FPSCR: break; |
| 1119 | //ZZ case OFFB_QFLAG32: break; |
| 1120 | //ZZ case OFFB_GEFLAG0: break; |
| 1121 | //ZZ case OFFB_GEFLAG1: break; |
| 1122 | //ZZ case OFFB_GEFLAG2: break; |
| 1123 | //ZZ case OFFB_GEFLAG3: break; |
| 1124 | //ZZ default: vassert(0); /* awaiting more cases */ |
| 1125 | //ZZ } |
| 1126 | //ZZ vassert(typeOfIRExpr(irsb->tyenv, e) == Ity_I32); |
| 1127 | //ZZ |
| 1128 | //ZZ if (guardT == IRTemp_INVALID) { |
| 1129 | //ZZ /* unconditional write */ |
| 1130 | //ZZ stmt(IRStmt_Put(gsoffset, e)); |
| 1131 | //ZZ } else { |
| 1132 | //ZZ stmt(IRStmt_Put( |
| 1133 | //ZZ gsoffset, |
| 1134 | //ZZ IRExpr_ITE( binop(Iop_CmpNE32, mkexpr(guardT), mkU32(0)), |
| 1135 | //ZZ e, IRExpr_Get(gsoffset, Ity_I32) ) |
| 1136 | //ZZ )); |
| 1137 | //ZZ } |
| 1138 | //ZZ } |
| 1139 | //ZZ |
| 1140 | //ZZ static IRTemp get_ITSTATE ( void ) |
| 1141 | //ZZ { |
| 1142 | //ZZ ASSERT_IS_THUMB; |
| 1143 | //ZZ IRTemp t = newTemp(Ity_I32); |
| 1144 | //ZZ assign(t, IRExpr_Get( OFFB_ITSTATE, Ity_I32)); |
| 1145 | //ZZ return t; |
| 1146 | //ZZ } |
| 1147 | //ZZ |
| 1148 | //ZZ static void put_ITSTATE ( IRTemp t ) |
| 1149 | //ZZ { |
| 1150 | //ZZ ASSERT_IS_THUMB; |
| 1151 | //ZZ stmt( IRStmt_Put( OFFB_ITSTATE, mkexpr(t)) ); |
| 1152 | //ZZ } |
| 1153 | //ZZ |
| 1154 | //ZZ static IRTemp get_QFLAG32 ( void ) |
| 1155 | //ZZ { |
| 1156 | //ZZ IRTemp t = newTemp(Ity_I32); |
| 1157 | //ZZ assign(t, IRExpr_Get( OFFB_QFLAG32, Ity_I32)); |
| 1158 | //ZZ return t; |
| 1159 | //ZZ } |
| 1160 | //ZZ |
| 1161 | //ZZ static void put_QFLAG32 ( IRTemp t, IRTemp condT ) |
| 1162 | //ZZ { |
| 1163 | //ZZ putMiscReg32( OFFB_QFLAG32, mkexpr(t), condT ); |
| 1164 | //ZZ } |
| 1165 | //ZZ |
| 1166 | //ZZ /* Stickily set the 'Q' flag (APSR bit 27) of the APSR (Application Program |
| 1167 | //ZZ Status Register) to indicate that overflow or saturation occurred. |
| 1168 | //ZZ Nb: t must be zero to denote no saturation, and any nonzero |
| 1169 | //ZZ value to indicate saturation. */ |
| 1170 | //ZZ static void or_into_QFLAG32 ( IRExpr* e, IRTemp condT ) |
| 1171 | //ZZ { |
| 1172 | //ZZ IRTemp old = get_QFLAG32(); |
| 1173 | //ZZ IRTemp nyu = newTemp(Ity_I32); |
| 1174 | //ZZ assign(nyu, binop(Iop_Or32, mkexpr(old), e) ); |
| 1175 | //ZZ put_QFLAG32(nyu, condT); |
| 1176 | //ZZ } |
| 1177 | |
| 1178 | |
| 1179 | /* ---------------- FPCR stuff ---------------- */ |
| 1180 | |
| 1181 | /* Generate IR to get hold of the rounding mode bits in FPCR, and |
| 1182 | convert them to IR format. Bind the final result to the |
| 1183 | returned temp. */ |
| 1184 | static IRTemp /* :: Ity_I32 */ mk_get_IR_rounding_mode ( void ) |
| 1185 | { |
| 1186 | /* The ARMvfp encoding for rounding mode bits is: |
| 1187 | 00 to nearest |
| 1188 | 01 to +infinity |
| 1189 | 10 to -infinity |
| 1190 | 11 to zero |
| 1191 | We need to convert that to the IR encoding: |
| 1192 | 00 to nearest (the default) |
| 1193 | 10 to +infinity |
| 1194 | 01 to -infinity |
| 1195 | 11 to zero |
| 1196 | Which can be done by swapping bits 0 and 1. |
| 1197 | The rmode bits are at 23:22 in FPSCR. |
| 1198 | */ |
| 1199 | IRTemp armEncd = newTemp(Ity_I32); |
| 1200 | IRTemp swapped = newTemp(Ity_I32); |
| 1201 | /* Fish FPCR[23:22] out, and slide to bottom. Doesn't matter that |
| 1202 | we don't zero out bits 24 and above, since the assignment to |
| 1203 | 'swapped' will mask them out anyway. */ |
| 1204 | assign(armEncd, |
| 1205 | binop(Iop_Shr32, IRExpr_Get(OFFB_FPCR, Ity_I32), mkU8(22))); |
| 1206 | /* Now swap them. */ |
| 1207 | assign(swapped, |
| 1208 | binop(Iop_Or32, |
| 1209 | binop(Iop_And32, |
| 1210 | binop(Iop_Shl32, mkexpr(armEncd), mkU8(1)), |
| 1211 | mkU32(2)), |
| 1212 | binop(Iop_And32, |
| 1213 | binop(Iop_Shr32, mkexpr(armEncd), mkU8(1)), |
| 1214 | mkU32(1)) |
| 1215 | )); |
| 1216 | return swapped; |
| 1217 | } |
| 1218 | |
| 1219 | |
| 1220 | /*------------------------------------------------------------*/ |
| 1221 | /*--- Helpers for flag handling and conditional insns ---*/ |
| 1222 | /*------------------------------------------------------------*/ |
| 1223 | |
| 1224 | static const HChar* nameARM64Condcode ( ARM64Condcode cond ) |
| 1225 | { |
| 1226 | switch (cond) { |
| 1227 | case ARM64CondEQ: return "eq"; |
| 1228 | case ARM64CondNE: return "ne"; |
| 1229 | case ARM64CondCS: return "cs"; // or 'hs' |
| 1230 | case ARM64CondCC: return "cc"; // or 'lo' |
| 1231 | case ARM64CondMI: return "mi"; |
| 1232 | case ARM64CondPL: return "pl"; |
| 1233 | case ARM64CondVS: return "vs"; |
| 1234 | case ARM64CondVC: return "vc"; |
| 1235 | case ARM64CondHI: return "hi"; |
| 1236 | case ARM64CondLS: return "ls"; |
| 1237 | case ARM64CondGE: return "ge"; |
| 1238 | case ARM64CondLT: return "lt"; |
| 1239 | case ARM64CondGT: return "gt"; |
| 1240 | case ARM64CondLE: return "le"; |
| 1241 | case ARM64CondAL: return "al"; |
| 1242 | case ARM64CondNV: return "nv"; |
| 1243 | default: vpanic("name_ARM64Condcode"); |
| 1244 | } |
| 1245 | } |
| 1246 | |
| 1247 | /* and a handy shorthand for it */ |
| 1248 | static const HChar* nameCC ( ARM64Condcode cond ) { |
| 1249 | return nameARM64Condcode(cond); |
| 1250 | } |
| 1251 | |
| 1252 | |
| 1253 | /* Build IR to calculate some particular condition from stored |
| 1254 | CC_OP/CC_DEP1/CC_DEP2/CC_NDEP. Returns an expression of type |
| 1255 | Ity_I64, suitable for narrowing. Although the return type is |
| 1256 | Ity_I64, the returned value is either 0 or 1. 'cond' must be |
| 1257 | :: Ity_I64 and must denote the condition to compute in |
| 1258 | bits 7:4, and be zero everywhere else. |
| 1259 | */ |
| 1260 | static IRExpr* mk_arm64g_calculate_condition_dyn ( IRExpr* cond ) |
| 1261 | { |
| 1262 | vassert(typeOfIRExpr(irsb->tyenv, cond) == Ity_I64); |
| 1263 | /* And 'cond' had better produce a value in which only bits 7:4 are |
| 1264 | nonzero. However, obviously we can't assert for that. */ |
| 1265 | |
| 1266 | /* So what we're constructing for the first argument is |
| 1267 | "(cond << 4) | stored-operation". |
| 1268 | However, as per comments above, 'cond' must be supplied |
| 1269 | pre-shifted to this function. |
| 1270 | |
| 1271 | This pairing scheme requires that the ARM64_CC_OP_ values all fit |
| 1272 | in 4 bits. Hence we are passing a (COND, OP) pair in the lowest |
| 1273 | 8 bits of the first argument. */ |
| 1274 | IRExpr** args |
| 1275 | = mkIRExprVec_4( |
| 1276 | binop(Iop_Or64, IRExpr_Get(OFFB_CC_OP, Ity_I64), cond), |
| 1277 | IRExpr_Get(OFFB_CC_DEP1, Ity_I64), |
| 1278 | IRExpr_Get(OFFB_CC_DEP2, Ity_I64), |
| 1279 | IRExpr_Get(OFFB_CC_NDEP, Ity_I64) |
| 1280 | ); |
| 1281 | IRExpr* call |
| 1282 | = mkIRExprCCall( |
| 1283 | Ity_I64, |
| 1284 | 0/*regparm*/, |
| 1285 | "arm64g_calculate_condition", &arm64g_calculate_condition, |
| 1286 | args |
| 1287 | ); |
| 1288 | |
| 1289 | /* Exclude the requested condition, OP and NDEP from definedness |
| 1290 | checking. We're only interested in DEP1 and DEP2. */ |
| 1291 | call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3); |
| 1292 | return call; |
| 1293 | } |
| 1294 | |
| 1295 | |
| 1296 | /* Build IR to calculate some particular condition from stored |
| 1297 | CC_OP/CC_DEP1/CC_DEP2/CC_NDEP. Returns an expression of type |
| 1298 | Ity_I64, suitable for narrowing. Although the return type is |
| 1299 | Ity_I64, the returned value is either 0 or 1. |
| 1300 | */ |
| 1301 | static IRExpr* mk_arm64g_calculate_condition ( ARM64Condcode cond ) |
| 1302 | { |
| 1303 | /* First arg is "(cond << 4) | condition". This requires that the |
| 1304 | ARM64_CC_OP_ values all fit in 4 bits. Hence we are passing a |
| 1305 | (COND, OP) pair in the lowest 8 bits of the first argument. */ |
| 1306 | vassert(cond >= 0 && cond <= 15); |
| 1307 | return mk_arm64g_calculate_condition_dyn( mkU64(cond << 4) ); |
| 1308 | } |
| 1309 | |
| 1310 | |
| 1311 | //ZZ /* Build IR to calculate just the carry flag from stored |
| 1312 | //ZZ CC_OP/CC_DEP1/CC_DEP2/CC_NDEP. Returns an expression :: |
| 1313 | //ZZ Ity_I32. */ |
| 1314 | //ZZ static IRExpr* mk_armg_calculate_flag_c ( void ) |
| 1315 | //ZZ { |
| 1316 | //ZZ IRExpr** args |
| 1317 | //ZZ = mkIRExprVec_4( IRExpr_Get(OFFB_CC_OP, Ity_I32), |
| 1318 | //ZZ IRExpr_Get(OFFB_CC_DEP1, Ity_I32), |
| 1319 | //ZZ IRExpr_Get(OFFB_CC_DEP2, Ity_I32), |
| 1320 | //ZZ IRExpr_Get(OFFB_CC_NDEP, Ity_I32) ); |
| 1321 | //ZZ IRExpr* call |
| 1322 | //ZZ = mkIRExprCCall( |
| 1323 | //ZZ Ity_I32, |
| 1324 | //ZZ 0/*regparm*/, |
| 1325 | //ZZ "armg_calculate_flag_c", &armg_calculate_flag_c, |
| 1326 | //ZZ args |
| 1327 | //ZZ ); |
| 1328 | //ZZ /* Exclude OP and NDEP from definedness checking. We're only |
| 1329 | //ZZ interested in DEP1 and DEP2. */ |
| 1330 | //ZZ call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3); |
| 1331 | //ZZ return call; |
| 1332 | //ZZ } |
| 1333 | //ZZ |
| 1334 | //ZZ |
| 1335 | //ZZ /* Build IR to calculate just the overflow flag from stored |
| 1336 | //ZZ CC_OP/CC_DEP1/CC_DEP2/CC_NDEP. Returns an expression :: |
| 1337 | //ZZ Ity_I32. */ |
| 1338 | //ZZ static IRExpr* mk_armg_calculate_flag_v ( void ) |
| 1339 | //ZZ { |
| 1340 | //ZZ IRExpr** args |
| 1341 | //ZZ = mkIRExprVec_4( IRExpr_Get(OFFB_CC_OP, Ity_I32), |
| 1342 | //ZZ IRExpr_Get(OFFB_CC_DEP1, Ity_I32), |
| 1343 | //ZZ IRExpr_Get(OFFB_CC_DEP2, Ity_I32), |
| 1344 | //ZZ IRExpr_Get(OFFB_CC_NDEP, Ity_I32) ); |
| 1345 | //ZZ IRExpr* call |
| 1346 | //ZZ = mkIRExprCCall( |
| 1347 | //ZZ Ity_I32, |
| 1348 | //ZZ 0/*regparm*/, |
| 1349 | //ZZ "armg_calculate_flag_v", &armg_calculate_flag_v, |
| 1350 | //ZZ args |
| 1351 | //ZZ ); |
| 1352 | //ZZ /* Exclude OP and NDEP from definedness checking. We're only |
| 1353 | //ZZ interested in DEP1 and DEP2. */ |
| 1354 | //ZZ call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3); |
| 1355 | //ZZ return call; |
| 1356 | //ZZ } |
| 1357 | |
| 1358 | |
| 1359 | /* Build IR to calculate N Z C V in bits 31:28 of the |
| 1360 | returned word. */ |
| 1361 | static IRExpr* mk_arm64g_calculate_flags_nzcv ( void ) |
| 1362 | { |
| 1363 | IRExpr** args |
| 1364 | = mkIRExprVec_4( IRExpr_Get(OFFB_CC_OP, Ity_I64), |
| 1365 | IRExpr_Get(OFFB_CC_DEP1, Ity_I64), |
| 1366 | IRExpr_Get(OFFB_CC_DEP2, Ity_I64), |
| 1367 | IRExpr_Get(OFFB_CC_NDEP, Ity_I64) ); |
| 1368 | IRExpr* call |
| 1369 | = mkIRExprCCall( |
| 1370 | Ity_I64, |
| 1371 | 0/*regparm*/, |
| 1372 | "arm64g_calculate_flags_nzcv", &arm64g_calculate_flags_nzcv, |
| 1373 | args |
| 1374 | ); |
| 1375 | /* Exclude OP and NDEP from definedness checking. We're only |
| 1376 | interested in DEP1 and DEP2. */ |
| 1377 | call->Iex.CCall.cee->mcx_mask = (1<<0) | (1<<3); |
| 1378 | return call; |
| 1379 | } |
| 1380 | |
| 1381 | |
| 1382 | /* Build IR to set the flags thunk, in the most general case. */ |
| 1383 | static |
| 1384 | void setFlags_D1_D2_ND ( UInt cc_op, |
| 1385 | IRTemp t_dep1, IRTemp t_dep2, IRTemp t_ndep ) |
| 1386 | { |
| 1387 | vassert(typeOfIRTemp(irsb->tyenv, t_dep1 == Ity_I64)); |
| 1388 | vassert(typeOfIRTemp(irsb->tyenv, t_dep2 == Ity_I64)); |
| 1389 | vassert(typeOfIRTemp(irsb->tyenv, t_ndep == Ity_I64)); |
| 1390 | vassert(cc_op >= ARM64G_CC_OP_COPY && cc_op < ARM64G_CC_OP_NUMBER); |
| 1391 | stmt( IRStmt_Put( OFFB_CC_OP, mkU64(cc_op) )); |
| 1392 | stmt( IRStmt_Put( OFFB_CC_DEP1, mkexpr(t_dep1) )); |
| 1393 | stmt( IRStmt_Put( OFFB_CC_DEP2, mkexpr(t_dep2) )); |
| 1394 | stmt( IRStmt_Put( OFFB_CC_NDEP, mkexpr(t_ndep) )); |
| 1395 | } |
| 1396 | |
| 1397 | /* Build IR to set the flags thunk after ADD or SUB. */ |
| 1398 | static |
| 1399 | void setFlags_ADD_SUB ( Bool is64, Bool isSUB, IRTemp argL, IRTemp argR ) |
| 1400 | { |
| 1401 | IRTemp argL64 = IRTemp_INVALID; |
| 1402 | IRTemp argR64 = IRTemp_INVALID; |
| 1403 | IRTemp z64 = newTemp(Ity_I64); |
| 1404 | if (is64) { |
| 1405 | argL64 = argL; |
| 1406 | argR64 = argR; |
| 1407 | } else { |
| 1408 | argL64 = newTemp(Ity_I64); |
| 1409 | argR64 = newTemp(Ity_I64); |
| 1410 | assign(argL64, unop(Iop_32Uto64, mkexpr(argL))); |
| 1411 | assign(argR64, unop(Iop_32Uto64, mkexpr(argR))); |
| 1412 | } |
| 1413 | assign(z64, mkU64(0)); |
| 1414 | UInt cc_op = ARM64G_CC_OP_NUMBER; |
| 1415 | /**/ if ( isSUB && is64) { cc_op = ARM64G_CC_OP_SUB64; } |
| 1416 | else if ( isSUB && !is64) { cc_op = ARM64G_CC_OP_SUB32; } |
| 1417 | else if (!isSUB && is64) { cc_op = ARM64G_CC_OP_ADD64; } |
| 1418 | else if (!isSUB && !is64) { cc_op = ARM64G_CC_OP_ADD32; } |
| 1419 | else { vassert(0); } |
| 1420 | setFlags_D1_D2_ND(cc_op, argL64, argR64, z64); |
| 1421 | } |
| 1422 | |
| 1423 | /* Build IR to set the flags thunk after ADD or SUB, if the given |
| 1424 | condition evaluates to True at run time. If not, the flags are set |
| 1425 | to the specified NZCV value. */ |
| 1426 | static |
| 1427 | void setFlags_ADD_SUB_conditionally ( |
| 1428 | Bool is64, Bool isSUB, |
| 1429 | IRTemp cond, IRTemp argL, IRTemp argR, UInt nzcv |
| 1430 | ) |
| 1431 | { |
| 1432 | /* Generate IR as follows: |
| 1433 | CC_OP = ITE(cond, OP_{ADD,SUB}{32,64}, OP_COPY) |
| 1434 | CC_DEP1 = ITE(cond, argL64, nzcv << 28) |
| 1435 | CC_DEP2 = ITE(cond, argR64, 0) |
| 1436 | CC_NDEP = 0 |
| 1437 | */ |
| 1438 | |
| 1439 | IRTemp z64 = newTemp(Ity_I64); |
| 1440 | assign(z64, mkU64(0)); |
| 1441 | |
| 1442 | /* Establish the operation and operands for the True case. */ |
| 1443 | IRTemp t_dep1 = IRTemp_INVALID; |
| 1444 | IRTemp t_dep2 = IRTemp_INVALID; |
| 1445 | UInt t_op = ARM64G_CC_OP_NUMBER; |
| 1446 | /**/ if ( isSUB && is64) { t_op = ARM64G_CC_OP_SUB64; } |
| 1447 | else if ( isSUB && !is64) { t_op = ARM64G_CC_OP_SUB32; } |
| 1448 | else if (!isSUB && is64) { t_op = ARM64G_CC_OP_ADD64; } |
| 1449 | else if (!isSUB && !is64) { t_op = ARM64G_CC_OP_ADD32; } |
| 1450 | else { vassert(0); } |
| 1451 | /* */ |
| 1452 | if (is64) { |
| 1453 | t_dep1 = argL; |
| 1454 | t_dep2 = argR; |
| 1455 | } else { |
| 1456 | t_dep1 = newTemp(Ity_I64); |
| 1457 | t_dep2 = newTemp(Ity_I64); |
| 1458 | assign(t_dep1, unop(Iop_32Uto64, mkexpr(argL))); |
| 1459 | assign(t_dep2, unop(Iop_32Uto64, mkexpr(argR))); |
| 1460 | } |
| 1461 | |
| 1462 | /* Establish the operation and operands for the False case. */ |
| 1463 | IRTemp f_dep1 = newTemp(Ity_I64); |
| 1464 | IRTemp f_dep2 = z64; |
| 1465 | UInt f_op = ARM64G_CC_OP_COPY; |
| 1466 | assign(f_dep1, mkU64(nzcv << 28)); |
| 1467 | |
| 1468 | /* Final thunk values */ |
| 1469 | IRTemp dep1 = newTemp(Ity_I64); |
| 1470 | IRTemp dep2 = newTemp(Ity_I64); |
| 1471 | IRTemp op = newTemp(Ity_I64); |
| 1472 | |
| 1473 | assign(op, IRExpr_ITE(mkexpr(cond), mkU64(t_op), mkU64(f_op))); |
| 1474 | assign(dep1, IRExpr_ITE(mkexpr(cond), mkexpr(t_dep1), mkexpr(f_dep1))); |
| 1475 | assign(dep2, IRExpr_ITE(mkexpr(cond), mkexpr(t_dep2), mkexpr(f_dep2))); |
| 1476 | |
| 1477 | /* finally .. */ |
| 1478 | stmt( IRStmt_Put( OFFB_CC_OP, mkexpr(op) )); |
| 1479 | stmt( IRStmt_Put( OFFB_CC_DEP1, mkexpr(dep1) )); |
| 1480 | stmt( IRStmt_Put( OFFB_CC_DEP2, mkexpr(dep2) )); |
| 1481 | stmt( IRStmt_Put( OFFB_CC_NDEP, mkexpr(z64) )); |
| 1482 | } |
| 1483 | |
| 1484 | /* Build IR to set the flags thunk after AND/OR/XOR or variants thereof. */ |
| 1485 | static |
| 1486 | void setFlags_LOGIC ( Bool is64, IRTemp res ) |
| 1487 | { |
| 1488 | IRTemp res64 = IRTemp_INVALID; |
| 1489 | IRTemp z64 = newTemp(Ity_I64); |
| 1490 | UInt cc_op = ARM64G_CC_OP_NUMBER; |
| 1491 | if (is64) { |
| 1492 | res64 = res; |
| 1493 | cc_op = ARM64G_CC_OP_LOGIC64; |
| 1494 | } else { |
| 1495 | res64 = newTemp(Ity_I64); |
| 1496 | assign(res64, unop(Iop_32Uto64, mkexpr(res))); |
| 1497 | cc_op = ARM64G_CC_OP_LOGIC32; |
| 1498 | } |
| 1499 | assign(z64, mkU64(0)); |
| 1500 | setFlags_D1_D2_ND(cc_op, res64, z64, z64); |
| 1501 | } |
| 1502 | |
| 1503 | /* Build IR to set the flags thunk to a given NZCV value. NZCV is |
| 1504 | located in bits 31:28 of the supplied value. */ |
| 1505 | static |
| 1506 | void setFlags_COPY ( IRTemp nzcv_28x0 ) |
| 1507 | { |
| 1508 | IRTemp z64 = newTemp(Ity_I64); |
| 1509 | assign(z64, mkU64(0)); |
| 1510 | setFlags_D1_D2_ND(ARM64G_CC_OP_COPY, nzcv_28x0, z64, z64); |
| 1511 | } |
| 1512 | |
| 1513 | |
| 1514 | //ZZ /* Minor variant of the above that sets NDEP to zero (if it |
| 1515 | //ZZ sets it at all) */ |
| 1516 | //ZZ static void setFlags_D1_D2 ( UInt cc_op, IRTemp t_dep1, |
| 1517 | //ZZ IRTemp t_dep2, |
| 1518 | //ZZ IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
| 1519 | //ZZ { |
| 1520 | //ZZ IRTemp z32 = newTemp(Ity_I32); |
| 1521 | //ZZ assign( z32, mkU32(0) ); |
| 1522 | //ZZ setFlags_D1_D2_ND( cc_op, t_dep1, t_dep2, z32, guardT ); |
| 1523 | //ZZ } |
| 1524 | //ZZ |
| 1525 | //ZZ |
| 1526 | //ZZ /* Minor variant of the above that sets DEP2 to zero (if it |
| 1527 | //ZZ sets it at all) */ |
| 1528 | //ZZ static void setFlags_D1_ND ( UInt cc_op, IRTemp t_dep1, |
| 1529 | //ZZ IRTemp t_ndep, |
| 1530 | //ZZ IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
| 1531 | //ZZ { |
| 1532 | //ZZ IRTemp z32 = newTemp(Ity_I32); |
| 1533 | //ZZ assign( z32, mkU32(0) ); |
| 1534 | //ZZ setFlags_D1_D2_ND( cc_op, t_dep1, z32, t_ndep, guardT ); |
| 1535 | //ZZ } |
| 1536 | //ZZ |
| 1537 | //ZZ |
| 1538 | //ZZ /* Minor variant of the above that sets DEP2 and NDEP to zero (if it |
| 1539 | //ZZ sets them at all) */ |
| 1540 | //ZZ static void setFlags_D1 ( UInt cc_op, IRTemp t_dep1, |
| 1541 | //ZZ IRTemp guardT /* :: Ity_I32, 0 or 1 */ ) |
| 1542 | //ZZ { |
| 1543 | //ZZ IRTemp z32 = newTemp(Ity_I32); |
| 1544 | //ZZ assign( z32, mkU32(0) ); |
| 1545 | //ZZ setFlags_D1_D2_ND( cc_op, t_dep1, z32, z32, guardT ); |
| 1546 | //ZZ } |
| 1547 | |
| 1548 | |
| 1549 | /*------------------------------------------------------------*/ |
| 1550 | /*--- Misc math helpers ---*/ |
| 1551 | /*------------------------------------------------------------*/ |
| 1552 | |
| 1553 | /* Generates a 64-bit byte swap. */ |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 1554 | static IRTemp math_BYTESWAP64 ( IRTemp src ) |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1555 | { |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1556 | IRTemp m8 = newTemp(Ity_I64); |
| 1557 | IRTemp s8 = newTemp(Ity_I64); |
| 1558 | IRTemp m16 = newTemp(Ity_I64); |
| 1559 | IRTemp s16 = newTemp(Ity_I64); |
| 1560 | IRTemp m32 = newTemp(Ity_I64); |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 1561 | IRTemp res = newTemp(Ity_I64); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1562 | assign( m8, mkU64(0xFF00FF00FF00FF00ULL) ); |
| 1563 | assign( s8, |
| 1564 | binop(Iop_Or64, |
| 1565 | binop(Iop_Shr64, |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 1566 | binop(Iop_And64,mkexpr(src),mkexpr(m8)), |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1567 | mkU8(8)), |
| 1568 | binop(Iop_And64, |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 1569 | binop(Iop_Shl64,mkexpr(src),mkU8(8)), |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1570 | mkexpr(m8)) |
| 1571 | ) |
| 1572 | ); |
| 1573 | |
| 1574 | assign( m16, mkU64(0xFFFF0000FFFF0000ULL) ); |
| 1575 | assign( s16, |
| 1576 | binop(Iop_Or64, |
| 1577 | binop(Iop_Shr64, |
| 1578 | binop(Iop_And64,mkexpr(s8),mkexpr(m16)), |
| 1579 | mkU8(16)), |
| 1580 | binop(Iop_And64, |
| 1581 | binop(Iop_Shl64,mkexpr(s8),mkU8(16)), |
| 1582 | mkexpr(m16)) |
| 1583 | ) |
| 1584 | ); |
| 1585 | |
| 1586 | assign( m32, mkU64(0xFFFFFFFF00000000ULL) ); |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 1587 | assign( res, |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1588 | binop(Iop_Or64, |
| 1589 | binop(Iop_Shr64, |
| 1590 | binop(Iop_And64,mkexpr(s16),mkexpr(m32)), |
| 1591 | mkU8(32)), |
| 1592 | binop(Iop_And64, |
| 1593 | binop(Iop_Shl64,mkexpr(s16),mkU8(32)), |
| 1594 | mkexpr(m32)) |
| 1595 | ) |
| 1596 | ); |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 1597 | return res; |
| 1598 | } |
| 1599 | |
| 1600 | |
| 1601 | /* Generates a 64-bit bit swap. */ |
| 1602 | static IRTemp math_BITSWAP64 ( IRTemp src ) |
| 1603 | { |
| 1604 | IRTemp m1 = newTemp(Ity_I64); |
| 1605 | IRTemp s1 = newTemp(Ity_I64); |
| 1606 | IRTemp m2 = newTemp(Ity_I64); |
| 1607 | IRTemp s2 = newTemp(Ity_I64); |
| 1608 | IRTemp m4 = newTemp(Ity_I64); |
| 1609 | IRTemp s4 = newTemp(Ity_I64); |
| 1610 | assign( m1, mkU64(0xAAAAAAAAAAAAAAAAULL) ); |
| 1611 | assign( s1, |
| 1612 | binop(Iop_Or64, |
| 1613 | binop(Iop_Shr64, |
| 1614 | binop(Iop_And64,mkexpr(src),mkexpr(m1)), |
| 1615 | mkU8(1)), |
| 1616 | binop(Iop_And64, |
| 1617 | binop(Iop_Shl64,mkexpr(src),mkU8(1)), |
| 1618 | mkexpr(m1)) |
| 1619 | ) |
| 1620 | ); |
| 1621 | |
| 1622 | assign( m2, mkU64(0xCCCCCCCCCCCCCCCCULL) ); |
| 1623 | assign( s2, |
| 1624 | binop(Iop_Or64, |
| 1625 | binop(Iop_Shr64, |
| 1626 | binop(Iop_And64,mkexpr(s1),mkexpr(m2)), |
| 1627 | mkU8(2)), |
| 1628 | binop(Iop_And64, |
| 1629 | binop(Iop_Shl64,mkexpr(s1),mkU8(2)), |
| 1630 | mkexpr(m2)) |
| 1631 | ) |
| 1632 | ); |
| 1633 | |
| 1634 | assign( m4, mkU64(0xF0F0F0F0F0F0F0F0ULL) ); |
| 1635 | assign( s4, |
| 1636 | binop(Iop_Or64, |
| 1637 | binop(Iop_Shr64, |
| 1638 | binop(Iop_And64,mkexpr(s2),mkexpr(m4)), |
| 1639 | mkU8(4)), |
| 1640 | binop(Iop_And64, |
| 1641 | binop(Iop_Shl64,mkexpr(s2),mkU8(4)), |
| 1642 | mkexpr(m4)) |
| 1643 | ) |
| 1644 | ); |
| 1645 | return math_BYTESWAP64(s4); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1646 | } |
| 1647 | |
| 1648 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 1649 | /* Duplicates the bits at the bottom of the given word to fill the |
| 1650 | whole word. src :: Ity_I64 is assumed to have zeroes everywhere |
| 1651 | except for the bottom bits. */ |
| 1652 | static IRTemp math_DUP_TO_64 ( IRTemp src, IRType srcTy ) |
| 1653 | { |
| 1654 | if (srcTy == Ity_I8) { |
| 1655 | IRTemp t16 = newTemp(Ity_I64); |
| 1656 | assign(t16, binop(Iop_Or64, mkexpr(src), |
| 1657 | binop(Iop_Shl64, mkexpr(src), mkU8(8)))); |
| 1658 | IRTemp t32 = newTemp(Ity_I64); |
| 1659 | assign(t32, binop(Iop_Or64, mkexpr(t16), |
| 1660 | binop(Iop_Shl64, mkexpr(t16), mkU8(16)))); |
| 1661 | IRTemp t64 = newTemp(Ity_I64); |
| 1662 | assign(t64, binop(Iop_Or64, mkexpr(t32), |
| 1663 | binop(Iop_Shl64, mkexpr(t32), mkU8(32)))); |
| 1664 | return t64; |
| 1665 | } |
| 1666 | if (srcTy == Ity_I16) { |
| 1667 | IRTemp t32 = newTemp(Ity_I64); |
| 1668 | assign(t32, binop(Iop_Or64, mkexpr(src), |
| 1669 | binop(Iop_Shl64, mkexpr(src), mkU8(16)))); |
| 1670 | IRTemp t64 = newTemp(Ity_I64); |
| 1671 | assign(t64, binop(Iop_Or64, mkexpr(t32), |
| 1672 | binop(Iop_Shl64, mkexpr(t32), mkU8(32)))); |
| 1673 | return t64; |
| 1674 | } |
| 1675 | if (srcTy == Ity_I32) { |
| 1676 | IRTemp t64 = newTemp(Ity_I64); |
| 1677 | assign(t64, binop(Iop_Or64, mkexpr(src), |
| 1678 | binop(Iop_Shl64, mkexpr(src), mkU8(32)))); |
| 1679 | return t64; |
| 1680 | } |
| 1681 | if (srcTy == Ity_I64) { |
| 1682 | return src; |
| 1683 | } |
| 1684 | vassert(0); |
| 1685 | } |
| 1686 | |
| 1687 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 1688 | /*------------------------------------------------------------*/ |
| 1689 | /*--- FP comparison helpers ---*/ |
| 1690 | /*------------------------------------------------------------*/ |
| 1691 | |
| 1692 | /* irRes :: Ity_I32 holds a floating point comparison result encoded |
| 1693 | as an IRCmpF64Result. Generate code to convert it to an |
| 1694 | ARM64-encoded (N,Z,C,V) group in the lowest 4 bits of an I64 value. |
| 1695 | Assign a new temp to hold that value, and return the temp. */ |
| 1696 | static |
| 1697 | IRTemp mk_convert_IRCmpF64Result_to_NZCV ( IRTemp irRes32 ) |
| 1698 | { |
| 1699 | IRTemp ix = newTemp(Ity_I64); |
| 1700 | IRTemp termL = newTemp(Ity_I64); |
| 1701 | IRTemp termR = newTemp(Ity_I64); |
| 1702 | IRTemp nzcv = newTemp(Ity_I64); |
| 1703 | IRTemp irRes = newTemp(Ity_I64); |
| 1704 | |
| 1705 | /* This is where the fun starts. We have to convert 'irRes' from |
| 1706 | an IR-convention return result (IRCmpF64Result) to an |
| 1707 | ARM-encoded (N,Z,C,V) group. The final result is in the bottom |
| 1708 | 4 bits of 'nzcv'. */ |
| 1709 | /* Map compare result from IR to ARM(nzcv) */ |
| 1710 | /* |
| 1711 | FP cmp result | IR | ARM(nzcv) |
| 1712 | -------------------------------- |
| 1713 | UN 0x45 0011 |
| 1714 | LT 0x01 1000 |
| 1715 | GT 0x00 0010 |
| 1716 | EQ 0x40 0110 |
| 1717 | */ |
| 1718 | /* Now since you're probably wondering WTF .. |
| 1719 | |
| 1720 | ix fishes the useful bits out of the IR value, bits 6 and 0, and |
| 1721 | places them side by side, giving a number which is 0, 1, 2 or 3. |
| 1722 | |
| 1723 | termL is a sequence cooked up by GNU superopt. It converts ix |
| 1724 | into an almost correct value NZCV value (incredibly), except |
| 1725 | for the case of UN, where it produces 0100 instead of the |
| 1726 | required 0011. |
| 1727 | |
| 1728 | termR is therefore a correction term, also computed from ix. It |
| 1729 | is 1 in the UN case and 0 for LT, GT and UN. Hence, to get |
| 1730 | the final correct value, we subtract termR from termL. |
| 1731 | |
| 1732 | Don't take my word for it. There's a test program at the bottom |
| 1733 | of guest_arm_toIR.c, to try this out with. |
| 1734 | */ |
| 1735 | assign(irRes, unop(Iop_32Uto64, mkexpr(irRes32))); |
| 1736 | |
| 1737 | assign( |
| 1738 | ix, |
| 1739 | binop(Iop_Or64, |
| 1740 | binop(Iop_And64, |
| 1741 | binop(Iop_Shr64, mkexpr(irRes), mkU8(5)), |
| 1742 | mkU64(3)), |
| 1743 | binop(Iop_And64, mkexpr(irRes), mkU64(1)))); |
| 1744 | |
| 1745 | assign( |
| 1746 | termL, |
| 1747 | binop(Iop_Add64, |
| 1748 | binop(Iop_Shr64, |
| 1749 | binop(Iop_Sub64, |
| 1750 | binop(Iop_Shl64, |
| 1751 | binop(Iop_Xor64, mkexpr(ix), mkU64(1)), |
| 1752 | mkU8(62)), |
| 1753 | mkU64(1)), |
| 1754 | mkU8(61)), |
| 1755 | mkU64(1))); |
| 1756 | |
| 1757 | assign( |
| 1758 | termR, |
| 1759 | binop(Iop_And64, |
| 1760 | binop(Iop_And64, |
| 1761 | mkexpr(ix), |
| 1762 | binop(Iop_Shr64, mkexpr(ix), mkU8(1))), |
| 1763 | mkU64(1))); |
| 1764 | |
| 1765 | assign(nzcv, binop(Iop_Sub64, mkexpr(termL), mkexpr(termR))); |
| 1766 | return nzcv; |
| 1767 | } |
| 1768 | |
| 1769 | |
| 1770 | /*------------------------------------------------------------*/ |
| 1771 | /*--- Data processing (immediate) ---*/ |
| 1772 | /*------------------------------------------------------------*/ |
| 1773 | |
| 1774 | /* Helper functions for supporting "DecodeBitMasks" */ |
| 1775 | |
| 1776 | static ULong dbm_ROR ( Int width, ULong x, Int rot ) |
| 1777 | { |
| 1778 | vassert(width > 0 && width <= 64); |
| 1779 | vassert(rot >= 0 && rot < width); |
| 1780 | if (rot == 0) return x; |
| 1781 | ULong res = x >> rot; |
| 1782 | res |= (x << (width - rot)); |
| 1783 | if (width < 64) |
| 1784 | res &= ((1ULL << width) - 1); |
| 1785 | return res; |
| 1786 | } |
| 1787 | |
| 1788 | static ULong dbm_RepTo64( Int esize, ULong x ) |
| 1789 | { |
| 1790 | switch (esize) { |
| 1791 | case 64: |
| 1792 | return x; |
| 1793 | case 32: |
| 1794 | x &= 0xFFFFFFFF; x |= (x << 32); |
| 1795 | return x; |
| 1796 | case 16: |
| 1797 | x &= 0xFFFF; x |= (x << 16); x |= (x << 32); |
| 1798 | return x; |
| 1799 | case 8: |
| 1800 | x &= 0xFF; x |= (x << 8); x |= (x << 16); x |= (x << 32); |
| 1801 | return x; |
| 1802 | case 4: |
| 1803 | x &= 0xF; x |= (x << 4); x |= (x << 8); |
| 1804 | x |= (x << 16); x |= (x << 32); |
| 1805 | return x; |
| 1806 | case 2: |
| 1807 | x &= 0x3; x |= (x << 2); x |= (x << 4); x |= (x << 8); |
| 1808 | x |= (x << 16); x |= (x << 32); |
| 1809 | return x; |
| 1810 | default: |
| 1811 | break; |
| 1812 | } |
| 1813 | vpanic("dbm_RepTo64"); |
| 1814 | /*NOTREACHED*/ |
| 1815 | return 0; |
| 1816 | } |
| 1817 | |
| 1818 | static Int dbm_highestSetBit ( ULong x ) |
| 1819 | { |
| 1820 | Int i; |
| 1821 | for (i = 63; i >= 0; i--) { |
| 1822 | if (x & (1ULL << i)) |
| 1823 | return i; |
| 1824 | } |
| 1825 | vassert(x == 0); |
| 1826 | return -1; |
| 1827 | } |
| 1828 | |
| 1829 | static |
| 1830 | Bool dbm_DecodeBitMasks ( /*OUT*/ULong* wmask, /*OUT*/ULong* tmask, |
| 1831 | ULong immN, ULong imms, ULong immr, Bool immediate, |
| 1832 | UInt M /*32 or 64*/) |
| 1833 | { |
| 1834 | vassert(immN < (1ULL << 1)); |
| 1835 | vassert(imms < (1ULL << 6)); |
| 1836 | vassert(immr < (1ULL << 6)); |
| 1837 | vassert(immediate == False || immediate == True); |
| 1838 | vassert(M == 32 || M == 64); |
| 1839 | |
| 1840 | Int len = dbm_highestSetBit( ((immN << 6) & 64) | ((~imms) & 63) ); |
| 1841 | if (len < 1) { /* printf("fail1\n"); */ return False; } |
| 1842 | vassert(len <= 6); |
| 1843 | vassert(M >= (1 << len)); |
| 1844 | |
| 1845 | vassert(len >= 1 && len <= 6); |
| 1846 | ULong levels = // (zeroes(6 - len) << (6-len)) | ones(len); |
| 1847 | (1 << len) - 1; |
| 1848 | vassert(levels >= 1 && levels <= 63); |
| 1849 | |
| 1850 | if (immediate && ((imms & levels) == levels)) { |
| 1851 | /* printf("fail2 imms %llu levels %llu len %d\n", imms, levels, len); */ |
| 1852 | return False; |
| 1853 | } |
| 1854 | |
| 1855 | ULong S = imms & levels; |
| 1856 | ULong R = immr & levels; |
| 1857 | Int diff = S - R; |
| 1858 | diff &= 63; |
| 1859 | Int esize = 1 << len; |
| 1860 | vassert(2 <= esize && esize <= 64); |
| 1861 | |
| 1862 | /* Be careful of these (1ULL << (S+1)) - 1 expressions, and the |
| 1863 | same below with d. S can be 63 in which case we have an out of |
| 1864 | range and hence undefined shift. */ |
| 1865 | vassert(S >= 0 && S <= 63); |
| 1866 | vassert(esize >= (S+1)); |
| 1867 | ULong elem_s = // Zeroes(esize-(S+1)):Ones(S+1) |
| 1868 | //(1ULL << (S+1)) - 1; |
| 1869 | ((1ULL << S) - 1) + (1ULL << S); |
| 1870 | |
| 1871 | Int d = // diff<len-1:0> |
| 1872 | diff & ((1 << len)-1); |
| 1873 | vassert(esize >= (d+1)); |
| 1874 | vassert(d >= 0 && d <= 63); |
| 1875 | |
| 1876 | ULong elem_d = // Zeroes(esize-(d+1)):Ones(d+1) |
| 1877 | //(1ULL << (d+1)) - 1; |
| 1878 | ((1ULL << d) - 1) + (1ULL << d); |
| 1879 | |
| 1880 | if (esize != 64) vassert(elem_s < (1ULL << esize)); |
| 1881 | if (esize != 64) vassert(elem_d < (1ULL << esize)); |
| 1882 | |
| 1883 | if (wmask) *wmask = dbm_RepTo64(esize, dbm_ROR(esize, elem_s, R)); |
| 1884 | if (tmask) *tmask = dbm_RepTo64(esize, elem_d); |
| 1885 | |
| 1886 | return True; |
| 1887 | } |
| 1888 | |
| 1889 | |
| 1890 | static |
| 1891 | Bool dis_ARM64_data_processing_immediate(/*MB_OUT*/DisResult* dres, |
| 1892 | UInt insn) |
| 1893 | { |
| 1894 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 1895 | |
| 1896 | /* insn[28:23] |
| 1897 | 10000x PC-rel addressing |
| 1898 | 10001x Add/subtract (immediate) |
| 1899 | 100100 Logical (immediate) |
| 1900 | 100101 Move Wide (immediate) |
| 1901 | 100110 Bitfield |
| 1902 | 100111 Extract |
| 1903 | */ |
| 1904 | |
| 1905 | /* ------------------ ADD/SUB{,S} imm12 ------------------ */ |
| 1906 | if (INSN(28,24) == BITS5(1,0,0,0,1)) { |
| 1907 | Bool is64 = INSN(31,31) == 1; |
| 1908 | Bool isSub = INSN(30,30) == 1; |
| 1909 | Bool setCC = INSN(29,29) == 1; |
| 1910 | UInt sh = INSN(23,22); |
| 1911 | UInt uimm12 = INSN(21,10); |
| 1912 | UInt nn = INSN(9,5); |
| 1913 | UInt dd = INSN(4,0); |
| 1914 | const HChar* nm = isSub ? "sub" : "add"; |
| 1915 | if (sh >= 2) { |
| 1916 | /* Invalid; fall through */ |
| 1917 | } else { |
| 1918 | vassert(sh <= 1); |
| 1919 | uimm12 <<= (12 * sh); |
| 1920 | if (is64) { |
| 1921 | IRTemp argL = newTemp(Ity_I64); |
| 1922 | IRTemp argR = newTemp(Ity_I64); |
| 1923 | IRTemp res = newTemp(Ity_I64); |
| 1924 | assign(argL, getIReg64orSP(nn)); |
| 1925 | assign(argR, mkU64(uimm12)); |
| 1926 | assign(res, binop(isSub ? Iop_Sub64 : Iop_Add64, |
| 1927 | mkexpr(argL), mkexpr(argR))); |
| 1928 | if (setCC) { |
| 1929 | putIReg64orZR(dd, mkexpr(res)); |
| 1930 | setFlags_ADD_SUB(True/*is64*/, isSub, argL, argR); |
| 1931 | DIP("%ss %s, %s, 0x%x\n", |
| 1932 | nm, nameIReg64orZR(dd), nameIReg64orSP(nn), uimm12); |
| 1933 | } else { |
| 1934 | putIReg64orSP(dd, mkexpr(res)); |
| 1935 | DIP("%s %s, %s, 0x%x\n", |
| 1936 | nm, nameIReg64orSP(dd), nameIReg64orSP(nn), uimm12); |
| 1937 | } |
| 1938 | } else { |
| 1939 | IRTemp argL = newTemp(Ity_I32); |
| 1940 | IRTemp argR = newTemp(Ity_I32); |
| 1941 | IRTemp res = newTemp(Ity_I32); |
| 1942 | assign(argL, getIReg32orSP(nn)); |
| 1943 | assign(argR, mkU32(uimm12)); |
| 1944 | assign(res, binop(isSub ? Iop_Sub32 : Iop_Add32, |
| 1945 | mkexpr(argL), mkexpr(argR))); |
| 1946 | if (setCC) { |
| 1947 | putIReg32orZR(dd, mkexpr(res)); |
| 1948 | setFlags_ADD_SUB(False/*!is64*/, isSub, argL, argR); |
| 1949 | DIP("%ss %s, %s, 0x%x\n", |
| 1950 | nm, nameIReg32orZR(dd), nameIReg32orSP(nn), uimm12); |
| 1951 | } else { |
| 1952 | putIReg32orSP(dd, mkexpr(res)); |
| 1953 | DIP("%s %s, %s, 0x%x\n", |
| 1954 | nm, nameIReg32orSP(dd), nameIReg32orSP(nn), uimm12); |
| 1955 | } |
| 1956 | } |
| 1957 | return True; |
| 1958 | } |
| 1959 | } |
| 1960 | |
| 1961 | /* -------------------- ADR/ADRP -------------------- */ |
| 1962 | if (INSN(28,24) == BITS5(1,0,0,0,0)) { |
| 1963 | UInt bP = INSN(31,31); |
| 1964 | UInt immLo = INSN(30,29); |
| 1965 | UInt immHi = INSN(23,5); |
| 1966 | UInt rD = INSN(4,0); |
| 1967 | ULong uimm = (immHi << 2) | immLo; |
| 1968 | ULong simm = sx_to_64(uimm, 21); |
| 1969 | ULong val; |
| 1970 | if (bP) { |
| 1971 | val = (guest_PC_curr_instr & 0xFFFFFFFFFFFFF000ULL) + (simm << 12); |
| 1972 | } else { |
| 1973 | val = guest_PC_curr_instr + simm; |
| 1974 | } |
| 1975 | putIReg64orZR(rD, mkU64(val)); |
| 1976 | DIP("adr%s %s, 0x%llx\n", bP ? "p" : "", nameIReg64orZR(rD), val); |
| 1977 | return True; |
| 1978 | } |
| 1979 | |
| 1980 | /* -------------------- LOGIC(imm) -------------------- */ |
| 1981 | if (INSN(28,23) == BITS6(1,0,0,1,0,0)) { |
| 1982 | /* 31 30 28 22 21 15 9 4 |
| 1983 | sf op 100100 N immr imms Rn Rd |
| 1984 | op=00: AND Rd|SP, Rn, #imm |
| 1985 | op=01: ORR Rd|SP, Rn, #imm |
| 1986 | op=10: EOR Rd|SP, Rn, #imm |
| 1987 | op=11: ANDS Rd|ZR, Rn, #imm |
| 1988 | */ |
| 1989 | Bool is64 = INSN(31,31) == 1; |
| 1990 | UInt op = INSN(30,29); |
| 1991 | UInt N = INSN(22,22); |
| 1992 | UInt immR = INSN(21,16); |
| 1993 | UInt immS = INSN(15,10); |
| 1994 | UInt nn = INSN(9,5); |
| 1995 | UInt dd = INSN(4,0); |
| 1996 | ULong imm = 0; |
| 1997 | Bool ok; |
| 1998 | if (N == 1 && !is64) |
| 1999 | goto after_logic_imm; /* not allowed; fall through */ |
| 2000 | ok = dbm_DecodeBitMasks(&imm, NULL, |
| 2001 | N, immS, immR, True, is64 ? 64 : 32); |
| 2002 | if (!ok) |
| 2003 | goto after_logic_imm; |
| 2004 | |
| 2005 | const HChar* names[4] = { "and", "orr", "eor", "ands" }; |
| 2006 | const IROp ops64[4] = { Iop_And64, Iop_Or64, Iop_Xor64, Iop_And64 }; |
| 2007 | const IROp ops32[4] = { Iop_And32, Iop_Or32, Iop_Xor32, Iop_And32 }; |
| 2008 | |
| 2009 | vassert(op < 4); |
| 2010 | if (is64) { |
| 2011 | IRExpr* argL = getIReg64orZR(nn); |
| 2012 | IRExpr* argR = mkU64(imm); |
| 2013 | IRTemp res = newTemp(Ity_I64); |
| 2014 | assign(res, binop(ops64[op], argL, argR)); |
| 2015 | if (op < 3) { |
| 2016 | putIReg64orSP(dd, mkexpr(res)); |
| 2017 | DIP("%s %s, %s, 0x%llx\n", names[op], |
| 2018 | nameIReg64orSP(dd), nameIReg64orZR(nn), imm); |
| 2019 | } else { |
| 2020 | putIReg64orZR(dd, mkexpr(res)); |
| 2021 | setFlags_LOGIC(True/*is64*/, res); |
| 2022 | DIP("%s %s, %s, 0x%llx\n", names[op], |
| 2023 | nameIReg64orZR(dd), nameIReg64orZR(nn), imm); |
| 2024 | } |
| 2025 | } else { |
| 2026 | IRExpr* argL = getIReg32orZR(nn); |
| 2027 | IRExpr* argR = mkU32((UInt)imm); |
| 2028 | IRTemp res = newTemp(Ity_I32); |
| 2029 | assign(res, binop(ops32[op], argL, argR)); |
| 2030 | if (op < 3) { |
| 2031 | putIReg32orSP(dd, mkexpr(res)); |
| 2032 | DIP("%s %s, %s, 0x%x\n", names[op], |
| 2033 | nameIReg32orSP(dd), nameIReg32orZR(nn), (UInt)imm); |
| 2034 | } else { |
| 2035 | putIReg32orZR(dd, mkexpr(res)); |
| 2036 | setFlags_LOGIC(False/*!is64*/, res); |
| 2037 | DIP("%s %s, %s, 0x%x\n", names[op], |
| 2038 | nameIReg32orZR(dd), nameIReg32orZR(nn), (UInt)imm); |
| 2039 | } |
| 2040 | } |
| 2041 | return True; |
| 2042 | } |
| 2043 | after_logic_imm: |
| 2044 | |
| 2045 | /* -------------------- MOV{Z,N,K} -------------------- */ |
| 2046 | if (INSN(28,23) == BITS6(1,0,0,1,0,1)) { |
| 2047 | /* 31 30 28 22 20 4 |
| 2048 | | | | | | | |
| 2049 | sf 10 100 101 hw imm16 Rd MOV(Z) Rd, (imm16 << (16*hw)) |
| 2050 | sf 00 100 101 hw imm16 Rd MOV(N) Rd, ~(imm16 << (16*hw)) |
| 2051 | sf 11 100 101 hw imm16 Rd MOV(K) Rd, (imm16 << (16*hw)) |
| 2052 | */ |
| 2053 | Bool is64 = INSN(31,31) == 1; |
| 2054 | UInt subopc = INSN(30,29); |
| 2055 | UInt hw = INSN(22,21); |
| 2056 | UInt imm16 = INSN(20,5); |
| 2057 | UInt dd = INSN(4,0); |
| 2058 | if (subopc == BITS2(0,1) || (!is64 && hw >= 2)) { |
| 2059 | /* invalid; fall through */ |
| 2060 | } else { |
| 2061 | ULong imm64 = ((ULong)imm16) << (16 * hw); |
| 2062 | if (!is64) |
| 2063 | vassert(imm64 < 0x100000000ULL); |
| 2064 | switch (subopc) { |
| 2065 | case BITS2(1,0): // MOVZ |
| 2066 | putIRegOrZR(is64, dd, is64 ? mkU64(imm64) : mkU32((UInt)imm64)); |
| 2067 | DIP("movz %s, 0x%llx\n", nameIRegOrZR(is64, dd), imm64); |
| 2068 | break; |
| 2069 | case BITS2(0,0): // MOVN |
| 2070 | imm64 = ~imm64; |
| 2071 | if (!is64) |
| 2072 | imm64 &= 0xFFFFFFFFULL; |
| 2073 | putIRegOrZR(is64, dd, is64 ? mkU64(imm64) : mkU32((UInt)imm64)); |
| 2074 | DIP("movn %s, 0x%llx\n", nameIRegOrZR(is64, dd), imm64); |
| 2075 | break; |
| 2076 | case BITS2(1,1): // MOVK |
| 2077 | /* This is more complex. We are inserting a slice into |
| 2078 | the destination register, so we need to have the old |
| 2079 | value of it. */ |
| 2080 | if (is64) { |
| 2081 | IRTemp old = newTemp(Ity_I64); |
| 2082 | assign(old, getIReg64orZR(dd)); |
| 2083 | ULong mask = 0xFFFFULL << (16 * hw); |
| 2084 | IRExpr* res |
| 2085 | = binop(Iop_Or64, |
| 2086 | binop(Iop_And64, mkexpr(old), mkU64(~mask)), |
| 2087 | mkU64(imm64)); |
| 2088 | putIReg64orZR(dd, res); |
| 2089 | DIP("movk %s, 0x%x, lsl %u\n", |
| 2090 | nameIReg64orZR(dd), imm16, 16*hw); |
| 2091 | } else { |
| 2092 | IRTemp old = newTemp(Ity_I32); |
| 2093 | assign(old, getIReg32orZR(dd)); |
| 2094 | vassert(hw <= 1); |
| 2095 | UInt mask = 0xFFFF << (16 * hw); |
| 2096 | IRExpr* res |
| 2097 | = binop(Iop_Or32, |
| 2098 | binop(Iop_And32, mkexpr(old), mkU32(~mask)), |
| 2099 | mkU32((UInt)imm64)); |
| 2100 | putIReg32orZR(dd, res); |
| 2101 | DIP("movk %s, 0x%x, lsl %u\n", |
| 2102 | nameIReg32orZR(dd), imm16, 16*hw); |
| 2103 | } |
| 2104 | break; |
| 2105 | default: |
| 2106 | vassert(0); |
| 2107 | } |
| 2108 | return True; |
| 2109 | } |
| 2110 | } |
| 2111 | |
| 2112 | /* -------------------- {U,S,}BFM -------------------- */ |
| 2113 | /* 30 28 22 21 15 9 4 |
| 2114 | |
| 2115 | sf 10 100110 N immr imms nn dd |
| 2116 | UBFM Wd, Wn, #immr, #imms when sf=0, N=0, immr[5]=0, imms[5]=0 |
| 2117 | UBFM Xd, Xn, #immr, #imms when sf=1, N=1 |
| 2118 | |
| 2119 | sf 00 100110 N immr imms nn dd |
| 2120 | SBFM Wd, Wn, #immr, #imms when sf=0, N=0, immr[5]=0, imms[5]=0 |
| 2121 | SBFM Xd, Xn, #immr, #imms when sf=1, N=1 |
| 2122 | |
| 2123 | sf 01 100110 N immr imms nn dd |
| 2124 | BFM Wd, Wn, #immr, #imms when sf=0, N=0, immr[5]=0, imms[5]=0 |
| 2125 | BFM Xd, Xn, #immr, #imms when sf=1, N=1 |
| 2126 | */ |
| 2127 | if (INSN(28,23) == BITS6(1,0,0,1,1,0)) { |
| 2128 | UInt sf = INSN(31,31); |
| 2129 | UInt opc = INSN(30,29); |
| 2130 | UInt N = INSN(22,22); |
| 2131 | UInt immR = INSN(21,16); |
| 2132 | UInt immS = INSN(15,10); |
| 2133 | UInt nn = INSN(9,5); |
| 2134 | UInt dd = INSN(4,0); |
| 2135 | Bool inZero = False; |
| 2136 | Bool extend = False; |
| 2137 | const HChar* nm = "???"; |
| 2138 | /* skip invalid combinations */ |
| 2139 | switch (opc) { |
| 2140 | case BITS2(0,0): |
| 2141 | inZero = True; extend = True; nm = "sbfm"; break; |
| 2142 | case BITS2(0,1): |
| 2143 | inZero = False; extend = False; nm = "bfm"; break; |
| 2144 | case BITS2(1,0): |
| 2145 | inZero = True; extend = False; nm = "ubfm"; break; |
| 2146 | case BITS2(1,1): |
| 2147 | goto after_bfm; /* invalid */ |
| 2148 | default: |
| 2149 | vassert(0); |
| 2150 | } |
| 2151 | if (sf == 1 && N != 1) goto after_bfm; |
| 2152 | if (sf == 0 && (N != 0 || ((immR >> 5) & 1) != 0 |
| 2153 | || ((immS >> 5) & 1) != 0)) goto after_bfm; |
| 2154 | ULong wmask = 0, tmask = 0; |
| 2155 | Bool ok = dbm_DecodeBitMasks(&wmask, &tmask, |
| 2156 | N, immS, immR, False, sf == 1 ? 64 : 32); |
| 2157 | if (!ok) goto after_bfm; /* hmmm */ |
| 2158 | |
| 2159 | Bool is64 = sf == 1; |
| 2160 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2161 | |
| 2162 | IRTemp dst = newTemp(ty); |
| 2163 | IRTemp src = newTemp(ty); |
| 2164 | IRTemp bot = newTemp(ty); |
| 2165 | IRTemp top = newTemp(ty); |
| 2166 | IRTemp res = newTemp(ty); |
| 2167 | assign(dst, inZero ? mkU(ty,0) : getIRegOrZR(is64, dd)); |
| 2168 | assign(src, getIRegOrZR(is64, nn)); |
| 2169 | /* perform bitfield move on low bits */ |
| 2170 | assign(bot, binop(mkOR(ty), |
| 2171 | binop(mkAND(ty), mkexpr(dst), mkU(ty, ~wmask)), |
| 2172 | binop(mkAND(ty), mkexpr(mathROR(ty, src, immR)), |
| 2173 | mkU(ty, wmask)))); |
| 2174 | /* determine extension bits (sign, zero or dest register) */ |
| 2175 | assign(top, mkexpr(extend ? mathREPLICATE(ty, src, immS) : dst)); |
| 2176 | /* combine extension bits and result bits */ |
| 2177 | assign(res, binop(mkOR(ty), |
| 2178 | binop(mkAND(ty), mkexpr(top), mkU(ty, ~tmask)), |
| 2179 | binop(mkAND(ty), mkexpr(bot), mkU(ty, tmask)))); |
| 2180 | putIRegOrZR(is64, dd, mkexpr(res)); |
| 2181 | DIP("%s %s, %s, immR=%u, immS=%u\n", |
| 2182 | nm, nameIRegOrZR(is64, dd), nameIRegOrZR(is64, nn), immR, immS); |
| 2183 | return True; |
| 2184 | } |
| 2185 | after_bfm: |
| 2186 | |
| 2187 | /* ---------------------- EXTR ---------------------- */ |
| 2188 | /* 30 28 22 20 15 9 4 |
| 2189 | 1 00 100111 10 m imm6 n d EXTR Xd, Xn, Xm, #imm6 |
| 2190 | 0 00 100111 00 m imm6 n d EXTR Wd, Wn, Wm, #imm6 when #imm6 < 32 |
| 2191 | */ |
| 2192 | if (INSN(30,23) == BITS8(0,0,1,0,0,1,1,1) && INSN(21,21) == 0) { |
| 2193 | Bool is64 = INSN(31,31) == 1; |
| 2194 | UInt mm = INSN(20,16); |
| 2195 | UInt imm6 = INSN(15,10); |
| 2196 | UInt nn = INSN(9,5); |
| 2197 | UInt dd = INSN(4,0); |
| 2198 | Bool valid = True; |
| 2199 | if (INSN(31,31) != INSN(22,22)) |
| 2200 | valid = False; |
| 2201 | if (!is64 && imm6 >= 32) |
| 2202 | valid = False; |
| 2203 | if (!valid) goto after_extr; |
| 2204 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2205 | IRTemp srcHi = newTemp(ty); |
| 2206 | IRTemp srcLo = newTemp(ty); |
| 2207 | IRTemp res = newTemp(ty); |
| 2208 | assign(srcHi, getIRegOrZR(is64, nn)); |
| 2209 | assign(srcLo, getIRegOrZR(is64, mm)); |
| 2210 | if (imm6 == 0) { |
| 2211 | assign(res, mkexpr(srcLo)); |
| 2212 | } else { |
| 2213 | UInt szBits = 8 * sizeofIRType(ty); |
| 2214 | vassert(imm6 > 0 && imm6 < szBits); |
| 2215 | assign(res, binop(mkOR(ty), |
| 2216 | binop(mkSHL(ty), mkexpr(srcHi), mkU8(szBits-imm6)), |
| 2217 | binop(mkSHR(ty), mkexpr(srcLo), mkU8(imm6)))); |
| 2218 | } |
| 2219 | putIRegOrZR(is64, dd, mkexpr(res)); |
| 2220 | DIP("extr %s, %s, %s, #%u\n", |
| 2221 | nameIRegOrZR(is64,dd), |
| 2222 | nameIRegOrZR(is64,nn), nameIRegOrZR(is64,mm), imm6); |
| 2223 | return True; |
| 2224 | } |
| 2225 | after_extr: |
| 2226 | |
| 2227 | vex_printf("ARM64 front end: data_processing_immediate\n"); |
| 2228 | return False; |
| 2229 | # undef INSN |
| 2230 | } |
| 2231 | |
| 2232 | |
| 2233 | /*------------------------------------------------------------*/ |
| 2234 | /*--- Data processing (register) instructions ---*/ |
| 2235 | /*------------------------------------------------------------*/ |
| 2236 | |
| 2237 | static const HChar* nameSH ( UInt sh ) { |
| 2238 | switch (sh) { |
| 2239 | case 0: return "lsl"; |
| 2240 | case 1: return "lsr"; |
| 2241 | case 2: return "asr"; |
| 2242 | case 3: return "ror"; |
| 2243 | default: vassert(0); |
| 2244 | } |
| 2245 | } |
| 2246 | |
| 2247 | /* Generate IR to get a register value, possibly shifted by an |
| 2248 | immediate. Returns either a 32- or 64-bit temporary holding the |
| 2249 | result. After the shift, the value can optionally be NOT-ed |
| 2250 | too. |
| 2251 | |
| 2252 | sh_how coding: 00=SHL, 01=SHR, 10=SAR, 11=ROR. sh_amt may only be |
| 2253 | in the range 0 to (is64 ? 64 : 32)-1. For some instructions, ROR |
| 2254 | isn't allowed, but it's the job of the caller to check that. |
| 2255 | */ |
| 2256 | static IRTemp getShiftedIRegOrZR ( Bool is64, |
| 2257 | UInt sh_how, UInt sh_amt, UInt regNo, |
| 2258 | Bool invert ) |
| 2259 | { |
| 2260 | vassert(sh_how < 4); |
| 2261 | vassert(sh_amt < (is64 ? 64 : 32)); |
| 2262 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2263 | IRTemp t0 = newTemp(ty); |
| 2264 | assign(t0, getIRegOrZR(is64, regNo)); |
| 2265 | IRTemp t1 = newTemp(ty); |
| 2266 | switch (sh_how) { |
| 2267 | case BITS2(0,0): |
| 2268 | assign(t1, binop(mkSHL(ty), mkexpr(t0), mkU8(sh_amt))); |
| 2269 | break; |
| 2270 | case BITS2(0,1): |
| 2271 | assign(t1, binop(mkSHR(ty), mkexpr(t0), mkU8(sh_amt))); |
| 2272 | break; |
| 2273 | case BITS2(1,0): |
| 2274 | assign(t1, binop(mkSAR(ty), mkexpr(t0), mkU8(sh_amt))); |
| 2275 | break; |
| 2276 | case BITS2(1,1): |
| 2277 | assign(t1, mkexpr(mathROR(ty, t0, sh_amt))); |
| 2278 | break; |
| 2279 | default: |
| 2280 | vassert(0); |
| 2281 | } |
| 2282 | if (invert) { |
| 2283 | IRTemp t2 = newTemp(ty); |
| 2284 | assign(t2, unop(mkNOT(ty), mkexpr(t1))); |
| 2285 | return t2; |
| 2286 | } else { |
| 2287 | return t1; |
| 2288 | } |
| 2289 | } |
| 2290 | |
| 2291 | |
| 2292 | static |
| 2293 | Bool dis_ARM64_data_processing_register(/*MB_OUT*/DisResult* dres, |
| 2294 | UInt insn) |
| 2295 | { |
| 2296 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 2297 | |
| 2298 | /* ------------------- ADD/SUB(reg) ------------------- */ |
| 2299 | /* x==0 => 32 bit op x==1 => 64 bit op |
| 2300 | sh: 00=LSL, 01=LSR, 10=ASR, 11=ROR(NOT ALLOWED) |
| 2301 | |
| 2302 | 31 30 29 28 23 21 20 15 9 4 |
| 2303 | | | | | | | | | | | |
| 2304 | x 0 0 01011 sh 0 Rm imm6 Rn Rd ADD Rd,Rn, sh(Rm,imm6) |
| 2305 | x 0 1 01011 sh 0 Rm imm6 Rn Rd ADDS Rd,Rn, sh(Rm,imm6) |
| 2306 | x 1 0 01011 sh 0 Rm imm6 Rn Rd SUB Rd,Rn, sh(Rm,imm6) |
| 2307 | x 1 1 01011 sh 0 Rm imm6 Rn Rd SUBS Rd,Rn, sh(Rm,imm6) |
| 2308 | */ |
| 2309 | if (INSN(28,24) == BITS5(0,1,0,1,1) && INSN(21,21) == 0) { |
| 2310 | UInt bX = INSN(31,31); |
| 2311 | UInt bOP = INSN(30,30); /* 0: ADD, 1: SUB */ |
| 2312 | UInt bS = INSN(29, 29); /* set flags? */ |
| 2313 | UInt sh = INSN(23,22); |
| 2314 | UInt rM = INSN(20,16); |
| 2315 | UInt imm6 = INSN(15,10); |
| 2316 | UInt rN = INSN(9,5); |
| 2317 | UInt rD = INSN(4,0); |
| 2318 | Bool isSUB = bOP == 1; |
| 2319 | Bool is64 = bX == 1; |
| 2320 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2321 | if ((!is64 && imm6 > 31) || sh == BITS2(1,1)) { |
| 2322 | /* invalid; fall through */ |
| 2323 | } else { |
| 2324 | IRTemp argL = newTemp(ty); |
| 2325 | assign(argL, getIRegOrZR(is64, rN)); |
| 2326 | IRTemp argR = getShiftedIRegOrZR(is64, sh, imm6, rM, False); |
| 2327 | IROp op = isSUB ? mkSUB(ty) : mkADD(ty); |
| 2328 | IRTemp res = newTemp(ty); |
| 2329 | assign(res, binop(op, mkexpr(argL), mkexpr(argR))); |
| 2330 | if (rD != 31) putIRegOrZR(is64, rD, mkexpr(res)); |
| 2331 | if (bS) { |
| 2332 | setFlags_ADD_SUB(is64, isSUB, argL, argR); |
| 2333 | } |
| 2334 | DIP("%s%s %s, %s, %s, %s #%u\n", |
| 2335 | bOP ? "sub" : "add", bS ? "s" : "", |
| 2336 | nameIRegOrZR(is64, rD), nameIRegOrZR(is64, rN), |
| 2337 | nameIRegOrZR(is64, rM), nameSH(sh), imm6); |
| 2338 | return True; |
| 2339 | } |
| 2340 | } |
| 2341 | |
| 2342 | /* -------------------- LOGIC(reg) -------------------- */ |
| 2343 | /* x==0 => 32 bit op x==1 => 64 bit op |
| 2344 | N==0 => inv? is no-op (no inversion) |
| 2345 | N==1 => inv? is NOT |
| 2346 | sh: 00=LSL, 01=LSR, 10=ASR, 11=ROR |
| 2347 | |
| 2348 | 31 30 28 23 21 20 15 9 4 |
| 2349 | | | | | | | | | | |
| 2350 | x 00 01010 sh N Rm imm6 Rn Rd AND Rd,Rn, inv?(sh(Rm,imm6)) |
| 2351 | x 01 01010 sh N Rm imm6 Rn Rd ORR Rd,Rn, inv?(sh(Rm,imm6)) |
| 2352 | x 10 01010 sh N Rm imm6 Rn Rd EOR Rd,Rn, inv?(sh(Rm,imm6)) |
| 2353 | x 11 01010 sh N Rm imm6 Rn Rd ANDS Rd,Rn, inv?(sh(Rm,imm6)) |
| 2354 | With N=1, the names are: BIC ORN EON BICS |
| 2355 | */ |
| 2356 | if (INSN(28,24) == BITS5(0,1,0,1,0)) { |
| 2357 | UInt bX = INSN(31,31); |
| 2358 | UInt sh = INSN(23,22); |
| 2359 | UInt bN = INSN(21,21); |
| 2360 | UInt rM = INSN(20,16); |
| 2361 | UInt imm6 = INSN(15,10); |
| 2362 | UInt rN = INSN(9,5); |
| 2363 | UInt rD = INSN(4,0); |
| 2364 | Bool is64 = bX == 1; |
| 2365 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2366 | if (!is64 && imm6 > 31) { |
| 2367 | /* invalid; fall though */ |
| 2368 | } else { |
| 2369 | IRTemp argL = newTemp(ty); |
| 2370 | assign(argL, getIRegOrZR(is64, rN)); |
| 2371 | IRTemp argR = getShiftedIRegOrZR(is64, sh, imm6, rM, bN == 1); |
| 2372 | IROp op = Iop_INVALID; |
| 2373 | switch (INSN(30,29)) { |
| 2374 | case BITS2(0,0): case BITS2(1,1): op = mkAND(ty); break; |
| 2375 | case BITS2(0,1): op = mkOR(ty); break; |
| 2376 | case BITS2(1,0): op = mkXOR(ty); break; |
| 2377 | default: vassert(0); |
| 2378 | } |
| 2379 | IRTemp res = newTemp(ty); |
| 2380 | assign(res, binop(op, mkexpr(argL), mkexpr(argR))); |
| 2381 | if (INSN(30,29) == BITS2(1,1)) { |
| 2382 | setFlags_LOGIC(is64, res); |
| 2383 | } |
| 2384 | putIRegOrZR(is64, rD, mkexpr(res)); |
| 2385 | |
| 2386 | static const HChar* names_op[8] |
| 2387 | = { "and", "orr", "eor", "ands", "bic", "orn", "eon", "bics" }; |
| 2388 | vassert(((bN << 2) | INSN(30,29)) < 8); |
| 2389 | const HChar* nm_op = names_op[(bN << 2) | INSN(30,29)]; |
| 2390 | /* Special-case the printing of "MOV" */ |
| 2391 | if (rN == 31/*zr*/ && sh == 0/*LSL*/ && imm6 == 0 && bN == 0) { |
| 2392 | DIP("mov %s, %s\n", nameIRegOrZR(is64, rD), |
| 2393 | nameIRegOrZR(is64, rM)); |
| 2394 | } else { |
| 2395 | DIP("%s %s, %s, %s, %s #%u\n", nm_op, |
| 2396 | nameIRegOrZR(is64, rD), nameIRegOrZR(is64, rN), |
| 2397 | nameIRegOrZR(is64, rM), nameSH(sh), imm6); |
| 2398 | } |
| 2399 | return True; |
| 2400 | } |
| 2401 | } |
| 2402 | |
| 2403 | /* -------------------- {U,S}MULH -------------------- */ |
| 2404 | /* 31 23 22 20 15 9 4 |
| 2405 | 10011011 1 10 Rm 011111 Rn Rd UMULH Xd,Xn,Xm |
| 2406 | 10011011 0 10 Rm 011111 Rn Rd SMULH Xd,Xn,Xm |
| 2407 | */ |
| 2408 | if (INSN(31,24) == BITS8(1,0,0,1,1,0,1,1) |
| 2409 | && INSN(22,21) == BITS2(1,0) && INSN(15,10) == BITS6(0,1,1,1,1,1) |
| 2410 | && INSN(23,23) == 1/*ATC*/) { |
| 2411 | Bool isU = INSN(23,23) == 1; |
| 2412 | UInt mm = INSN(20,16); |
| 2413 | UInt nn = INSN(9,5); |
| 2414 | UInt dd = INSN(4,0); |
| 2415 | putIReg64orZR(dd, unop(Iop_128HIto64, |
| 2416 | binop(isU ? Iop_MullU64 : Iop_MullS64, |
| 2417 | getIReg64orZR(nn), getIReg64orZR(mm)))); |
| 2418 | DIP("%cmulh %s, %s, %s\n", |
| 2419 | isU ? 'u' : 's', |
| 2420 | nameIReg64orZR(dd), nameIReg64orZR(nn), nameIReg64orZR(mm)); |
| 2421 | return True; |
| 2422 | } |
| 2423 | |
| 2424 | /* -------------------- M{ADD,SUB} -------------------- */ |
| 2425 | /* 31 30 20 15 14 9 4 |
| 2426 | sf 00 11011 000 m 0 a n r MADD Rd,Rn,Rm,Ra d = a+m*n |
| 2427 | sf 00 11011 000 m 1 a n r MADD Rd,Rn,Rm,Ra d = a-m*n |
| 2428 | */ |
| 2429 | if (INSN(30,21) == BITS10(0,0,1,1,0,1,1,0,0,0)) { |
| 2430 | Bool is64 = INSN(31,31) == 1; |
| 2431 | UInt mm = INSN(20,16); |
| 2432 | Bool isAdd = INSN(15,15) == 0; |
| 2433 | UInt aa = INSN(14,10); |
| 2434 | UInt nn = INSN(9,5); |
| 2435 | UInt dd = INSN(4,0); |
| 2436 | if (is64) { |
| 2437 | putIReg64orZR( |
| 2438 | dd, |
| 2439 | binop(isAdd ? Iop_Add64 : Iop_Sub64, |
| 2440 | getIReg64orZR(aa), |
| 2441 | binop(Iop_Mul64, getIReg64orZR(mm), getIReg64orZR(nn)))); |
| 2442 | } else { |
| 2443 | putIReg32orZR( |
| 2444 | dd, |
| 2445 | binop(isAdd ? Iop_Add32 : Iop_Sub32, |
| 2446 | getIReg32orZR(aa), |
| 2447 | binop(Iop_Mul32, getIReg32orZR(mm), getIReg32orZR(nn)))); |
| 2448 | } |
| 2449 | DIP("%s %s, %s, %s, %s\n", |
| 2450 | isAdd ? "madd" : "msub", |
| 2451 | nameIRegOrZR(is64, dd), nameIRegOrZR(is64, nn), |
| 2452 | nameIRegOrZR(is64, mm), nameIRegOrZR(is64, aa)); |
| 2453 | return True; |
| 2454 | } |
| 2455 | |
| 2456 | /* ---------------- CS{EL,INC,INV,NEG} ---------------- */ |
| 2457 | /* 31 30 28 20 15 11 9 4 |
| 2458 | sf 00 1101 0100 mm cond 00 nn dd CSEL Rd,Rn,Rm |
| 2459 | sf 00 1101 0100 mm cond 01 nn dd CSINC Rd,Rn,Rm |
| 2460 | sf 10 1101 0100 mm cond 00 nn dd CSINV Rd,Rn,Rm |
| 2461 | sf 10 1101 0100 mm cond 01 nn dd CSNEG Rd,Rn,Rm |
| 2462 | In all cases, the operation is: Rd = if cond then Rn else OP(Rm) |
| 2463 | */ |
| 2464 | if (INSN(29,21) == BITS9(0, 1,1,0,1, 0,1,0,0) && INSN(11,11) == 0) { |
| 2465 | Bool is64 = INSN(31,31) == 1; |
| 2466 | UInt b30 = INSN(30,30); |
| 2467 | UInt mm = INSN(20,16); |
| 2468 | UInt cond = INSN(15,12); |
| 2469 | UInt b10 = INSN(10,10); |
| 2470 | UInt nn = INSN(9,5); |
| 2471 | UInt dd = INSN(4,0); |
| 2472 | UInt op = (b30 << 1) | b10; /* 00=id 01=inc 10=inv 11=neg */ |
| 2473 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2474 | IRExpr* argL = getIRegOrZR(is64, nn); |
| 2475 | IRExpr* argR = getIRegOrZR(is64, mm); |
| 2476 | switch (op) { |
| 2477 | case BITS2(0,0): |
| 2478 | break; |
| 2479 | case BITS2(0,1): |
| 2480 | argR = binop(mkADD(ty), argR, mkU(ty,1)); |
| 2481 | break; |
| 2482 | case BITS2(1,0): |
| 2483 | argR = unop(mkNOT(ty), argR); |
| 2484 | break; |
| 2485 | case BITS2(1,1): |
| 2486 | argR = binop(mkSUB(ty), mkU(ty,0), argR); |
| 2487 | break; |
| 2488 | default: |
| 2489 | vassert(0); |
| 2490 | } |
| 2491 | putIRegOrZR( |
| 2492 | is64, dd, |
| 2493 | IRExpr_ITE(unop(Iop_64to1, mk_arm64g_calculate_condition(cond)), |
| 2494 | argL, argR) |
| 2495 | ); |
| 2496 | const HChar* op_nm[4] = { "csel", "csinc", "csinv", "csneg" }; |
| 2497 | DIP("%s %s, %s, %s, %s\n", op_nm[op], |
| 2498 | nameIRegOrZR(is64, dd), nameIRegOrZR(is64, nn), |
| 2499 | nameIRegOrZR(is64, mm), nameCC(cond)); |
| 2500 | return True; |
| 2501 | } |
| 2502 | |
| 2503 | /* -------------- ADD/SUB(extended reg) -------------- */ |
| 2504 | /* 28 20 15 12 9 4 |
| 2505 | 000 01011 00 1 m opt imm3 n d ADD Wd|SP, Wn|SP, Wm ext&lsld |
| 2506 | 100 01011 00 1 m opt imm3 n d ADD Xd|SP, Xn|SP, Rm ext&lsld |
| 2507 | |
| 2508 | 001 01011 00 1 m opt imm3 n d ADDS Wd, Wn|SP, Wm ext&lsld |
| 2509 | 101 01011 00 1 m opt imm3 n d ADDS Xd, Xn|SP, Rm ext&lsld |
| 2510 | |
| 2511 | 010 01011 00 1 m opt imm3 n d SUB Wd|SP, Wn|SP, Wm ext&lsld |
| 2512 | 110 01011 00 1 m opt imm3 n d SUB Xd|SP, Xn|SP, Rm ext&lsld |
| 2513 | |
| 2514 | 011 01011 00 1 m opt imm3 n d SUBS Wd, Wn|SP, Wm ext&lsld |
| 2515 | 111 01011 00 1 m opt imm3 n d SUBS Xd, Xn|SP, Rm ext&lsld |
| 2516 | |
| 2517 | The 'm' operand is extended per opt, thusly: |
| 2518 | |
| 2519 | 000 Xm & 0xFF UXTB |
| 2520 | 001 Xm & 0xFFFF UXTH |
| 2521 | 010 Xm & (2^32)-1 UXTW |
| 2522 | 011 Xm UXTX |
| 2523 | |
| 2524 | 100 Xm sx from bit 7 SXTB |
| 2525 | 101 Xm sx from bit 15 SXTH |
| 2526 | 110 Xm sx from bit 31 SXTW |
| 2527 | 111 Xm SXTX |
| 2528 | |
| 2529 | In the 64 bit case (bit31 == 1), UXTX and SXTX are the identity |
| 2530 | operation on Xm. In the 32 bit case, UXTW, UXTX, SXTW and SXTX |
| 2531 | are the identity operation on Wm. |
| 2532 | |
| 2533 | After extension, the value is shifted left by imm3 bits, which |
| 2534 | may only be in the range 0 .. 4 inclusive. |
| 2535 | */ |
| 2536 | if (INSN(28,21) == BITS8(0,1,0,1,1,0,0,1) && INSN(12,10) <= 4) { |
| 2537 | Bool is64 = INSN(31,31) == 1; |
| 2538 | Bool isSub = INSN(30,30) == 1; |
| 2539 | Bool setCC = INSN(29,29) == 1; |
| 2540 | UInt mm = INSN(20,16); |
| 2541 | UInt opt = INSN(15,13); |
| 2542 | UInt imm3 = INSN(12,10); |
| 2543 | UInt nn = INSN(9,5); |
| 2544 | UInt dd = INSN(4,0); |
| 2545 | const HChar* nameExt[8] = { "uxtb", "uxth", "uxtw", "uxtx", |
| 2546 | "sxtb", "sxth", "sxtw", "sxtx" }; |
| 2547 | /* Do almost the same thing in the 32- and 64-bit cases. */ |
| 2548 | IRTemp xN = newTemp(Ity_I64); |
| 2549 | IRTemp xM = newTemp(Ity_I64); |
| 2550 | assign(xN, getIReg64orSP(nn)); |
| 2551 | assign(xM, getIReg64orZR(mm)); |
| 2552 | IRExpr* xMw = mkexpr(xM); /* "xM widened" */ |
| 2553 | Int shSX = 0; |
| 2554 | /* widen Xm .. */ |
| 2555 | switch (opt) { |
| 2556 | case BITS3(0,0,0): // UXTB |
| 2557 | xMw = binop(Iop_And64, xMw, mkU64(0xFF)); break; |
| 2558 | case BITS3(0,0,1): // UXTH |
| 2559 | xMw = binop(Iop_And64, xMw, mkU64(0xFFFF)); break; |
| 2560 | case BITS3(0,1,0): // UXTW -- noop for the 32bit case |
| 2561 | if (is64) { |
| 2562 | xMw = unop(Iop_32Uto64, unop(Iop_64to32, xMw)); |
| 2563 | } |
| 2564 | break; |
| 2565 | case BITS3(0,1,1): // UXTX -- always a noop |
| 2566 | break; |
| 2567 | case BITS3(1,0,0): // SXTB |
| 2568 | shSX = 56; goto sxTo64; |
| 2569 | case BITS3(1,0,1): // SXTH |
| 2570 | shSX = 48; goto sxTo64; |
| 2571 | case BITS3(1,1,0): // SXTW -- noop for the 32bit case |
| 2572 | if (is64) { |
| 2573 | shSX = 32; goto sxTo64; |
| 2574 | } |
| 2575 | break; |
| 2576 | case BITS3(1,1,1): // SXTX -- always a noop |
| 2577 | break; |
| 2578 | sxTo64: |
| 2579 | vassert(shSX >= 32); |
| 2580 | xMw = binop(Iop_Sar64, binop(Iop_Shl64, xMw, mkU8(shSX)), |
| 2581 | mkU8(shSX)); |
| 2582 | break; |
| 2583 | default: |
| 2584 | vassert(0); |
| 2585 | } |
| 2586 | /* and now shift */ |
| 2587 | IRTemp argL = xN; |
| 2588 | IRTemp argR = newTemp(Ity_I64); |
| 2589 | assign(argR, binop(Iop_Shl64, xMw, mkU8(imm3))); |
| 2590 | IRTemp res = newTemp(Ity_I64); |
| 2591 | assign(res, binop(isSub ? Iop_Sub64 : Iop_Add64, |
| 2592 | mkexpr(argL), mkexpr(argR))); |
| 2593 | if (is64) { |
| 2594 | if (setCC) { |
| 2595 | putIReg64orZR(dd, mkexpr(res)); |
| 2596 | setFlags_ADD_SUB(True/*is64*/, isSub, argL, argR); |
| 2597 | } else { |
| 2598 | putIReg64orSP(dd, mkexpr(res)); |
| 2599 | } |
| 2600 | } else { |
| 2601 | if (setCC) { |
| 2602 | IRTemp argL32 = newTemp(Ity_I32); |
| 2603 | IRTemp argR32 = newTemp(Ity_I32); |
| 2604 | putIReg32orZR(dd, unop(Iop_64to32, mkexpr(res))); |
| 2605 | assign(argL32, unop(Iop_64to32, mkexpr(argL))); |
| 2606 | assign(argR32, unop(Iop_64to32, mkexpr(argR))); |
| 2607 | setFlags_ADD_SUB(False/*!is64*/, isSub, argL32, argR32); |
| 2608 | } else { |
| 2609 | putIReg32orSP(dd, unop(Iop_64to32, mkexpr(res))); |
| 2610 | } |
| 2611 | } |
| 2612 | DIP("%s%s %s, %s, %s %s lsl %u\n", |
| 2613 | isSub ? "sub" : "add", setCC ? "s" : "", |
| 2614 | setCC ? nameIRegOrZR(is64, dd) : nameIRegOrSP(is64, dd), |
| 2615 | nameIRegOrSP(is64, nn), nameIRegOrSP(is64, mm), |
| 2616 | nameExt[opt], imm3); |
| 2617 | return True; |
| 2618 | } |
| 2619 | |
| 2620 | /* ---------------- CCMP/CCMN(imm) ---------------- */ |
| 2621 | /* Bizarrely, these appear in the "data processing register" |
| 2622 | category, even though they are operations against an |
| 2623 | immediate. */ |
| 2624 | /* 31 29 20 15 11 9 3 |
| 2625 | sf 1 111010010 imm5 cond 10 Rn 0 nzcv CCMP Rn, #imm5, #nzcv, cond |
| 2626 | sf 0 111010010 imm5 cond 10 Rn 0 nzcv CCMN Rn, #imm5, #nzcv, cond |
| 2627 | |
| 2628 | Operation is: |
| 2629 | (CCMP) flags = if cond then flags-after-sub(Rn,imm5) else nzcv |
| 2630 | (CCMN) flags = if cond then flags-after-add(Rn,imm5) else nzcv |
| 2631 | */ |
| 2632 | if (INSN(29,21) == BITS9(1,1,1,0,1,0,0,1,0) |
| 2633 | && INSN(11,10) == BITS2(1,0) && INSN(4,4) == 0) { |
| 2634 | Bool is64 = INSN(31,31) == 1; |
| 2635 | Bool isSUB = INSN(30,30) == 1; |
| 2636 | UInt imm5 = INSN(20,16); |
| 2637 | UInt cond = INSN(15,12); |
| 2638 | UInt nn = INSN(9,5); |
| 2639 | UInt nzcv = INSN(3,0); |
| 2640 | |
| 2641 | IRTemp condT = newTemp(Ity_I1); |
| 2642 | assign(condT, unop(Iop_64to1, mk_arm64g_calculate_condition(cond))); |
| 2643 | |
| 2644 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2645 | IRTemp argL = newTemp(ty); |
| 2646 | IRTemp argR = newTemp(ty); |
| 2647 | |
| 2648 | if (is64) { |
| 2649 | assign(argL, getIReg64orZR(nn)); |
| 2650 | assign(argR, mkU64(imm5)); |
| 2651 | } else { |
| 2652 | assign(argL, getIReg32orZR(nn)); |
| 2653 | assign(argR, mkU32(imm5)); |
| 2654 | } |
| 2655 | setFlags_ADD_SUB_conditionally(is64, isSUB, condT, argL, argR, nzcv); |
| 2656 | |
| 2657 | DIP("ccm%c %s, #%u, #%u, %s\n", |
| 2658 | isSUB ? 'p' : 'n', nameIRegOrZR(is64, nn), |
| 2659 | imm5, nzcv, nameCC(cond)); |
| 2660 | return True; |
| 2661 | } |
| 2662 | |
| 2663 | /* ---------------- CCMP/CCMN(reg) ---------------- */ |
| 2664 | /* 31 29 20 15 11 9 3 |
| 2665 | sf 1 111010010 Rm cond 00 Rn 0 nzcv CCMP Rn, Rm, #nzcv, cond |
| 2666 | sf 0 111010010 Rm cond 00 Rn 0 nzcv CCMN Rn, Rm, #nzcv, cond |
| 2667 | Operation is: |
| 2668 | (CCMP) flags = if cond then flags-after-sub(Rn,Rm) else nzcv |
| 2669 | (CCMN) flags = if cond then flags-after-add(Rn,Rm) else nzcv |
| 2670 | */ |
| 2671 | if (INSN(29,21) == BITS9(1,1,1,0,1,0,0,1,0) |
| 2672 | && INSN(11,10) == BITS2(0,0) && INSN(4,4) == 0) { |
| 2673 | Bool is64 = INSN(31,31) == 1; |
| 2674 | Bool isSUB = INSN(30,30) == 1; |
| 2675 | UInt mm = INSN(20,16); |
| 2676 | UInt cond = INSN(15,12); |
| 2677 | UInt nn = INSN(9,5); |
| 2678 | UInt nzcv = INSN(3,0); |
| 2679 | |
| 2680 | IRTemp condT = newTemp(Ity_I1); |
| 2681 | assign(condT, unop(Iop_64to1, mk_arm64g_calculate_condition(cond))); |
| 2682 | |
| 2683 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2684 | IRTemp argL = newTemp(ty); |
| 2685 | IRTemp argR = newTemp(ty); |
| 2686 | |
| 2687 | if (is64) { |
| 2688 | assign(argL, getIReg64orZR(nn)); |
| 2689 | assign(argR, getIReg64orZR(mm)); |
| 2690 | } else { |
| 2691 | assign(argL, getIReg32orZR(nn)); |
| 2692 | assign(argR, getIReg32orZR(mm)); |
| 2693 | } |
| 2694 | setFlags_ADD_SUB_conditionally(is64, isSUB, condT, argL, argR, nzcv); |
| 2695 | |
| 2696 | DIP("ccm%c %s, %s, #%u, %s\n", |
| 2697 | isSUB ? 'p' : 'n', nameIRegOrZR(is64, nn), |
| 2698 | nameIRegOrZR(is64, mm), nzcv, nameCC(cond)); |
| 2699 | return True; |
| 2700 | } |
| 2701 | |
| 2702 | |
| 2703 | /* -------------- REV/REV16/REV32/RBIT -------------- */ |
| 2704 | /* 31 30 28 20 15 11 9 4 |
| 2705 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 2706 | 1 10 11010110 00000 0000 11 n d (1) REV Xd, Xn |
| 2707 | 0 10 11010110 00000 0000 10 n d (2) REV Wd, Wn |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 2708 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 2709 | 1 10 11010110 00000 0000 00 n d (3) RBIT Xd, Xn |
| 2710 | 0 10 11010110 00000 0000 00 n d (4) RBIT Wd, Wn |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 2711 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 2712 | 1 10 11010110 00000 0000 01 n d (5) REV16 Xd, Xn |
| 2713 | 0 10 11010110 00000 0000 01 n d (6) REV16 Wd, Wn |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 2714 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 2715 | 1 10 11010110 00000 0000 10 n d (7) REV32 Xd, Xn |
| 2716 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 2717 | */ |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 2718 | /* Only REV and RBIT are currently implemented. */ |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 2719 | 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^] | 2720 | && INSN(20,12) == BITS9(0,0,0,0,0,0,0,0,0)) { |
| 2721 | UInt b31 = INSN(31,31); |
| 2722 | UInt opc = INSN(11,10); |
| 2723 | |
| 2724 | UInt ix = 0; |
| 2725 | /**/ if (b31 == 1 && opc == BITS2(1,1)) ix = 1; |
| 2726 | else if (b31 == 0 && opc == BITS2(1,0)) ix = 2; |
| 2727 | else if (b31 == 1 && opc == BITS2(0,0)) ix = 3; |
| 2728 | else if (b31 == 0 && opc == BITS2(0,0)) ix = 4; |
| 2729 | else if (b31 == 1 && opc == BITS2(0,1)) ix = 5; |
| 2730 | else if (b31 == 0 && opc == BITS2(0,1)) ix = 6; |
| 2731 | else if (b31 == 1 && opc == BITS2(1,0)) ix = 7; |
| 2732 | if (ix >= 1 && ix <= 4) { |
| 2733 | Bool is64 = ix == 1 || ix == 3; |
| 2734 | Bool isBIT = ix == 3 || ix == 4; |
| 2735 | UInt nn = INSN(9,5); |
| 2736 | UInt dd = INSN(4,0); |
| 2737 | IRTemp src = newTemp(Ity_I64); |
| 2738 | IRTemp dst = IRTemp_INVALID; |
| 2739 | if (is64) { |
| 2740 | assign(src, getIReg64orZR(nn)); |
| 2741 | dst = isBIT ? math_BITSWAP64(src) : math_BYTESWAP64(src); |
| 2742 | putIReg64orZR(dd, mkexpr(dst)); |
| 2743 | } else { |
| 2744 | assign(src, binop(Iop_Shl64, getIReg64orZR(nn), mkU8(32))); |
| 2745 | dst = isBIT ? math_BITSWAP64(src) : math_BYTESWAP64(src); |
| 2746 | putIReg32orZR(dd, unop(Iop_64to32, mkexpr(dst))); |
| 2747 | } |
| 2748 | DIP("%s %s, %s\n", isBIT ? "rbit" : "rev", |
| 2749 | nameIRegOrZR(is64,dd), nameIRegOrZR(is64,nn)); |
| 2750 | return True; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 2751 | } |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 2752 | /* else fall through */ |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 2753 | } |
| 2754 | |
| 2755 | /* -------------------- CLZ/CLS -------------------- */ |
| 2756 | /* 30 28 24 20 15 9 4 |
| 2757 | sf 10 1101 0110 00000 00010 0 n d CLZ Rd, Rn |
| 2758 | sf 10 1101 0110 00000 00010 1 n d CLS Rd, Rn |
| 2759 | */ |
| 2760 | if (INSN(30,21) == BITS10(1,0,1,1,0,1,0,1,1,0) |
| 2761 | && INSN(20,11) == BITS10(0,0,0,0,0,0,0,0,1,0)) { |
| 2762 | Bool is64 = INSN(31,31) == 1; |
| 2763 | Bool isCLS = INSN(10,10) == 1; |
| 2764 | UInt nn = INSN(9,5); |
| 2765 | UInt dd = INSN(4,0); |
| 2766 | IRTemp src = newTemp(Ity_I64); |
| 2767 | IRTemp dst = newTemp(Ity_I64); |
| 2768 | if (!isCLS) { // CLS not yet supported |
| 2769 | if (is64) { |
| 2770 | assign(src, getIReg64orZR(nn)); |
| 2771 | assign(dst, IRExpr_ITE(binop(Iop_CmpEQ64, mkexpr(src), mkU64(0)), |
| 2772 | mkU64(64), |
| 2773 | unop(Iop_Clz64, mkexpr(src)))); |
| 2774 | putIReg64orZR(dd, mkexpr(dst)); |
| 2775 | } else { |
| 2776 | assign(src, binop(Iop_Shl64, |
| 2777 | unop(Iop_32Uto64, getIReg32orZR(nn)), mkU8(32))); |
| 2778 | assign(dst, IRExpr_ITE(binop(Iop_CmpEQ64, mkexpr(src), mkU64(0)), |
| 2779 | mkU64(32), |
| 2780 | unop(Iop_Clz64, mkexpr(src)))); |
| 2781 | putIReg32orZR(dd, unop(Iop_64to32, mkexpr(dst))); |
| 2782 | } |
| 2783 | DIP("cl%c %s, %s\n", |
| 2784 | isCLS ? 's' : 'z', nameIRegOrZR(is64, dd), nameIRegOrZR(is64, nn)); |
| 2785 | return True; |
| 2786 | } |
| 2787 | } |
| 2788 | |
| 2789 | /* -------------------- LSLV/LSRV/ASRV -------------------- */ |
| 2790 | /* 30 28 20 15 11 9 4 |
| 2791 | sf 00 1101 0110 m 0010 00 n d LSLV Rd,Rn,Rm |
| 2792 | sf 00 1101 0110 m 0010 01 n d LSRV Rd,Rn,Rm |
| 2793 | sf 00 1101 0110 m 0010 10 n d ASRV Rd,Rn,Rm |
| 2794 | */ |
| 2795 | if (INSN(30,21) == BITS10(0,0,1,1,0,1,0,1,1,0) |
| 2796 | && INSN(15,12) == BITS4(0,0,1,0) && INSN(11,10) < BITS2(1,1)) { |
| 2797 | Bool is64 = INSN(31,31) == 1; |
| 2798 | UInt mm = INSN(20,16); |
| 2799 | UInt op = INSN(11,10); |
| 2800 | UInt nn = INSN(9,5); |
| 2801 | UInt dd = INSN(4,0); |
| 2802 | IRType ty = is64 ? Ity_I64 : Ity_I32; |
| 2803 | IRTemp srcL = newTemp(ty); |
| 2804 | IRTemp srcR = newTemp(Ity_I8); |
| 2805 | IRTemp res = newTemp(ty); |
| 2806 | IROp iop = Iop_INVALID; |
| 2807 | assign(srcL, getIRegOrZR(is64, nn)); |
| 2808 | assign(srcR, |
| 2809 | unop(Iop_64to8, |
| 2810 | binop(Iop_And64, |
| 2811 | getIReg64orZR(mm), mkU64(is64 ? 63 : 31)))); |
| 2812 | switch (op) { |
| 2813 | case BITS2(0,0): iop = mkSHL(ty); break; |
| 2814 | case BITS2(0,1): iop = mkSHR(ty); break; |
| 2815 | case BITS2(1,0): iop = mkSAR(ty); break; |
| 2816 | default: vassert(0); |
| 2817 | } |
| 2818 | assign(res, binop(iop, mkexpr(srcL), mkexpr(srcR))); |
| 2819 | putIRegOrZR(is64, dd, mkexpr(res)); |
| 2820 | vassert(op < 3); |
| 2821 | const HChar* names[3] = { "lslv", "lsrv", "asrv" }; |
| 2822 | DIP("%s %s, %s, %s\n", |
| 2823 | names[op], nameIRegOrZR(is64,dd), |
| 2824 | nameIRegOrZR(is64,nn), nameIRegOrZR(is64,mm)); |
| 2825 | return True; |
| 2826 | } |
| 2827 | |
| 2828 | /* -------------------- SDIV/UDIV -------------------- */ |
| 2829 | /* 30 28 20 15 10 9 4 |
| 2830 | sf 00 1101 0110 m 00001 1 n d SDIV Rd,Rn,Rm |
| 2831 | sf 00 1101 0110 m 00001 0 n d UDIV Rd,Rn,Rm |
| 2832 | */ |
| 2833 | if (INSN(30,21) == BITS10(0,0,1,1,0,1,0,1,1,0) |
| 2834 | && INSN(15,11) == BITS5(0,0,0,0,1)) { |
| 2835 | Bool is64 = INSN(31,31) == 1; |
| 2836 | UInt mm = INSN(20,16); |
| 2837 | Bool isS = INSN(10,10) == 1; |
| 2838 | UInt nn = INSN(9,5); |
| 2839 | UInt dd = INSN(4,0); |
| 2840 | if (isS) { |
| 2841 | putIRegOrZR(is64, dd, binop(is64 ? Iop_DivS64 : Iop_DivS32, |
| 2842 | getIRegOrZR(is64, nn), |
| 2843 | getIRegOrZR(is64, mm))); |
| 2844 | } else { |
| 2845 | putIRegOrZR(is64, dd, binop(is64 ? Iop_DivU64 : Iop_DivU32, |
| 2846 | getIRegOrZR(is64, nn), |
| 2847 | getIRegOrZR(is64, mm))); |
| 2848 | } |
| 2849 | DIP("%cdiv %s, %s, %s\n", isS ? 's' : 'u', |
| 2850 | nameIRegOrZR(is64, dd), |
| 2851 | nameIRegOrZR(is64, nn), nameIRegOrZR(is64, mm)); |
| 2852 | return True; |
| 2853 | } |
| 2854 | |
| 2855 | /* ------------------ {S,U}M{ADD,SUB}L ------------------ */ |
| 2856 | /* 31 23 20 15 14 9 4 |
| 2857 | 1001 1011 101 m 0 a n d UMADDL Xd,Wn,Wm,Xa |
| 2858 | 1001 1011 001 m 0 a n d SMADDL Xd,Wn,Wm,Xa |
| 2859 | 1001 1011 101 m 1 a n d UMSUBL Xd,Wn,Wm,Xa |
| 2860 | 1001 1011 001 m 1 a n d SMSUBL Xd,Wn,Wm,Xa |
| 2861 | with operation |
| 2862 | Xd = Xa +/- (Wn *u/s Wm) |
| 2863 | */ |
| 2864 | if (INSN(31,24) == BITS8(1,0,0,1,1,0,1,1) && INSN(22,21) == BITS2(0,1)) { |
| 2865 | Bool isU = INSN(23,23) == 1; |
| 2866 | UInt mm = INSN(20,16); |
| 2867 | Bool isAdd = INSN(15,15) == 0; |
| 2868 | UInt aa = INSN(14,10); |
| 2869 | UInt nn = INSN(9,5); |
| 2870 | UInt dd = INSN(4,0); |
| 2871 | IRTemp wN = newTemp(Ity_I32); |
| 2872 | IRTemp wM = newTemp(Ity_I32); |
| 2873 | IRTemp xA = newTemp(Ity_I64); |
| 2874 | IRTemp muld = newTemp(Ity_I64); |
| 2875 | IRTemp res = newTemp(Ity_I64); |
| 2876 | assign(wN, getIReg32orZR(nn)); |
| 2877 | assign(wM, getIReg32orZR(mm)); |
| 2878 | assign(xA, getIReg64orZR(aa)); |
| 2879 | assign(muld, binop(isU ? Iop_MullU32 : Iop_MullS32, |
| 2880 | mkexpr(wN), mkexpr(wM))); |
| 2881 | assign(res, binop(isAdd ? Iop_Add64 : Iop_Sub64, |
| 2882 | mkexpr(xA), mkexpr(muld))); |
| 2883 | putIReg64orZR(dd, mkexpr(res)); |
| 2884 | DIP("%cm%sl %s, %s, %s, %s\n", isU ? 'u' : 's', isAdd ? "add" : "sub", |
| 2885 | nameIReg64orZR(dd), nameIReg32orZR(nn), |
| 2886 | nameIReg32orZR(mm), nameIReg64orZR(aa)); |
| 2887 | return True; |
| 2888 | } |
| 2889 | vex_printf("ARM64 front end: data_processing_register\n"); |
| 2890 | return False; |
| 2891 | # undef INSN |
| 2892 | } |
| 2893 | |
| 2894 | |
| 2895 | /*------------------------------------------------------------*/ |
| 2896 | /*--- Load and Store instructions ---*/ |
| 2897 | /*------------------------------------------------------------*/ |
| 2898 | |
| 2899 | /* Generate the EA for a "reg + reg" style amode. This is done from |
| 2900 | parts of the insn, but for sanity checking sake it takes the whole |
| 2901 | insn. This appears to depend on insn[15:12], with opt=insn[15:13] |
| 2902 | and S=insn[12]: |
| 2903 | |
| 2904 | The possible forms, along with their opt:S values, are: |
| 2905 | 011:0 Xn|SP + Xm |
| 2906 | 111:0 Xn|SP + Xm |
| 2907 | 011:1 Xn|SP + Xm * transfer_szB |
| 2908 | 111:1 Xn|SP + Xm * transfer_szB |
| 2909 | 010:0 Xn|SP + 32Uto64(Wm) |
| 2910 | 010:1 Xn|SP + 32Uto64(Wm) * transfer_szB |
| 2911 | 110:0 Xn|SP + 32Sto64(Wm) |
| 2912 | 110:1 Xn|SP + 32Sto64(Wm) * transfer_szB |
| 2913 | |
| 2914 | Rm is insn[20:16]. Rn is insn[9:5]. Rt is insn[4:0]. Log2 of |
| 2915 | the transfer size is insn[23,31,30]. For integer loads/stores, |
| 2916 | insn[23] is zero, hence szLg2 can be at most 3 in such cases. |
| 2917 | |
| 2918 | If the decoding fails, it returns IRTemp_INVALID. |
| 2919 | |
| 2920 | isInt is True iff this is decoding is for transfers to/from integer |
| 2921 | registers. If False it is for transfers to/from vector registers. |
| 2922 | */ |
| 2923 | static IRTemp gen_indexed_EA ( /*OUT*/HChar* buf, UInt insn, Bool isInt ) |
| 2924 | { |
| 2925 | UInt optS = SLICE_UInt(insn, 15, 12); |
| 2926 | UInt mm = SLICE_UInt(insn, 20, 16); |
| 2927 | UInt nn = SLICE_UInt(insn, 9, 5); |
| 2928 | UInt szLg2 = (isInt ? 0 : (SLICE_UInt(insn, 23, 23) << 2)) |
| 2929 | | SLICE_UInt(insn, 31, 30); // Log2 of the size |
| 2930 | |
| 2931 | buf[0] = 0; |
| 2932 | |
| 2933 | /* Sanity checks, that this really is a load/store insn. */ |
| 2934 | if (SLICE_UInt(insn, 11, 10) != BITS2(1,0)) |
| 2935 | goto fail; |
| 2936 | |
| 2937 | if (isInt |
| 2938 | && SLICE_UInt(insn, 29, 21) != BITS9(1,1,1,0,0,0,0,1,1)/*LDR*/ |
| 2939 | && SLICE_UInt(insn, 29, 21) != BITS9(1,1,1,0,0,0,0,0,1)/*STR*/ |
| 2940 | && SLICE_UInt(insn, 29, 21) != BITS9(1,1,1,0,0,0,1,0,1)/*LDRSbhw Xt*/ |
| 2941 | && SLICE_UInt(insn, 29, 21) != BITS9(1,1,1,0,0,0,1,1,1))/*LDRSbhw Wt*/ |
| 2942 | goto fail; |
| 2943 | |
| 2944 | if (!isInt |
| 2945 | && SLICE_UInt(insn, 29, 24) != BITS6(1,1,1,1,0,0)) /*LDR/STR*/ |
| 2946 | goto fail; |
| 2947 | |
| 2948 | /* Throw out non-verified but possibly valid cases. */ |
| 2949 | switch (szLg2) { |
| 2950 | case BITS3(0,0,0): break; // 8 bit, valid for both int and vec |
| 2951 | case BITS3(0,0,1): break; // 16 bit, valid for both int and vec |
| 2952 | case BITS3(0,1,0): break; // 32 bit, valid for both int and vec |
| 2953 | case BITS3(0,1,1): break; // 64 bit, valid for both int and vec |
| 2954 | case BITS3(1,0,0): // can only ever be valid for the vector case |
| 2955 | if (isInt) goto fail; else goto fail; |
| 2956 | case BITS3(1,0,1): // these sizes are never valid |
| 2957 | case BITS3(1,1,0): |
| 2958 | case BITS3(1,1,1): goto fail; |
| 2959 | |
| 2960 | default: vassert(0); |
| 2961 | } |
| 2962 | |
| 2963 | IRExpr* rhs = NULL; |
| 2964 | switch (optS) { |
| 2965 | case BITS4(1,1,1,0): goto fail; //ATC |
| 2966 | case BITS4(0,1,1,0): |
| 2967 | rhs = getIReg64orZR(mm); |
| 2968 | vex_sprintf(buf, "[%s, %s]", |
| 2969 | nameIReg64orZR(nn), nameIReg64orZR(mm)); |
| 2970 | break; |
| 2971 | case BITS4(1,1,1,1): goto fail; //ATC |
| 2972 | case BITS4(0,1,1,1): |
| 2973 | rhs = binop(Iop_Shl64, getIReg64orZR(mm), mkU8(szLg2)); |
| 2974 | vex_sprintf(buf, "[%s, %s lsl %u]", |
| 2975 | nameIReg64orZR(nn), nameIReg64orZR(mm), szLg2); |
| 2976 | break; |
| 2977 | case BITS4(0,1,0,0): |
| 2978 | rhs = unop(Iop_32Uto64, getIReg32orZR(mm)); |
| 2979 | vex_sprintf(buf, "[%s, %s uxtx]", |
| 2980 | nameIReg64orZR(nn), nameIReg32orZR(mm)); |
| 2981 | break; |
| 2982 | case BITS4(0,1,0,1): |
| 2983 | rhs = binop(Iop_Shl64, |
| 2984 | unop(Iop_32Uto64, getIReg32orZR(mm)), mkU8(szLg2)); |
| 2985 | vex_sprintf(buf, "[%s, %s uxtx, lsl %u]", |
| 2986 | nameIReg64orZR(nn), nameIReg32orZR(mm), szLg2); |
| 2987 | break; |
| 2988 | case BITS4(1,1,0,0): |
| 2989 | rhs = unop(Iop_32Sto64, getIReg32orZR(mm)); |
| 2990 | vex_sprintf(buf, "[%s, %s sxtx]", |
| 2991 | nameIReg64orZR(nn), nameIReg32orZR(mm)); |
| 2992 | break; |
| 2993 | case BITS4(1,1,0,1): |
| 2994 | rhs = binop(Iop_Shl64, |
| 2995 | unop(Iop_32Sto64, getIReg32orZR(mm)), mkU8(szLg2)); |
| 2996 | vex_sprintf(buf, "[%s, %s sxtx, lsl %u]", |
| 2997 | nameIReg64orZR(nn), nameIReg32orZR(mm), szLg2); |
| 2998 | break; |
| 2999 | default: |
| 3000 | /* The rest appear to be genuinely invalid */ |
| 3001 | goto fail; |
| 3002 | } |
| 3003 | |
| 3004 | vassert(rhs); |
| 3005 | IRTemp res = newTemp(Ity_I64); |
| 3006 | assign(res, binop(Iop_Add64, getIReg64orSP(nn), rhs)); |
| 3007 | return res; |
| 3008 | |
| 3009 | fail: |
| 3010 | vex_printf("gen_indexed_EA: unhandled case optS == 0x%x\n", optS); |
| 3011 | return IRTemp_INVALID; |
| 3012 | } |
| 3013 | |
| 3014 | |
| 3015 | /* Generate an 8/16/32/64 bit integer store to ADDR for the lowest |
| 3016 | bits of DATAE :: Ity_I64. */ |
| 3017 | static void gen_narrowing_store ( UInt szB, IRTemp addr, IRExpr* dataE ) |
| 3018 | { |
| 3019 | IRExpr* addrE = mkexpr(addr); |
| 3020 | switch (szB) { |
| 3021 | case 8: |
| 3022 | storeLE(addrE, dataE); |
| 3023 | break; |
| 3024 | case 4: |
| 3025 | storeLE(addrE, unop(Iop_64to32, dataE)); |
| 3026 | break; |
| 3027 | case 2: |
| 3028 | storeLE(addrE, unop(Iop_64to16, dataE)); |
| 3029 | break; |
| 3030 | case 1: |
| 3031 | storeLE(addrE, unop(Iop_64to8, dataE)); |
| 3032 | break; |
| 3033 | default: |
| 3034 | vassert(0); |
| 3035 | } |
| 3036 | } |
| 3037 | |
| 3038 | |
| 3039 | /* Generate an 8/16/32/64 bit unsigned widening load from ADDR, |
| 3040 | placing the result in an Ity_I64 temporary. */ |
| 3041 | static IRTemp gen_zwidening_load ( UInt szB, IRTemp addr ) |
| 3042 | { |
| 3043 | IRTemp res = newTemp(Ity_I64); |
| 3044 | IRExpr* addrE = mkexpr(addr); |
| 3045 | switch (szB) { |
| 3046 | case 8: |
| 3047 | assign(res, loadLE(Ity_I64,addrE)); |
| 3048 | break; |
| 3049 | case 4: |
| 3050 | assign(res, unop(Iop_32Uto64, loadLE(Ity_I32,addrE))); |
| 3051 | break; |
| 3052 | case 2: |
| 3053 | assign(res, unop(Iop_16Uto64, loadLE(Ity_I16,addrE))); |
| 3054 | break; |
| 3055 | case 1: |
| 3056 | assign(res, unop(Iop_8Uto64, loadLE(Ity_I8,addrE))); |
| 3057 | break; |
| 3058 | default: |
| 3059 | vassert(0); |
| 3060 | } |
| 3061 | return res; |
| 3062 | } |
| 3063 | |
| 3064 | |
| 3065 | static |
| 3066 | Bool dis_ARM64_load_store(/*MB_OUT*/DisResult* dres, UInt insn) |
| 3067 | { |
| 3068 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 3069 | |
| 3070 | /* ------------ LDR,STR (immediate, uimm12) ----------- */ |
| 3071 | /* uimm12 is scaled by the transfer size |
| 3072 | |
| 3073 | 31 29 26 21 9 4 |
| 3074 | | | | | | | |
| 3075 | 11 111 00100 imm12 nn tt STR Xt, [Xn|SP, #imm12 * 8] |
| 3076 | 11 111 00101 imm12 nn tt LDR Xt, [Xn|SP, #imm12 * 8] |
| 3077 | |
| 3078 | 10 111 00100 imm12 nn tt STR Wt, [Xn|SP, #imm12 * 4] |
| 3079 | 10 111 00101 imm12 nn tt LDR Wt, [Xn|SP, #imm12 * 4] |
| 3080 | |
| 3081 | 01 111 00100 imm12 nn tt STRH Wt, [Xn|SP, #imm12 * 2] |
| 3082 | 01 111 00101 imm12 nn tt LDRH Wt, [Xn|SP, #imm12 * 2] |
| 3083 | |
| 3084 | 00 111 00100 imm12 nn tt STRB Wt, [Xn|SP, #imm12 * 1] |
| 3085 | 00 111 00101 imm12 nn tt LDRB Wt, [Xn|SP, #imm12 * 1] |
| 3086 | */ |
| 3087 | if (INSN(29,23) == BITS7(1,1,1,0,0,1,0)) { |
| 3088 | UInt szLg2 = INSN(31,30); |
| 3089 | UInt szB = 1 << szLg2; |
| 3090 | Bool isLD = INSN(22,22) == 1; |
| 3091 | UInt offs = INSN(21,10) * szB; |
| 3092 | UInt nn = INSN(9,5); |
| 3093 | UInt tt = INSN(4,0); |
| 3094 | IRTemp ta = newTemp(Ity_I64); |
| 3095 | assign(ta, binop(Iop_Add64, getIReg64orSP(nn), mkU64(offs))); |
| 3096 | if (nn == 31) { /* FIXME generate stack alignment check */ } |
| 3097 | vassert(szLg2 < 4); |
| 3098 | if (isLD) { |
| 3099 | putIReg64orZR(tt, mkexpr(gen_zwidening_load(szB, ta))); |
| 3100 | } else { |
| 3101 | gen_narrowing_store(szB, ta, getIReg64orZR(tt)); |
| 3102 | } |
| 3103 | const HChar* ld_name[4] = { "ldrb", "ldrh", "ldr", "ldr" }; |
| 3104 | const HChar* st_name[4] = { "strb", "strh", "str", "str" }; |
| 3105 | DIP("%s %s, [%s, #%u]\n", |
| 3106 | (isLD ? ld_name : st_name)[szLg2], nameIRegOrZR(szB == 8, tt), |
| 3107 | nameIReg64orSP(nn), offs); |
| 3108 | return True; |
| 3109 | } |
| 3110 | |
| 3111 | /* ------------ LDUR,STUR (immediate, simm9) ----------- */ |
| 3112 | /* |
| 3113 | 31 29 26 20 11 9 4 |
| 3114 | | | | | | | | |
| 3115 | (at-Rn-then-Rn=EA) | | | |
| 3116 | sz 111 00000 0 imm9 01 Rn Rt STR Rt, [Xn|SP], #simm9 |
| 3117 | sz 111 00001 0 imm9 01 Rn Rt LDR Rt, [Xn|SP], #simm9 |
| 3118 | |
| 3119 | (at-EA-then-Rn=EA) |
| 3120 | sz 111 00000 0 imm9 11 Rn Rt STR Rt, [Xn|SP, #simm9]! |
| 3121 | sz 111 00001 0 imm9 11 Rn Rt LDR Rt, [Xn|SP, #simm9]! |
| 3122 | |
| 3123 | (at-EA) |
| 3124 | sz 111 00000 0 imm9 00 Rn Rt STR Rt, [Xn|SP, #simm9] |
| 3125 | sz 111 00001 0 imm9 00 Rn Rt LDR Rt, [Xn|SP, #simm9] |
| 3126 | |
| 3127 | simm9 is unscaled. |
| 3128 | |
| 3129 | The case 'wback && Rn == Rt && Rt != 31' is disallowed. In the |
| 3130 | load case this is because would create two competing values for |
| 3131 | Rt. In the store case the reason is unclear, but the spec |
| 3132 | disallows it anyway. |
| 3133 | |
| 3134 | Stores are narrowing, loads are unsigned widening. sz encodes |
| 3135 | the transfer size in the normal way: 00=1, 01=2, 10=4, 11=8. |
| 3136 | */ |
| 3137 | if ((INSN(29,21) & BITS9(1,1,1, 1,1,1,1,0, 1)) |
| 3138 | == BITS9(1,1,1, 0,0,0,0,0, 0)) { |
| 3139 | UInt szLg2 = INSN(31,30); |
| 3140 | UInt szB = 1 << szLg2; |
| 3141 | Bool isLoad = INSN(22,22) == 1; |
| 3142 | UInt imm9 = INSN(20,12); |
| 3143 | UInt nn = INSN(9,5); |
| 3144 | UInt tt = INSN(4,0); |
| 3145 | Bool wBack = INSN(10,10) == 1; |
| 3146 | UInt how = INSN(11,10); |
| 3147 | if (how == BITS2(1,0) || (wBack && nn == tt && tt != 31)) { |
| 3148 | /* undecodable; fall through */ |
| 3149 | } else { |
| 3150 | if (nn == 31) { /* FIXME generate stack alignment check */ } |
| 3151 | |
| 3152 | // Compute the transfer address TA and the writeback address WA. |
| 3153 | IRTemp tRN = newTemp(Ity_I64); |
| 3154 | assign(tRN, getIReg64orSP(nn)); |
| 3155 | IRTemp tEA = newTemp(Ity_I64); |
| 3156 | Long simm9 = (Long)sx_to_64(imm9, 9); |
| 3157 | assign(tEA, binop(Iop_Add64, mkexpr(tRN), mkU64(simm9))); |
| 3158 | |
| 3159 | IRTemp tTA = newTemp(Ity_I64); |
| 3160 | IRTemp tWA = newTemp(Ity_I64); |
| 3161 | switch (how) { |
| 3162 | case BITS2(0,1): |
| 3163 | assign(tTA, mkexpr(tRN)); assign(tWA, mkexpr(tEA)); break; |
| 3164 | case BITS2(1,1): |
| 3165 | assign(tTA, mkexpr(tEA)); assign(tWA, mkexpr(tEA)); break; |
| 3166 | case BITS2(0,0): |
| 3167 | assign(tTA, mkexpr(tEA)); /* tWA is unused */ break; |
| 3168 | default: |
| 3169 | vassert(0); /* NOTREACHED */ |
| 3170 | } |
| 3171 | |
| 3172 | if (isLoad) { |
| 3173 | putIReg64orZR(tt, mkexpr(gen_zwidening_load(szB, tTA))); |
| 3174 | } else { |
| 3175 | gen_narrowing_store(szB, tTA, getIReg64orZR(tt)); |
| 3176 | } |
| 3177 | |
| 3178 | if (wBack) |
| 3179 | putIReg64orSP(nn, mkexpr(tEA)); |
| 3180 | |
| 3181 | const HChar* ld_name[4] = { "ldurb", "ldurh", "ldur", "ldur" }; |
| 3182 | const HChar* st_name[4] = { "sturb", "sturh", "stur", "stur" }; |
| 3183 | const HChar* fmt_str = NULL; |
| 3184 | switch (how) { |
| 3185 | case BITS2(0,1): |
| 3186 | fmt_str = "%s %s, [%s], #%lld (at-Rn-then-Rn=EA)\n"; |
| 3187 | break; |
| 3188 | case BITS2(1,1): |
| 3189 | fmt_str = "%s %s, [%s, #%lld]! (at-EA-then-Rn=EA)\n"; |
| 3190 | break; |
| 3191 | case BITS2(0,0): |
| 3192 | fmt_str = "%s %s, [%s, #%lld] (at-Rn)\n"; |
| 3193 | break; |
| 3194 | default: |
| 3195 | vassert(0); |
| 3196 | } |
| 3197 | DIP(fmt_str, (isLoad ? ld_name : st_name)[szLg2], |
| 3198 | nameIRegOrZR(szB == 8, tt), |
| 3199 | nameIReg64orSP(nn), simm9); |
| 3200 | return True; |
| 3201 | } |
| 3202 | } |
| 3203 | |
| 3204 | /* -------- LDP,STP (immediate, simm7) (INT REGS) -------- */ |
| 3205 | /* L==1 => mm==LD |
| 3206 | L==0 => mm==ST |
| 3207 | x==0 => 32 bit transfers, and zero extended loads |
| 3208 | x==1 => 64 bit transfers |
| 3209 | simm7 is scaled by the (single-register) transfer size |
| 3210 | |
| 3211 | (at-Rn-then-Rn=EA) |
| 3212 | x0 101 0001 L imm7 Rt2 Rn Rt1 mmP Rt1,Rt2, [Xn|SP], #imm |
| 3213 | |
| 3214 | (at-EA-then-Rn=EA) |
| 3215 | x0 101 0011 L imm7 Rt2 Rn Rt1 mmP Rt1,Rt2, [Xn|SP, #imm]! |
| 3216 | |
| 3217 | (at-EA) |
| 3218 | x0 101 0010 L imm7 Rt2 Rn Rt1 mmP Rt1,Rt2, [Xn|SP, #imm] |
| 3219 | */ |
| 3220 | |
| 3221 | UInt insn_30_23 = INSN(30,23); |
| 3222 | if (insn_30_23 == BITS8(0,1,0,1,0,0,0,1) |
| 3223 | || insn_30_23 == BITS8(0,1,0,1,0,0,1,1) |
| 3224 | || insn_30_23 == BITS8(0,1,0,1,0,0,1,0)) { |
| 3225 | UInt bL = INSN(22,22); |
| 3226 | UInt bX = INSN(31,31); |
| 3227 | UInt bWBack = INSN(23,23); |
| 3228 | UInt rT1 = INSN(4,0); |
| 3229 | UInt rN = INSN(9,5); |
| 3230 | UInt rT2 = INSN(14,10); |
| 3231 | Long simm7 = (Long)sx_to_64(INSN(21,15), 7); |
| 3232 | if ((bWBack && (rT1 == rN || rT2 == rN) && rN != 31) |
| 3233 | || (bL && rT1 == rT2)) { |
| 3234 | /* undecodable; fall through */ |
| 3235 | } else { |
| 3236 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 3237 | |
| 3238 | // Compute the transfer address TA and the writeback address WA. |
| 3239 | IRTemp tRN = newTemp(Ity_I64); |
| 3240 | assign(tRN, getIReg64orSP(rN)); |
| 3241 | IRTemp tEA = newTemp(Ity_I64); |
| 3242 | simm7 = (bX ? 8 : 4) * simm7; |
| 3243 | assign(tEA, binop(Iop_Add64, mkexpr(tRN), mkU64(simm7))); |
| 3244 | |
| 3245 | IRTemp tTA = newTemp(Ity_I64); |
| 3246 | IRTemp tWA = newTemp(Ity_I64); |
| 3247 | switch (INSN(24,23)) { |
| 3248 | case BITS2(0,1): |
| 3249 | assign(tTA, mkexpr(tRN)); assign(tWA, mkexpr(tEA)); break; |
| 3250 | case BITS2(1,1): |
| 3251 | assign(tTA, mkexpr(tEA)); assign(tWA, mkexpr(tEA)); break; |
| 3252 | case BITS2(1,0): |
| 3253 | assign(tTA, mkexpr(tEA)); /* tWA is unused */ break; |
| 3254 | default: |
| 3255 | vassert(0); /* NOTREACHED */ |
| 3256 | } |
| 3257 | |
| 3258 | /* Normally rN would be updated after the transfer. However, in |
| 3259 | the special case typifed by |
| 3260 | stp x29, x30, [sp,#-112]! |
| 3261 | it is necessary to update SP before the transfer, (1) |
| 3262 | because Memcheck will otherwise complain about a write |
| 3263 | below the stack pointer, and (2) because the segfault |
| 3264 | stack extension mechanism will otherwise extend the stack |
| 3265 | only down to SP before the instruction, which might not be |
| 3266 | far enough, if the -112 bit takes the actual access |
| 3267 | address to the next page. |
| 3268 | */ |
| 3269 | Bool earlyWBack |
| 3270 | = bWBack && simm7 < 0 |
| 3271 | && INSN(24,23) == BITS2(1,1) && rN == 31 && bL == 0; |
| 3272 | |
| 3273 | if (bWBack && earlyWBack) |
| 3274 | putIReg64orSP(rN, mkexpr(tEA)); |
| 3275 | |
| 3276 | /**/ if (bL == 1 && bX == 1) { |
| 3277 | // 64 bit load |
| 3278 | putIReg64orZR(rT1, loadLE(Ity_I64, |
| 3279 | binop(Iop_Add64,mkexpr(tTA),mkU64(0)))); |
| 3280 | putIReg64orZR(rT2, loadLE(Ity_I64, |
| 3281 | binop(Iop_Add64,mkexpr(tTA),mkU64(8)))); |
| 3282 | } else if (bL == 1 && bX == 0) { |
| 3283 | vassert(0); //ATC |
| 3284 | // 32 bit load |
| 3285 | putIReg32orZR(rT1, loadLE(Ity_I32, |
| 3286 | binop(Iop_Add64,mkexpr(tTA),mkU64(0)))); |
| 3287 | putIReg32orZR(rT2, loadLE(Ity_I32, |
| 3288 | binop(Iop_Add64,mkexpr(tTA),mkU64(4)))); |
| 3289 | } else if (bL == 0 && bX == 1) { |
| 3290 | // 64 bit store |
| 3291 | storeLE(binop(Iop_Add64,mkexpr(tTA),mkU64(0)), |
| 3292 | getIReg64orZR(rT1)); |
| 3293 | storeLE(binop(Iop_Add64,mkexpr(tTA),mkU64(8)), |
| 3294 | getIReg64orZR(rT2)); |
| 3295 | } else { |
| 3296 | vassert(bL == 0 && bX == 0); |
| 3297 | vassert(0); //ATC |
| 3298 | // 32 bit store |
| 3299 | storeLE(binop(Iop_Add64,mkexpr(tTA),mkU64(0)), |
| 3300 | getIReg32orZR(rT1)); |
| 3301 | storeLE(binop(Iop_Add64,mkexpr(tTA),mkU64(4)), |
| 3302 | getIReg32orZR(rT2)); |
| 3303 | } |
| 3304 | |
| 3305 | if (bWBack && !earlyWBack) |
| 3306 | putIReg64orSP(rN, mkexpr(tEA)); |
| 3307 | |
| 3308 | const HChar* fmt_str = NULL; |
| 3309 | switch (INSN(24,23)) { |
| 3310 | case BITS2(0,1): |
| 3311 | fmt_str = "%sp %s, %s, [%s], #%lld (at-Rn-then-Rn=EA)\n"; |
| 3312 | break; |
| 3313 | case BITS2(1,1): |
| 3314 | fmt_str = "%sp %s, %s, [%s, #%lld]! (at-EA-then-Rn=EA)\n"; |
| 3315 | break; |
| 3316 | case BITS2(1,0): |
| 3317 | fmt_str = "%sp %s, %s, [%s, #%lld] (at-Rn)\n"; |
| 3318 | break; |
| 3319 | default: |
| 3320 | vassert(0); |
| 3321 | } |
| 3322 | DIP(fmt_str, bL == 0 ? "st" : "ld", |
| 3323 | nameIRegOrZR(bX == 1, rT1), |
| 3324 | nameIRegOrZR(bX == 1, rT2), |
| 3325 | nameIReg64orSP(rN), simm7); |
| 3326 | return True; |
| 3327 | } |
| 3328 | } |
| 3329 | |
| 3330 | /* ---------------- LDR (literal, int reg) ---------------- */ |
| 3331 | /* 31 29 23 4 |
| 3332 | 00 011 000 imm19 Rt LDR Wt, [PC + sxTo64(imm19 << 2)] |
| 3333 | 01 011 000 imm19 Rt LDR Xt, [PC + sxTo64(imm19 << 2)] |
| 3334 | 10 011 000 imm19 Rt LDRSW Xt, [PC + sxTo64(imm19 << 2)] |
| 3335 | 11 011 000 imm19 Rt prefetch [PC + sxTo64(imm19 << 2)] |
| 3336 | Just handles the first two cases for now. |
| 3337 | */ |
| 3338 | if (INSN(29,24) == BITS6(0,1,1,0,0,0) && INSN(31,31) == 0) { |
| 3339 | UInt imm19 = INSN(23,5); |
| 3340 | UInt rT = INSN(4,0); |
| 3341 | UInt bX = INSN(30,30); |
| 3342 | ULong ea = guest_PC_curr_instr + sx_to_64(imm19 << 2, 21); |
| 3343 | if (bX) { |
| 3344 | putIReg64orZR(rT, loadLE(Ity_I64, mkU64(ea))); |
| 3345 | } else { |
| 3346 | putIReg32orZR(rT, loadLE(Ity_I32, mkU64(ea))); |
| 3347 | } |
| 3348 | DIP("ldr %s, 0x%llx (literal)\n", nameIRegOrZR(bX == 1, rT), ea); |
| 3349 | return True; |
| 3350 | } |
| 3351 | |
| 3352 | /* -------------- {LD,ST}R (integer register) --------------- */ |
| 3353 | /* 31 29 20 15 12 11 9 4 |
| 3354 | | | | | | | | | |
| 3355 | 11 111000011 Rm option S 10 Rn Rt LDR Xt, [Xn|SP, R<m>{ext/sh}] |
| 3356 | 10 111000011 Rm option S 10 Rn Rt LDR Wt, [Xn|SP, R<m>{ext/sh}] |
| 3357 | 01 111000011 Rm option S 10 Rn Rt LDRH Wt, [Xn|SP, R<m>{ext/sh}] |
| 3358 | 00 111000011 Rm option S 10 Rn Rt LDRB Wt, [Xn|SP, R<m>{ext/sh}] |
| 3359 | |
| 3360 | 11 111000001 Rm option S 10 Rn Rt STR Xt, [Xn|SP, R<m>{ext/sh}] |
| 3361 | 10 111000001 Rm option S 10 Rn Rt STR Wt, [Xn|SP, R<m>{ext/sh}] |
| 3362 | 01 111000001 Rm option S 10 Rn Rt STRH Wt, [Xn|SP, R<m>{ext/sh}] |
| 3363 | 00 111000001 Rm option S 10 Rn Rt STRB Wt, [Xn|SP, R<m>{ext/sh}] |
| 3364 | */ |
| 3365 | if (INSN(29,23) == BITS7(1,1,1,0,0,0,0) |
| 3366 | && INSN(21,21) == 1 && INSN(11,10) == BITS2(1,0)) { |
| 3367 | HChar dis_buf[64]; |
| 3368 | UInt szLg2 = INSN(31,30); |
| 3369 | Bool isLD = INSN(22,22) == 1; |
| 3370 | UInt tt = INSN(4,0); |
| 3371 | IRTemp ea = gen_indexed_EA(dis_buf, insn, True/*to/from int regs*/); |
| 3372 | if (ea != IRTemp_INVALID) { |
| 3373 | switch (szLg2) { |
| 3374 | case 3: /* 64 bit */ |
| 3375 | if (isLD) { |
| 3376 | putIReg64orZR(tt, loadLE(Ity_I64, mkexpr(ea))); |
| 3377 | DIP("ldr %s, %s\n", nameIReg64orZR(tt), dis_buf); |
| 3378 | } else { |
| 3379 | storeLE(mkexpr(ea), getIReg64orZR(tt)); |
| 3380 | DIP("str %s, %s\n", nameIReg64orZR(tt), dis_buf); |
| 3381 | } |
| 3382 | break; |
| 3383 | case 2: /* 32 bit */ |
| 3384 | if (isLD) { |
| 3385 | putIReg32orZR(tt, loadLE(Ity_I32, mkexpr(ea))); |
| 3386 | DIP("ldr %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3387 | } else { |
| 3388 | storeLE(mkexpr(ea), getIReg32orZR(tt)); |
| 3389 | DIP("str %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3390 | } |
| 3391 | break; |
| 3392 | case 1: /* 16 bit */ |
| 3393 | if (isLD) { |
| 3394 | putIReg64orZR(tt, unop(Iop_16Uto64, |
| 3395 | loadLE(Ity_I16, mkexpr(ea)))); |
| 3396 | DIP("ldruh %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3397 | } else { |
| 3398 | storeLE(mkexpr(ea), unop(Iop_64to16, getIReg64orZR(tt))); |
| 3399 | DIP("strh %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3400 | } |
| 3401 | break; |
| 3402 | case 0: /* 8 bit */ |
| 3403 | if (isLD) { |
| 3404 | putIReg64orZR(tt, unop(Iop_8Uto64, |
| 3405 | loadLE(Ity_I8, mkexpr(ea)))); |
| 3406 | DIP("ldrub %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3407 | } else { |
| 3408 | storeLE(mkexpr(ea), unop(Iop_64to8, getIReg64orZR(tt))); |
| 3409 | DIP("strb %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3410 | } |
| 3411 | break; |
| 3412 | default: |
| 3413 | vassert(0); |
| 3414 | } |
| 3415 | return True; |
| 3416 | } |
| 3417 | } |
| 3418 | |
| 3419 | /* -------------- LDRS{B,H,W} (uimm12) -------------- */ |
| 3420 | /* 31 29 26 23 21 9 4 |
| 3421 | 10 111 001 10 imm12 n t LDRSW Xt, [Xn|SP, #pimm12 * 4] |
| 3422 | 01 111 001 1x imm12 n t LDRSH Rt, [Xn|SP, #pimm12 * 2] |
| 3423 | 00 111 001 1x imm12 n t LDRSB Rt, [Xn|SP, #pimm12 * 1] |
| 3424 | where |
| 3425 | Rt is Wt when x==1, Xt when x==0 |
| 3426 | */ |
| 3427 | if (INSN(29,23) == BITS7(1,1,1,0,0,1,1)) { |
| 3428 | /* Further checks on bits 31:30 and 22 */ |
| 3429 | Bool valid = False; |
| 3430 | switch ((INSN(31,30) << 1) | INSN(22,22)) { |
| 3431 | case BITS3(1,0,0): |
| 3432 | case BITS3(0,1,0): case BITS3(0,1,1): |
| 3433 | case BITS3(0,0,0): case BITS3(0,0,1): |
| 3434 | valid = True; |
| 3435 | break; |
| 3436 | } |
| 3437 | if (valid) { |
| 3438 | UInt szLg2 = INSN(31,30); |
| 3439 | UInt bitX = INSN(22,22); |
| 3440 | UInt imm12 = INSN(21,10); |
| 3441 | UInt nn = INSN(9,5); |
| 3442 | UInt tt = INSN(4,0); |
| 3443 | UInt szB = 1 << szLg2; |
| 3444 | IRExpr* ea = binop(Iop_Add64, |
| 3445 | getIReg64orSP(nn), mkU64(imm12 * szB)); |
| 3446 | switch (szB) { |
| 3447 | case 4: |
| 3448 | vassert(bitX == 0); |
| 3449 | putIReg64orZR(tt, unop(Iop_32Sto64, loadLE(Ity_I32, ea))); |
| 3450 | DIP("ldrsw %s, [%s, #%u]\n", nameIReg64orZR(tt), |
| 3451 | nameIReg64orSP(nn), imm12 * szB); |
| 3452 | break; |
| 3453 | case 2: |
| 3454 | if (bitX == 1) { |
| 3455 | putIReg32orZR(tt, unop(Iop_16Sto32, loadLE(Ity_I16, ea))); |
| 3456 | } else { |
| 3457 | putIReg64orZR(tt, unop(Iop_16Sto64, loadLE(Ity_I16, ea))); |
| 3458 | } |
| 3459 | DIP("ldrsh %s, [%s, #%u]\n", |
| 3460 | nameIRegOrZR(bitX == 0, tt), |
| 3461 | nameIReg64orSP(nn), imm12 * szB); |
| 3462 | break; |
| 3463 | case 1: |
| 3464 | if (bitX == 1) { |
| 3465 | putIReg32orZR(tt, unop(Iop_8Sto32, loadLE(Ity_I8, ea))); |
| 3466 | } else { |
| 3467 | putIReg64orZR(tt, unop(Iop_8Sto64, loadLE(Ity_I8, ea))); |
| 3468 | } |
| 3469 | DIP("ldrsb %s, [%s, #%u]\n", |
| 3470 | nameIRegOrZR(bitX == 0, tt), |
| 3471 | nameIReg64orSP(nn), imm12 * szB); |
| 3472 | break; |
| 3473 | default: |
| 3474 | vassert(0); |
| 3475 | } |
| 3476 | return True; |
| 3477 | } |
| 3478 | /* else fall through */ |
| 3479 | } |
| 3480 | |
| 3481 | /* -------------- LDRS{B,H,W} (simm9, upd) -------------- */ |
| 3482 | /* (at-Rn-then-Rn=EA) |
| 3483 | 31 29 23 21 20 11 9 4 |
| 3484 | 00 111 000 1x 0 imm9 01 n t LDRSB Rt, [Xn|SP], #simm9 |
| 3485 | 01 111 000 1x 0 imm9 01 n t LDRSH Rt, [Xn|SP], #simm9 |
| 3486 | 10 111 000 10 0 imm9 01 n t LDRSW Xt, [Xn|SP], #simm9 |
| 3487 | |
| 3488 | (at-EA-then-Rn=EA) |
| 3489 | 00 111 000 1x 0 imm9 11 n t LDRSB Rt, [Xn|SP, #simm9]! |
| 3490 | 01 111 000 1x 0 imm9 11 n t LDRSH Rt, [Xn|SP, #simm9]! |
| 3491 | 10 111 000 10 0 imm9 11 n t LDRSW Xt, [Xn|SP, #simm9]! |
| 3492 | where |
| 3493 | Rt is Wt when x==1, Xt when x==0 |
| 3494 | transfer-at-Rn when [11]==0, at EA when [11]==1 |
| 3495 | */ |
| 3496 | if (INSN(29,23) == BITS7(1,1,1,0,0,0,1) |
| 3497 | && INSN(21,21) == 0 && INSN(10,10) == 1) { |
| 3498 | /* Further checks on bits 31:30 and 22 */ |
| 3499 | Bool valid = False; |
| 3500 | switch ((INSN(31,30) << 1) | INSN(22,22)) { |
| 3501 | case BITS3(1,0,0): // LDRSW Xt |
| 3502 | case BITS3(0,1,0): case BITS3(0,1,1): // LDRSH Xt, Wt |
| 3503 | case BITS3(0,0,0): case BITS3(0,0,1): // LDRSB Xt, Wt |
| 3504 | valid = True; |
| 3505 | break; |
| 3506 | } |
| 3507 | if (valid) { |
| 3508 | UInt szLg2 = INSN(31,30); |
| 3509 | UInt imm9 = INSN(20,12); |
| 3510 | Bool atRN = INSN(11,11) == 0; |
| 3511 | UInt nn = INSN(9,5); |
| 3512 | UInt tt = INSN(4,0); |
| 3513 | IRTemp tRN = newTemp(Ity_I64); |
| 3514 | IRTemp tEA = newTemp(Ity_I64); |
| 3515 | IRTemp tTA = IRTemp_INVALID; |
| 3516 | ULong simm9 = sx_to_64(imm9, 9); |
| 3517 | Bool is64 = INSN(22,22) == 0; |
| 3518 | assign(tRN, getIReg64orSP(nn)); |
| 3519 | assign(tEA, binop(Iop_Add64, mkexpr(tRN), mkU64(simm9))); |
| 3520 | tTA = atRN ? tRN : tEA; |
| 3521 | HChar ch = '?'; |
| 3522 | /* There are 5 cases: |
| 3523 | byte load, SX to 64 |
| 3524 | byte load, SX to 32, ZX to 64 |
| 3525 | halfword load, SX to 64 |
| 3526 | halfword load, SX to 32, ZX to 64 |
| 3527 | word load, SX to 64 |
| 3528 | The ifs below handle them in the listed order. |
| 3529 | */ |
| 3530 | if (szLg2 == 0) { |
| 3531 | ch = 'b'; |
| 3532 | if (is64) { |
| 3533 | putIReg64orZR(tt, unop(Iop_8Sto64, |
| 3534 | loadLE(Ity_I8, mkexpr(tTA)))); |
| 3535 | } else { |
| 3536 | putIReg32orZR(tt, unop(Iop_8Sto32, |
| 3537 | loadLE(Ity_I8, mkexpr(tTA)))); |
| 3538 | } |
| 3539 | } |
| 3540 | else if (szLg2 == 1) { |
| 3541 | ch = 'h'; |
| 3542 | if (is64) { |
| 3543 | putIReg64orZR(tt, unop(Iop_16Sto64, |
| 3544 | loadLE(Ity_I16, mkexpr(tTA)))); |
| 3545 | } else { |
| 3546 | putIReg32orZR(tt, unop(Iop_16Sto32, |
| 3547 | loadLE(Ity_I16, mkexpr(tTA)))); |
| 3548 | } |
| 3549 | } |
| 3550 | else if (szLg2 == 2 && is64) { |
| 3551 | ch = 'w'; |
| 3552 | putIReg64orZR(tt, unop(Iop_32Sto64, |
| 3553 | loadLE(Ity_I32, mkexpr(tTA)))); |
| 3554 | } |
| 3555 | else { |
| 3556 | vassert(0); |
| 3557 | } |
| 3558 | putIReg64orSP(nn, mkexpr(tEA)); |
| 3559 | DIP(atRN ? "ldrs%c %s, [%s], #%lld\n" : "ldrs%c %s, [%s, #%lld]!", |
| 3560 | ch, nameIRegOrZR(is64, tt), nameIReg64orSP(nn), simm9); |
| 3561 | return True; |
| 3562 | } |
| 3563 | /* else fall through */ |
| 3564 | } |
| 3565 | |
| 3566 | /* -------------- LDRS{B,H,W} (simm9, noUpd) -------------- */ |
| 3567 | /* 31 29 23 21 20 11 9 4 |
| 3568 | 00 111 000 1x 0 imm9 00 n t LDURSB Rt, [Xn|SP, #simm9] |
| 3569 | 01 111 000 1x 0 imm9 00 n t LDURSH Rt, [Xn|SP, #simm9] |
| 3570 | 10 111 000 10 0 imm9 00 n t LDURSW Xt, [Xn|SP, #simm9] |
| 3571 | where |
| 3572 | Rt is Wt when x==1, Xt when x==0 |
| 3573 | */ |
| 3574 | if (INSN(29,23) == BITS7(1,1,1,0,0,0,1) |
| 3575 | && INSN(21,21) == 0 && INSN(11,10) == BITS2(0,0)) { |
| 3576 | /* Further checks on bits 31:30 and 22 */ |
| 3577 | Bool valid = False; |
| 3578 | switch ((INSN(31,30) << 1) | INSN(22,22)) { |
| 3579 | case BITS3(1,0,0): // LDURSW Xt |
| 3580 | case BITS3(0,1,0): case BITS3(0,1,1): // LDURSH Xt, Wt |
| 3581 | case BITS3(0,0,0): case BITS3(0,0,1): // LDURSB Xt, Wt |
| 3582 | valid = True; |
| 3583 | break; |
| 3584 | } |
| 3585 | if (valid) { |
| 3586 | UInt szLg2 = INSN(31,30); |
| 3587 | UInt imm9 = INSN(20,12); |
| 3588 | UInt nn = INSN(9,5); |
| 3589 | UInt tt = INSN(4,0); |
| 3590 | IRTemp tRN = newTemp(Ity_I64); |
| 3591 | IRTemp tEA = newTemp(Ity_I64); |
| 3592 | ULong simm9 = sx_to_64(imm9, 9); |
| 3593 | Bool is64 = INSN(22,22) == 0; |
| 3594 | assign(tRN, getIReg64orSP(nn)); |
| 3595 | assign(tEA, binop(Iop_Add64, mkexpr(tRN), mkU64(simm9))); |
| 3596 | HChar ch = '?'; |
| 3597 | /* There are 5 cases: |
| 3598 | byte load, SX to 64 |
| 3599 | byte load, SX to 32, ZX to 64 |
| 3600 | halfword load, SX to 64 |
| 3601 | halfword load, SX to 32, ZX to 64 |
| 3602 | word load, SX to 64 |
| 3603 | The ifs below handle them in the listed order. |
| 3604 | */ |
| 3605 | if (szLg2 == 0) { |
| 3606 | ch = 'b'; |
| 3607 | if (is64) { |
| 3608 | putIReg64orZR(tt, unop(Iop_8Sto64, |
| 3609 | loadLE(Ity_I8, mkexpr(tEA)))); |
| 3610 | } else { |
| 3611 | putIReg32orZR(tt, unop(Iop_8Sto32, |
| 3612 | loadLE(Ity_I8, mkexpr(tEA)))); |
| 3613 | } |
| 3614 | } |
| 3615 | else if (szLg2 == 1) { |
| 3616 | ch = 'h'; |
| 3617 | if (is64) { |
| 3618 | putIReg64orZR(tt, unop(Iop_16Sto64, |
| 3619 | loadLE(Ity_I16, mkexpr(tEA)))); |
| 3620 | } else { |
| 3621 | putIReg32orZR(tt, unop(Iop_16Sto32, |
| 3622 | loadLE(Ity_I16, mkexpr(tEA)))); |
| 3623 | } |
| 3624 | } |
| 3625 | else if (szLg2 == 2 && is64) { |
| 3626 | ch = 'w'; |
| 3627 | putIReg64orZR(tt, unop(Iop_32Sto64, |
| 3628 | loadLE(Ity_I32, mkexpr(tEA)))); |
| 3629 | } |
| 3630 | else { |
| 3631 | vassert(0); |
| 3632 | } |
| 3633 | DIP("ldurs%c %s, [%s, #%lld]", |
| 3634 | ch, nameIRegOrZR(is64, tt), nameIReg64orSP(nn), simm9); |
| 3635 | return True; |
| 3636 | } |
| 3637 | /* else fall through */ |
| 3638 | } |
| 3639 | |
| 3640 | /* -------- LDP,STP (immediate, simm7) (FP&VEC) -------- */ |
| 3641 | /* L==1 => mm==LD |
| 3642 | L==0 => mm==ST |
| 3643 | sz==00 => 32 bit (S) transfers |
| 3644 | sz==01 => 64 bit (D) transfers |
| 3645 | sz==10 => 128 bit (Q) transfers |
| 3646 | sz==11 isn't allowed |
| 3647 | simm7 is scaled by the (single-register) transfer size |
| 3648 | |
| 3649 | 31 29 22 21 14 9 4 |
| 3650 | sz 101 1001 L imm7 t2 n t1 mmP SDQt1, SDQt2, [Xn|SP], #imm |
| 3651 | (at-Rn-then-Rn=EA) |
| 3652 | |
| 3653 | sz 101 1011 L imm7 t2 n t1 mmP SDQt1, SDQt2, [Xn|SP, #imm]! |
| 3654 | (at-EA-then-Rn=EA) |
| 3655 | |
| 3656 | sz 101 1010 L imm7 t2 n t1 mmP SDQt1, SDQt2, [Xn|SP, #imm] |
| 3657 | (at-EA) |
| 3658 | */ |
| 3659 | |
| 3660 | UInt insn_29_23 = INSN(29,23); |
| 3661 | if (insn_29_23 == BITS7(1,0,1,1,0,0,1) |
| 3662 | || insn_29_23 == BITS7(1,0,1,1,0,1,1) |
| 3663 | || insn_29_23 == BITS7(1,0,1,1,0,1,0)) { |
| 3664 | UInt szSlg2 = INSN(31,30); // log2 of the xfer size in 32-bit units |
| 3665 | Bool isLD = INSN(22,22) == 1; |
| 3666 | Bool wBack = INSN(23,23) == 1; |
| 3667 | Long simm7 = (Long)sx_to_64(INSN(21,15), 7); |
| 3668 | UInt tt2 = INSN(14,10); |
| 3669 | UInt nn = INSN(9,5); |
| 3670 | UInt tt1 = INSN(4,0); |
| 3671 | if (szSlg2 == BITS2(1,1) || (isLD && tt1 == tt2)) { |
| 3672 | /* undecodable; fall through */ |
| 3673 | } else { |
| 3674 | if (nn == 31) { /* FIXME generate stack alignment check */ } |
| 3675 | |
| 3676 | // Compute the transfer address TA and the writeback address WA. |
| 3677 | UInt szB = 4 << szSlg2; /* szB is the per-register size */ |
| 3678 | IRTemp tRN = newTemp(Ity_I64); |
| 3679 | assign(tRN, getIReg64orSP(nn)); |
| 3680 | IRTemp tEA = newTemp(Ity_I64); |
| 3681 | simm7 = szB * simm7; |
| 3682 | assign(tEA, binop(Iop_Add64, mkexpr(tRN), mkU64(simm7))); |
| 3683 | |
| 3684 | IRTemp tTA = newTemp(Ity_I64); |
| 3685 | IRTemp tWA = newTemp(Ity_I64); |
| 3686 | switch (INSN(24,23)) { |
| 3687 | case BITS2(0,1): |
| 3688 | assign(tTA, mkexpr(tRN)); assign(tWA, mkexpr(tEA)); break; |
| 3689 | case BITS2(1,1): |
| 3690 | assign(tTA, mkexpr(tEA)); assign(tWA, mkexpr(tEA)); break; |
| 3691 | case BITS2(1,0): |
| 3692 | assign(tTA, mkexpr(tEA)); /* tWA is unused */ break; |
| 3693 | default: |
| 3694 | vassert(0); /* NOTREACHED */ |
| 3695 | } |
| 3696 | |
| 3697 | IRType ty = Ity_INVALID; |
| 3698 | switch (szB) { |
| 3699 | case 4: ty = Ity_F32; break; |
| 3700 | case 8: ty = Ity_F64; break; |
| 3701 | case 16: ty = Ity_V128; break; |
| 3702 | default: vassert(0); |
| 3703 | } |
| 3704 | |
| 3705 | if (isLD) { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3706 | putQRegLO(tt1, |
| 3707 | loadLE(ty, binop(Iop_Add64, mkexpr(tTA), mkU64(0)))); |
| 3708 | putQRegLO(tt2, |
| 3709 | loadLE(ty, binop(Iop_Add64, mkexpr(tTA), mkU64(szB)))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3710 | } else { |
| 3711 | storeLE(binop(Iop_Add64, mkexpr(tTA), mkU64(0)), |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3712 | getQRegLO(tt1, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3713 | storeLE(binop(Iop_Add64, mkexpr(tTA), mkU64(szB)), |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3714 | getQRegLO(tt2, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3715 | } |
| 3716 | |
| 3717 | if (wBack) |
| 3718 | putIReg64orSP(nn, mkexpr(tEA)); |
| 3719 | |
| 3720 | const HChar* fmt_str = NULL; |
| 3721 | switch (INSN(24,23)) { |
| 3722 | case BITS2(0,1): |
| 3723 | fmt_str = "%sp %s, %s, [%s], #%lld (at-Rn-then-Rn=EA)\n"; |
| 3724 | break; |
| 3725 | case BITS2(1,1): |
| 3726 | fmt_str = "%sp %s, %s, [%s, #%lld]! (at-EA-then-Rn=EA)\n"; |
| 3727 | break; |
| 3728 | case BITS2(1,0): |
| 3729 | fmt_str = "%sp %s, %s, [%s, #%lld] (at-Rn)\n"; |
| 3730 | break; |
| 3731 | default: |
| 3732 | vassert(0); |
| 3733 | } |
| 3734 | DIP(fmt_str, isLD ? "ld" : "st", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3735 | nameQRegLO(tt1, ty), nameQRegLO(tt2, ty), |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3736 | nameIReg64orSP(nn), simm7); |
| 3737 | return True; |
| 3738 | } |
| 3739 | } |
| 3740 | |
| 3741 | /* -------------- {LD,ST}R (vector register) --------------- */ |
| 3742 | /* 31 29 23 20 15 12 11 9 4 |
| 3743 | | | | | | | | | | |
| 3744 | 00 111100 011 Rm option S 10 Rn Rt LDR Bt, [Xn|SP, R<m>{ext/sh}] |
| 3745 | 01 111100 011 Rm option S 10 Rn Rt LDR Ht, [Xn|SP, R<m>{ext/sh}] |
| 3746 | 10 111100 011 Rm option S 10 Rn Rt LDR St, [Xn|SP, R<m>{ext/sh}] |
| 3747 | 11 111100 011 Rm option S 10 Rn Rt LDR Dt, [Xn|SP, R<m>{ext/sh}] |
| 3748 | 00 111100 111 Rm option S 10 Rn Rt LDR Qt, [Xn|SP, R<m>{ext/sh}] |
| 3749 | |
| 3750 | 00 111100 001 Rm option S 10 Rn Rt STR Bt, [Xn|SP, R<m>{ext/sh}] |
| 3751 | 01 111100 001 Rm option S 10 Rn Rt STR Ht, [Xn|SP, R<m>{ext/sh}] |
| 3752 | 10 111100 001 Rm option S 10 Rn Rt STR St, [Xn|SP, R<m>{ext/sh}] |
| 3753 | 11 111100 001 Rm option S 10 Rn Rt STR Dt, [Xn|SP, R<m>{ext/sh}] |
| 3754 | 00 111100 101 Rm option S 10 Rn Rt STR Qt, [Xn|SP, R<m>{ext/sh}] |
| 3755 | */ |
| 3756 | if (INSN(29,24) == BITS6(1,1,1,1,0,0) |
| 3757 | && INSN(21,21) == 1 && INSN(11,10) == BITS2(1,0)) { |
| 3758 | HChar dis_buf[64]; |
| 3759 | UInt szLg2 = (INSN(23,23) << 2) | INSN(31,30); |
| 3760 | Bool isLD = INSN(22,22) == 1; |
| 3761 | UInt tt = INSN(4,0); |
| 3762 | if (szLg2 >= 4) goto after_LDR_STR_vector_register; |
| 3763 | IRTemp ea = gen_indexed_EA(dis_buf, insn, False/*to/from vec regs*/); |
| 3764 | if (ea == IRTemp_INVALID) goto after_LDR_STR_vector_register; |
| 3765 | switch (szLg2) { |
| 3766 | case 0: /* 8 bit */ |
| 3767 | if (isLD) { |
| 3768 | putQReg128(tt, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3769 | putQRegLO(tt, loadLE(Ity_I8, mkexpr(ea))); |
| 3770 | DIP("ldr %s, %s\n", nameQRegLO(tt, Ity_I8), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3771 | } else { |
| 3772 | vassert(0); //ATC |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3773 | storeLE(mkexpr(ea), getQRegLO(tt, Ity_I8)); |
| 3774 | DIP("str %s, %s\n", nameQRegLO(tt, Ity_I8), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3775 | } |
| 3776 | break; |
| 3777 | case 1: |
| 3778 | if (isLD) { |
| 3779 | putQReg128(tt, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3780 | putQRegLO(tt, loadLE(Ity_I16, mkexpr(ea))); |
| 3781 | DIP("ldr %s, %s\n", nameQRegLO(tt, Ity_I16), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3782 | } else { |
| 3783 | vassert(0); //ATC |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3784 | storeLE(mkexpr(ea), getQRegLO(tt, Ity_I16)); |
| 3785 | DIP("str %s, %s\n", nameQRegLO(tt, Ity_I16), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3786 | } |
| 3787 | break; |
| 3788 | case 2: /* 32 bit */ |
| 3789 | if (isLD) { |
| 3790 | putQReg128(tt, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3791 | putQRegLO(tt, loadLE(Ity_I32, mkexpr(ea))); |
| 3792 | DIP("ldr %s, %s\n", nameQRegLO(tt, Ity_I32), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3793 | } else { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3794 | storeLE(mkexpr(ea), getQRegLO(tt, Ity_I32)); |
| 3795 | DIP("str %s, %s\n", nameQRegLO(tt, Ity_I32), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3796 | } |
| 3797 | break; |
| 3798 | case 3: /* 64 bit */ |
| 3799 | if (isLD) { |
| 3800 | putQReg128(tt, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3801 | putQRegLO(tt, loadLE(Ity_I64, mkexpr(ea))); |
| 3802 | DIP("ldr %s, %s\n", nameQRegLO(tt, Ity_I64), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3803 | } else { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3804 | storeLE(mkexpr(ea), getQRegLO(tt, Ity_I64)); |
| 3805 | DIP("str %s, %s\n", nameQRegLO(tt, Ity_I64), dis_buf); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3806 | } |
| 3807 | break; |
| 3808 | case 4: return False; //ATC |
| 3809 | default: vassert(0); |
| 3810 | } |
| 3811 | return True; |
| 3812 | } |
| 3813 | after_LDR_STR_vector_register: |
| 3814 | |
| 3815 | /* ---------- LDRS{B,H,W} (integer register, SX) ---------- */ |
| 3816 | /* 31 29 22 20 15 12 11 9 4 |
| 3817 | | | | | | | | | | |
| 3818 | 10 1110001 01 Rm opt S 10 Rn Rt LDRSW Xt, [Xn|SP, R<m>{ext/sh}] |
| 3819 | |
| 3820 | 01 1110001 01 Rm opt S 10 Rn Rt LDRSH Xt, [Xn|SP, R<m>{ext/sh}] |
| 3821 | 01 1110001 11 Rm opt S 10 Rn Rt LDRSH Wt, [Xn|SP, R<m>{ext/sh}] |
| 3822 | |
| 3823 | 00 1110001 01 Rm opt S 10 Rn Rt LDRSB Xt, [Xn|SP, R<m>{ext/sh}] |
| 3824 | 00 1110001 11 Rm opt S 10 Rn Rt LDRSB Wt, [Xn|SP, R<m>{ext/sh}] |
| 3825 | */ |
| 3826 | if (INSN(29,23) == BITS7(1,1,1,0,0,0,1) |
| 3827 | && INSN(21,21) == 1 && INSN(11,10) == BITS2(1,0)) { |
| 3828 | HChar dis_buf[64]; |
| 3829 | UInt szLg2 = INSN(31,30); |
| 3830 | Bool sxTo64 = INSN(22,22) == 0; // else sx to 32 and zx to 64 |
| 3831 | UInt tt = INSN(4,0); |
| 3832 | if (szLg2 == 3) goto after_LDRS_integer_register; |
| 3833 | IRTemp ea = gen_indexed_EA(dis_buf, insn, True/*to/from int regs*/); |
| 3834 | if (ea == IRTemp_INVALID) goto after_LDRS_integer_register; |
| 3835 | /* Enumerate the 5 variants explicitly. */ |
| 3836 | if (szLg2 == 2/*32 bit*/ && sxTo64) { |
| 3837 | putIReg64orZR(tt, unop(Iop_32Sto64, loadLE(Ity_I32, mkexpr(ea)))); |
| 3838 | DIP("ldrsw %s, %s\n", nameIReg64orZR(tt), dis_buf); |
| 3839 | return True; |
| 3840 | } |
| 3841 | else |
| 3842 | if (szLg2 == 1/*16 bit*/) { |
| 3843 | if (sxTo64) { |
| 3844 | putIReg64orZR(tt, unop(Iop_16Sto64, loadLE(Ity_I16, mkexpr(ea)))); |
| 3845 | DIP("ldrsh %s, %s\n", nameIReg64orZR(tt), dis_buf); |
| 3846 | } else { |
| 3847 | putIReg32orZR(tt, unop(Iop_16Sto32, loadLE(Ity_I16, mkexpr(ea)))); |
| 3848 | DIP("ldrsh %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3849 | } |
| 3850 | return True; |
| 3851 | } |
| 3852 | else |
| 3853 | if (szLg2 == 0/*8 bit*/) { |
| 3854 | if (sxTo64) { |
| 3855 | putIReg64orZR(tt, unop(Iop_8Sto64, loadLE(Ity_I8, mkexpr(ea)))); |
| 3856 | DIP("ldrsb %s, %s\n", nameIReg64orZR(tt), dis_buf); |
| 3857 | } else { |
| 3858 | putIReg32orZR(tt, unop(Iop_8Sto32, loadLE(Ity_I8, mkexpr(ea)))); |
| 3859 | DIP("ldrsb %s, %s\n", nameIReg32orZR(tt), dis_buf); |
| 3860 | } |
| 3861 | return True; |
| 3862 | } |
| 3863 | /* else it's an invalid combination */ |
| 3864 | } |
| 3865 | after_LDRS_integer_register: |
| 3866 | |
| 3867 | /* -------- LDR/STR (immediate, SIMD&FP, unsigned offset) -------- */ |
| 3868 | /* This is the Unsigned offset variant only. The Post-Index and |
| 3869 | Pre-Index variants are below. |
| 3870 | |
| 3871 | 31 29 23 21 9 4 |
| 3872 | 00 111 101 01 imm12 n t LDR Bt, [Xn|SP + imm12 * 1] |
| 3873 | 01 111 101 01 imm12 n t LDR Ht, [Xn|SP + imm12 * 2] |
| 3874 | 10 111 101 01 imm12 n t LDR St, [Xn|SP + imm12 * 4] |
| 3875 | 11 111 101 01 imm12 n t LDR Dt, [Xn|SP + imm12 * 8] |
| 3876 | 00 111 101 11 imm12 n t LDR Qt, [Xn|SP + imm12 * 16] |
| 3877 | |
| 3878 | 00 111 101 00 imm12 n t STR Bt, [Xn|SP + imm12 * 1] |
| 3879 | 01 111 101 00 imm12 n t STR Ht, [Xn|SP + imm12 * 2] |
| 3880 | 10 111 101 00 imm12 n t STR St, [Xn|SP + imm12 * 4] |
| 3881 | 11 111 101 00 imm12 n t STR Dt, [Xn|SP + imm12 * 8] |
| 3882 | 00 111 101 10 imm12 n t STR Qt, [Xn|SP + imm12 * 16] |
| 3883 | */ |
| 3884 | if (INSN(29,24) == BITS6(1,1,1,1,0,1) |
| 3885 | && ((INSN(23,23) << 2) | INSN(31,30)) <= 4) { |
| 3886 | UInt szLg2 = (INSN(23,23) << 2) | INSN(31,30); |
| 3887 | Bool isLD = INSN(22,22) == 1; |
| 3888 | UInt pimm12 = INSN(21,10) << szLg2; |
| 3889 | UInt nn = INSN(9,5); |
| 3890 | UInt tt = INSN(4,0); |
| 3891 | IRTemp tEA = newTemp(Ity_I64); |
| 3892 | IRType ty = preferredVectorSubTypeFromSize(1 << szLg2); |
| 3893 | assign(tEA, binop(Iop_Add64, getIReg64orSP(nn), mkU64(pimm12))); |
| 3894 | if (isLD) { |
| 3895 | if (szLg2 < 4) { |
| 3896 | putQReg128(tt, mkV128(0x0000)); |
| 3897 | } |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3898 | putQRegLO(tt, loadLE(ty, mkexpr(tEA))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3899 | } else { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3900 | storeLE(mkexpr(tEA), getQRegLO(tt, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3901 | } |
| 3902 | DIP("%s %s, [%s, #%u]\n", |
| 3903 | isLD ? "ldr" : "str", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3904 | nameQRegLO(tt, ty), nameIReg64orSP(nn), pimm12); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3905 | return True; |
| 3906 | } |
| 3907 | |
| 3908 | /* -------- LDR/STR (immediate, SIMD&FP, pre/post index) -------- */ |
| 3909 | /* These are the Post-Index and Pre-Index variants. |
| 3910 | |
| 3911 | 31 29 23 20 11 9 4 |
| 3912 | (at-Rn-then-Rn=EA) |
| 3913 | 00 111 100 01 0 imm9 01 n t LDR Bt, [Xn|SP], #simm |
| 3914 | 01 111 100 01 0 imm9 01 n t LDR Ht, [Xn|SP], #simm |
| 3915 | 10 111 100 01 0 imm9 01 n t LDR St, [Xn|SP], #simm |
| 3916 | 11 111 100 01 0 imm9 01 n t LDR Dt, [Xn|SP], #simm |
| 3917 | 00 111 100 11 0 imm9 01 n t LDR Qt, [Xn|SP], #simm |
| 3918 | |
| 3919 | (at-EA-then-Rn=EA) |
| 3920 | 00 111 100 01 0 imm9 11 n t LDR Bt, [Xn|SP, #simm]! |
| 3921 | 01 111 100 01 0 imm9 11 n t LDR Ht, [Xn|SP, #simm]! |
| 3922 | 10 111 100 01 0 imm9 11 n t LDR St, [Xn|SP, #simm]! |
| 3923 | 11 111 100 01 0 imm9 11 n t LDR Dt, [Xn|SP, #simm]! |
| 3924 | 00 111 100 11 0 imm9 11 n t LDR Qt, [Xn|SP, #simm]! |
| 3925 | |
| 3926 | Stores are the same except with bit 22 set to 0. |
| 3927 | */ |
| 3928 | if (INSN(29,24) == BITS6(1,1,1,1,0,0) |
| 3929 | && ((INSN(23,23) << 2) | INSN(31,30)) <= 4 |
| 3930 | && INSN(21,21) == 0 && INSN(10,10) == 1) { |
| 3931 | UInt szLg2 = (INSN(23,23) << 2) | INSN(31,30); |
| 3932 | Bool isLD = INSN(22,22) == 1; |
| 3933 | UInt imm9 = INSN(20,12); |
| 3934 | Bool atRN = INSN(11,11) == 0; |
| 3935 | UInt nn = INSN(9,5); |
| 3936 | UInt tt = INSN(4,0); |
| 3937 | IRTemp tRN = newTemp(Ity_I64); |
| 3938 | IRTemp tEA = newTemp(Ity_I64); |
| 3939 | IRTemp tTA = IRTemp_INVALID; |
| 3940 | IRType ty = preferredVectorSubTypeFromSize(1 << szLg2); |
| 3941 | ULong simm9 = sx_to_64(imm9, 9); |
| 3942 | assign(tRN, getIReg64orSP(nn)); |
| 3943 | assign(tEA, binop(Iop_Add64, mkexpr(tRN), mkU64(simm9))); |
| 3944 | tTA = atRN ? tRN : tEA; |
| 3945 | if (isLD) { |
| 3946 | if (szLg2 < 4) { |
| 3947 | putQReg128(tt, mkV128(0x0000)); |
| 3948 | } |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3949 | putQRegLO(tt, loadLE(ty, mkexpr(tTA))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3950 | } else { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3951 | storeLE(mkexpr(tTA), getQRegLO(tt, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3952 | } |
| 3953 | putIReg64orSP(nn, mkexpr(tEA)); |
| 3954 | DIP(atRN ? "%s %s, [%s], #%lld\n" : "%s %s, [%s, #%lld]!\n", |
| 3955 | isLD ? "ldr" : "str", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3956 | nameQRegLO(tt, ty), nameIReg64orSP(nn), simm9); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3957 | return True; |
| 3958 | } |
| 3959 | |
| 3960 | /* -------- LDUR/STUR (unscaled offset, SIMD&FP) -------- */ |
| 3961 | /* 31 29 23 20 11 9 4 |
| 3962 | 00 111 100 01 0 imm9 00 n t LDR Bt, [Xn|SP, #simm] |
| 3963 | 01 111 100 01 0 imm9 00 n t LDR Ht, [Xn|SP, #simm] |
| 3964 | 10 111 100 01 0 imm9 00 n t LDR St, [Xn|SP, #simm] |
| 3965 | 11 111 100 01 0 imm9 00 n t LDR Dt, [Xn|SP, #simm] |
| 3966 | 00 111 100 11 0 imm9 00 n t LDR Qt, [Xn|SP, #simm] |
| 3967 | |
| 3968 | 00 111 100 00 0 imm9 00 n t STR Bt, [Xn|SP, #simm] |
| 3969 | 01 111 100 00 0 imm9 00 n t STR Ht, [Xn|SP, #simm] |
| 3970 | 10 111 100 00 0 imm9 00 n t STR St, [Xn|SP, #simm] |
| 3971 | 11 111 100 00 0 imm9 00 n t STR Dt, [Xn|SP, #simm] |
| 3972 | 00 111 100 10 0 imm9 00 n t STR Qt, [Xn|SP, #simm] |
| 3973 | */ |
| 3974 | if (INSN(29,24) == BITS6(1,1,1,1,0,0) |
| 3975 | && ((INSN(23,23) << 2) | INSN(31,30)) <= 4 |
| 3976 | && INSN(21,21) == 0 && INSN(11,10) == BITS2(0,0)) { |
| 3977 | UInt szLg2 = (INSN(23,23) << 2) | INSN(31,30); |
| 3978 | Bool isLD = INSN(22,22) == 1; |
| 3979 | UInt imm9 = INSN(20,12); |
| 3980 | UInt nn = INSN(9,5); |
| 3981 | UInt tt = INSN(4,0); |
| 3982 | ULong simm9 = sx_to_64(imm9, 9); |
| 3983 | IRTemp tEA = newTemp(Ity_I64); |
| 3984 | IRType ty = preferredVectorSubTypeFromSize(1 << szLg2); |
| 3985 | assign(tEA, binop(Iop_Add64, getIReg64orSP(nn), mkU64(simm9))); |
| 3986 | if (isLD) { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3987 | if (szLg2 < 4) { |
| 3988 | putQReg128(tt, mkV128(0x0000)); |
| 3989 | } |
| 3990 | putQRegLO(tt, loadLE(ty, mkexpr(tEA))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3991 | } else { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3992 | storeLE(mkexpr(tEA), getQRegLO(tt, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3993 | } |
| 3994 | DIP("%s %s, [%s, #%lld]\n", |
| 3995 | isLD ? "ldur" : "stur", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 3996 | nameQRegLO(tt, ty), nameIReg64orSP(nn), (Long)simm9); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 3997 | return True; |
| 3998 | } |
| 3999 | |
| 4000 | /* ---------------- LDR (literal, SIMD&FP) ---------------- */ |
| 4001 | /* 31 29 23 4 |
| 4002 | 00 011 100 imm19 t LDR St, [PC + sxTo64(imm19 << 2)] |
| 4003 | 01 011 100 imm19 t LDR Dt, [PC + sxTo64(imm19 << 2)] |
| 4004 | 10 011 100 imm19 t LDR Qt, [PC + sxTo64(imm19 << 2)] |
| 4005 | */ |
| 4006 | if (INSN(29,24) == BITS6(0,1,1,1,0,0) && INSN(31,30) < BITS2(1,1)) { |
| 4007 | UInt szB = 4 << INSN(31,30); |
| 4008 | UInt imm19 = INSN(23,5); |
| 4009 | UInt tt = INSN(4,0); |
| 4010 | ULong ea = guest_PC_curr_instr + sx_to_64(imm19 << 2, 21); |
| 4011 | IRType ty = preferredVectorSubTypeFromSize(szB); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4012 | putQReg128(tt, mkV128(0x0000)); |
| 4013 | putQRegLO(tt, loadLE(ty, mkU64(ea))); |
| 4014 | DIP("ldr %s, 0x%llx (literal)\n", nameQRegLO(tt, ty), ea); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4015 | return True; |
| 4016 | } |
| 4017 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4018 | /* ---------- LD1/ST1 (single structure, no offset) ---------- */ |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4019 | /* 31 23 |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4020 | 0100 1100 0100 0000 0111 11 N T LD1 {vT.2d}, [Xn|SP] |
| 4021 | 0100 1100 0000 0000 0111 11 N T ST1 {vT.2d}, [Xn|SP] |
| 4022 | 0100 1100 0100 0000 0111 10 N T LD1 {vT.4s}, [Xn|SP] |
| 4023 | 0100 1100 0000 0000 0111 10 N T ST1 {vT.4s}, [Xn|SP] |
| 4024 | 0100 1100 0100 0000 0111 01 N T LD1 {vT.8h}, [Xn|SP] |
| 4025 | 0100 1100 0000 0000 0111 01 N T ST1 {vT.8h}, [Xn|SP] |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4026 | 0100 1100 0100 0000 0111 00 N T LD1 {vT.16b}, [Xn|SP] |
| 4027 | 0100 1100 0000 0000 0111 00 N T ST1 {vT.16b}, [Xn|SP] |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4028 | FIXME does this assume that the host is little endian? |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4029 | */ |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4030 | if ( (insn & 0xFFFFF000) == 0x4C407000 // LD1 cases |
| 4031 | || (insn & 0xFFFFF000) == 0x4C007000 // ST1 cases |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4032 | ) { |
| 4033 | Bool isLD = INSN(22,22) == 1; |
| 4034 | UInt rN = INSN(9,5); |
| 4035 | UInt vT = INSN(4,0); |
| 4036 | IRTemp tEA = newTemp(Ity_I64); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4037 | const HChar* names[4] = { "2d", "4s", "8h", "16b" }; |
| 4038 | const HChar* name = names[INSN(11,10)]; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4039 | assign(tEA, getIReg64orSP(rN)); |
| 4040 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 4041 | if (isLD) { |
| 4042 | putQReg128(vT, loadLE(Ity_V128, mkexpr(tEA))); |
| 4043 | } else { |
| 4044 | storeLE(mkexpr(tEA), getQReg128(vT)); |
| 4045 | } |
| 4046 | DIP("%s {v%u.%s}, [%s]\n", isLD ? "ld1" : "st1", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4047 | vT, name, nameIReg64orSP(rN)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4048 | return True; |
| 4049 | } |
| 4050 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4051 | /* 31 23 |
| 4052 | 0000 1100 0100 0000 0111 11 N T LD1 {vT.1d}, [Xn|SP] |
| 4053 | 0000 1100 0000 0000 0111 11 N T ST1 {vT.1d}, [Xn|SP] |
| 4054 | 0000 1100 0100 0000 0111 10 N T LD1 {vT.2s}, [Xn|SP] |
| 4055 | 0000 1100 0000 0000 0111 10 N T ST1 {vT.2s}, [Xn|SP] |
| 4056 | 0000 1100 0100 0000 0111 01 N T LD1 {vT.4h}, [Xn|SP] |
| 4057 | 0000 1100 0000 0000 0111 01 N T ST1 {vT.4h}, [Xn|SP] |
| 4058 | 0000 1100 0100 0000 0111 00 N T LD1 {vT.8b}, [Xn|SP] |
| 4059 | 0000 1100 0000 0000 0111 00 N T ST1 {vT.8b}, [Xn|SP] |
| 4060 | FIXME does this assume that the host is little endian? |
| 4061 | */ |
| 4062 | if ( (insn & 0xFFFFF000) == 0x0C407000 // LD1 cases |
| 4063 | || (insn & 0xFFFFF000) == 0x0C007000 // ST1 cases |
| 4064 | ) { |
| 4065 | Bool isLD = INSN(22,22) == 1; |
| 4066 | UInt rN = INSN(9,5); |
| 4067 | UInt vT = INSN(4,0); |
| 4068 | IRTemp tEA = newTemp(Ity_I64); |
| 4069 | const HChar* names[4] = { "1d", "2s", "4h", "8b" }; |
| 4070 | const HChar* name = names[INSN(11,10)]; |
| 4071 | assign(tEA, getIReg64orSP(rN)); |
| 4072 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 4073 | if (isLD) { |
| 4074 | putQRegLane(vT, 0, loadLE(Ity_I64, mkexpr(tEA))); |
| 4075 | putQRegLane(vT, 1, mkU64(0)); |
| 4076 | } else { |
| 4077 | storeLE(mkexpr(tEA), getQRegLane(vT, 0, Ity_I64)); |
| 4078 | } |
| 4079 | DIP("%s {v%u.%s}, [%s]\n", isLD ? "ld1" : "st1", |
| 4080 | vT, name, nameIReg64orSP(rN)); |
| 4081 | return True; |
| 4082 | } |
| 4083 | |
| 4084 | /* ---------- LD1/ST1 (single structure, post index) ---------- */ |
| 4085 | /* 31 23 |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 4086 | 0100 1100 1001 1111 0111 11 N T ST1 {vT.2d}, [xN|SP], #16 |
| 4087 | 0100 1100 1101 1111 0111 11 N T LD1 {vT.2d}, [xN|SP], #16 |
| 4088 | 0100 1100 1001 1111 0111 10 N T ST1 {vT.4s}, [xN|SP], #16 |
| 4089 | 0100 1100 1101 1111 0111 10 N T LD1 {vT.4s}, [xN|SP], #16 |
| 4090 | 0100 1100 1001 1111 0111 01 N T ST1 {vT.8h}, [xN|SP], #16 |
| 4091 | 0100 1100 1101 1111 0111 01 N T LD1 {vT.8h}, [xN|SP], #16 |
| 4092 | 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] | 4093 | 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] | 4094 | Note that #16 is implied and cannot be any other value. |
| 4095 | FIXME does this assume that the host is little endian? |
| 4096 | */ |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 4097 | if ( (insn & 0xFFFFF000) == 0x4CDF7000 // LD1 cases |
| 4098 | || (insn & 0xFFFFF000) == 0x4C9F7000 // ST1 cases |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4099 | ) { |
| 4100 | Bool isLD = INSN(22,22) == 1; |
| 4101 | UInt rN = INSN(9,5); |
| 4102 | UInt vT = INSN(4,0); |
| 4103 | IRTemp tEA = newTemp(Ity_I64); |
| 4104 | const HChar* names[4] = { "2d", "4s", "8h", "16b" }; |
| 4105 | const HChar* name = names[INSN(11,10)]; |
| 4106 | assign(tEA, getIReg64orSP(rN)); |
| 4107 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 4108 | if (isLD) { |
| 4109 | putQReg128(vT, loadLE(Ity_V128, mkexpr(tEA))); |
| 4110 | } else { |
| 4111 | storeLE(mkexpr(tEA), getQReg128(vT)); |
| 4112 | } |
| 4113 | putIReg64orSP(rN, binop(Iop_Add64, mkexpr(tEA), mkU64(16))); |
| 4114 | DIP("%s {v%u.%s}, [%s], #16\n", isLD ? "ld1" : "st1", |
| 4115 | vT, name, nameIReg64orSP(rN)); |
| 4116 | return True; |
| 4117 | } |
| 4118 | |
| 4119 | /* |
| 4120 | 0000 1100 1001 1111 0111 10 N T ST1 {vT.2s}, [xN|SP], #8 |
sewardj | f5b0891 | 2014-02-06 12:57:58 +0000 | [diff] [blame] | 4121 | 0000 1100 1001 1111 0111 01 N T ST1 {vT.4h}, [xN|SP], #8 |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4122 | Note that #8 is implied and cannot be any other value. |
| 4123 | FIXME does this assume that the host is little endian? |
| 4124 | */ |
| 4125 | if ( (insn & 0xFFFFFC00) == 0x0C9F7800 // st1 {vT.2s}, [xN|SP], #8 |
sewardj | f5b0891 | 2014-02-06 12:57:58 +0000 | [diff] [blame] | 4126 | || (insn & 0xFFFFFC00) == 0x0C9F7400 // st1 {vT.4h}, [xN|SP], #8 |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4127 | ) { |
| 4128 | UInt rN = INSN(9,5); |
| 4129 | UInt vT = INSN(4,0); |
| 4130 | IRTemp tEA = newTemp(Ity_I64); |
| 4131 | const HChar* names[4] = { "1d", "2s", "4h", "8b" }; |
| 4132 | const HChar* name = names[INSN(11,10)]; |
| 4133 | assign(tEA, getIReg64orSP(rN)); |
| 4134 | if (rN == 31) { /* FIXME generate stack alignment check */ } |
| 4135 | storeLE(mkexpr(tEA), getQRegLane(vT, 0, Ity_I64)); |
| 4136 | putIReg64orSP(rN, binop(Iop_Add64, mkexpr(tEA), mkU64(8))); |
| 4137 | DIP("st1 {v%u.%s}, [%s], #8\n", vT, name, nameIReg64orSP(rN)); |
| 4138 | return True; |
| 4139 | } |
| 4140 | |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 4141 | /* ------------------ LD{,A}X{R,RH,RB} ------------------ */ |
| 4142 | /* ------------------ ST{,L}X{R,RH,RB} ------------------ */ |
| 4143 | /* 31 29 23 20 14 9 4 |
| 4144 | sz 001000 010 11111 0 11111 n t LDX{R,RH,RB} Rt, [Xn|SP] |
| 4145 | sz 001000 010 11111 1 11111 n t LDAX{R,RH,RB} Rt, [Xn|SP] |
| 4146 | sz 001000 000 s 0 11111 n t STX{R,RH,RB} Ws, Rt, [Xn|SP] |
| 4147 | 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] | 4148 | */ |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 4149 | if (INSN(29,23) == BITS7(0,0,1,0,0,0,0) |
| 4150 | && (INSN(23,21) & BITS3(1,0,1)) == BITS3(0,0,0) |
| 4151 | && INSN(14,10) == BITS5(1,1,1,1,1)) { |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 4152 | UInt szBlg2 = INSN(31,30); |
| 4153 | Bool isLD = INSN(22,22) == 1; |
| 4154 | Bool isAcqOrRel = INSN(15,15) == 1; |
| 4155 | UInt ss = INSN(20,16); |
| 4156 | UInt nn = INSN(9,5); |
| 4157 | UInt tt = INSN(4,0); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4158 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 4159 | vassert(szBlg2 < 4); |
| 4160 | UInt szB = 1 << szBlg2; /* 1, 2, 4 or 8 */ |
| 4161 | IRType ty = integerIRTypeOfSize(szB); |
| 4162 | const HChar* suffix[4] = { "rb", "rh", "r", "r" }; |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 4163 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 4164 | IRTemp ea = newTemp(Ity_I64); |
| 4165 | assign(ea, getIReg64orSP(nn)); |
| 4166 | /* FIXME generate check that ea is szB-aligned */ |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 4167 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 4168 | if (isLD && ss == BITS5(1,1,1,1,1)) { |
| 4169 | IRTemp res = newTemp(ty); |
| 4170 | stmt(IRStmt_LLSC(Iend_LE, res, mkexpr(ea), NULL/*LL*/)); |
| 4171 | putIReg64orZR(tt, widenUto64(ty, mkexpr(res))); |
| 4172 | if (isAcqOrRel) { |
| 4173 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 4174 | } |
| 4175 | DIP("ld%sx%s %s, [%s]\n", isAcqOrRel ? "a" : "", suffix[szBlg2], |
| 4176 | nameIRegOrZR(szB == 8, tt), nameIReg64orSP(nn)); |
| 4177 | return True; |
| 4178 | } |
| 4179 | if (!isLD) { |
| 4180 | if (isAcqOrRel) { |
| 4181 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 4182 | } |
| 4183 | IRTemp res = newTemp(Ity_I1); |
| 4184 | IRExpr* data = narrowFrom64(ty, getIReg64orZR(tt)); |
| 4185 | stmt(IRStmt_LLSC(Iend_LE, res, mkexpr(ea), data)); |
| 4186 | /* IR semantics: res is 1 if store succeeds, 0 if it fails. |
| 4187 | Need to set rS to 1 on failure, 0 on success. */ |
| 4188 | putIReg64orZR(ss, binop(Iop_Xor64, unop(Iop_1Uto64, mkexpr(res)), |
| 4189 | mkU64(1))); |
| 4190 | DIP("st%sx%s %s, %s, [%s]\n", isAcqOrRel ? "a" : "", suffix[szBlg2], |
| 4191 | nameIRegOrZR(False, ss), |
| 4192 | nameIRegOrZR(szB == 8, tt), nameIReg64orSP(nn)); |
| 4193 | return True; |
| 4194 | } |
| 4195 | /* else fall through */ |
| 4196 | } |
| 4197 | |
| 4198 | /* ------------------ LDA{R,RH,RB} ------------------ */ |
| 4199 | /* ------------------ STL{R,RH,RB} ------------------ */ |
| 4200 | /* 31 29 23 20 14 9 4 |
| 4201 | sz 001000 110 11111 1 11111 n t LDAR<sz> Rt, [Xn|SP] |
| 4202 | sz 001000 100 11111 1 11111 n t STLR<sz> Rt, [Xn|SP] |
| 4203 | */ |
| 4204 | if (INSN(29,23) == BITS7(0,0,1,0,0,0,1) |
| 4205 | && INSN(21,10) == BITS12(0,1,1,1,1,1,1,1,1,1,1,1)) { |
| 4206 | UInt szBlg2 = INSN(31,30); |
| 4207 | Bool isLD = INSN(22,22) == 1; |
| 4208 | UInt nn = INSN(9,5); |
| 4209 | UInt tt = INSN(4,0); |
| 4210 | |
| 4211 | vassert(szBlg2 < 4); |
| 4212 | UInt szB = 1 << szBlg2; /* 1, 2, 4 or 8 */ |
| 4213 | IRType ty = integerIRTypeOfSize(szB); |
| 4214 | const HChar* suffix[4] = { "rb", "rh", "r", "r" }; |
| 4215 | |
| 4216 | IRTemp ea = newTemp(Ity_I64); |
| 4217 | assign(ea, getIReg64orSP(nn)); |
| 4218 | /* FIXME generate check that ea is szB-aligned */ |
| 4219 | |
| 4220 | if (isLD) { |
| 4221 | IRTemp res = newTemp(ty); |
| 4222 | assign(res, loadLE(ty, mkexpr(ea))); |
| 4223 | putIReg64orZR(tt, widenUto64(ty, mkexpr(res))); |
| 4224 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 4225 | DIP("lda%s %s, [%s]\n", suffix[szBlg2], |
| 4226 | nameIRegOrZR(szB == 8, tt), nameIReg64orSP(nn)); |
| 4227 | } else { |
| 4228 | stmt(IRStmt_MBE(Imbe_Fence)); |
| 4229 | IRExpr* data = narrowFrom64(ty, getIReg64orZR(tt)); |
| 4230 | storeLE(mkexpr(ea), data); |
| 4231 | DIP("stl%s %s, [%s]\n", suffix[szBlg2], |
| 4232 | nameIRegOrZR(szB == 8, tt), nameIReg64orSP(nn)); |
| 4233 | } |
| 4234 | return True; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4235 | } |
| 4236 | |
| 4237 | vex_printf("ARM64 front end: load_store\n"); |
| 4238 | return False; |
| 4239 | # undef INSN |
| 4240 | } |
| 4241 | |
| 4242 | |
| 4243 | /*------------------------------------------------------------*/ |
| 4244 | /*--- Control flow and misc instructions ---*/ |
| 4245 | /*------------------------------------------------------------*/ |
| 4246 | |
| 4247 | static |
| 4248 | Bool dis_ARM64_branch_etc(/*MB_OUT*/DisResult* dres, UInt insn) |
| 4249 | { |
| 4250 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 4251 | |
| 4252 | /* ---------------------- B cond ----------------------- */ |
| 4253 | /* 31 24 4 3 |
| 4254 | 0101010 0 imm19 0 cond */ |
| 4255 | if (INSN(31,24) == BITS8(0,1,0,1,0,1,0,0) && INSN(4,4) == 0) { |
| 4256 | UInt cond = INSN(3,0); |
| 4257 | ULong uimm64 = INSN(23,5) << 2; |
| 4258 | Long simm64 = (Long)sx_to_64(uimm64, 21); |
| 4259 | vassert(dres->whatNext == Dis_Continue); |
| 4260 | vassert(dres->len == 4); |
| 4261 | vassert(dres->continueAt == 0); |
| 4262 | vassert(dres->jk_StopHere == Ijk_INVALID); |
| 4263 | stmt( IRStmt_Exit(unop(Iop_64to1, mk_arm64g_calculate_condition(cond)), |
| 4264 | Ijk_Boring, |
| 4265 | IRConst_U64(guest_PC_curr_instr + simm64), |
| 4266 | OFFB_PC) ); |
| 4267 | putPC(mkU64(guest_PC_curr_instr + 4)); |
| 4268 | dres->whatNext = Dis_StopHere; |
| 4269 | dres->jk_StopHere = Ijk_Boring; |
| 4270 | DIP("b.%s 0x%llx\n", nameCC(cond), guest_PC_curr_instr + simm64); |
| 4271 | return True; |
| 4272 | } |
| 4273 | |
| 4274 | /* -------------------- B{L} uncond -------------------- */ |
| 4275 | if (INSN(30,26) == BITS5(0,0,1,0,1)) { |
| 4276 | /* 000101 imm26 B (PC + sxTo64(imm26 << 2)) |
| 4277 | 100101 imm26 B (PC + sxTo64(imm26 << 2)) |
| 4278 | */ |
| 4279 | UInt bLink = INSN(31,31); |
| 4280 | ULong uimm64 = INSN(25,0) << 2; |
| 4281 | Long simm64 = (Long)sx_to_64(uimm64, 28); |
| 4282 | if (bLink) { |
| 4283 | putIReg64orSP(30, mkU64(guest_PC_curr_instr + 4)); |
| 4284 | } |
| 4285 | putPC(mkU64(guest_PC_curr_instr + simm64)); |
| 4286 | dres->whatNext = Dis_StopHere; |
| 4287 | dres->jk_StopHere = Ijk_Call; |
| 4288 | DIP("b%s 0x%llx\n", bLink == 1 ? "l" : "", |
| 4289 | guest_PC_curr_instr + simm64); |
| 4290 | return True; |
| 4291 | } |
| 4292 | |
| 4293 | /* --------------------- B{L} reg --------------------- */ |
| 4294 | /* 31 24 22 20 15 9 4 |
| 4295 | 1101011 00 10 11111 000000 nn 00000 RET Rn |
| 4296 | 1101011 00 01 11111 000000 nn 00000 CALL Rn |
| 4297 | 1101011 00 00 11111 000000 nn 00000 JMP Rn |
| 4298 | */ |
| 4299 | if (INSN(31,23) == BITS9(1,1,0,1,0,1,1,0,0) |
| 4300 | && INSN(20,16) == BITS5(1,1,1,1,1) |
| 4301 | && INSN(15,10) == BITS6(0,0,0,0,0,0) |
| 4302 | && INSN(4,0) == BITS5(0,0,0,0,0)) { |
| 4303 | UInt branch_type = INSN(22,21); |
| 4304 | UInt nn = INSN(9,5); |
| 4305 | if (branch_type == BITS2(1,0) /* RET */) { |
| 4306 | putPC(getIReg64orZR(nn)); |
| 4307 | dres->whatNext = Dis_StopHere; |
| 4308 | dres->jk_StopHere = Ijk_Ret; |
| 4309 | DIP("ret %s\n", nameIReg64orZR(nn)); |
| 4310 | return True; |
| 4311 | } |
| 4312 | if (branch_type == BITS2(0,1) /* CALL */) { |
| 4313 | putIReg64orSP(30, mkU64(guest_PC_curr_instr + 4)); |
| 4314 | putPC(getIReg64orZR(nn)); |
| 4315 | dres->whatNext = Dis_StopHere; |
| 4316 | dres->jk_StopHere = Ijk_Call; |
| 4317 | DIP("blr %s\n", nameIReg64orZR(nn)); |
| 4318 | return True; |
| 4319 | } |
| 4320 | if (branch_type == BITS2(0,0) /* JMP */) { |
| 4321 | putPC(getIReg64orZR(nn)); |
| 4322 | dres->whatNext = Dis_StopHere; |
| 4323 | dres->jk_StopHere = Ijk_Boring; |
| 4324 | DIP("jmp %s\n", nameIReg64orZR(nn)); |
| 4325 | return True; |
| 4326 | } |
| 4327 | } |
| 4328 | |
| 4329 | /* -------------------- CB{N}Z -------------------- */ |
| 4330 | /* sf 011 010 1 imm19 Rt CBNZ Xt|Wt, (PC + sxTo64(imm19 << 2)) |
| 4331 | sf 011 010 0 imm19 Rt CBZ Xt|Wt, (PC + sxTo64(imm19 << 2)) |
| 4332 | */ |
| 4333 | if (INSN(30,25) == BITS6(0,1,1,0,1,0)) { |
| 4334 | Bool is64 = INSN(31,31) == 1; |
| 4335 | Bool bIfZ = INSN(24,24) == 0; |
| 4336 | ULong uimm64 = INSN(23,5) << 2; |
| 4337 | UInt rT = INSN(4,0); |
| 4338 | Long simm64 = (Long)sx_to_64(uimm64, 21); |
| 4339 | IRExpr* cond = NULL; |
| 4340 | if (is64) { |
| 4341 | cond = binop(bIfZ ? Iop_CmpEQ64 : Iop_CmpNE64, |
| 4342 | getIReg64orZR(rT), mkU64(0)); |
| 4343 | } else { |
| 4344 | cond = binop(bIfZ ? Iop_CmpEQ32 : Iop_CmpNE32, |
| 4345 | getIReg32orZR(rT), mkU32(0)); |
| 4346 | } |
| 4347 | stmt( IRStmt_Exit(cond, |
| 4348 | Ijk_Boring, |
| 4349 | IRConst_U64(guest_PC_curr_instr + simm64), |
| 4350 | OFFB_PC) ); |
| 4351 | putPC(mkU64(guest_PC_curr_instr + 4)); |
| 4352 | dres->whatNext = Dis_StopHere; |
| 4353 | dres->jk_StopHere = Ijk_Boring; |
| 4354 | DIP("cb%sz %s, 0x%llx\n", |
| 4355 | bIfZ ? "" : "n", nameIRegOrZR(is64, rT), |
| 4356 | guest_PC_curr_instr + simm64); |
| 4357 | return True; |
| 4358 | } |
| 4359 | |
| 4360 | /* -------------------- TB{N}Z -------------------- */ |
| 4361 | /* 31 30 24 23 18 5 4 |
| 4362 | b5 011 011 1 b40 imm14 t TBNZ Xt, #(b5:b40), (PC + sxTo64(imm14 << 2)) |
| 4363 | b5 011 011 0 b40 imm14 t TBZ Xt, #(b5:b40), (PC + sxTo64(imm14 << 2)) |
| 4364 | */ |
| 4365 | if (INSN(30,25) == BITS6(0,1,1,0,1,1)) { |
| 4366 | UInt b5 = INSN(31,31); |
| 4367 | Bool bIfZ = INSN(24,24) == 0; |
| 4368 | UInt b40 = INSN(23,19); |
| 4369 | UInt imm14 = INSN(18,5); |
| 4370 | UInt tt = INSN(4,0); |
| 4371 | UInt bitNo = (b5 << 5) | b40; |
| 4372 | ULong uimm64 = imm14 << 2; |
| 4373 | Long simm64 = sx_to_64(uimm64, 16); |
| 4374 | IRExpr* cond |
| 4375 | = binop(bIfZ ? Iop_CmpEQ64 : Iop_CmpNE64, |
| 4376 | binop(Iop_And64, |
| 4377 | binop(Iop_Shr64, getIReg64orZR(tt), mkU8(bitNo)), |
| 4378 | mkU64(1)), |
| 4379 | mkU64(0)); |
| 4380 | stmt( IRStmt_Exit(cond, |
| 4381 | Ijk_Boring, |
| 4382 | IRConst_U64(guest_PC_curr_instr + simm64), |
| 4383 | OFFB_PC) ); |
| 4384 | putPC(mkU64(guest_PC_curr_instr + 4)); |
| 4385 | dres->whatNext = Dis_StopHere; |
| 4386 | dres->jk_StopHere = Ijk_Boring; |
| 4387 | DIP("tb%sz %s, #%u, 0x%llx\n", |
| 4388 | bIfZ ? "" : "n", nameIReg64orZR(tt), bitNo, |
| 4389 | guest_PC_curr_instr + simm64); |
| 4390 | return True; |
| 4391 | } |
| 4392 | |
| 4393 | /* -------------------- SVC -------------------- */ |
| 4394 | /* 11010100 000 imm16 000 01 |
| 4395 | Don't bother with anything except the imm16==0 case. |
| 4396 | */ |
| 4397 | if (INSN(31,0) == 0xD4000001) { |
| 4398 | putPC(mkU64(guest_PC_curr_instr + 4)); |
| 4399 | dres->whatNext = Dis_StopHere; |
| 4400 | dres->jk_StopHere = Ijk_Sys_syscall; |
| 4401 | DIP("svc #0\n"); |
| 4402 | return True; |
| 4403 | } |
| 4404 | |
| 4405 | /* ------------------ M{SR,RS} ------------------ */ |
| 4406 | /* Only handles the case where the system register is TPIDR_EL0. |
| 4407 | 0xD51BD0 010 Rt MSR tpidr_el0, rT |
| 4408 | 0xD53BD0 010 Rt MRS rT, tpidr_el0 |
| 4409 | */ |
| 4410 | if ( (INSN(31,0) & 0xFFFFFFE0) == 0xD51BD040 /*MSR*/ |
| 4411 | || (INSN(31,0) & 0xFFFFFFE0) == 0xD53BD040 /*MRS*/) { |
| 4412 | Bool toSys = INSN(21,21) == 0; |
| 4413 | UInt tt = INSN(4,0); |
| 4414 | if (toSys) { |
| 4415 | stmt( IRStmt_Put( OFFB_TPIDR_EL0, getIReg64orZR(tt)) ); |
| 4416 | DIP("msr tpidr_el0, %s\n", nameIReg64orZR(tt)); |
| 4417 | } else { |
| 4418 | putIReg64orZR(tt, IRExpr_Get( OFFB_TPIDR_EL0, Ity_I64 )); |
| 4419 | DIP("mrs %s, tpidr_el0\n", nameIReg64orZR(tt)); |
| 4420 | } |
| 4421 | return True; |
| 4422 | } |
| 4423 | /* Cases for FPCR |
| 4424 | 0xD51B44 000 Rt MSR fpcr, rT |
| 4425 | 0xD53B44 000 Rt MSR rT, fpcr |
| 4426 | */ |
| 4427 | if ( (INSN(31,0) & 0xFFFFFFE0) == 0xD51B4400 /*MSR*/ |
| 4428 | || (INSN(31,0) & 0xFFFFFFE0) == 0xD53B4400 /*MRS*/) { |
| 4429 | Bool toSys = INSN(21,21) == 0; |
| 4430 | UInt tt = INSN(4,0); |
| 4431 | if (toSys) { |
| 4432 | stmt( IRStmt_Put( OFFB_FPCR, getIReg32orZR(tt)) ); |
| 4433 | DIP("msr fpcr, %s\n", nameIReg64orZR(tt)); |
| 4434 | } else { |
| 4435 | putIReg32orZR(tt, IRExpr_Get(OFFB_FPCR, Ity_I32)); |
| 4436 | DIP("mrs %s, fpcr\n", nameIReg64orZR(tt)); |
| 4437 | } |
| 4438 | return True; |
| 4439 | } |
| 4440 | /* Cases for FPSR |
sewardj | 7d00913 | 2014-02-20 17:43:38 +0000 | [diff] [blame] | 4441 | 0xD51B44 001 Rt MSR fpsr, rT |
| 4442 | 0xD53B44 001 Rt MSR rT, fpsr |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4443 | */ |
| 4444 | if ( (INSN(31,0) & 0xFFFFFFE0) == 0xD51B4420 /*MSR*/ |
| 4445 | || (INSN(31,0) & 0xFFFFFFE0) == 0xD53B4420 /*MRS*/) { |
| 4446 | Bool toSys = INSN(21,21) == 0; |
| 4447 | UInt tt = INSN(4,0); |
| 4448 | if (toSys) { |
| 4449 | stmt( IRStmt_Put( OFFB_FPSR, getIReg32orZR(tt)) ); |
| 4450 | DIP("msr fpsr, %s\n", nameIReg64orZR(tt)); |
| 4451 | } else { |
| 4452 | putIReg32orZR(tt, IRExpr_Get(OFFB_FPSR, Ity_I32)); |
| 4453 | DIP("mrs %s, fpsr\n", nameIReg64orZR(tt)); |
| 4454 | } |
| 4455 | return True; |
| 4456 | } |
| 4457 | /* Cases for NZCV |
| 4458 | D51B42 000 Rt MSR nzcv, rT |
| 4459 | D53B42 000 Rt MRS rT, nzcv |
| 4460 | */ |
| 4461 | if ( (INSN(31,0) & 0xFFFFFFE0) == 0xD51B4200 /*MSR*/ |
| 4462 | || (INSN(31,0) & 0xFFFFFFE0) == 0xD53B4200 /*MRS*/) { |
| 4463 | Bool toSys = INSN(21,21) == 0; |
| 4464 | UInt tt = INSN(4,0); |
| 4465 | if (toSys) { |
| 4466 | IRTemp t = newTemp(Ity_I64); |
| 4467 | assign(t, binop(Iop_And64, getIReg64orZR(tt), mkU64(0xF0000000ULL))); |
| 4468 | setFlags_COPY(t); |
| 4469 | DIP("msr %s, nzcv\n", nameIReg32orZR(tt)); |
| 4470 | } else { |
| 4471 | IRTemp res = newTemp(Ity_I64); |
| 4472 | assign(res, mk_arm64g_calculate_flags_nzcv()); |
| 4473 | putIReg32orZR(tt, unop(Iop_64to32, mkexpr(res))); |
| 4474 | DIP("mrs %s, nzcv\n", nameIReg64orZR(tt)); |
| 4475 | } |
| 4476 | return True; |
| 4477 | } |
sewardj | d512d10 | 2014-02-21 14:49:44 +0000 | [diff] [blame] | 4478 | /* Cases for DCZID_EL0 |
| 4479 | Don't support arbitrary reads and writes to this register. Just |
| 4480 | return the value 16, which indicates that the DC ZVA instruction |
| 4481 | is not permitted, so we don't have to emulate it. |
| 4482 | D5 3B 00 111 Rt MRS rT, dczid_el0 |
| 4483 | */ |
| 4484 | if ((INSN(31,0) & 0xFFFFFFE0) == 0xD53B00E0) { |
| 4485 | UInt tt = INSN(4,0); |
| 4486 | putIReg64orZR(tt, mkU64(1<<4)); |
| 4487 | DIP("mrs %s, dczid_el0 (FAKED)\n", nameIReg64orZR(tt)); |
| 4488 | return True; |
| 4489 | } |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4490 | |
sewardj | d512d10 | 2014-02-21 14:49:44 +0000 | [diff] [blame] | 4491 | /* ------------------ ISB, DSB ------------------ */ |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4492 | if (INSN(31,0) == 0xD5033FDF) { |
sewardj | d512d10 | 2014-02-21 14:49:44 +0000 | [diff] [blame] | 4493 | stmt(IRStmt_MBE(Imbe_Fence)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4494 | DIP("isb\n"); |
| 4495 | return True; |
| 4496 | } |
| 4497 | if (INSN(31,0) == 0xD5033BBF) { |
sewardj | d512d10 | 2014-02-21 14:49:44 +0000 | [diff] [blame] | 4498 | stmt(IRStmt_MBE(Imbe_Fence)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4499 | DIP("dmb ish\n"); |
| 4500 | return True; |
| 4501 | } |
| 4502 | |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 4503 | /* -------------------- NOP -------------------- */ |
| 4504 | if (INSN(31,0) == 0xD503201F) { |
| 4505 | DIP("nop\n"); |
| 4506 | return True; |
| 4507 | } |
| 4508 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4509 | //fail: |
| 4510 | vex_printf("ARM64 front end: branch_etc\n"); |
| 4511 | return False; |
| 4512 | # undef INSN |
| 4513 | } |
| 4514 | |
| 4515 | |
| 4516 | /*------------------------------------------------------------*/ |
| 4517 | /*--- SIMD and FP instructions ---*/ |
| 4518 | /*------------------------------------------------------------*/ |
| 4519 | |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 4520 | /* begin FIXME -- rm temp scaffolding */ |
| 4521 | static IRExpr* mk_CatEvenLanes64x2 ( IRTemp, IRTemp ); |
| 4522 | static IRExpr* mk_CatOddLanes64x2 ( IRTemp, IRTemp ); |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 4523 | |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 4524 | static IRExpr* mk_CatEvenLanes32x4 ( IRTemp, IRTemp ); |
| 4525 | static IRExpr* mk_CatOddLanes32x4 ( IRTemp, IRTemp ); |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 4526 | static IRExpr* mk_InterleaveLO32x4 ( IRTemp, IRTemp ); |
| 4527 | static IRExpr* mk_InterleaveHI32x4 ( IRTemp, IRTemp ); |
| 4528 | |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 4529 | static IRExpr* mk_CatEvenLanes16x8 ( IRTemp, IRTemp ); |
| 4530 | static IRExpr* mk_CatOddLanes16x8 ( IRTemp, IRTemp ); |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 4531 | static IRExpr* mk_InterleaveLO16x8 ( IRTemp, IRTemp ); |
| 4532 | static IRExpr* mk_InterleaveHI16x8 ( IRTemp, IRTemp ); |
| 4533 | |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 4534 | static IRExpr* mk_CatEvenLanes8x16 ( IRTemp, IRTemp ); |
| 4535 | static IRExpr* mk_CatOddLanes8x16 ( IRTemp, IRTemp ); |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 4536 | static IRExpr* mk_InterleaveLO8x16 ( IRTemp, IRTemp ); |
| 4537 | static IRExpr* mk_InterleaveHI8x16 ( IRTemp, IRTemp ); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 4538 | /* end FIXME -- rm temp scaffolding */ |
| 4539 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4540 | /* Generate N copies of |bit| in the bottom of a ULong. */ |
| 4541 | static ULong Replicate ( ULong bit, Int N ) |
| 4542 | { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4543 | vassert(bit <= 1 && N >= 1 && N < 64); |
| 4544 | if (bit == 0) { |
| 4545 | return 0; |
| 4546 | } else { |
| 4547 | /* Careful. This won't work for N == 64. */ |
| 4548 | return (1ULL << N) - 1; |
| 4549 | } |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4550 | } |
| 4551 | |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 4552 | static ULong Replicate32x2 ( ULong bits32 ) |
| 4553 | { |
| 4554 | vassert(0 == (bits32 & ~0xFFFFFFFFULL)); |
| 4555 | return (bits32 << 32) | bits32; |
| 4556 | } |
| 4557 | |
| 4558 | static ULong Replicate16x4 ( ULong bits16 ) |
| 4559 | { |
| 4560 | vassert(0 == (bits16 & ~0xFFFFULL)); |
| 4561 | return Replicate32x2((bits16 << 16) | bits16); |
| 4562 | } |
| 4563 | |
| 4564 | static ULong Replicate8x8 ( ULong bits8 ) |
| 4565 | { |
| 4566 | vassert(0 == (bits8 & ~0xFFULL)); |
| 4567 | return Replicate16x4((bits8 << 8) | bits8); |
| 4568 | } |
| 4569 | |
| 4570 | /* Expand the VFPExpandImm-style encoding in the bottom 8 bits of |
| 4571 | |imm8| to either a 32-bit value if N is 32 or a 64 bit value if N |
| 4572 | is 64. In the former case, the upper 32 bits of the returned value |
| 4573 | are guaranteed to be zero. */ |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4574 | static ULong VFPExpandImm ( ULong imm8, Int N ) |
| 4575 | { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4576 | vassert(imm8 <= 0xFF); |
| 4577 | vassert(N == 32 || N == 64); |
| 4578 | Int E = ((N == 32) ? 8 : 11) - 2; // The spec incorrectly omits the -2. |
| 4579 | Int F = N - E - 1; |
| 4580 | ULong imm8_6 = (imm8 >> 6) & 1; |
| 4581 | /* sign: 1 bit */ |
| 4582 | /* exp: E bits */ |
| 4583 | /* frac: F bits */ |
| 4584 | ULong sign = (imm8 >> 7) & 1; |
| 4585 | ULong exp = ((imm8_6 ^ 1) << (E-1)) | Replicate(imm8_6, E-1); |
| 4586 | ULong frac = ((imm8 & 63) << (F-6)) | Replicate(0, F-6); |
| 4587 | vassert(sign < (1ULL << 1)); |
| 4588 | vassert(exp < (1ULL << E)); |
| 4589 | vassert(frac < (1ULL << F)); |
| 4590 | vassert(1 + E + F == N); |
| 4591 | ULong res = (sign << (E+F)) | (exp << F) | frac; |
| 4592 | return res; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4593 | } |
| 4594 | |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 4595 | /* Expand an AdvSIMDExpandImm-style encoding into a 64-bit value. |
| 4596 | This might fail, as indicated by the returned Bool. Page 2530 of |
| 4597 | the manual. */ |
| 4598 | static Bool AdvSIMDExpandImm ( /*OUT*/ULong* res, |
| 4599 | UInt op, UInt cmode, UInt imm8 ) |
| 4600 | { |
| 4601 | vassert(op <= 1); |
| 4602 | vassert(cmode <= 15); |
| 4603 | vassert(imm8 <= 255); |
| 4604 | |
| 4605 | *res = 0; /* will overwrite iff returning True */ |
| 4606 | |
| 4607 | ULong imm64 = 0; |
| 4608 | Bool testimm8 = False; |
| 4609 | |
| 4610 | switch (cmode >> 1) { |
| 4611 | case 0: |
| 4612 | testimm8 = False; imm64 = Replicate32x2(imm8); break; |
| 4613 | case 1: |
| 4614 | testimm8 = True; imm64 = Replicate32x2(imm8 << 8); break; |
| 4615 | case 2: |
| 4616 | testimm8 = True; imm64 = Replicate32x2(imm8 << 16); break; |
| 4617 | case 3: |
| 4618 | testimm8 = True; imm64 = Replicate32x2(imm8 << 24); break; |
| 4619 | case 4: |
| 4620 | testimm8 = False; imm64 = Replicate16x4(imm8); break; |
| 4621 | case 5: |
| 4622 | testimm8 = True; imm64 = Replicate16x4(imm8 << 8); break; |
| 4623 | case 6: |
| 4624 | testimm8 = True; |
| 4625 | if ((cmode & 1) == 0) |
| 4626 | imm64 = Replicate32x2((imm8 << 8) | 0xFF); |
| 4627 | else |
| 4628 | imm64 = Replicate32x2((imm8 << 16) | 0xFFFF); |
| 4629 | break; |
| 4630 | case 7: |
| 4631 | testimm8 = False; |
| 4632 | if ((cmode & 1) == 0 && op == 0) |
| 4633 | imm64 = Replicate8x8(imm8); |
| 4634 | if ((cmode & 1) == 0 && op == 1) { |
| 4635 | imm64 = 0; imm64 |= (imm8 & 0x80) ? 0xFF : 0x00; |
| 4636 | imm64 <<= 8; imm64 |= (imm8 & 0x40) ? 0xFF : 0x00; |
| 4637 | imm64 <<= 8; imm64 |= (imm8 & 0x20) ? 0xFF : 0x00; |
| 4638 | imm64 <<= 8; imm64 |= (imm8 & 0x10) ? 0xFF : 0x00; |
| 4639 | imm64 <<= 8; imm64 |= (imm8 & 0x08) ? 0xFF : 0x00; |
| 4640 | imm64 <<= 8; imm64 |= (imm8 & 0x04) ? 0xFF : 0x00; |
| 4641 | imm64 <<= 8; imm64 |= (imm8 & 0x02) ? 0xFF : 0x00; |
| 4642 | imm64 <<= 8; imm64 |= (imm8 & 0x01) ? 0xFF : 0x00; |
| 4643 | } |
| 4644 | if ((cmode & 1) == 1 && op == 0) { |
| 4645 | ULong imm8_7 = (imm8 >> 7) & 1; |
| 4646 | ULong imm8_6 = (imm8 >> 6) & 1; |
| 4647 | ULong imm8_50 = imm8 & 63; |
| 4648 | ULong imm32 = (imm8_7 << (1 + 5 + 6 + 19)) |
| 4649 | | ((imm8_6 ^ 1) << (5 + 6 + 19)) |
| 4650 | | (Replicate(imm8_6, 5) << (6 + 19)) |
| 4651 | | (imm8_50 << 19); |
| 4652 | imm64 = Replicate32x2(imm32); |
| 4653 | } |
| 4654 | if ((cmode & 1) == 1 && op == 1) { |
| 4655 | // imm64 = imm8<7>:NOT(imm8<6>) |
| 4656 | // :Replicate(imm8<6>,8):imm8<5:0>:Zeros(48); |
| 4657 | ULong imm8_7 = (imm8 >> 7) & 1; |
| 4658 | ULong imm8_6 = (imm8 >> 6) & 1; |
| 4659 | ULong imm8_50 = imm8 & 63; |
| 4660 | imm64 = (imm8_7 << 63) | ((imm8_6 ^ 1) << 62) |
| 4661 | | (Replicate(imm8_6, 8) << 54) |
| 4662 | | (imm8_50 << 48); |
| 4663 | } |
| 4664 | break; |
| 4665 | default: |
| 4666 | vassert(0); |
| 4667 | } |
| 4668 | |
| 4669 | if (testimm8 && imm8 == 0) |
| 4670 | return False; |
| 4671 | |
| 4672 | *res = imm64; |
| 4673 | return True; |
| 4674 | } |
| 4675 | |
| 4676 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 4677 | /* Help a bit for decoding laneage for vector operations that can be |
| 4678 | of the form 4x32, 2x64 or 2x32-and-zero-upper-half, as encoded by Q |
| 4679 | and SZ bits, typically for vector floating point. */ |
| 4680 | static Bool getLaneInfo_Q_SZ ( /*OUT*/IRType* tyI, /*OUT*/IRType* tyF, |
| 4681 | /*OUT*/UInt* nLanes, /*OUT*/Bool* zeroUpper, |
| 4682 | /*OUT*/const HChar** arrSpec, |
| 4683 | Bool bitQ, Bool bitSZ ) |
| 4684 | { |
| 4685 | vassert(bitQ == True || bitQ == False); |
| 4686 | vassert(bitSZ == True || bitSZ == False); |
| 4687 | if (bitQ && bitSZ) { // 2x64 |
| 4688 | if (tyI) *tyI = Ity_I64; |
| 4689 | if (tyF) *tyF = Ity_F64; |
| 4690 | if (nLanes) *nLanes = 2; |
| 4691 | if (zeroUpper) *zeroUpper = False; |
| 4692 | if (arrSpec) *arrSpec = "2d"; |
| 4693 | return True; |
| 4694 | } |
| 4695 | if (bitQ && !bitSZ) { // 4x32 |
| 4696 | if (tyI) *tyI = Ity_I32; |
| 4697 | if (tyF) *tyF = Ity_F32; |
| 4698 | if (nLanes) *nLanes = 4; |
| 4699 | if (zeroUpper) *zeroUpper = False; |
| 4700 | if (arrSpec) *arrSpec = "4s"; |
| 4701 | return True; |
| 4702 | } |
| 4703 | if (!bitQ && !bitSZ) { // 2x32 |
| 4704 | if (tyI) *tyI = Ity_I32; |
| 4705 | if (tyF) *tyF = Ity_F32; |
| 4706 | if (nLanes) *nLanes = 2; |
| 4707 | if (zeroUpper) *zeroUpper = True; |
| 4708 | if (arrSpec) *arrSpec = "2s"; |
| 4709 | return True; |
| 4710 | } |
| 4711 | // Else impliedly 1x64, which isn't allowed. |
| 4712 | return False; |
| 4713 | } |
| 4714 | |
| 4715 | /* Helper for decoding laneage for simple vector operations, |
| 4716 | eg integer add. */ |
| 4717 | static Bool getLaneInfo_SIMPLE ( /*OUT*/Bool* zeroUpper, |
| 4718 | /*OUT*/const HChar** arrSpec, |
| 4719 | Bool bitQ, UInt szBlg2 ) |
| 4720 | { |
| 4721 | vassert(bitQ == True || bitQ == False); |
| 4722 | vassert(szBlg2 < 4); |
| 4723 | Bool zu = False; |
| 4724 | const HChar* as = NULL; |
| 4725 | switch ((szBlg2 << 1) | (bitQ ? 1 : 0)) { |
| 4726 | case 0: zu = True; as = "8b"; break; |
| 4727 | case 1: zu = False; as = "16b"; break; |
| 4728 | case 2: zu = True; as = "4h"; break; |
| 4729 | case 3: zu = False; as = "8h"; break; |
| 4730 | case 4: zu = True; as = "2s"; break; |
| 4731 | case 5: zu = False; as = "4s"; break; |
| 4732 | case 6: return False; // impliedly 1x64 |
| 4733 | case 7: zu = False; as = "2d"; break; |
| 4734 | default: vassert(0); |
| 4735 | } |
| 4736 | vassert(as); |
| 4737 | if (arrSpec) *arrSpec = as; |
| 4738 | if (zeroUpper) *zeroUpper = zu; |
| 4739 | return True; |
| 4740 | } |
| 4741 | |
| 4742 | |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 4743 | /* Helper for decoding laneage for shift-style vector operations |
| 4744 | that involve an immediate shift amount. */ |
| 4745 | static Bool getLaneInfo_IMMH_IMMB ( /*OUT*/UInt* shift, /*OUT*/UInt* szBlg2, |
| 4746 | UInt immh, UInt immb ) |
| 4747 | { |
| 4748 | vassert(immh < (1<<4)); |
| 4749 | vassert(immb < (1<<3)); |
| 4750 | UInt immhb = (immh << 3) | immb; |
| 4751 | if (immh & 8) { |
| 4752 | if (shift) *shift = 128 - immhb; |
| 4753 | if (szBlg2) *szBlg2 = 3; |
| 4754 | return True; |
| 4755 | } |
| 4756 | if (immh & 4) { |
| 4757 | if (shift) *shift = 64 - immhb; |
| 4758 | if (szBlg2) *szBlg2 = 2; |
| 4759 | return True; |
| 4760 | } |
| 4761 | if (immh & 2) { |
| 4762 | if (shift) *shift = 32 - immhb; |
| 4763 | if (szBlg2) *szBlg2 = 1; |
| 4764 | return True; |
| 4765 | } |
| 4766 | if (immh & 1) { |
| 4767 | if (shift) *shift = 16 - immhb; |
| 4768 | if (szBlg2) *szBlg2 = 0; |
| 4769 | return True; |
| 4770 | } |
| 4771 | return False; |
| 4772 | } |
| 4773 | |
| 4774 | |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 4775 | /* Generate IR to fold all lanes of the V128 value in 'src' as |
| 4776 | characterised by the operator 'op', and return the result in the |
| 4777 | bottom bits of a V128, with all other bits set to zero. */ |
| 4778 | static IRTemp math_MINMAXV ( IRTemp src, IROp op ) |
| 4779 | { |
| 4780 | /* The basic idea is to use repeated applications of Iop_CatEven* |
| 4781 | and Iop_CatOdd* operators to 'src' so as to clone each lane into |
| 4782 | a complete vector. Then fold all those vectors with 'op' and |
| 4783 | zero out all but the least significant lane. */ |
| 4784 | switch (op) { |
| 4785 | case Iop_Min8Sx16: case Iop_Min8Ux16: |
| 4786 | case Iop_Max8Sx16: case Iop_Max8Ux16: { |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 4787 | /* NB: temp naming here is misleading -- the naming is for 8 |
| 4788 | lanes of 16 bit, whereas what is being operated on is 16 |
| 4789 | lanes of 8 bits. */ |
| 4790 | IRTemp x76543210 = src; |
| 4791 | IRTemp x76547654 = newTemp(Ity_V128); |
| 4792 | IRTemp x32103210 = newTemp(Ity_V128); |
| 4793 | assign(x76547654, mk_CatOddLanes64x2 (x76543210, x76543210)); |
| 4794 | assign(x32103210, mk_CatEvenLanes64x2(x76543210, x76543210)); |
| 4795 | IRTemp x76767676 = newTemp(Ity_V128); |
| 4796 | IRTemp x54545454 = newTemp(Ity_V128); |
| 4797 | IRTemp x32323232 = newTemp(Ity_V128); |
| 4798 | IRTemp x10101010 = newTemp(Ity_V128); |
| 4799 | assign(x76767676, mk_CatOddLanes32x4 (x76547654, x76547654)); |
| 4800 | assign(x54545454, mk_CatEvenLanes32x4(x76547654, x76547654)); |
| 4801 | assign(x32323232, mk_CatOddLanes32x4 (x32103210, x32103210)); |
| 4802 | assign(x10101010, mk_CatEvenLanes32x4(x32103210, x32103210)); |
| 4803 | IRTemp x77777777 = newTemp(Ity_V128); |
| 4804 | IRTemp x66666666 = newTemp(Ity_V128); |
| 4805 | IRTemp x55555555 = newTemp(Ity_V128); |
| 4806 | IRTemp x44444444 = newTemp(Ity_V128); |
| 4807 | IRTemp x33333333 = newTemp(Ity_V128); |
| 4808 | IRTemp x22222222 = newTemp(Ity_V128); |
| 4809 | IRTemp x11111111 = newTemp(Ity_V128); |
| 4810 | IRTemp x00000000 = newTemp(Ity_V128); |
| 4811 | assign(x77777777, mk_CatOddLanes16x8 (x76767676, x76767676)); |
| 4812 | assign(x66666666, mk_CatEvenLanes16x8(x76767676, x76767676)); |
| 4813 | assign(x55555555, mk_CatOddLanes16x8 (x54545454, x54545454)); |
| 4814 | assign(x44444444, mk_CatEvenLanes16x8(x54545454, x54545454)); |
| 4815 | assign(x33333333, mk_CatOddLanes16x8 (x32323232, x32323232)); |
| 4816 | assign(x22222222, mk_CatEvenLanes16x8(x32323232, x32323232)); |
| 4817 | assign(x11111111, mk_CatOddLanes16x8 (x10101010, x10101010)); |
| 4818 | assign(x00000000, mk_CatEvenLanes16x8(x10101010, x10101010)); |
| 4819 | /* Naming not misleading after here. */ |
| 4820 | IRTemp xAllF = newTemp(Ity_V128); |
| 4821 | IRTemp xAllE = newTemp(Ity_V128); |
| 4822 | IRTemp xAllD = newTemp(Ity_V128); |
| 4823 | IRTemp xAllC = newTemp(Ity_V128); |
| 4824 | IRTemp xAllB = newTemp(Ity_V128); |
| 4825 | IRTemp xAllA = newTemp(Ity_V128); |
| 4826 | IRTemp xAll9 = newTemp(Ity_V128); |
| 4827 | IRTemp xAll8 = newTemp(Ity_V128); |
| 4828 | IRTemp xAll7 = newTemp(Ity_V128); |
| 4829 | IRTemp xAll6 = newTemp(Ity_V128); |
| 4830 | IRTemp xAll5 = newTemp(Ity_V128); |
| 4831 | IRTemp xAll4 = newTemp(Ity_V128); |
| 4832 | IRTemp xAll3 = newTemp(Ity_V128); |
| 4833 | IRTemp xAll2 = newTemp(Ity_V128); |
| 4834 | IRTemp xAll1 = newTemp(Ity_V128); |
| 4835 | IRTemp xAll0 = newTemp(Ity_V128); |
| 4836 | assign(xAllF, mk_CatOddLanes8x16 (x77777777, x77777777)); |
| 4837 | assign(xAllE, mk_CatEvenLanes8x16(x77777777, x77777777)); |
| 4838 | assign(xAllD, mk_CatOddLanes8x16 (x66666666, x66666666)); |
| 4839 | assign(xAllC, mk_CatEvenLanes8x16(x66666666, x66666666)); |
| 4840 | assign(xAllB, mk_CatOddLanes8x16 (x55555555, x55555555)); |
| 4841 | assign(xAllA, mk_CatEvenLanes8x16(x55555555, x55555555)); |
| 4842 | assign(xAll9, mk_CatOddLanes8x16 (x44444444, x44444444)); |
| 4843 | assign(xAll8, mk_CatEvenLanes8x16(x44444444, x44444444)); |
| 4844 | assign(xAll7, mk_CatOddLanes8x16 (x33333333, x33333333)); |
| 4845 | assign(xAll6, mk_CatEvenLanes8x16(x33333333, x33333333)); |
| 4846 | assign(xAll5, mk_CatOddLanes8x16 (x22222222, x22222222)); |
| 4847 | assign(xAll4, mk_CatEvenLanes8x16(x22222222, x22222222)); |
| 4848 | assign(xAll3, mk_CatOddLanes8x16 (x11111111, x11111111)); |
| 4849 | assign(xAll2, mk_CatEvenLanes8x16(x11111111, x11111111)); |
| 4850 | assign(xAll1, mk_CatOddLanes8x16 (x00000000, x00000000)); |
| 4851 | assign(xAll0, mk_CatEvenLanes8x16(x00000000, x00000000)); |
| 4852 | IRTemp maxFE = newTemp(Ity_V128); |
| 4853 | IRTemp maxDC = newTemp(Ity_V128); |
| 4854 | IRTemp maxBA = newTemp(Ity_V128); |
| 4855 | IRTemp max98 = newTemp(Ity_V128); |
| 4856 | IRTemp max76 = newTemp(Ity_V128); |
| 4857 | IRTemp max54 = newTemp(Ity_V128); |
| 4858 | IRTemp max32 = newTemp(Ity_V128); |
| 4859 | IRTemp max10 = newTemp(Ity_V128); |
| 4860 | assign(maxFE, binop(op, mkexpr(xAllF), mkexpr(xAllE))); |
| 4861 | assign(maxDC, binop(op, mkexpr(xAllD), mkexpr(xAllC))); |
| 4862 | assign(maxBA, binop(op, mkexpr(xAllB), mkexpr(xAllA))); |
| 4863 | assign(max98, binop(op, mkexpr(xAll9), mkexpr(xAll8))); |
| 4864 | assign(max76, binop(op, mkexpr(xAll7), mkexpr(xAll6))); |
| 4865 | assign(max54, binop(op, mkexpr(xAll5), mkexpr(xAll4))); |
| 4866 | assign(max32, binop(op, mkexpr(xAll3), mkexpr(xAll2))); |
| 4867 | assign(max10, binop(op, mkexpr(xAll1), mkexpr(xAll0))); |
| 4868 | IRTemp maxFEDC = newTemp(Ity_V128); |
| 4869 | IRTemp maxBA98 = newTemp(Ity_V128); |
| 4870 | IRTemp max7654 = newTemp(Ity_V128); |
| 4871 | IRTemp max3210 = newTemp(Ity_V128); |
| 4872 | assign(maxFEDC, binop(op, mkexpr(maxFE), mkexpr(maxDC))); |
| 4873 | assign(maxBA98, binop(op, mkexpr(maxBA), mkexpr(max98))); |
| 4874 | assign(max7654, binop(op, mkexpr(max76), mkexpr(max54))); |
| 4875 | assign(max3210, binop(op, mkexpr(max32), mkexpr(max10))); |
| 4876 | IRTemp maxFEDCBA98 = newTemp(Ity_V128); |
| 4877 | IRTemp max76543210 = newTemp(Ity_V128); |
| 4878 | assign(maxFEDCBA98, binop(op, mkexpr(maxFEDC), mkexpr(maxBA98))); |
| 4879 | assign(max76543210, binop(op, mkexpr(max7654), mkexpr(max3210))); |
| 4880 | IRTemp maxAllLanes = newTemp(Ity_V128); |
| 4881 | assign(maxAllLanes, binop(op, mkexpr(maxFEDCBA98), |
| 4882 | mkexpr(max76543210))); |
| 4883 | IRTemp res = newTemp(Ity_V128); |
| 4884 | assign(res, unop(Iop_ZeroHI120ofV128, mkexpr(maxAllLanes))); |
| 4885 | return res; |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 4886 | } |
| 4887 | case Iop_Min16Sx8: case Iop_Min16Ux8: |
| 4888 | case Iop_Max16Sx8: case Iop_Max16Ux8: { |
| 4889 | IRTemp x76543210 = src; |
| 4890 | IRTemp x76547654 = newTemp(Ity_V128); |
| 4891 | IRTemp x32103210 = newTemp(Ity_V128); |
| 4892 | assign(x76547654, mk_CatOddLanes64x2 (x76543210, x76543210)); |
| 4893 | assign(x32103210, mk_CatEvenLanes64x2(x76543210, x76543210)); |
| 4894 | IRTemp x76767676 = newTemp(Ity_V128); |
| 4895 | IRTemp x54545454 = newTemp(Ity_V128); |
| 4896 | IRTemp x32323232 = newTemp(Ity_V128); |
| 4897 | IRTemp x10101010 = newTemp(Ity_V128); |
| 4898 | assign(x76767676, mk_CatOddLanes32x4 (x76547654, x76547654)); |
| 4899 | assign(x54545454, mk_CatEvenLanes32x4(x76547654, x76547654)); |
| 4900 | assign(x32323232, mk_CatOddLanes32x4 (x32103210, x32103210)); |
| 4901 | assign(x10101010, mk_CatEvenLanes32x4(x32103210, x32103210)); |
| 4902 | IRTemp x77777777 = newTemp(Ity_V128); |
| 4903 | IRTemp x66666666 = newTemp(Ity_V128); |
| 4904 | IRTemp x55555555 = newTemp(Ity_V128); |
| 4905 | IRTemp x44444444 = newTemp(Ity_V128); |
| 4906 | IRTemp x33333333 = newTemp(Ity_V128); |
| 4907 | IRTemp x22222222 = newTemp(Ity_V128); |
| 4908 | IRTemp x11111111 = newTemp(Ity_V128); |
| 4909 | IRTemp x00000000 = newTemp(Ity_V128); |
| 4910 | assign(x77777777, mk_CatOddLanes16x8 (x76767676, x76767676)); |
| 4911 | assign(x66666666, mk_CatEvenLanes16x8(x76767676, x76767676)); |
| 4912 | assign(x55555555, mk_CatOddLanes16x8 (x54545454, x54545454)); |
| 4913 | assign(x44444444, mk_CatEvenLanes16x8(x54545454, x54545454)); |
| 4914 | assign(x33333333, mk_CatOddLanes16x8 (x32323232, x32323232)); |
| 4915 | assign(x22222222, mk_CatEvenLanes16x8(x32323232, x32323232)); |
| 4916 | assign(x11111111, mk_CatOddLanes16x8 (x10101010, x10101010)); |
| 4917 | assign(x00000000, mk_CatEvenLanes16x8(x10101010, x10101010)); |
| 4918 | IRTemp max76 = newTemp(Ity_V128); |
| 4919 | IRTemp max54 = newTemp(Ity_V128); |
| 4920 | IRTemp max32 = newTemp(Ity_V128); |
| 4921 | IRTemp max10 = newTemp(Ity_V128); |
| 4922 | assign(max76, binop(op, mkexpr(x77777777), mkexpr(x66666666))); |
| 4923 | assign(max54, binop(op, mkexpr(x55555555), mkexpr(x44444444))); |
| 4924 | assign(max32, binop(op, mkexpr(x33333333), mkexpr(x22222222))); |
| 4925 | assign(max10, binop(op, mkexpr(x11111111), mkexpr(x00000000))); |
| 4926 | IRTemp max7654 = newTemp(Ity_V128); |
| 4927 | IRTemp max3210 = newTemp(Ity_V128); |
| 4928 | assign(max7654, binop(op, mkexpr(max76), mkexpr(max54))); |
| 4929 | assign(max3210, binop(op, mkexpr(max32), mkexpr(max10))); |
| 4930 | IRTemp max76543210 = newTemp(Ity_V128); |
| 4931 | assign(max76543210, binop(op, mkexpr(max7654), mkexpr(max3210))); |
| 4932 | IRTemp res = newTemp(Ity_V128); |
| 4933 | assign(res, unop(Iop_ZeroHI112ofV128, mkexpr(max76543210))); |
| 4934 | return res; |
| 4935 | } |
| 4936 | case Iop_Min32Sx4: case Iop_Min32Ux4: |
| 4937 | case Iop_Max32Sx4: case Iop_Max32Ux4: { |
| 4938 | IRTemp x3210 = src; |
| 4939 | IRTemp x3232 = newTemp(Ity_V128); |
| 4940 | IRTemp x1010 = newTemp(Ity_V128); |
| 4941 | assign(x3232, mk_CatOddLanes64x2 (x3210, x3210)); |
| 4942 | assign(x1010, mk_CatEvenLanes64x2(x3210, x3210)); |
| 4943 | IRTemp x3333 = newTemp(Ity_V128); |
| 4944 | IRTemp x2222 = newTemp(Ity_V128); |
| 4945 | IRTemp x1111 = newTemp(Ity_V128); |
| 4946 | IRTemp x0000 = newTemp(Ity_V128); |
| 4947 | assign(x3333, mk_CatOddLanes32x4 (x3232, x3232)); |
| 4948 | assign(x2222, mk_CatEvenLanes32x4(x3232, x3232)); |
| 4949 | assign(x1111, mk_CatOddLanes32x4 (x1010, x1010)); |
| 4950 | assign(x0000, mk_CatEvenLanes32x4(x1010, x1010)); |
| 4951 | IRTemp max32 = newTemp(Ity_V128); |
| 4952 | IRTemp max10 = newTemp(Ity_V128); |
| 4953 | assign(max32, binop(op, mkexpr(x3333), mkexpr(x2222))); |
| 4954 | assign(max10, binop(op, mkexpr(x1111), mkexpr(x0000))); |
| 4955 | IRTemp max3210 = newTemp(Ity_V128); |
| 4956 | assign(max3210, binop(op, mkexpr(max32), mkexpr(max10))); |
| 4957 | IRTemp res = newTemp(Ity_V128); |
| 4958 | assign(res, unop(Iop_ZeroHI96ofV128, mkexpr(max3210))); |
| 4959 | return res; |
| 4960 | } |
| 4961 | default: |
| 4962 | vassert(0); |
| 4963 | } |
| 4964 | } |
| 4965 | |
| 4966 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 4967 | static |
| 4968 | Bool dis_ARM64_simd_and_fp(/*MB_OUT*/DisResult* dres, UInt insn) |
| 4969 | { |
| 4970 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 4971 | |
| 4972 | /* ---------------- FMOV (general) ---------------- */ |
| 4973 | /* case 30 23 20 18 15 9 4 |
| 4974 | (1) 0 00 11110 00 1 00 111 000000 n d FMOV Sd, Wn |
| 4975 | (2) 1 00 11110 01 1 00 111 000000 n d FMOV Dd, Xn |
| 4976 | (3) 1 00 11110 10 1 01 111 000000 n d FMOV Vd.D[1], Xn |
| 4977 | |
| 4978 | (4) 0 00 11110 00 1 00 110 000000 n d FMOV Wd, Sn |
| 4979 | (5) 1 00 11110 01 1 00 110 000000 n d FMOV Xd, Dn |
| 4980 | (6) 1 00 11110 10 1 01 110 000000 n d FMOV Xd, Vn.D[1] |
| 4981 | */ |
| 4982 | if (INSN(30,24) == BITS7(0,0,1,1,1,1,0) |
| 4983 | && INSN(21,21) == 1 && INSN(15,10) == BITS6(0,0,0,0,0,0)) { |
| 4984 | UInt sf = INSN(31,31); |
| 4985 | UInt ty = INSN(23,22); // type |
| 4986 | UInt rm = INSN(20,19); // rmode |
| 4987 | UInt op = INSN(18,16); // opcode |
| 4988 | UInt nn = INSN(9,5); |
| 4989 | UInt dd = INSN(4,0); |
| 4990 | UInt ix = 0; // case |
| 4991 | if (sf == 0) { |
| 4992 | if (ty == BITS2(0,0) && rm == BITS2(0,0) && op == BITS3(1,1,1)) |
| 4993 | ix = 1; |
| 4994 | else |
| 4995 | if (ty == BITS2(0,0) && rm == BITS2(0,0) && op == BITS3(1,1,0)) |
| 4996 | ix = 4; |
| 4997 | } else { |
| 4998 | vassert(sf == 1); |
| 4999 | if (ty == BITS2(0,1) && rm == BITS2(0,0) && op == BITS3(1,1,1)) |
| 5000 | ix = 2; |
| 5001 | else |
| 5002 | if (ty == BITS2(0,1) && rm == BITS2(0,0) && op == BITS3(1,1,0)) |
| 5003 | ix = 5; |
| 5004 | else |
| 5005 | if (ty == BITS2(1,0) && rm == BITS2(0,1) && op == BITS3(1,1,1)) |
| 5006 | ix = 3; |
| 5007 | else |
| 5008 | if (ty == BITS2(1,0) && rm == BITS2(0,1) && op == BITS3(1,1,0)) |
| 5009 | ix = 6; |
| 5010 | } |
| 5011 | if (ix > 0) { |
| 5012 | switch (ix) { |
| 5013 | case 1: |
| 5014 | putQReg128(dd, mkV128(0)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5015 | putQRegLO(dd, getIReg32orZR(nn)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5016 | DIP("fmov s%u, w%u\n", dd, nn); |
| 5017 | break; |
| 5018 | case 2: |
| 5019 | putQReg128(dd, mkV128(0)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5020 | putQRegLO(dd, getIReg64orZR(nn)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5021 | DIP("fmov d%u, x%u\n", dd, nn); |
| 5022 | break; |
| 5023 | case 3: |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5024 | putQRegHI64(dd, getIReg64orZR(nn)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5025 | DIP("fmov v%u.d[1], x%u\n", dd, nn); |
| 5026 | break; |
| 5027 | case 4: |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5028 | putIReg32orZR(dd, getQRegLO(nn, Ity_I32)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5029 | DIP("fmov w%u, s%u\n", dd, nn); |
| 5030 | break; |
| 5031 | case 5: |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5032 | putIReg64orZR(dd, getQRegLO(nn, Ity_I64)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5033 | DIP("fmov x%u, d%u\n", dd, nn); |
| 5034 | break; |
| 5035 | case 6: |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5036 | putIReg64orZR(dd, getQRegHI64(nn)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5037 | DIP("fmov x%u, v%u.d[1]\n", dd, nn); |
| 5038 | break; |
| 5039 | default: |
| 5040 | vassert(0); |
| 5041 | } |
| 5042 | return True; |
| 5043 | } |
| 5044 | /* undecodable; fall through */ |
| 5045 | } |
| 5046 | |
| 5047 | /* -------------- FMOV (scalar, immediate) -------------- */ |
| 5048 | /* 31 28 23 20 12 9 4 |
| 5049 | 000 11110 00 1 imm8 100 00000 d FMOV Sd, #imm |
| 5050 | 000 11110 01 1 imm8 100 00000 d FMOV Dd, #imm |
| 5051 | */ |
| 5052 | if (INSN(31,23) == BITS9(0,0,0,1,1,1,1,0,0) |
| 5053 | && INSN(21,21) == 1 && INSN(12,5) == BITS8(1,0,0,0,0,0,0,0)) { |
| 5054 | Bool isD = INSN(22,22) == 1; |
| 5055 | UInt imm8 = INSN(20,13); |
| 5056 | UInt dd = INSN(4,0); |
| 5057 | ULong imm = VFPExpandImm(imm8, isD ? 64 : 32); |
| 5058 | if (!isD) { |
sewardj | aeeb31d | 2014-01-12 18:23:45 +0000 | [diff] [blame] | 5059 | vassert(0 == (imm & 0xFFFFFFFF00000000ULL)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5060 | } |
| 5061 | putQReg128(dd, mkV128(0)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5062 | putQRegLO(dd, isD ? mkU64(imm) : mkU32(imm & 0xFFFFFFFFULL)); |
| 5063 | DIP("fmov %s, #0x%llx\n", |
| 5064 | nameQRegLO(dd, isD ? Ity_F64 : Ity_F32), imm); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5065 | return True; |
| 5066 | } |
| 5067 | |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5068 | #if 0 |
| 5069 | /* -------------- FMOV (vector, immediate) -------------- */ |
| 5070 | /* 31 28 18 15 9 4 |
| 5071 | 011 01111 00000 abc 111101 defgh d FMOV Vd.2d, #imm |
| 5072 | 0q0 01111 00000 abc 111101 defgh d FMOV Vd.2s, #imm (q=0) |
| 5073 | FMOV Vd.4s, #imm (q=1) |
| 5074 | */ |
| 5075 | if (INSN(31,31) == 0 |
| 5076 | && INSN(28,19) == BITS10(0,1,1,1,1,0,0,0,0,0) |
| 5077 | && INSN(15,10) == BITS6(1,1,1,1,0,1) |
| 5078 | && INSN(30,29) != BITS2(0,1)) { |
| 5079 | UInt bitQ = INSN(30,30); |
| 5080 | UInt bitOP = INSN(29,29); |
| 5081 | UInt cmode = INSN(15,12); |
| 5082 | UInt imm8 = (INSN(18,16) << 5) | INSN(9,5); |
| 5083 | UInt dd = INSN(4,0); |
| 5084 | ULong imm64lo = 0; |
| 5085 | Bool ok = AdvSIMDExpandImm(&imm64lo, bitOP, cmode, imm8); |
| 5086 | vassert(! (bitOP == 1 && bitQ == 0) ); |
| 5087 | if (ok) { |
| 5088 | ULong imm64hi = (bitQ == 0 && bitOP == 0) ? 0 : imm64lo; |
| 5089 | putQReg128(dd, binop(Iop_64HLtoV128, mkU64(imm64hi), mkU64(imm64lo))); |
| 5090 | const HChar* ar[4] = { "2s", "??", "4s", "2d" }; |
| 5091 | DIP("fmov %s.%s, #0x%llx\n", |
| 5092 | nameQReg128(dd), ar[INSN(30,29)], imm64lo); |
| 5093 | return True; |
| 5094 | } |
| 5095 | /* else fall through */ |
| 5096 | } |
| 5097 | #else |
| 5098 | /* -------------- {FMOV,MOVI} (vector, immediate) -------------- */ |
| 5099 | /* 31 28 18 15 11 9 4 |
| 5100 | 0q op 01111 00000 abc cmode 01 defgh d MOV Dd, #imm (q=0) |
| 5101 | MOV Vd.2d #imm (q=1) |
| 5102 | Allowable op:cmode |
| 5103 | FMOV = 1:1111 |
| 5104 | MOVI = 0:xx00, 1:0x00, 1:10x0, 1:110x, 11110 |
| 5105 | */ |
| 5106 | if (INSN(31,31) == 0 |
| 5107 | && INSN(28,19) == BITS10(0,1,1,1,1,0,0,0,0,0) |
| 5108 | && INSN(11,10) == BITS2(0,1)) { |
| 5109 | UInt bitQ = INSN(30,30); |
| 5110 | UInt bitOP = INSN(29,29); |
| 5111 | UInt cmode = INSN(15,12); |
| 5112 | UInt imm8 = (INSN(18,16) << 5) | INSN(9,5); |
| 5113 | UInt dd = INSN(4,0); |
| 5114 | ULong imm64lo = 0; |
| 5115 | UInt op_cmode = (bitOP << 4) | cmode; |
| 5116 | Bool ok = False; |
| 5117 | switch (op_cmode) { |
| 5118 | case BITS5(1,1,1,1,1): // 1:1111 |
| 5119 | case BITS5(0,0,0,0,0): case BITS5(0,0,1,0,0): |
| 5120 | case BITS5(0,1,0,0,0): case BITS5(0,1,1,0,0): // 0:xx00 |
| 5121 | case BITS5(1,0,0,0,0): case BITS5(1,0,1,0,0): // 1:0x00 |
| 5122 | case BITS5(1,1,0,0,0): case BITS5(1,1,0,1,0): // 1:10x0 |
| 5123 | case BITS5(1,1,1,0,0): case BITS5(1,1,1,0,1): // 1:110x |
| 5124 | case BITS5(1,1,1,1,0): // 1:1110 |
| 5125 | ok = True; break; |
| 5126 | default: |
| 5127 | break; |
| 5128 | } |
| 5129 | if (ok) { |
| 5130 | ok = AdvSIMDExpandImm(&imm64lo, bitOP, cmode, imm8); |
| 5131 | } |
| 5132 | if (ok) { |
| 5133 | ULong imm64hi = (bitQ == 0 && bitOP == 0) ? 0 : imm64lo; |
| 5134 | putQReg128(dd, binop(Iop_64HLtoV128, mkU64(imm64hi), mkU64(imm64lo))); |
| 5135 | DIP("mov %s, #0x016%llx'%016llx\n", nameQReg128(dd), imm64hi, imm64lo); |
| 5136 | return True; |
| 5137 | } |
| 5138 | /* else fall through */ |
| 5139 | } |
| 5140 | #endif |
| 5141 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5142 | /* -------------- {S,U}CVTF (scalar, integer) -------------- */ |
| 5143 | /* 31 28 23 21 20 18 15 9 4 ix |
| 5144 | 000 11110 00 1 00 010 000000 n d SCVTF Sd, Wn 0 |
| 5145 | 000 11110 01 1 00 010 000000 n d SCVTF Dd, Wn 1 |
| 5146 | 100 11110 00 1 00 010 000000 n d SCVTF Sd, Xn 2 |
| 5147 | 100 11110 01 1 00 010 000000 n d SCVTF Dd, Xn 3 |
| 5148 | |
| 5149 | 000 11110 00 1 00 011 000000 n d UCVTF Sd, Wn 4 |
| 5150 | 000 11110 01 1 00 011 000000 n d UCVTF Dd, Wn 5 |
| 5151 | 100 11110 00 1 00 011 000000 n d UCVTF Sd, Xn 6 |
| 5152 | 100 11110 01 1 00 011 000000 n d UCVTF Dd, Xn 7 |
| 5153 | |
| 5154 | These are signed/unsigned conversion from integer registers to |
| 5155 | FP registers, all 4 32/64-bit combinations, rounded per FPCR. |
| 5156 | */ |
| 5157 | if (INSN(30,23) == BITS8(0,0,1,1,1,1,0,0) && INSN(21,17) == BITS5(1,0,0,0,1) |
| 5158 | && INSN(15,10) == BITS6(0,0,0,0,0,0)) { |
| 5159 | Bool isI64 = INSN(31,31) == 1; |
| 5160 | Bool isF64 = INSN(22,22) == 1; |
| 5161 | Bool isU = INSN(16,16) == 1; |
| 5162 | UInt nn = INSN(9,5); |
| 5163 | UInt dd = INSN(4,0); |
| 5164 | UInt ix = (isU ? 4 : 0) | (isI64 ? 2 : 0) | (isF64 ? 1 : 0); |
| 5165 | const IROp ops[8] |
| 5166 | = { Iop_I32StoF32, Iop_I32StoF64, Iop_I64StoF32, Iop_I64StoF64, |
| 5167 | Iop_I32UtoF32, Iop_I32UtoF64, Iop_I64UtoF32, Iop_I64UtoF64 }; |
| 5168 | IRExpr* src = getIRegOrZR(isI64, nn); |
| 5169 | IRExpr* res = (isF64 && !isI64) |
| 5170 | ? unop(ops[ix], src) |
| 5171 | : binop(ops[ix], mkexpr(mk_get_IR_rounding_mode()), src); |
| 5172 | putQReg128(dd, mkV128(0)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5173 | putQRegLO(dd, res); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5174 | DIP("%ccvtf %s, %s\n", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5175 | isU ? 'u' : 's', nameQRegLO(dd, isF64 ? Ity_F64 : Ity_F32), |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5176 | nameIRegOrZR(isI64, nn)); |
| 5177 | return True; |
| 5178 | } |
| 5179 | |
| 5180 | /* -------------- F{ADD,SUB,MUL,DIV} (scalar) -------------- */ |
| 5181 | /* 31 23 20 15 11 9 4 |
| 5182 | ---------------- 0000 ------ FMUL -------- |
| 5183 | 000 11110 001 m 0001 10 n d FDIV Sd,Sn,Sm |
| 5184 | 000 11110 011 m 0001 10 n d FDIV Dd,Dn,Dm |
| 5185 | ---------------- 0010 ------ FADD -------- |
| 5186 | ---------------- 0011 ------ FSUB -------- |
| 5187 | ---------------- 1000 ------ FNMUL -------- |
| 5188 | */ |
| 5189 | if (INSN(31,23) == BITS9(0,0,0,1,1,1,1,0,0) |
| 5190 | && INSN(21,21) == 1 && INSN(11,10) == BITS2(1,0)) { |
| 5191 | Bool isD = INSN(22,22) == 1; |
| 5192 | UInt mm = INSN(20,16); |
| 5193 | UInt op = INSN(15,12); |
| 5194 | UInt nn = INSN(9,5); |
| 5195 | UInt dd = INSN(4,0); |
| 5196 | IROp iop = Iop_INVALID; |
| 5197 | IRType ty = isD ? Ity_F64 : Ity_F32; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5198 | Bool neg = False; |
| 5199 | const HChar* nm = "???"; |
| 5200 | switch (op) { |
| 5201 | case BITS4(0,0,0,0): nm = "fmul"; iop = mkMULF(ty); break; |
| 5202 | case BITS4(0,0,0,1): nm = "fdiv"; iop = mkDIVF(ty); break; |
| 5203 | case BITS4(0,0,1,0): nm = "fadd"; iop = mkADDF(ty); break; |
| 5204 | case BITS4(0,0,1,1): nm = "fsub"; iop = mkSUBF(ty); break; |
| 5205 | case BITS4(1,0,0,0): nm = "fnmul"; iop = mkMULF(ty); |
| 5206 | neg = True; break; |
| 5207 | default: return False; |
| 5208 | } |
| 5209 | vassert(iop != Iop_INVALID); |
| 5210 | IRExpr* resE = triop(iop, mkexpr(mk_get_IR_rounding_mode()), |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5211 | getQRegLO(nn, ty), getQRegLO(mm, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5212 | IRTemp res = newTemp(ty); |
| 5213 | assign(res, neg ? unop(mkNEGF(ty),resE) : resE); |
| 5214 | putQReg128(dd, mkV128(0)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5215 | putQRegLO(dd, mkexpr(res)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5216 | DIP("%s %s, %s, %s\n", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5217 | nm, nameQRegLO(dd, ty), nameQRegLO(nn, ty), nameQRegLO(mm, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5218 | return True; |
| 5219 | } |
| 5220 | |
| 5221 | /* ------------ F{MOV,ABS,NEG,SQRT} D/D or S/S ------------ */ |
| 5222 | /* 31 23 21 16 14 9 4 |
| 5223 | 000 11110 00 10000 00 10000 n d FMOV Sd, Sn |
| 5224 | 000 11110 01 10000 00 10000 n d FMOV Dd, Dn |
| 5225 | ------------------ 01 --------- FABS ------ |
| 5226 | ------------------ 10 --------- FNEG ------ |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5227 | ------------------ 11 --------- FSQRT ----- |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5228 | */ |
| 5229 | if (INSN(31,23) == BITS9(0,0,0,1,1,1,1,0,0) |
| 5230 | && INSN(21,17) == BITS5(1,0,0,0,0) |
| 5231 | && INSN(14,10) == BITS5(1,0,0,0,0)) { |
| 5232 | Bool isD = INSN(22,22) == 1; |
| 5233 | UInt opc = INSN(16,15); |
| 5234 | UInt nn = INSN(9,5); |
| 5235 | UInt dd = INSN(4,0); |
| 5236 | IRType ty = isD ? Ity_F64 : Ity_F32; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5237 | IRTemp res = newTemp(ty); |
| 5238 | if (opc == BITS2(0,0)) { |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5239 | assign(res, getQRegLO(nn, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5240 | putQReg128(dd, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5241 | putQRegLO(dd, mkexpr(res)); |
| 5242 | DIP("fmov %s, %s\n", |
| 5243 | nameQRegLO(dd, ty), nameQRegLO(nn, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5244 | return True; |
| 5245 | } |
| 5246 | if (opc == BITS2(1,0) || opc == BITS2(0,1)) { |
| 5247 | Bool isAbs = opc == BITS2(0,1); |
| 5248 | IROp op = isAbs ? mkABSF(ty) : mkNEGF(ty); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5249 | assign(res, unop(op, getQRegLO(nn, ty))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5250 | putQReg128(dd, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5251 | putQRegLO(dd, mkexpr(res)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5252 | DIP("%s %s, %s\n", isAbs ? "fabs" : "fneg", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5253 | nameQRegLO(dd, ty), nameQRegLO(nn, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5254 | return True; |
| 5255 | } |
| 5256 | if (opc == BITS2(1,1)) { |
| 5257 | assign(res, |
| 5258 | binop(mkSQRTF(ty), |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5259 | mkexpr(mk_get_IR_rounding_mode()), getQRegLO(nn, ty))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5260 | putQReg128(dd, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5261 | putQRegLO(dd, mkexpr(res)); |
| 5262 | DIP("fsqrt %s, %s\n", nameQRegLO(dd, ty), nameQRegLO(nn, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5263 | return True; |
| 5264 | } |
| 5265 | /* else fall through; other cases are ATC */ |
| 5266 | } |
| 5267 | |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5268 | /* ---------------- F{ABS,NEG} (vector) ---------------- */ |
| 5269 | /* 31 28 22 21 16 9 4 |
| 5270 | 0q0 01110 1 sz 10000 01111 10 n d FABS Vd.T, Vn.T |
| 5271 | 0q1 01110 1 sz 10000 01111 10 n d FNEG Vd.T, Vn.T |
| 5272 | */ |
| 5273 | if (INSN(31,31) == 0 && INSN(28,23) == BITS6(0,1,1,1,0,1) |
| 5274 | && INSN(21,17) == BITS5(1,0,0,0,0) |
| 5275 | && INSN(16,10) == BITS7(0,1,1,1,1,1,0)) { |
| 5276 | UInt bitQ = INSN(30,30); |
| 5277 | UInt bitSZ = INSN(22,22); |
| 5278 | Bool isFNEG = INSN(29,29) == 1; |
| 5279 | UInt nn = INSN(9,5); |
| 5280 | UInt dd = INSN(4,0); |
| 5281 | const HChar* ar = "??"; |
| 5282 | IRType tyF = Ity_INVALID; |
| 5283 | Bool zeroHI = False; |
| 5284 | Bool ok = getLaneInfo_Q_SZ(NULL, &tyF, NULL, &zeroHI, &ar, |
| 5285 | (Bool)bitQ, (Bool)bitSZ); |
| 5286 | if (ok) { |
| 5287 | vassert(tyF == Ity_F64 || tyF == Ity_I32); |
| 5288 | IROp op = (tyF == Ity_F64) ? (isFNEG ? Iop_Neg64Fx2 : Iop_Abs64Fx2) |
| 5289 | : (isFNEG ? Iop_Neg32Fx4 : Iop_Abs32Fx4); |
| 5290 | IRTemp res = newTemp(Ity_V128); |
| 5291 | assign(res, unop(op, getQReg128(nn))); |
| 5292 | putQReg128(dd, zeroHI ? unop(Iop_ZeroHI64ofV128, mkexpr(res)) |
| 5293 | : mkexpr(res)); |
| 5294 | DIP("%s %s.%s, %s.%s\n", isFNEG ? "fneg" : "fabs", |
| 5295 | nameQReg128(dd), ar, nameQReg128(nn), ar); |
| 5296 | return True; |
| 5297 | } |
| 5298 | /* else fall through */ |
| 5299 | } |
| 5300 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5301 | /* -------------------- FCMP,FCMPE -------------------- */ |
| 5302 | /* 31 23 20 15 9 4 |
| 5303 | 000 11110 01 1 m 00 1000 n 10 000 FCMPE Dn, Dm |
| 5304 | 000 11110 01 1 00000 00 1000 n 11 000 FCMPE Dn, #0.0 |
| 5305 | 000 11110 01 1 m 00 1000 n 00 000 FCMP Dn, Dm |
| 5306 | 000 11110 01 1 00000 00 1000 n 01 000 FCMP Dn, #0.0 |
| 5307 | |
| 5308 | 000 11110 00 1 m 00 1000 n 10 000 FCMPE Sn, Sm |
| 5309 | 000 11110 00 1 00000 00 1000 n 11 000 FCMPE Sn, #0.0 |
| 5310 | 000 11110 00 1 m 00 1000 n 00 000 FCMP Sn, Sm |
| 5311 | 000 11110 00 1 00000 00 1000 n 01 000 FCMP Sn, #0.0 |
| 5312 | |
| 5313 | FCMPE generates Invalid Operation exn if either arg is any kind |
| 5314 | of NaN. FCMP generates Invalid Operation exn if either arg is a |
| 5315 | signalling NaN. We ignore this detail here and produce the same |
| 5316 | IR for both. |
| 5317 | */ |
| 5318 | if (INSN(31,23) == BITS9(0,0,0,1,1,1,1,0,0) && INSN(21,21) == 1 |
| 5319 | && INSN(15,10) == BITS6(0,0,1,0,0,0) && INSN(2,0) == BITS3(0,0,0)) { |
| 5320 | Bool isD = INSN(22,22) == 1; |
| 5321 | UInt mm = INSN(20,16); |
| 5322 | UInt nn = INSN(9,5); |
| 5323 | Bool isCMPE = INSN(4,4) == 1; |
| 5324 | Bool cmpZero = INSN(3,3) == 1; |
| 5325 | IRType ty = isD ? Ity_F64 : Ity_F32; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5326 | Bool valid = True; |
| 5327 | if (cmpZero && mm != 0) valid = False; |
| 5328 | if (valid) { |
| 5329 | IRTemp argL = newTemp(ty); |
| 5330 | IRTemp argR = newTemp(ty); |
| 5331 | IRTemp irRes = newTemp(Ity_I32); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5332 | assign(argL, getQRegLO(nn, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5333 | assign(argR, |
| 5334 | cmpZero |
| 5335 | ? (IRExpr_Const(isD ? IRConst_F64i(0) : IRConst_F32i(0))) |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5336 | : getQRegLO(mm, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5337 | assign(irRes, binop(isD ? Iop_CmpF64 : Iop_CmpF32, |
| 5338 | mkexpr(argL), mkexpr(argR))); |
| 5339 | IRTemp nzcv = mk_convert_IRCmpF64Result_to_NZCV(irRes); |
| 5340 | IRTemp nzcv_28x0 = newTemp(Ity_I64); |
| 5341 | assign(nzcv_28x0, binop(Iop_Shl64, mkexpr(nzcv), mkU8(28))); |
| 5342 | setFlags_COPY(nzcv_28x0); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5343 | DIP("fcmp%s %s, %s\n", isCMPE ? "e" : "", nameQRegLO(nn, ty), |
| 5344 | cmpZero ? "#0.0" : nameQRegLO(mm, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5345 | return True; |
| 5346 | } |
| 5347 | } |
| 5348 | |
| 5349 | /* -------------------- F{N}M{ADD,SUB} -------------------- */ |
| 5350 | /* 31 22 20 15 14 9 4 ix |
| 5351 | 000 11111 0 sz 0 m 0 a n d 0 FMADD Fd,Fn,Fm,Fa |
| 5352 | 000 11111 0 sz 0 m 1 a n d 1 FMSUB Fd,Fn,Fm,Fa |
| 5353 | 000 11111 0 sz 1 m 0 a n d 2 FNMADD Fd,Fn,Fm,Fa |
| 5354 | 000 11111 0 sz 1 m 1 a n d 3 FNMSUB Fd,Fn,Fm,Fa |
| 5355 | where Fx=Dx when sz=1, Fx=Sx when sz=0 |
| 5356 | |
| 5357 | -----SPEC------ ----IMPL---- |
| 5358 | fmadd a + n * m a + n * m |
| 5359 | fmsub a + (-n) * m a - n * m |
| 5360 | fnmadd (-a) + (-n) * m -(a + n * m) |
| 5361 | fnmsub (-a) + n * m -(a - n * m) |
| 5362 | */ |
| 5363 | if (INSN(31,23) == BITS9(0,0,0,1,1,1,1,1,0)) { |
| 5364 | Bool isD = INSN(22,22) == 1; |
| 5365 | UInt mm = INSN(20,16); |
| 5366 | UInt aa = INSN(14,10); |
| 5367 | UInt nn = INSN(9,5); |
| 5368 | UInt dd = INSN(4,0); |
| 5369 | UInt ix = (INSN(21,21) << 1) | INSN(15,15); |
| 5370 | IRType ty = isD ? Ity_F64 : Ity_F32; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5371 | IROp opADD = mkADDF(ty); |
| 5372 | IROp opSUB = mkSUBF(ty); |
| 5373 | IROp opMUL = mkMULF(ty); |
| 5374 | IROp opNEG = mkNEGF(ty); |
| 5375 | IRTemp res = newTemp(ty); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5376 | IRExpr* eA = getQRegLO(aa, ty); |
| 5377 | IRExpr* eN = getQRegLO(nn, ty); |
| 5378 | IRExpr* eM = getQRegLO(mm, ty); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5379 | IRExpr* rm = mkexpr(mk_get_IR_rounding_mode()); |
| 5380 | IRExpr* eNxM = triop(opMUL, rm, eN, eM); |
| 5381 | switch (ix) { |
| 5382 | case 0: assign(res, triop(opADD, rm, eA, eNxM)); break; |
| 5383 | case 1: assign(res, triop(opSUB, rm, eA, eNxM)); break; |
| 5384 | case 2: assign(res, unop(opNEG, triop(opADD, rm, eA, eNxM))); break; |
| 5385 | case 3: assign(res, unop(opNEG, triop(opSUB, rm, eA, eNxM))); break; |
| 5386 | default: vassert(0); |
| 5387 | } |
| 5388 | putQReg128(dd, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5389 | putQRegLO(dd, mkexpr(res)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5390 | const HChar* names[4] = { "fmadd", "fmsub", "fnmadd", "fnmsub" }; |
| 5391 | DIP("%s %s, %s, %s, %s\n", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5392 | names[ix], nameQRegLO(dd, ty), nameQRegLO(nn, ty), |
| 5393 | nameQRegLO(mm, ty), nameQRegLO(aa, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5394 | return True; |
| 5395 | } |
| 5396 | |
| 5397 | /* -------- FCVT{N,P,M,Z}{S,U} (scalar, integer) -------- */ |
| 5398 | /* 30 23 20 18 15 9 4 |
| 5399 | sf 00 11110 0x 1 00 000 000000 n d FCVTNS Rd, Fn (round to |
| 5400 | sf 00 11110 0x 1 00 001 000000 n d FCVTNU Rd, Fn nearest) |
| 5401 | ---------------- 01 -------------- FCVTP-------- (round to +inf) |
| 5402 | ---------------- 10 -------------- FCVTM-------- (round to -inf) |
| 5403 | ---------------- 11 -------------- FCVTZ-------- (round to zero) |
| 5404 | |
| 5405 | Rd is Xd when sf==1, Wd when sf==0 |
| 5406 | Fn is Dn when x==1, Sn when x==0 |
| 5407 | 20:19 carry the rounding mode, using the same encoding as FPCR |
| 5408 | */ |
| 5409 | if (INSN(30,23) == BITS8(0,0,1,1,1,1,0,0) && INSN(21,21) == 1 |
| 5410 | && INSN(18,17) == BITS2(0,0) && INSN(15,10) == BITS6(0,0,0,0,0,0)) { |
| 5411 | Bool isI64 = INSN(31,31) == 1; |
| 5412 | Bool isF64 = INSN(22,22) == 1; |
| 5413 | UInt rm = INSN(20,19); |
| 5414 | Bool isU = INSN(16,16) == 1; |
| 5415 | UInt nn = INSN(9,5); |
| 5416 | UInt dd = INSN(4,0); |
| 5417 | /* Decide on the IR rounding mode to use. */ |
| 5418 | IRRoundingMode irrm = 8; /*impossible*/ |
| 5419 | HChar ch = '?'; |
| 5420 | switch (rm) { |
| 5421 | case BITS2(0,0): ch = 'n'; irrm = Irrm_NEAREST; break; |
| 5422 | case BITS2(0,1): ch = 'p'; irrm = Irrm_PosINF; break; |
| 5423 | case BITS2(1,0): ch = 'm'; irrm = Irrm_NegINF; break; |
| 5424 | case BITS2(1,1): ch = 'z'; irrm = Irrm_ZERO; break; |
| 5425 | default: vassert(0); |
| 5426 | } |
| 5427 | vassert(irrm != 8); |
| 5428 | /* Decide on the conversion primop, based on the source size, |
| 5429 | dest size and signedness (8 possibilities). Case coding: |
| 5430 | F32 ->s I32 0 |
| 5431 | F32 ->u I32 1 |
| 5432 | F32 ->s I64 2 |
| 5433 | F32 ->u I64 3 |
| 5434 | F64 ->s I32 4 |
| 5435 | F64 ->u I32 5 |
| 5436 | F64 ->s I64 6 |
| 5437 | F64 ->u I64 7 |
| 5438 | */ |
| 5439 | UInt ix = (isF64 ? 4 : 0) | (isI64 ? 2 : 0) | (isU ? 1 : 0); |
| 5440 | vassert(ix < 8); |
| 5441 | const IROp ops[8] |
| 5442 | = { Iop_F32toI32S, Iop_F32toI32U, Iop_F32toI64S, Iop_F32toI64U, |
| 5443 | Iop_F64toI32S, Iop_F64toI32U, Iop_F64toI64S, Iop_F64toI64U }; |
| 5444 | IROp op = ops[ix]; |
| 5445 | // A bit of ATCery: bounce all cases we haven't seen an example of. |
| 5446 | if (/* F32toI32S */ |
| 5447 | (op == Iop_F32toI32S && irrm == Irrm_ZERO) /* FCVTZS Wd,Sn */ |
| 5448 | /* F32toI32U */ |
| 5449 | /* F32toI64S */ |
| 5450 | /* F32toI64U */ |
| 5451 | || (op == Iop_F32toI64U && irrm == Irrm_ZERO) /* FCVTZU Xd,Sn */ |
| 5452 | /* F64toI32S */ |
| 5453 | || (op == Iop_F64toI32S && irrm == Irrm_ZERO) /* FCVTZS Wd,Dn */ |
| 5454 | || (op == Iop_F64toI32S && irrm == Irrm_NegINF) /* FCVTMS Wd,Dn */ |
| 5455 | || (op == Iop_F64toI32S && irrm == Irrm_PosINF) /* FCVTPS Wd,Dn */ |
| 5456 | /* F64toI32U */ |
| 5457 | || (op == Iop_F64toI32U && irrm == Irrm_NegINF) /* FCVTMU Wd,Dn */ |
| 5458 | || (op == Iop_F64toI32U && irrm == Irrm_ZERO) /* FCVTZU Wd,Dn */ |
| 5459 | /* F64toI64S */ |
| 5460 | || (op == Iop_F64toI64S && irrm == Irrm_ZERO) /* FCVTZS Xd,Dn */ |
| 5461 | /* F64toI64U */ |
| 5462 | || (op == Iop_F64toI64U && irrm == Irrm_ZERO) /* FCVTZU Xd,Dn */ |
| 5463 | ) { |
| 5464 | /* validated */ |
| 5465 | } else { |
| 5466 | return False; |
| 5467 | } |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5468 | IRType srcTy = isF64 ? Ity_F64 : Ity_F32; |
| 5469 | IRType dstTy = isI64 ? Ity_I64 : Ity_I32; |
| 5470 | IRTemp src = newTemp(srcTy); |
| 5471 | IRTemp dst = newTemp(dstTy); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5472 | assign(src, getQRegLO(nn, srcTy)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5473 | assign(dst, binop(op, mkU32(irrm), mkexpr(src))); |
| 5474 | putIRegOrZR(isI64, dd, mkexpr(dst)); |
| 5475 | DIP("fcvt%c%c %s, %s\n", ch, isU ? 'u' : 's', |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5476 | nameIRegOrZR(isI64, dd), nameQRegLO(nn, srcTy)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5477 | return True; |
| 5478 | } |
| 5479 | |
| 5480 | /* ---------------- FRINT{I,M,P,Z} (scalar) ---------------- */ |
| 5481 | /* 31 23 21 17 14 9 4 |
| 5482 | 000 11110 0x 1001 111 10000 n d FRINTI Fd, Fm (round per FPCR) |
| 5483 | rm |
| 5484 | x==0 => S-registers, x==1 => D-registers |
| 5485 | rm (17:15) encodings: |
| 5486 | 111 per FPCR (FRINTI) |
| 5487 | 001 +inf (FRINTP) |
| 5488 | 010 -inf (FRINTM) |
| 5489 | 011 zero (FRINTZ) |
| 5490 | 000 tieeven |
| 5491 | 100 tieaway |
| 5492 | 110 per FPCR + "exact = TRUE" |
| 5493 | 101 unallocated |
| 5494 | */ |
| 5495 | if (INSN(31,23) == BITS9(0,0,0,1,1,1,1,0,0) |
| 5496 | && INSN(21,18) == BITS4(1,0,0,1) && INSN(14,10) == BITS5(1,0,0,0,0)) { |
| 5497 | Bool isD = INSN(22,22) == 1; |
| 5498 | UInt rm = INSN(17,15); |
| 5499 | UInt nn = INSN(9,5); |
| 5500 | UInt dd = INSN(4,0); |
| 5501 | IRType ty = isD ? Ity_F64 : Ity_F32; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5502 | IRExpr* irrmE = NULL; |
| 5503 | UChar ch = '?'; |
| 5504 | switch (rm) { |
| 5505 | case BITS3(0,1,1): ch = 'z'; irrmE = mkU32(Irrm_ZERO); break; |
| 5506 | case BITS3(0,1,0): ch = 'm'; irrmE = mkU32(Irrm_NegINF); break; |
| 5507 | case BITS3(0,0,1): ch = 'p'; irrmE = mkU32(Irrm_PosINF); break; |
| 5508 | default: break; |
| 5509 | } |
| 5510 | if (irrmE) { |
| 5511 | IRTemp src = newTemp(ty); |
| 5512 | IRTemp dst = newTemp(ty); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5513 | assign(src, getQRegLO(nn, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5514 | assign(dst, binop(isD ? Iop_RoundF64toInt : Iop_RoundF32toInt, |
| 5515 | irrmE, mkexpr(src))); |
| 5516 | putQReg128(dd, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5517 | putQRegLO(dd, mkexpr(dst)); |
| 5518 | DIP("frint%c %s, %s\n", |
| 5519 | ch, nameQRegLO(dd, ty), nameQRegLO(nn, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5520 | return True; |
| 5521 | } |
| 5522 | /* else unhandled rounding mode case -- fall through */ |
| 5523 | } |
| 5524 | |
| 5525 | /* ------------------ FCVT (scalar) ------------------ */ |
| 5526 | /* 31 23 21 16 14 9 4 |
| 5527 | 000 11110 11 10001 00 10000 n d FCVT Sd, Hn (unimp) |
| 5528 | --------- 11 ----- 01 --------- FCVT Dd, Hn (unimp) |
| 5529 | --------- 00 ----- 11 --------- FCVT Hd, Sn (unimp) |
| 5530 | --------- 00 ----- 01 --------- FCVT Dd, Sn (unimp) |
| 5531 | --------- 01 ----- 11 --------- FCVT Hd, Dn (unimp) |
| 5532 | --------- 01 ----- 00 --------- FCVT Sd, Dn (unimp) |
| 5533 | Rounding, when dst is smaller than src, is per the FPCR. |
| 5534 | */ |
| 5535 | if (INSN(31,24) == BITS8(0,0,0,1,1,1,1,0) |
| 5536 | && INSN(21,17) == BITS5(1,0,0,0,1) |
| 5537 | && INSN(14,10) == BITS5(1,0,0,0,0)) { |
| 5538 | UInt b2322 = INSN(23,22); |
| 5539 | UInt b1615 = INSN(16,15); |
| 5540 | UInt nn = INSN(9,5); |
| 5541 | UInt dd = INSN(4,0); |
| 5542 | if (b2322 == BITS2(0,0) && b1615 == BITS2(0,1)) { |
| 5543 | /* Convert S to D */ |
| 5544 | IRTemp res = newTemp(Ity_F64); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5545 | assign(res, unop(Iop_F32toF64, getQRegLO(nn, Ity_F32))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5546 | putQReg128(dd, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5547 | putQRegLO(dd, mkexpr(res)); |
| 5548 | DIP("fcvt %s, %s\n", |
| 5549 | nameQRegLO(dd, Ity_F64), nameQRegLO(nn, Ity_F32)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5550 | return True; |
| 5551 | } |
| 5552 | if (b2322 == BITS2(0,1) && b1615 == BITS2(0,0)) { |
| 5553 | /* Convert D to S */ |
| 5554 | IRTemp res = newTemp(Ity_F32); |
| 5555 | assign(res, binop(Iop_F64toF32, mkexpr(mk_get_IR_rounding_mode()), |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5556 | getQRegLO(nn, Ity_F64))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5557 | putQReg128(dd, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5558 | putQRegLO(dd, mkexpr(res)); |
| 5559 | DIP("fcvt %s, %s\n", |
| 5560 | nameQRegLO(dd, Ity_F32), nameQRegLO(nn, Ity_F64)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5561 | return True; |
| 5562 | } |
| 5563 | /* else unhandled */ |
| 5564 | } |
| 5565 | |
| 5566 | /* ------------------ FABD (scalar) ------------------ */ |
| 5567 | /* 31 23 20 15 9 4 |
| 5568 | 011 11110 111 m 110101 n d FABD Dd, Dn, Dm |
| 5569 | 011 11110 101 m 110101 n d FABD Sd, Sn, Sm |
| 5570 | */ |
| 5571 | if (INSN(31,23) == BITS9(0,1,1,1,1,1,1,0,1) && INSN(21,21) == 1 |
| 5572 | && INSN(15,10) == BITS6(1,1,0,1,0,1)) { |
| 5573 | Bool isD = INSN(22,22) == 1; |
| 5574 | UInt mm = INSN(20,16); |
| 5575 | UInt nn = INSN(9,5); |
| 5576 | UInt dd = INSN(4,0); |
| 5577 | IRType ty = isD ? Ity_F64 : Ity_F32; |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5578 | IRTemp res = newTemp(ty); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5579 | assign(res, unop(mkABSF(ty), |
| 5580 | triop(mkSUBF(ty), |
| 5581 | mkexpr(mk_get_IR_rounding_mode()), |
| 5582 | getQRegLO(nn,ty), getQRegLO(mm,ty)))); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5583 | putQReg128(dd, mkV128(0x0000)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5584 | putQRegLO(dd, mkexpr(res)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5585 | DIP("fabd %s, %s, %s\n", |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5586 | nameQRegLO(dd, ty), nameQRegLO(nn, ty), nameQRegLO(mm, ty)); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 5587 | return True; |
| 5588 | } |
| 5589 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5590 | /* -------------- {S,U}CVTF (vector, integer) -------------- */ |
| 5591 | /* 31 28 22 21 15 9 4 |
| 5592 | 0q0 01110 0 sz 1 00001 110110 n d SCVTF Vd, Vn |
| 5593 | 0q1 01110 0 sz 1 00001 110110 n d UCVTF Vd, Vn |
| 5594 | with laneage: |
| 5595 | case sz:Q of 00 -> 2S, zero upper, 01 -> 4S, 10 -> illegal, 11 -> 2D |
| 5596 | */ |
| 5597 | if (INSN(31,31) == 0 && INSN(28,23) == BITS6(0,1,1,1,0,0) |
| 5598 | && INSN(21,16) == BITS6(1,0,0,0,0,1) |
| 5599 | && INSN(15,10) == BITS6(1,1,0,1,1,0)) { |
| 5600 | Bool isQ = INSN(30,30) == 1; |
| 5601 | Bool isU = INSN(29,29) == 1; |
| 5602 | Bool isF64 = INSN(22,22) == 1; |
| 5603 | UInt nn = INSN(9,5); |
| 5604 | UInt dd = INSN(4,0); |
| 5605 | if (isQ || !isF64) { |
| 5606 | IRType tyF = Ity_INVALID, tyI = Ity_INVALID; |
| 5607 | UInt nLanes = 0; |
| 5608 | Bool zeroHI = False; |
| 5609 | const HChar* arrSpec = NULL; |
| 5610 | Bool ok = getLaneInfo_Q_SZ(&tyI, &tyF, &nLanes, &zeroHI, &arrSpec, |
| 5611 | isQ, isF64 ); |
| 5612 | IROp op = isU ? (isF64 ? Iop_I64UtoF64 : Iop_I32UtoF32) |
| 5613 | : (isF64 ? Iop_I64StoF64 : Iop_I32StoF32); |
| 5614 | IRTemp rm = mk_get_IR_rounding_mode(); |
| 5615 | UInt i; |
| 5616 | vassert(ok); /* the 'if' above should ensure this */ |
| 5617 | for (i = 0; i < nLanes; i++) { |
| 5618 | putQRegLane(dd, i, |
| 5619 | binop(op, mkexpr(rm), getQRegLane(nn, i, tyI))); |
| 5620 | } |
| 5621 | if (zeroHI) { |
| 5622 | putQRegLane(dd, 1, mkU64(0)); |
| 5623 | } |
| 5624 | DIP("%ccvtf %s.%s, %s.%s\n", isU ? 'u' : 's', |
| 5625 | nameQReg128(dd), arrSpec, nameQReg128(nn), arrSpec); |
| 5626 | return True; |
| 5627 | } |
| 5628 | /* else fall through */ |
| 5629 | } |
| 5630 | |
| 5631 | /* ---------- F{ADD,SUB,MUL,DIV,MLA,MLS} (vector) ---------- */ |
| 5632 | /* 31 28 22 21 20 15 9 4 case |
| 5633 | 0q0 01110 0 sz 1 m 110101 n d FADD Vd,Vn,Vm 1 |
| 5634 | 0q0 01110 1 sz 1 m 110101 n d FSUB Vd,Vn,Vm 2 |
| 5635 | 0q1 01110 0 sz 1 m 110111 n d FMUL Vd,Vn,Vm 3 |
| 5636 | 0q1 01110 0 sz 1 m 111111 n d FDIV Vd,Vn,Vm 4 |
| 5637 | 0q0 01110 0 sz 1 m 110011 n d FMLA Vd,Vn,Vm 5 |
| 5638 | 0q0 01110 1 sz 1 m 110011 n d FMLS Vd,Vn,Vm 6 |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 5639 | 0q1 01110 1 sz 1 m 110101 n d FABD Vd,Vn,Vm 7 |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5640 | */ |
| 5641 | if (INSN(31,31) == 0 |
| 5642 | && INSN(28,24) == BITS5(0,1,1,1,0) && INSN(21,21) == 1) { |
| 5643 | Bool isQ = INSN(30,30) == 1; |
| 5644 | UInt b29 = INSN(29,29); |
| 5645 | UInt b23 = INSN(23,23); |
| 5646 | Bool isF64 = INSN(22,22) == 1; |
| 5647 | UInt mm = INSN(20,16); |
| 5648 | UInt b1510 = INSN(15,10); |
| 5649 | UInt nn = INSN(9,5); |
| 5650 | UInt dd = INSN(4,0); |
| 5651 | UInt ix = 0; |
| 5652 | /**/ if (b29 == 0 && b23 == 0 && b1510 == BITS6(1,1,0,1,0,1)) ix = 1; |
| 5653 | else if (b29 == 0 && b23 == 1 && b1510 == BITS6(1,1,0,1,0,1)) ix = 2; |
| 5654 | else if (b29 == 1 && b23 == 0 && b1510 == BITS6(1,1,0,1,1,1)) ix = 3; |
| 5655 | else if (b29 == 1 && b23 == 0 && b1510 == BITS6(1,1,1,1,1,1)) ix = 4; |
| 5656 | else if (b29 == 0 && b23 == 0 && b1510 == BITS6(1,1,0,0,1,1)) ix = 5; |
| 5657 | else if (b29 == 0 && b23 == 1 && b1510 == BITS6(1,1,0,0,1,1)) ix = 6; |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 5658 | else if (b29 == 1 && b23 == 1 && b1510 == BITS6(1,1,0,1,0,1)) ix = 7; |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5659 | IRType laneTy = Ity_INVALID; |
| 5660 | Bool zeroHI = False; |
| 5661 | const HChar* arr = "??"; |
| 5662 | Bool ok |
| 5663 | = getLaneInfo_Q_SZ(NULL, &laneTy, NULL, &zeroHI, &arr, isQ, isF64); |
| 5664 | /* Skip MLA/MLS for the time being */ |
| 5665 | if (ok && ix >= 1 && ix <= 4) { |
| 5666 | const IROp ops64[4] |
| 5667 | = { Iop_Add64Fx2, Iop_Sub64Fx2, Iop_Mul64Fx2, Iop_Div64Fx2 }; |
| 5668 | const IROp ops32[4] |
| 5669 | = { Iop_Add32Fx4, Iop_Sub32Fx4, Iop_Mul32Fx4, Iop_Div32Fx4 }; |
| 5670 | const HChar* names[4] |
| 5671 | = { "fadd", "fsub", "fmul", "fdiv" }; |
| 5672 | IROp op = laneTy==Ity_F64 ? ops64[ix-1] : ops32[ix-1]; |
| 5673 | IRTemp rm = mk_get_IR_rounding_mode(); |
| 5674 | IRTemp t1 = newTemp(Ity_V128); |
| 5675 | IRTemp t2 = newTemp(Ity_V128); |
| 5676 | assign(t1, triop(op, mkexpr(rm), getQReg128(nn), getQReg128(mm))); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5677 | assign(t2, zeroHI ? unop(Iop_ZeroHI64ofV128, mkexpr(t1)) |
| 5678 | : mkexpr(t1)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5679 | putQReg128(dd, mkexpr(t2)); |
| 5680 | DIP("%s %s.%s, %s.%s, %s.%s\n", names[ix-1], |
| 5681 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 5682 | return True; |
| 5683 | } |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5684 | if (ok && ix >= 5 && ix <= 6) { |
| 5685 | IROp opADD = laneTy==Ity_F64 ? Iop_Add64Fx2 : Iop_Add32Fx4; |
| 5686 | IROp opSUB = laneTy==Ity_F64 ? Iop_Sub64Fx2 : Iop_Sub32Fx4; |
| 5687 | IROp opMUL = laneTy==Ity_F64 ? Iop_Mul64Fx2 : Iop_Mul32Fx4; |
| 5688 | IRTemp rm = mk_get_IR_rounding_mode(); |
| 5689 | IRTemp t1 = newTemp(Ity_V128); |
| 5690 | IRTemp t2 = newTemp(Ity_V128); |
| 5691 | // FIXME: double rounding; use FMA primops instead |
| 5692 | assign(t1, triop(opMUL, |
| 5693 | mkexpr(rm), getQReg128(nn), getQReg128(mm))); |
| 5694 | assign(t2, triop(ix == 5 ? opADD : opSUB, |
| 5695 | mkexpr(rm), getQReg128(dd), mkexpr(t1))); |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 5696 | putQReg128(dd, zeroHI ? unop(Iop_ZeroHI64ofV128, mkexpr(t2)) |
| 5697 | : mkexpr(t2)); |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5698 | DIP("%s %s.%s, %s.%s, %s.%s\n", ix == 5 ? "fmla" : "fmls", |
| 5699 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 5700 | return True; |
| 5701 | } |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 5702 | if (ok && ix == 7) { |
| 5703 | IROp opSUB = laneTy==Ity_F64 ? Iop_Sub64Fx2 : Iop_Sub32Fx4; |
| 5704 | IROp opABS = laneTy==Ity_F64 ? Iop_Abs64Fx2 : Iop_Abs32Fx4; |
| 5705 | IRTemp rm = mk_get_IR_rounding_mode(); |
| 5706 | IRTemp t1 = newTemp(Ity_V128); |
| 5707 | IRTemp t2 = newTemp(Ity_V128); |
| 5708 | // FIXME: use Abd primop instead? |
| 5709 | assign(t1, triop(opSUB, |
| 5710 | mkexpr(rm), getQReg128(nn), getQReg128(mm))); |
| 5711 | assign(t2, unop(opABS, mkexpr(t1))); |
| 5712 | putQReg128(dd, zeroHI ? unop(Iop_ZeroHI64ofV128, mkexpr(t2)) |
| 5713 | : mkexpr(t2)); |
| 5714 | DIP("fabd %s.%s, %s.%s, %s.%s\n", |
| 5715 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 5716 | return True; |
| 5717 | } |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5718 | } |
| 5719 | |
| 5720 | /* ---------------- ADD/SUB (vector) ---------------- */ |
| 5721 | /* 31 28 23 21 20 15 9 4 |
| 5722 | 0q0 01110 size 1 m 100001 n d ADD Vd.T, Vn.T, Vm.T |
| 5723 | 0q1 01110 size 1 m 100001 n d SUB Vd.T, Vn.T, Vm.T |
| 5724 | */ |
| 5725 | if (INSN(31,31) == 0 && INSN(28,24) == BITS5(0,1,1,1,0) |
| 5726 | && INSN(21,21) == 1 && INSN(15,10) == BITS6(1,0,0,0,0,1)) { |
| 5727 | Bool isQ = INSN(30,30) == 1; |
| 5728 | UInt szBlg2 = INSN(23,22); |
| 5729 | Bool isSUB = INSN(29,29) == 1; |
| 5730 | UInt mm = INSN(20,16); |
| 5731 | UInt nn = INSN(9,5); |
| 5732 | UInt dd = INSN(4,0); |
| 5733 | Bool zeroHI = False; |
| 5734 | const HChar* arrSpec = ""; |
| 5735 | Bool ok = getLaneInfo_SIMPLE(&zeroHI, &arrSpec, isQ, szBlg2 ); |
| 5736 | if (ok) { |
sewardj | f5b0891 | 2014-02-06 12:57:58 +0000 | [diff] [blame] | 5737 | const IROp opsADD[4] |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5738 | = { Iop_Add8x16, Iop_Add16x8, Iop_Add32x4, Iop_Add64x2 }; |
sewardj | f5b0891 | 2014-02-06 12:57:58 +0000 | [diff] [blame] | 5739 | const IROp opsSUB[4] |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5740 | = { Iop_Sub8x16, Iop_Sub16x8, Iop_Sub32x4, Iop_Sub64x2 }; |
| 5741 | vassert(szBlg2 < 4); |
sewardj | f5b0891 | 2014-02-06 12:57:58 +0000 | [diff] [blame] | 5742 | IROp op = isSUB ? opsSUB[szBlg2] : opsADD[szBlg2]; |
| 5743 | IRTemp t = newTemp(Ity_V128); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5744 | assign(t, binop(op, getQReg128(nn), getQReg128(mm))); |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5745 | putQReg128(dd, zeroHI ? unop(Iop_ZeroHI64ofV128, mkexpr(t)) |
| 5746 | : mkexpr(t)); |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 5747 | const HChar* nm = isSUB ? "sub" : "add"; |
| 5748 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 5749 | nameQReg128(dd), arrSpec, |
| 5750 | nameQReg128(nn), arrSpec, nameQReg128(mm), arrSpec); |
| 5751 | return True; |
| 5752 | } |
| 5753 | /* else fall through */ |
| 5754 | } |
| 5755 | |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5756 | /* ---------------- ADD/SUB (scalar) ---------------- */ |
| 5757 | /* 31 28 23 21 20 15 9 4 |
| 5758 | 010 11110 11 1 m 100001 n d ADD Dd, Dn, Dm |
| 5759 | 011 11110 11 1 m 100001 n d SUB Dd, Dn, Dm |
| 5760 | */ |
| 5761 | if (INSN(31,30) == BITS2(0,1) && INSN(28,21) == BITS8(1,1,1,1,0,1,1,1) |
| 5762 | && INSN(15,10) == BITS6(1,0,0,0,0,1)) { |
| 5763 | Bool isSUB = INSN(29,29) == 1; |
| 5764 | UInt mm = INSN(20,16); |
| 5765 | UInt nn = INSN(9,5); |
| 5766 | UInt dd = INSN(4,0); |
| 5767 | IRTemp res = newTemp(Ity_I64); |
| 5768 | assign(res, binop(isSUB ? Iop_Sub64 : Iop_Add64, |
| 5769 | getQRegLane(nn, 0, Ity_I64), |
| 5770 | getQRegLane(mm, 0, Ity_I64))); |
| 5771 | putQRegLane(dd, 0, mkexpr(res)); |
| 5772 | putQRegLane(dd, 1, mkU64(0)); |
| 5773 | DIP("%s %s, %s, %s\n", isSUB ? "sub" : "add", |
| 5774 | nameQRegLO(dd, Ity_I64), |
| 5775 | nameQRegLO(nn, Ity_I64), nameQRegLO(mm, Ity_I64)); |
| 5776 | return True; |
| 5777 | } |
| 5778 | |
sewardj | f5b0891 | 2014-02-06 12:57:58 +0000 | [diff] [blame] | 5779 | /* ------------ MUL/PMUL/MLA/MLS (vector) ------------ */ |
| 5780 | /* 31 28 23 21 20 15 9 4 |
| 5781 | 0q0 01110 size 1 m 100111 n d MUL Vd.T, Vn.T, Vm.T B/H/S only |
| 5782 | 0q1 01110 size 1 m 100111 n d PMUL Vd.T, Vn.T, Vm.T B only |
| 5783 | 0q0 01110 size 1 m 100101 n d MLA Vd.T, Vn.T, Vm.T B/H/S only |
| 5784 | 0q1 01110 size 1 m 100101 n d MLS Vd.T, Vn.T, Vm.T B/H/S only |
| 5785 | */ |
| 5786 | if (INSN(31,31) == 0 && INSN(28,24) == BITS5(0,1,1,1,0) |
| 5787 | && INSN(21,21) == 1 |
| 5788 | && (INSN(15,10) & BITS6(1,1,1,1,0,1)) == BITS6(1,0,0,1,0,1)) { |
| 5789 | Bool isQ = INSN(30,30) == 1; |
| 5790 | UInt szBlg2 = INSN(23,22); |
| 5791 | UInt bit29 = INSN(29,29); |
| 5792 | UInt mm = INSN(20,16); |
| 5793 | UInt nn = INSN(9,5); |
| 5794 | UInt dd = INSN(4,0); |
| 5795 | Bool isMLAS = INSN(11,11) == 0; |
| 5796 | const IROp opsADD[4] |
| 5797 | = { Iop_Add8x16, Iop_Add16x8, Iop_Add32x4, Iop_INVALID }; |
| 5798 | const IROp opsSUB[4] |
| 5799 | = { Iop_Sub8x16, Iop_Sub16x8, Iop_Sub32x4, Iop_INVALID }; |
| 5800 | const IROp opsMUL[4] |
| 5801 | = { Iop_Mul8x16, Iop_Mul16x8, Iop_Mul32x4, Iop_INVALID }; |
| 5802 | const IROp opsPMUL[4] |
| 5803 | = { Iop_PolynomialMul8x16, Iop_INVALID, Iop_INVALID, Iop_INVALID }; |
| 5804 | /* Set opMUL and, if necessary, opACC. A result value of |
| 5805 | Iop_INVALID for opMUL indicates that the instruction is |
| 5806 | invalid. */ |
| 5807 | Bool zeroHI = False; |
| 5808 | const HChar* arrSpec = ""; |
| 5809 | Bool ok = getLaneInfo_SIMPLE(&zeroHI, &arrSpec, isQ, szBlg2 ); |
| 5810 | vassert(szBlg2 < 4); |
| 5811 | IROp opACC = Iop_INVALID; |
| 5812 | IROp opMUL = Iop_INVALID; |
| 5813 | if (ok) { |
| 5814 | opMUL = (bit29 == 1 && !isMLAS) ? opsPMUL[szBlg2] |
| 5815 | : opsMUL[szBlg2]; |
| 5816 | opACC = isMLAS ? (bit29 == 1 ? opsSUB[szBlg2] : opsADD[szBlg2]) |
| 5817 | : Iop_INVALID; |
| 5818 | } |
| 5819 | if (ok && opMUL != Iop_INVALID) { |
| 5820 | IRTemp t1 = newTemp(Ity_V128); |
| 5821 | assign(t1, binop(opMUL, getQReg128(nn), getQReg128(mm))); |
| 5822 | IRTemp t2 = newTemp(Ity_V128); |
| 5823 | assign(t2, opACC == Iop_INVALID |
| 5824 | ? mkexpr(t1) |
| 5825 | : binop(opACC, getQReg128(dd), mkexpr(t1))); |
| 5826 | putQReg128(dd, zeroHI ? unop(Iop_ZeroHI64ofV128, mkexpr(t2)) |
| 5827 | : mkexpr(t2)); |
| 5828 | const HChar* nm = isMLAS ? (bit29 == 1 ? "mls" : "mla") |
| 5829 | : (bit29 == 1 ? "pmul" : "mul"); |
| 5830 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 5831 | nameQReg128(dd), arrSpec, |
| 5832 | nameQReg128(nn), arrSpec, nameQReg128(mm), arrSpec); |
| 5833 | return True; |
| 5834 | } |
| 5835 | /* else fall through */ |
| 5836 | } |
| 5837 | |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 5838 | /* ---------------- {S,U}{MIN,MAX} (vector) ---------------- */ |
| 5839 | /* 31 28 23 21 20 15 9 4 |
| 5840 | 0q0 01110 size 1 m 011011 n d SMIN Vd.T, Vn.T, Vm.T |
| 5841 | 0q1 01110 size 1 m 011011 n d UMIN Vd.T, Vn.T, Vm.T |
| 5842 | 0q0 01110 size 1 m 011001 n d SMAX Vd.T, Vn.T, Vm.T |
| 5843 | 0q1 01110 size 1 m 011001 n d UMAX Vd.T, Vn.T, Vm.T |
| 5844 | */ |
| 5845 | if (INSN(31,31) == 0 && INSN(28,24) == BITS5(0,1,1,1,0) |
| 5846 | && INSN(21,21) == 1 |
| 5847 | && ((INSN(15,10) & BITS6(1,1,1,1,0,1)) == BITS6(0,1,1,0,0,1))) { |
| 5848 | Bool isQ = INSN(30,30) == 1; |
| 5849 | Bool isU = INSN(29,29) == 1; |
| 5850 | UInt szBlg2 = INSN(23,22); |
| 5851 | Bool isMAX = INSN(12,12) == 0; |
| 5852 | UInt mm = INSN(20,16); |
| 5853 | UInt nn = INSN(9,5); |
| 5854 | UInt dd = INSN(4,0); |
| 5855 | Bool zeroHI = False; |
| 5856 | const HChar* arrSpec = ""; |
| 5857 | Bool ok = getLaneInfo_SIMPLE(&zeroHI, &arrSpec, isQ, szBlg2 ); |
| 5858 | if (ok) { |
| 5859 | const IROp opMINS[4] |
| 5860 | = { Iop_Min8Sx16, Iop_Min16Sx8, Iop_Min32Sx4, Iop_Min64Sx2 }; |
| 5861 | const IROp opMINU[4] |
| 5862 | = { Iop_Min8Ux16, Iop_Min16Ux8, Iop_Min32Ux4, Iop_Min64Ux2 }; |
| 5863 | const IROp opMAXS[4] |
| 5864 | = { Iop_Max8Sx16, Iop_Max16Sx8, Iop_Max32Sx4, Iop_Max64Sx2 }; |
| 5865 | const IROp opMAXU[4] |
| 5866 | = { Iop_Max8Ux16, Iop_Max16Ux8, Iop_Max32Ux4, Iop_Max64Ux2 }; |
| 5867 | vassert(szBlg2 < 4); |
| 5868 | IROp op = isMAX ? (isU ? opMAXU[szBlg2] : opMAXS[szBlg2]) |
| 5869 | : (isU ? opMINU[szBlg2] : opMINS[szBlg2]); |
| 5870 | IRTemp t = newTemp(Ity_V128); |
| 5871 | assign(t, binop(op, getQReg128(nn), getQReg128(mm))); |
| 5872 | putQReg128(dd, zeroHI ? unop(Iop_ZeroHI64ofV128, mkexpr(t)) |
| 5873 | : mkexpr(t)); |
| 5874 | const HChar* nm = isMAX ? (isU ? "umax" : "smax") |
| 5875 | : (isU ? "umin" : "smin"); |
| 5876 | DIP("%s %s.%s, %s.%s, %s.%s\n", nm, |
| 5877 | nameQReg128(dd), arrSpec, |
| 5878 | nameQReg128(nn), arrSpec, nameQReg128(mm), arrSpec); |
| 5879 | return True; |
| 5880 | } |
| 5881 | /* else fall through */ |
| 5882 | } |
| 5883 | |
| 5884 | /* -------------------- {S,U}{MIN,MAX}V -------------------- */ |
| 5885 | /* 31 28 23 21 16 15 9 4 |
| 5886 | 0q0 01110 size 11000 1 101010 n d SMINV Vd, Vn.T |
| 5887 | 0q1 01110 size 11000 1 101010 n d UMINV Vd, Vn.T |
| 5888 | 0q0 01110 size 11000 0 101010 n d SMAXV Vd, Vn.T |
| 5889 | 0q1 01110 size 11000 0 101010 n d UMAXV Vd, Vn.T |
| 5890 | */ |
| 5891 | if (INSN(31,31) == 0 && INSN(28,24) == BITS5(0,1,1,1,0) |
| 5892 | && INSN(21,17) == BITS5(1,1,0,0,0) |
| 5893 | && INSN(15,10) == BITS6(1,0,1,0,1,0)) { |
| 5894 | Bool isQ = INSN(30,30) == 1; |
| 5895 | Bool isU = INSN(29,29) == 1; |
| 5896 | UInt szBlg2 = INSN(23,22); |
| 5897 | Bool isMAX = INSN(16,16) == 0; |
| 5898 | UInt nn = INSN(9,5); |
| 5899 | UInt dd = INSN(4,0); |
| 5900 | Bool zeroHI = False; |
| 5901 | const HChar* arrSpec = ""; |
| 5902 | Bool ok = getLaneInfo_SIMPLE(&zeroHI, &arrSpec, isQ, szBlg2); |
| 5903 | if (ok) { |
| 5904 | if (szBlg2 == 3) ok = False; |
| 5905 | if (szBlg2 == 2 && !isQ) ok = False; |
| 5906 | } |
| 5907 | if (ok) { |
| 5908 | const IROp opMINS[3] |
| 5909 | = { Iop_Min8Sx16, Iop_Min16Sx8, Iop_Min32Sx4 }; |
| 5910 | const IROp opMINU[3] |
| 5911 | = { Iop_Min8Ux16, Iop_Min16Ux8, Iop_Min32Ux4 }; |
| 5912 | const IROp opMAXS[3] |
| 5913 | = { Iop_Max8Sx16, Iop_Max16Sx8, Iop_Max32Sx4 }; |
| 5914 | const IROp opMAXU[3] |
| 5915 | = { Iop_Max8Ux16, Iop_Max16Ux8, Iop_Max32Ux4 }; |
| 5916 | vassert(szBlg2 < 3); |
| 5917 | IROp op = isMAX ? (isU ? opMAXU[szBlg2] : opMAXS[szBlg2]) |
| 5918 | : (isU ? opMINU[szBlg2] : opMINS[szBlg2]); |
| 5919 | IRTemp tN1 = newTemp(Ity_V128); |
| 5920 | assign(tN1, getQReg128(nn)); |
| 5921 | /* If Q == 0, we're just folding lanes in the lower half of |
| 5922 | the value. In which case, copy the lower half of the |
| 5923 | source into the upper half, so we can then treat it the |
| 5924 | same as the full width case. */ |
| 5925 | IRTemp tN2 = newTemp(Ity_V128); |
| 5926 | assign(tN2, zeroHI ? mk_CatOddLanes64x2(tN1,tN1) : mkexpr(tN1)); |
| 5927 | IRTemp res = math_MINMAXV(tN2, op); |
| 5928 | if (res == IRTemp_INVALID) |
| 5929 | return False; /* means math_MINMAXV |
| 5930 | doesn't handle this case yet */ |
| 5931 | putQReg128(dd, mkexpr(res)); |
| 5932 | const HChar* nm = isMAX ? (isU ? "umaxv" : "smaxv") |
| 5933 | : (isU ? "uminv" : "sminv"); |
| 5934 | const IRType tys[3] = { Ity_I8, Ity_I16, Ity_I32 }; |
| 5935 | IRType laneTy = tys[szBlg2]; |
| 5936 | DIP("%s %s, %s.%s\n", nm, |
| 5937 | nameQRegLO(dd, laneTy), nameQReg128(nn), arrSpec); |
| 5938 | return True; |
| 5939 | } |
| 5940 | /* else fall through */ |
| 5941 | } |
| 5942 | |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 5943 | /* ------------ {AND,BIC,ORR,ORN} (vector) ------------ */ |
| 5944 | /* 31 28 23 20 15 9 4 |
| 5945 | 0q0 01110 001 m 000111 n d AND Vd.T, Vn.T, Vm.T |
| 5946 | 0q0 01110 011 m 000111 n d BIC Vd.T, Vn.T, Vm.T |
| 5947 | 0q0 01110 101 m 000111 n d ORR Vd.T, Vn.T, Vm.T |
| 5948 | 0q0 01110 111 m 000111 n d ORN Vd.T, Vn.T, Vm.T |
| 5949 | T is 16b when q==1, 8b when q==0 |
| 5950 | */ |
| 5951 | if (INSN(31,31) == 0 && INSN(29,24) == BITS6(0,0,1,1,1,0) |
| 5952 | && INSN(21,21) == 1 && INSN(15,10) == BITS6(0,0,0,1,1,1)) { |
| 5953 | Bool isQ = INSN(30,30) == 1; |
| 5954 | Bool isORR = INSN(23,23) == 1; |
| 5955 | Bool invert = INSN(22,22) == 1; |
| 5956 | UInt mm = INSN(20,16); |
| 5957 | UInt nn = INSN(9,5); |
| 5958 | UInt dd = INSN(4,0); |
| 5959 | IRTemp res = newTemp(Ity_V128); |
| 5960 | assign(res, binop(isORR ? Iop_OrV128 : Iop_AndV128, |
| 5961 | getQReg128(nn), |
| 5962 | invert ? unop(Iop_NotV128, getQReg128(mm)) |
| 5963 | : getQReg128(mm))); |
| 5964 | putQReg128(dd, isQ ? mkexpr(res) |
| 5965 | : unop(Iop_ZeroHI64ofV128, mkexpr(res))); |
| 5966 | const HChar* names[4] = { "and", "bic", "orr", "orn" }; |
| 5967 | const HChar* ar = isQ ? "16b" : "8b"; |
| 5968 | DIP("%s %s.%s, %s.%s, %s.%s\n", names[INSN(23,22)], |
| 5969 | nameQReg128(dd), ar, nameQReg128(nn), ar, nameQReg128(mm), ar); |
| 5970 | return True; |
| 5971 | } |
| 5972 | |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 5973 | /* ---------- CM{EQ,HI,HS,GE,GT,TST,LE,LT} (vector) ---------- */ |
| 5974 | /* 31 28 23 21 15 9 4 ix |
| 5975 | 0q1 01110 size 1 m 100011 n d CMEQ Vd.T, Vn.T, Vm.T (1) == |
| 5976 | 0q0 01110 size 1 m 100011 n d CMTST Vd.T, Vn.T, Vm.T (2) &, == 0 |
| 5977 | |
| 5978 | 0q1 01110 size 1 m 001101 n d CMHI Vd.T, Vn.T, Vm.T (3) >u |
| 5979 | 0q0 01110 size 1 m 001101 n d CMGT Vd.T, Vn.T, Vm.T (4) >s |
| 5980 | |
| 5981 | 0q1 01110 size 1 m 001111 n d CMHS Vd.T, Vn.T, Vm.T (5) >=u |
| 5982 | 0q0 01110 size 1 m 001111 n d CMGE Vd.T, Vn.T, Vm.T (6) >=s |
| 5983 | |
| 5984 | 0q1 01110 size 100000 100010 n d CMGE Vd.T, Vn.T, #0 (7) >=s 0 |
| 5985 | 0q0 01110 size 100000 100010 n d CMGT Vd.T, Vn.T, #0 (8) >s 0 |
| 5986 | |
| 5987 | 0q1 01110 size 100000 100110 n d CMLE Vd.T, Vn.T, #0 (9) <=s 0 |
| 5988 | 0q0 01110 size 100000 100110 n d CMEQ Vd.T, Vn.T, #0 (10) == 0 |
| 5989 | |
| 5990 | 0q0 01110 size 100000 101010 n d CMLT Vd.T, Vn.T, #0 (11) <s 0 |
| 5991 | */ |
| 5992 | if (INSN(31,31) == 0 |
| 5993 | && INSN(28,24) == BITS5(0,1,1,1,0) && INSN(21,21) == 1) { |
| 5994 | Bool isQ = INSN(30,30) == 1; |
| 5995 | UInt bit29 = INSN(29,29); |
| 5996 | UInt szBlg2 = INSN(23,22); |
| 5997 | UInt mm = INSN(20,16); |
| 5998 | UInt b1510 = INSN(15,10); |
| 5999 | UInt nn = INSN(9,5); |
| 6000 | UInt dd = INSN(4,0); |
| 6001 | const IROp opsEQ[4] |
| 6002 | = { Iop_CmpEQ8x16, Iop_CmpEQ16x8, Iop_CmpEQ32x4, Iop_CmpEQ64x2 }; |
| 6003 | const IROp opsGTS[4] |
| 6004 | = { Iop_CmpGT8Sx16, Iop_CmpGT16Sx8, Iop_CmpGT32Sx4, Iop_CmpGT64Sx2 }; |
| 6005 | const IROp opsGTU[4] |
| 6006 | = { Iop_CmpGT8Ux16, Iop_CmpGT16Ux8, Iop_CmpGT32Ux4, Iop_CmpGT64Ux2 }; |
| 6007 | Bool zeroHI = False; |
| 6008 | const HChar* arrSpec = "??"; |
| 6009 | Bool ok = getLaneInfo_SIMPLE(&zeroHI, &arrSpec, isQ, szBlg2); |
| 6010 | UInt ix = 0; |
| 6011 | if (ok) { |
| 6012 | switch (b1510) { |
| 6013 | case BITS6(1,0,0,0,1,1): ix = bit29 ? 1 : 2; break; |
| 6014 | case BITS6(0,0,1,1,0,1): ix = bit29 ? 3 : 4; break; |
| 6015 | case BITS6(0,0,1,1,1,1): ix = bit29 ? 5 : 6; break; |
| 6016 | case BITS6(1,0,0,0,1,0): |
| 6017 | if (mm == 0) { ix = bit29 ? 7 : 8; }; break; |
| 6018 | case BITS6(1,0,0,1,1,0): |
| 6019 | if (mm == 0) { ix = bit29 ? 9 : 10; }; break; |
| 6020 | case BITS6(1,0,1,0,1,0): |
| 6021 | if (mm == 0 && bit29 == 0) { ix = 11; }; break; |
| 6022 | default: break; |
| 6023 | } |
| 6024 | } |
| 6025 | if (ix != 0) { |
| 6026 | vassert(ok && szBlg2 < 4); |
| 6027 | IRExpr* argL = getQReg128(nn); |
| 6028 | IRExpr* argR = (ix <= 6) ? getQReg128(mm) : mkV128(0x0000); |
| 6029 | IRExpr* res = NULL; |
| 6030 | /* Some useful identities: |
| 6031 | x > y can be expressed directly |
| 6032 | x < y == y > x |
| 6033 | x <= y == not (x > y) |
| 6034 | x >= y == not (y > x) |
| 6035 | */ |
| 6036 | switch (ix) { |
| 6037 | case 1: res = binop(opsEQ[szBlg2], argL, argR); break; |
| 6038 | case 2: binop(opsEQ[szBlg2], |
| 6039 | binop(Iop_AndV128, argL, argR), |
| 6040 | mkV128(0x0000)); |
| 6041 | break; |
| 6042 | case 3: res = binop(opsGTU[szBlg2], argL, argR); break; |
| 6043 | case 4: res = binop(opsGTS[szBlg2], argL, argR); break; |
| 6044 | case 5: res = unop(Iop_NotV128, binop(opsGTU[szBlg2], argR, argL)); |
| 6045 | break; |
| 6046 | case 6: res = unop(Iop_NotV128, binop(opsGTS[szBlg2], argR, argL)); |
| 6047 | break; |
| 6048 | case 7: res = unop(Iop_NotV128, binop(opsGTS[szBlg2], argR, argL)); |
| 6049 | break; |
| 6050 | case 8: res = binop(opsGTS[szBlg2], argL, argR); break; |
| 6051 | case 9: res = unop(Iop_NotV128, |
| 6052 | binop(opsGTS[szBlg2], argL, argR)); |
| 6053 | break; |
| 6054 | case 10: res = binop(opsEQ[szBlg2], argL, argR); break; |
| 6055 | case 11: res = binop(opsGTS[szBlg2], argR, argL); break; |
| 6056 | default: vassert(0); |
| 6057 | } |
| 6058 | vassert(res); |
| 6059 | putQReg128(dd, zeroHI ? unop(Iop_ZeroHI64ofV128, res) : res); |
| 6060 | const HChar* nms[11] = { "eq", "tst", "hi", "gt", "hs", "ge", |
| 6061 | "ge", "gt", "le", "eq", "lt" }; |
| 6062 | if (ix <= 6) { |
| 6063 | DIP("cm%s %s.%s, %s.%s, %s.%s\n", nms[ix-1], |
| 6064 | nameQReg128(dd), arrSpec, |
| 6065 | nameQReg128(nn), arrSpec, nameQReg128(mm), arrSpec); |
| 6066 | } else { |
| 6067 | DIP("cm%s %s.%s, %s.%s, #0\n", nms[ix-1], |
| 6068 | nameQReg128(dd), arrSpec, nameQReg128(nn), arrSpec); |
| 6069 | } |
| 6070 | return True; |
| 6071 | } |
| 6072 | /* else fall through */ |
| 6073 | } |
| 6074 | |
| 6075 | /* -------------- {EOR,BSL,BIT,BIF} (vector) -------------- */ |
| 6076 | /* 31 28 23 20 15 9 4 |
| 6077 | 0q1 01110 00 1 m 000111 n d EOR Vd.T, Vm.T, Vn.T |
| 6078 | 0q1 01110 01 1 m 000111 n d BSL Vd.T, Vm.T, Vn.T |
| 6079 | 0q1 01110 10 1 m 000111 n d BIT Vd.T, Vm.T, Vn.T |
| 6080 | 0q1 01110 11 1 m 000111 n d BIF Vd.T, Vm.T, Vn.T |
| 6081 | */ |
| 6082 | if (INSN(31,31) == 0 && INSN(29,24) == BITS6(1,0,1,1,1,0) |
| 6083 | && INSN(21,21) == 1 && INSN(15,10) == BITS6(0,0,0,1,1,1)) { |
| 6084 | Bool isQ = INSN(30,30) == 1; |
| 6085 | UInt op = INSN(23,22); |
| 6086 | UInt mm = INSN(20,16); |
| 6087 | UInt nn = INSN(9,5); |
| 6088 | UInt dd = INSN(4,0); |
| 6089 | IRTemp argD = newTemp(Ity_V128); |
| 6090 | IRTemp argN = newTemp(Ity_V128); |
| 6091 | IRTemp argM = newTemp(Ity_V128); |
| 6092 | assign(argD, getQReg128(dd)); |
| 6093 | assign(argN, getQReg128(nn)); |
| 6094 | assign(argM, getQReg128(mm)); |
| 6095 | const IROp opXOR = Iop_XorV128; |
| 6096 | const IROp opAND = Iop_AndV128; |
| 6097 | const IROp opNOT = Iop_NotV128; |
| 6098 | IRExpr* res = NULL; |
| 6099 | switch (op) { |
| 6100 | case BITS2(0,0): |
| 6101 | res = binop(opXOR, mkexpr(argM), mkexpr(argN)); |
| 6102 | break; |
| 6103 | case BITS2(0,1): |
| 6104 | res = binop(opXOR, mkexpr(argM), |
| 6105 | binop(opAND, |
| 6106 | binop(opXOR, mkexpr(argM), mkexpr(argN)), |
| 6107 | mkexpr(argD))); |
| 6108 | break; |
| 6109 | case BITS2(1,0): |
| 6110 | res = binop(opXOR, mkexpr(argD), |
| 6111 | binop(opAND, |
| 6112 | binop(opXOR, mkexpr(argD), mkexpr(argN)), |
| 6113 | mkexpr(argM))); |
| 6114 | break; |
| 6115 | case BITS2(1,1): |
| 6116 | res = binop(opXOR, mkexpr(argD), |
| 6117 | binop(opAND, |
| 6118 | binop(opXOR, mkexpr(argD), mkexpr(argN)), |
| 6119 | unop(opNOT, mkexpr(argM)))); |
| 6120 | break; |
| 6121 | default: |
| 6122 | vassert(0); |
| 6123 | } |
| 6124 | vassert(res); |
| 6125 | putQReg128(dd, isQ ? res : unop(Iop_ZeroHI64ofV128, res)); |
| 6126 | const HChar* nms[4] = { "eor", "bsl", "bit", "bif" }; |
| 6127 | const HChar* arr = isQ ? "16b" : "8b"; |
| 6128 | vassert(op < 4); |
| 6129 | DIP("%s %s.%s, %s.%s, %s.%s\n", nms[op], |
| 6130 | nameQReg128(dd), arr, nameQReg128(nn), arr, nameQReg128(mm), arr); |
| 6131 | return True; |
| 6132 | } |
| 6133 | |
| 6134 | /* ------------ {USHR,SSHR} (vector, immediate) ------------ */ |
| 6135 | /* 31 28 22 18 15 9 4 |
| 6136 | 0q1 011110 immh immb 000001 n d USHR Vd.T, Vn.T, #shift |
| 6137 | 0q0 011110 immh immb 000001 n d SSHR Vd.T, Vn.T, #shift |
| 6138 | laneTy, shift = case immh:immb of |
| 6139 | 0001:xxx -> B, 8-xxx |
| 6140 | 001x:xxx -> H, 16-xxxx |
| 6141 | 01xx:xxx -> S, 32-xxxxx |
| 6142 | 1xxx:xxx -> D, 64-xxxxxx |
| 6143 | other -> invalid |
| 6144 | As usual the case laneTy==D && q==0 is not allowed. |
| 6145 | */ |
| 6146 | if (INSN(31,31) == 0 && INSN(28,23) == BITS6(0,1,1,1,1,0) |
| 6147 | && INSN(15,10) == BITS6(0,0,0,0,0,1)) { |
| 6148 | Bool isQ = INSN(30,30) == 1; |
| 6149 | Bool isU = INSN(29,29) == 1; |
| 6150 | UInt immh = INSN(22,19); |
| 6151 | UInt immb = INSN(18,16); |
| 6152 | UInt nn = INSN(9,5); |
| 6153 | UInt dd = INSN(4,0); |
| 6154 | const IROp opsSHRN[4] |
| 6155 | = { Iop_ShrN8x16, Iop_ShrN16x8, Iop_ShrN32x4, Iop_ShrN64x2 }; |
| 6156 | const IROp opsSARN[4] |
| 6157 | = { Iop_SarN8x16, Iop_SarN16x8, Iop_SarN32x4, Iop_SarN64x2 }; |
| 6158 | UInt szBlg2 = 0; |
| 6159 | UInt shift = 0; |
| 6160 | Bool ok = getLaneInfo_IMMH_IMMB(&shift, &szBlg2, immh, immb); |
| 6161 | if (ok && szBlg2 < 4 && shift > 0 && shift < (8 << szBlg2) |
| 6162 | && !(szBlg2 == 3/*64bit*/ && !isQ)) { |
| 6163 | IROp op = isU ? opsSHRN[szBlg2] : opsSARN[szBlg2]; |
| 6164 | IRExpr* src = getQReg128(nn); |
| 6165 | IRExpr* res = binop(op, src, mkU8(shift)); |
| 6166 | putQReg128(dd, isQ ? res : unop(Iop_ZeroHI64ofV128, res)); |
| 6167 | HChar laneCh = "bhsd"[szBlg2]; |
| 6168 | UInt nLanes = (isQ ? 128 : 64) / (8 << szBlg2); |
| 6169 | DIP("%s %s.%u%c, %s.%u%c, #%u\n", isU ? "ushr" : "sshr", |
| 6170 | nameQReg128(dd), nLanes, laneCh, |
| 6171 | nameQReg128(nn), nLanes, laneCh, shift); |
| 6172 | return True; |
| 6173 | } |
| 6174 | /* else fall through */ |
| 6175 | } |
| 6176 | |
| 6177 | /* -------------------- {U,S}SHLL{,2} -------------------- */ |
| 6178 | /* 31 28 22 18 15 9 4 |
| 6179 | 0q0 011110 immh immb 101001 n d SSHLL Vd.Ta, Vn.Tb, #sh |
| 6180 | 0q1 011110 immh immb 101001 n d USHLL Vd.Ta, Vn.Tb, #sh |
| 6181 | where Ta,Tb,sh |
| 6182 | = case immh of 0001 -> 8h, 8b(q0)/16b(q1), immh:immb - 8 (0..7) |
| 6183 | 001x -> 4s, 4h(q0)/8h(q1), immh:immb - 16 (0..15) |
| 6184 | 01xx -> 2d, 2s(q0)/4s(q1), immh:immb - 32 (0..31) |
| 6185 | */ |
| 6186 | if (INSN(31,31) == 0 && INSN(28,23) == BITS6(0,1,1,1,1,0) |
| 6187 | && INSN(15,10) == BITS6(1,0,1,0,0,1)) { |
| 6188 | Bool isQ = INSN(30,30) == 1; |
| 6189 | Bool isU = INSN(29,29) == 1; |
| 6190 | UInt immh = INSN(22,19); |
| 6191 | UInt immb = INSN(18,16); |
| 6192 | UInt nn = INSN(9,5); |
| 6193 | UInt dd = INSN(4,0); |
| 6194 | UInt immhb = (immh << 3) | immb; |
| 6195 | IRTemp src = newTemp(Ity_V128); |
| 6196 | IRTemp zero = newTemp(Ity_V128); |
| 6197 | IRExpr* res = NULL; |
| 6198 | UInt sh = 0; |
| 6199 | const HChar* ta = "??"; |
| 6200 | const HChar* tb = "??"; |
| 6201 | assign(src, getQReg128(nn)); |
| 6202 | assign(zero, mkV128(0x0000)); |
| 6203 | if (immh & 1) { |
| 6204 | sh = immhb - 8; |
| 6205 | vassert(sh < 8); /* so 8-sh is 1..8 */ |
| 6206 | ta = "8h"; |
| 6207 | tb = isQ ? "16b" : "8b"; |
| 6208 | IRExpr* tmp = isQ ? mk_InterleaveHI8x16(src, zero) |
| 6209 | : mk_InterleaveLO8x16(src, zero); |
| 6210 | res = binop(isU ? Iop_ShrN16x8 : Iop_SarN16x8, tmp, mkU8(8-sh)); |
| 6211 | } |
| 6212 | else if (immh & 2) { |
| 6213 | sh = immhb - 16; |
| 6214 | vassert(sh < 16); /* so 16-sh is 1..16 */ |
| 6215 | ta = "4s"; |
| 6216 | tb = isQ ? "8h" : "4h"; |
| 6217 | IRExpr* tmp = isQ ? mk_InterleaveHI16x8(src, zero) |
| 6218 | : mk_InterleaveLO16x8(src, zero); |
| 6219 | res = binop(isU ? Iop_ShrN32x4 : Iop_SarN32x4, tmp, mkU8(16-sh)); |
| 6220 | } |
| 6221 | else if (immh & 4) { |
| 6222 | sh = immhb - 32; |
| 6223 | vassert(sh < 32); /* so 32-sh is 1..32 */ |
| 6224 | ta = "2d"; |
| 6225 | tb = isQ ? "4s" : "2s"; |
| 6226 | IRExpr* tmp = isQ ? mk_InterleaveHI32x4(src, zero) |
| 6227 | : mk_InterleaveLO32x4(src, zero); |
| 6228 | res = binop(isU ? Iop_ShrN64x2 : Iop_SarN64x2, tmp, mkU8(32-sh)); |
| 6229 | } |
| 6230 | /* */ |
| 6231 | if (res) { |
| 6232 | putQReg128(dd, res); |
| 6233 | DIP("%cshll%s %s.%s, %s.%s, #%d\n", |
| 6234 | isU ? 'u' : 's', isQ ? "2" : "", |
| 6235 | nameQReg128(dd), ta, nameQReg128(nn), tb, sh); |
| 6236 | return True; |
| 6237 | } |
| 6238 | /* else fall through */ |
| 6239 | } |
| 6240 | |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 6241 | /* -------------------- XTN{,2} -------------------- */ |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 6242 | /* 31 28 23 21 15 9 4 XTN{,2} Vd.Tb, Vn.Ta |
sewardj | 606c4ba | 2014-01-26 19:11:14 +0000 | [diff] [blame] | 6243 | 0q0 01110 size 100001 001010 n d |
| 6244 | */ |
| 6245 | if (INSN(31,31) == 0 && INSN(29,24) == BITS6(0,0,1,1,1,0) |
| 6246 | && INSN(21,16) == BITS6(1,0,0,0,0,1) |
| 6247 | && INSN(15,10) == BITS6(0,0,1,0,1,0)) { |
| 6248 | Bool isQ = INSN(30,30) == 1; |
| 6249 | UInt size = INSN(23,22); |
| 6250 | UInt nn = INSN(9,5); |
| 6251 | UInt dd = INSN(4,0); |
| 6252 | IROp op = Iop_INVALID; |
| 6253 | const HChar* tb = NULL; |
| 6254 | const HChar* ta = NULL; |
| 6255 | switch ((size << 1) | (isQ ? 1 : 0)) { |
| 6256 | case 0: tb = "8b"; ta = "8h"; op = Iop_NarrowUn16to8x8; break; |
| 6257 | case 1: tb = "16b"; ta = "8h"; op = Iop_NarrowUn16to8x8; break; |
| 6258 | case 2: tb = "4h"; ta = "4s"; op = Iop_NarrowUn32to16x4; break; |
| 6259 | case 3: tb = "8h"; ta = "4s"; op = Iop_NarrowUn32to16x4; break; |
| 6260 | case 4: tb = "2s"; ta = "2d"; op = Iop_NarrowUn64to32x2; break; |
| 6261 | case 5: tb = "4s"; ta = "2d"; op = Iop_NarrowUn64to32x2; break; |
| 6262 | case 6: break; |
| 6263 | case 7: break; |
| 6264 | default: vassert(0); |
| 6265 | } |
| 6266 | if (op != Iop_INVALID) { |
| 6267 | if (!isQ) { |
| 6268 | putQRegLane(dd, 1, mkU64(0)); |
| 6269 | } |
| 6270 | putQRegLane(dd, isQ ? 1 : 0, unop(op, getQReg128(nn))); |
| 6271 | DIP("xtn%s %s.%s, %s.%s\n", isQ ? "2" : "", |
| 6272 | nameQReg128(dd), tb, nameQReg128(nn), ta); |
| 6273 | return True; |
| 6274 | } |
| 6275 | /* else fall through */ |
| 6276 | } |
| 6277 | |
| 6278 | /* ---------------- DUP (element, vector) ---------------- */ |
| 6279 | /* 31 28 20 15 9 4 |
| 6280 | 0q0 01110000 imm5 000001 n d DUP Vd.T, Vn.Ts[index] |
| 6281 | */ |
| 6282 | if (INSN(31,31) == 0 && INSN(29,21) == BITS9(0,0,1,1,1,0,0,0,0) |
| 6283 | && INSN(15,10) == BITS6(0,0,0,0,0,1)) { |
| 6284 | Bool isQ = INSN(30,30) == 1; |
| 6285 | UInt imm5 = INSN(20,16); |
| 6286 | UInt nn = INSN(9,5); |
| 6287 | UInt dd = INSN(4,0); |
| 6288 | IRTemp w0 = newTemp(Ity_I64); |
| 6289 | const HChar* arT = "??"; |
| 6290 | const HChar* arTs = "??"; |
| 6291 | IRType laneTy = Ity_INVALID; |
| 6292 | UInt laneNo = 16; /* invalid */ |
| 6293 | if (imm5 & 1) { |
| 6294 | arT = isQ ? "16b" : "8b"; |
| 6295 | arTs = "b"; |
| 6296 | laneNo = (imm5 >> 1) & 15; |
| 6297 | laneTy = Ity_I8; |
| 6298 | assign(w0, unop(Iop_8Uto64, getQRegLane(nn, laneNo, laneTy))); |
| 6299 | } |
| 6300 | else if (imm5 & 2) { |
| 6301 | arT = isQ ? "8h" : "4h"; |
| 6302 | arTs = "h"; |
| 6303 | laneNo = (imm5 >> 2) & 7; |
| 6304 | laneTy = Ity_I16; |
| 6305 | assign(w0, unop(Iop_16Uto64, getQRegLane(nn, laneNo, laneTy))); |
| 6306 | } |
| 6307 | else if (imm5 & 4) { |
| 6308 | arT = isQ ? "4s" : "2s"; |
| 6309 | arTs = "s"; |
| 6310 | laneNo = (imm5 >> 3) & 3; |
| 6311 | laneTy = Ity_I32; |
| 6312 | assign(w0, unop(Iop_32Uto64, getQRegLane(nn, laneNo, laneTy))); |
| 6313 | } |
| 6314 | else if ((imm5 & 8) && isQ) { |
| 6315 | arT = "2d"; |
| 6316 | arTs = "d"; |
| 6317 | laneNo = (imm5 >> 4) & 1; |
| 6318 | laneTy = Ity_I64; |
| 6319 | assign(w0, getQRegLane(nn, laneNo, laneTy)); |
| 6320 | } |
| 6321 | else { |
| 6322 | /* invalid; leave laneTy unchanged. */ |
| 6323 | } |
| 6324 | /* */ |
| 6325 | if (laneTy != Ity_INVALID) { |
| 6326 | vassert(laneNo < 16); |
| 6327 | IRTemp w1 = math_DUP_TO_64(w0, laneTy); |
| 6328 | putQReg128(dd, binop(Iop_64HLtoV128, |
| 6329 | isQ ? mkexpr(w1) : mkU64(0), mkexpr(w1))); |
| 6330 | DIP("dup %s.%s, %s.%s[%u]\n", |
| 6331 | nameQReg128(dd), arT, nameQReg128(nn), arTs, laneNo); |
| 6332 | return True; |
| 6333 | } |
| 6334 | /* else fall through */ |
| 6335 | } |
| 6336 | |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 6337 | /* ---------------- DUP (general, vector) ---------------- */ |
| 6338 | /* 31 28 23 20 15 9 4 |
| 6339 | 0q0 01110 000 imm5 000011 n d DUP Vd.T, Rn |
| 6340 | Q=0 writes 64, Q=1 writes 128 |
| 6341 | imm5: xxxx1 8B(q=0) or 16b(q=1), R=W |
| 6342 | xxx10 4H(q=0) or 8H(q=1), R=W |
| 6343 | xx100 2S(q=0) or 4S(q=1), R=W |
| 6344 | x1000 Invalid(q=0) or 2D(q=1), R=X |
| 6345 | x0000 Invalid(q=0) or Invalid(q=1) |
| 6346 | */ |
| 6347 | if (INSN(31,31) == 0 && INSN(29,21) == BITS9(0,0,1,1,1,0,0,0,0) |
| 6348 | && INSN(15,10) == BITS6(0,0,0,0,1,1)) { |
| 6349 | Bool isQ = INSN(30,30) == 1; |
| 6350 | UInt imm5 = INSN(20,16); |
| 6351 | UInt nn = INSN(9,5); |
| 6352 | UInt dd = INSN(4,0); |
| 6353 | IRTemp w0 = newTemp(Ity_I64); |
| 6354 | const HChar* arT = "??"; |
| 6355 | IRType laneTy = Ity_INVALID; |
| 6356 | if (imm5 & 1) { |
| 6357 | arT = isQ ? "16b" : "8b"; |
| 6358 | laneTy = Ity_I8; |
| 6359 | assign(w0, unop(Iop_8Uto64, unop(Iop_64to8, getIReg64orZR(nn)))); |
| 6360 | } |
| 6361 | else if (imm5 & 2) { |
| 6362 | arT = isQ ? "8h" : "4h"; |
| 6363 | laneTy = Ity_I16; |
| 6364 | assign(w0, unop(Iop_16Uto64, unop(Iop_64to16, getIReg64orZR(nn)))); |
| 6365 | } |
| 6366 | else if (imm5 & 4) { |
| 6367 | arT = isQ ? "4s" : "2s"; |
| 6368 | laneTy = Ity_I32; |
| 6369 | assign(w0, unop(Iop_32Uto64, unop(Iop_64to32, getIReg64orZR(nn)))); |
| 6370 | } |
| 6371 | else if ((imm5 & 8) && isQ) { |
| 6372 | arT = "2d"; |
| 6373 | laneTy = Ity_I64; |
| 6374 | assign(w0, getIReg64orZR(nn)); |
| 6375 | } |
| 6376 | else { |
| 6377 | /* invalid; leave laneTy unchanged. */ |
| 6378 | } |
| 6379 | /* */ |
| 6380 | if (laneTy != Ity_INVALID) { |
| 6381 | IRTemp w1 = math_DUP_TO_64(w0, laneTy); |
| 6382 | putQReg128(dd, binop(Iop_64HLtoV128, |
| 6383 | isQ ? mkexpr(w1) : mkU64(0), mkexpr(w1))); |
| 6384 | DIP("dup %s.%s, %s\n", |
| 6385 | nameQReg128(dd), arT, nameIRegOrZR(laneTy == Ity_I64, nn)); |
| 6386 | return True; |
| 6387 | } |
| 6388 | /* else fall through */ |
| 6389 | } |
| 6390 | |
sewardj | f5b0891 | 2014-02-06 12:57:58 +0000 | [diff] [blame] | 6391 | /* ---------------------- {S,U}MOV ---------------------- */ |
| 6392 | /* 31 28 20 15 9 4 |
| 6393 | 0q0 01110 000 imm5 001111 n d UMOV Xd/Wd, Vn.Ts[index] |
| 6394 | 0q0 01110 000 imm5 001011 n d SMOV Xd/Wd, Vn.Ts[index] |
| 6395 | dest is Xd when q==1, Wd when q==0 |
| 6396 | UMOV: |
| 6397 | Ts,index,ops = case q:imm5 of |
| 6398 | 0:xxxx1 -> B, xxxx, 8Uto64 |
| 6399 | 1:xxxx1 -> invalid |
| 6400 | 0:xxx10 -> H, xxx, 16Uto64 |
| 6401 | 1:xxx10 -> invalid |
| 6402 | 0:xx100 -> S, xx, 32Uto64 |
| 6403 | 1:xx100 -> invalid |
| 6404 | 1:x1000 -> D, x, copy64 |
| 6405 | other -> invalid |
| 6406 | SMOV: |
| 6407 | Ts,index,ops = case q:imm5 of |
| 6408 | 0:xxxx1 -> B, xxxx, (32Uto64 . 8Sto32) |
| 6409 | 1:xxxx1 -> B, xxxx, 8Sto64 |
| 6410 | 0:xxx10 -> H, xxx, (32Uto64 . 16Sto32) |
| 6411 | 1:xxx10 -> H, xxx, 16Sto64 |
| 6412 | 0:xx100 -> invalid |
| 6413 | 1:xx100 -> S, xx, 32Sto64 |
| 6414 | 1:x1000 -> invalid |
| 6415 | other -> invalid |
| 6416 | */ |
| 6417 | if (INSN(31,31) == 0 && INSN(29,21) == BITS9(0,0,1,1,1,0,0,0,0) |
| 6418 | && (INSN(15,10) & BITS6(1,1,1,0,1,1)) == BITS6(0,0,1,0,1,1)) { |
| 6419 | UInt bitQ = INSN(30,30) == 1; |
| 6420 | UInt imm5 = INSN(20,16); |
| 6421 | UInt nn = INSN(9,5); |
| 6422 | UInt dd = INSN(4,0); |
| 6423 | Bool isU = INSN(12,12) == 1; |
| 6424 | const HChar* arTs = "??"; |
| 6425 | UInt laneNo = 16; /* invalid */ |
| 6426 | // Setting 'res' to non-NULL determines valid/invalid |
| 6427 | IRExpr* res = NULL; |
| 6428 | if (!bitQ && (imm5 & 1)) { // 0:xxxx1 |
| 6429 | laneNo = (imm5 >> 1) & 15; |
| 6430 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I8); |
| 6431 | res = isU ? unop(Iop_8Uto64, lane) |
| 6432 | : unop(Iop_32Uto64, unop(Iop_8Sto32, lane)); |
| 6433 | arTs = "b"; |
| 6434 | } |
| 6435 | else if (bitQ && (imm5 & 1)) { // 1:xxxx1 |
| 6436 | laneNo = (imm5 >> 1) & 15; |
| 6437 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I8); |
| 6438 | res = isU ? NULL |
| 6439 | : unop(Iop_8Sto64, lane); |
| 6440 | arTs = "b"; |
| 6441 | } |
| 6442 | else if (!bitQ && (imm5 & 2)) { // 0:xxx10 |
| 6443 | laneNo = (imm5 >> 2) & 7; |
| 6444 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I16); |
| 6445 | res = isU ? unop(Iop_16Uto64, lane) |
| 6446 | : unop(Iop_32Uto64, unop(Iop_16Sto32, lane)); |
| 6447 | arTs = "h"; |
| 6448 | } |
| 6449 | else if (bitQ && (imm5 & 2)) { // 1:xxx10 |
| 6450 | laneNo = (imm5 >> 2) & 7; |
| 6451 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I16); |
| 6452 | res = isU ? NULL |
| 6453 | : unop(Iop_16Sto64, lane); |
| 6454 | arTs = "h"; |
| 6455 | } |
| 6456 | else if (!bitQ && (imm5 & 4)) { // 0:xx100 |
| 6457 | laneNo = (imm5 >> 3) & 3; |
| 6458 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I32); |
| 6459 | res = isU ? unop(Iop_32Uto64, lane) |
| 6460 | : NULL; |
| 6461 | arTs = "s"; |
| 6462 | } |
| 6463 | else if (bitQ && (imm5 & 4)) { // 1:xxx10 |
| 6464 | laneNo = (imm5 >> 3) & 3; |
| 6465 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I32); |
| 6466 | res = isU ? NULL |
| 6467 | : unop(Iop_32Sto64, lane); |
| 6468 | arTs = "s"; |
| 6469 | } |
| 6470 | else if (bitQ && (imm5 & 8)) { // 1:x1000 |
| 6471 | laneNo = (imm5 >> 4) & 1; |
| 6472 | IRExpr* lane = getQRegLane(nn, laneNo, Ity_I64); |
| 6473 | res = isU ? lane |
| 6474 | : NULL; |
| 6475 | arTs = "d"; |
| 6476 | } |
| 6477 | /* */ |
| 6478 | if (res) { |
| 6479 | vassert(laneNo < 16); |
| 6480 | putIReg64orZR(dd, res); |
| 6481 | DIP("%cmov %s, %s.%s[%u]\n", isU ? 'u' : 's', |
| 6482 | nameIRegOrZR(bitQ == 1, dd), |
| 6483 | nameQReg128(nn), arTs, laneNo); |
| 6484 | return True; |
| 6485 | } |
| 6486 | /* else fall through */ |
| 6487 | } |
| 6488 | |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 6489 | /* -------------------- INS (general) -------------------- */ |
| 6490 | /* 31 28 20 15 9 4 |
| 6491 | 010 01110000 imm5 000111 n d INS Vd.Ts[ix], Rn |
| 6492 | where Ts,ix = case imm5 of xxxx1 -> B, xxxx |
| 6493 | xxx10 -> H, xxx |
| 6494 | xx100 -> S, xx |
| 6495 | x1000 -> D, x |
| 6496 | */ |
| 6497 | if (INSN(31,21) == BITS11(0,1,0,0,1,1,1,0,0,0,0) |
| 6498 | && INSN(15,10) == BITS6(0,0,0,1,1,1)) { |
| 6499 | UInt imm5 = INSN(20,16); |
| 6500 | UInt nn = INSN(9,5); |
| 6501 | UInt dd = INSN(4,0); |
| 6502 | HChar ts = '?'; |
| 6503 | UInt laneNo = 16; |
| 6504 | IRExpr* src = NULL; |
| 6505 | if (imm5 & 1) { |
| 6506 | src = unop(Iop_64to8, getIReg64orZR(nn)); |
| 6507 | laneNo = (imm5 >> 1) & 15; |
| 6508 | ts = 'b'; |
| 6509 | } |
| 6510 | else if (imm5 & 2) { |
| 6511 | src = unop(Iop_64to16, getIReg64orZR(nn)); |
| 6512 | laneNo = (imm5 >> 2) & 7; |
| 6513 | ts = 'h'; |
| 6514 | } |
| 6515 | else if (imm5 & 4) { |
| 6516 | src = unop(Iop_64to32, getIReg64orZR(nn)); |
| 6517 | laneNo = (imm5 >> 3) & 3; |
| 6518 | ts = 's'; |
| 6519 | } |
| 6520 | else if (imm5 & 8) { |
| 6521 | src = getIReg64orZR(nn); |
| 6522 | laneNo = (imm5 >> 4) & 1; |
| 6523 | ts = 'd'; |
| 6524 | } |
| 6525 | /* */ |
| 6526 | if (src) { |
| 6527 | vassert(laneNo < 16); |
| 6528 | putQRegLane(dd, laneNo, src); |
| 6529 | DIP("ins %s.%c[%u], %s\n", |
| 6530 | nameQReg128(dd), ts, laneNo, nameIReg64orZR(nn)); |
| 6531 | return True; |
| 6532 | } |
| 6533 | /* else invalid; fall through */ |
| 6534 | } |
| 6535 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 6536 | /* FIXME Temporary hacks to get through ld.so FIXME */ |
| 6537 | |
| 6538 | /* ------------------ movi vD.4s, #0x0 ------------------ */ |
| 6539 | /* 0x4F 0x00 0x04 000 vD */ |
| 6540 | if ((insn & 0xFFFFFFE0) == 0x4F000400) { |
| 6541 | UInt vD = INSN(4,0); |
| 6542 | putQReg128(vD, mkV128(0x0000)); |
| 6543 | DIP("movi v%u.4s, #0x0\n", vD); |
| 6544 | return True; |
| 6545 | } |
| 6546 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 6547 | /* ---------------- MOV vD.16b, vN.16b ---------------- */ |
| 6548 | /* 31 23 20 15 9 4 |
| 6549 | 010 01110 101 m 000111 n d ORR vD.16b, vN.16b, vM.16b |
| 6550 | This only handles the N == M case. |
| 6551 | */ |
| 6552 | if (INSN(31,24) == BITS8(0,1,0,0,1,1,1,0) |
| 6553 | && INSN(23,21) == BITS3(1,0,1) && INSN(15,10) == BITS6(0,0,0,1,1,1)) { |
| 6554 | UInt mm = INSN(20,16); |
| 6555 | UInt nn = INSN(9,5); |
| 6556 | UInt dd = INSN(4,0); |
| 6557 | if (mm == nn) { |
| 6558 | putQReg128(dd, getQReg128(nn)); |
| 6559 | DIP("mov v%u.16b, v%u.16b\n", dd, nn); |
| 6560 | return True; |
| 6561 | } |
| 6562 | /* else it's really an ORR; fall through. */ |
| 6563 | } |
| 6564 | |
| 6565 | vex_printf("ARM64 front end: simd_and_fp\n"); |
| 6566 | return False; |
| 6567 | # undef INSN |
| 6568 | } |
| 6569 | |
| 6570 | |
| 6571 | /*------------------------------------------------------------*/ |
| 6572 | /*--- Disassemble a single ARM64 instruction ---*/ |
| 6573 | /*------------------------------------------------------------*/ |
| 6574 | |
| 6575 | /* Disassemble a single ARM64 instruction into IR. The instruction |
| 6576 | has is located at |guest_instr| and has guest IP of |
| 6577 | |guest_PC_curr_instr|, which will have been set before the call |
| 6578 | here. Returns True iff the instruction was decoded, in which case |
| 6579 | *dres will be set accordingly, or False, in which case *dres should |
| 6580 | be ignored by the caller. */ |
| 6581 | |
| 6582 | static |
| 6583 | Bool disInstr_ARM64_WRK ( |
| 6584 | /*MB_OUT*/DisResult* dres, |
| 6585 | Bool (*resteerOkFn) ( /*opaque*/void*, Addr64 ), |
| 6586 | Bool resteerCisOk, |
| 6587 | void* callback_opaque, |
| 6588 | UChar* guest_instr, |
| 6589 | VexArchInfo* archinfo, |
| 6590 | VexAbiInfo* abiinfo |
| 6591 | ) |
| 6592 | { |
| 6593 | // A macro to fish bits out of 'insn'. |
| 6594 | # define INSN(_bMax,_bMin) SLICE_UInt(insn, (_bMax), (_bMin)) |
| 6595 | |
| 6596 | //ZZ DisResult dres; |
| 6597 | //ZZ UInt insn; |
| 6598 | //ZZ //Bool allow_VFP = False; |
| 6599 | //ZZ //UInt hwcaps = archinfo->hwcaps; |
| 6600 | //ZZ IRTemp condT; /* :: Ity_I32 */ |
| 6601 | //ZZ UInt summary; |
| 6602 | //ZZ HChar dis_buf[128]; // big enough to hold LDMIA etc text |
| 6603 | //ZZ |
| 6604 | //ZZ /* What insn variants are we supporting today? */ |
| 6605 | //ZZ //allow_VFP = (0 != (hwcaps & VEX_HWCAPS_ARM_VFP)); |
| 6606 | //ZZ // etc etc |
| 6607 | |
| 6608 | /* Set result defaults. */ |
| 6609 | dres->whatNext = Dis_Continue; |
| 6610 | dres->len = 4; |
| 6611 | dres->continueAt = 0; |
| 6612 | dres->jk_StopHere = Ijk_INVALID; |
| 6613 | |
| 6614 | /* At least this is simple on ARM64: insns are all 4 bytes long, and |
| 6615 | 4-aligned. So just fish the whole thing out of memory right now |
| 6616 | and have done. */ |
| 6617 | UInt insn = getUIntLittleEndianly( guest_instr ); |
| 6618 | |
| 6619 | if (0) vex_printf("insn: 0x%x\n", insn); |
| 6620 | |
| 6621 | DIP("\t(arm64) 0x%llx: ", (ULong)guest_PC_curr_instr); |
| 6622 | |
| 6623 | vassert(0 == (guest_PC_curr_instr & 3ULL)); |
| 6624 | |
| 6625 | /* ----------------------------------------------------------- */ |
| 6626 | |
| 6627 | /* Spot "Special" instructions (see comment at top of file). */ |
| 6628 | { |
| 6629 | UChar* code = (UChar*)guest_instr; |
| 6630 | /* Spot the 16-byte preamble: |
| 6631 | 93CC0D8C ror x12, x12, #3 |
| 6632 | 93CC358C ror x12, x12, #13 |
| 6633 | 93CCCD8C ror x12, x12, #51 |
| 6634 | 93CCF58C ror x12, x12, #61 |
| 6635 | */ |
| 6636 | UInt word1 = 0x93CC0D8C; |
| 6637 | UInt word2 = 0x93CC358C; |
| 6638 | UInt word3 = 0x93CCCD8C; |
| 6639 | UInt word4 = 0x93CCF58C; |
| 6640 | if (getUIntLittleEndianly(code+ 0) == word1 && |
| 6641 | getUIntLittleEndianly(code+ 4) == word2 && |
| 6642 | getUIntLittleEndianly(code+ 8) == word3 && |
| 6643 | getUIntLittleEndianly(code+12) == word4) { |
| 6644 | /* Got a "Special" instruction preamble. Which one is it? */ |
| 6645 | if (getUIntLittleEndianly(code+16) == 0xAA0A014A |
| 6646 | /* orr x10,x10,x10 */) { |
| 6647 | /* X3 = client_request ( X4 ) */ |
| 6648 | DIP("x3 = client_request ( x4 )\n"); |
| 6649 | putPC(mkU64( guest_PC_curr_instr + 20 )); |
| 6650 | dres->jk_StopHere = Ijk_ClientReq; |
| 6651 | dres->whatNext = Dis_StopHere; |
| 6652 | return True; |
| 6653 | } |
| 6654 | else |
| 6655 | if (getUIntLittleEndianly(code+16) == 0xAA0B016B |
| 6656 | /* orr x11,x11,x11 */) { |
| 6657 | /* X3 = guest_NRADDR */ |
| 6658 | DIP("x3 = guest_NRADDR\n"); |
| 6659 | dres->len = 20; |
| 6660 | putIReg64orZR(3, IRExpr_Get( OFFB_NRADDR, Ity_I64 )); |
| 6661 | return True; |
| 6662 | } |
| 6663 | else |
| 6664 | if (getUIntLittleEndianly(code+16) == 0xAA0C018C |
| 6665 | /* orr x12,x12,x12 */) { |
| 6666 | /* branch-and-link-to-noredir X8 */ |
| 6667 | DIP("branch-and-link-to-noredir x8\n"); |
| 6668 | putIReg64orZR(30, mkU64(guest_PC_curr_instr + 20)); |
| 6669 | putPC(getIReg64orZR(8)); |
| 6670 | dres->jk_StopHere = Ijk_NoRedir; |
| 6671 | dres->whatNext = Dis_StopHere; |
| 6672 | return True; |
| 6673 | } |
| 6674 | else |
| 6675 | if (getUIntLittleEndianly(code+16) == 0xAA090129 |
| 6676 | /* orr x9,x9,x9 */) { |
| 6677 | /* IR injection */ |
| 6678 | DIP("IR injection\n"); |
| 6679 | vex_inject_ir(irsb, Iend_LE); |
| 6680 | // Invalidate the current insn. The reason is that the IRop we're |
| 6681 | // injecting here can change. In which case the translation has to |
| 6682 | // be redone. For ease of handling, we simply invalidate all the |
| 6683 | // time. |
| 6684 | stmt(IRStmt_Put(OFFB_TISTART, mkU64(guest_PC_curr_instr))); |
| 6685 | stmt(IRStmt_Put(OFFB_TILEN, mkU64(20))); |
| 6686 | putPC(mkU64( guest_PC_curr_instr + 20 )); |
| 6687 | dres->whatNext = Dis_StopHere; |
| 6688 | dres->jk_StopHere = Ijk_TInval; |
| 6689 | return True; |
| 6690 | } |
| 6691 | /* We don't know what it is. */ |
| 6692 | return False; |
| 6693 | /*NOTREACHED*/ |
| 6694 | } |
| 6695 | } |
| 6696 | |
| 6697 | /* ----------------------------------------------------------- */ |
| 6698 | |
| 6699 | /* Main ARM64 instruction decoder starts here. */ |
| 6700 | |
| 6701 | Bool ok = False; |
| 6702 | |
| 6703 | /* insn[28:25] determines the top-level grouping, so let's start |
| 6704 | off with that. |
| 6705 | |
| 6706 | For all of these dis_ARM64_ functions, we pass *dres with the |
| 6707 | normal default results "insn OK, 4 bytes long, keep decoding" so |
| 6708 | they don't need to change it. However, decodes of control-flow |
| 6709 | insns may cause *dres to change. |
| 6710 | */ |
| 6711 | switch (INSN(28,25)) { |
| 6712 | case BITS4(1,0,0,0): case BITS4(1,0,0,1): |
| 6713 | // Data processing - immediate |
| 6714 | ok = dis_ARM64_data_processing_immediate(dres, insn); |
| 6715 | break; |
| 6716 | case BITS4(1,0,1,0): case BITS4(1,0,1,1): |
| 6717 | // Branch, exception generation and system instructions |
| 6718 | ok = dis_ARM64_branch_etc(dres, insn); |
| 6719 | break; |
| 6720 | case BITS4(0,1,0,0): case BITS4(0,1,1,0): |
| 6721 | case BITS4(1,1,0,0): case BITS4(1,1,1,0): |
| 6722 | // Loads and stores |
| 6723 | ok = dis_ARM64_load_store(dres, insn); |
| 6724 | break; |
| 6725 | case BITS4(0,1,0,1): case BITS4(1,1,0,1): |
| 6726 | // Data processing - register |
| 6727 | ok = dis_ARM64_data_processing_register(dres, insn); |
| 6728 | break; |
| 6729 | case BITS4(0,1,1,1): case BITS4(1,1,1,1): |
| 6730 | // Data processing - SIMD and floating point |
| 6731 | ok = dis_ARM64_simd_and_fp(dres, insn); |
| 6732 | break; |
| 6733 | case BITS4(0,0,0,0): case BITS4(0,0,0,1): |
| 6734 | case BITS4(0,0,1,0): case BITS4(0,0,1,1): |
| 6735 | // UNALLOCATED |
| 6736 | break; |
| 6737 | default: |
| 6738 | vassert(0); /* Can't happen */ |
| 6739 | } |
| 6740 | |
| 6741 | /* If the next-level down decoders failed, make sure |dres| didn't |
| 6742 | get changed. */ |
| 6743 | if (!ok) { |
| 6744 | vassert(dres->whatNext == Dis_Continue); |
| 6745 | vassert(dres->len == 4); |
| 6746 | vassert(dres->continueAt == 0); |
| 6747 | vassert(dres->jk_StopHere == Ijk_INVALID); |
| 6748 | } |
| 6749 | |
| 6750 | return ok; |
| 6751 | |
| 6752 | # undef INSN |
| 6753 | } |
| 6754 | |
| 6755 | |
| 6756 | /*------------------------------------------------------------*/ |
| 6757 | /*--- Top-level fn ---*/ |
| 6758 | /*------------------------------------------------------------*/ |
| 6759 | |
| 6760 | /* Disassemble a single instruction into IR. The instruction |
| 6761 | is located in host memory at &guest_code[delta]. */ |
| 6762 | |
| 6763 | DisResult disInstr_ARM64 ( IRSB* irsb_IN, |
| 6764 | Bool (*resteerOkFn) ( void*, Addr64 ), |
| 6765 | Bool resteerCisOk, |
| 6766 | void* callback_opaque, |
| 6767 | UChar* guest_code_IN, |
| 6768 | Long delta_IN, |
| 6769 | Addr64 guest_IP, |
| 6770 | VexArch guest_arch, |
| 6771 | VexArchInfo* archinfo, |
| 6772 | VexAbiInfo* abiinfo, |
| 6773 | Bool host_bigendian_IN, |
| 6774 | Bool sigill_diag_IN ) |
| 6775 | { |
| 6776 | DisResult dres; |
| 6777 | vex_bzero(&dres, sizeof(dres)); |
| 6778 | |
| 6779 | /* Set globals (see top of this file) */ |
| 6780 | vassert(guest_arch == VexArchARM64); |
| 6781 | |
| 6782 | irsb = irsb_IN; |
| 6783 | host_is_bigendian = host_bigendian_IN; |
| 6784 | guest_PC_curr_instr = (Addr64)guest_IP; |
| 6785 | |
| 6786 | /* Try to decode */ |
| 6787 | Bool ok = disInstr_ARM64_WRK( &dres, |
| 6788 | resteerOkFn, resteerCisOk, callback_opaque, |
| 6789 | (UChar*)&guest_code_IN[delta_IN], |
| 6790 | archinfo, abiinfo ); |
| 6791 | if (ok) { |
| 6792 | /* All decode successes end up here. */ |
sewardj | dc9259c | 2014-02-27 11:10:19 +0000 | [diff] [blame^] | 6793 | vassert(dres.len == 4 || dres.len == 20); |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 6794 | switch (dres.whatNext) { |
| 6795 | case Dis_Continue: |
| 6796 | putPC( mkU64(dres.len + guest_PC_curr_instr) ); |
| 6797 | break; |
| 6798 | case Dis_ResteerU: |
| 6799 | case Dis_ResteerC: |
| 6800 | putPC(mkU64(dres.continueAt)); |
| 6801 | break; |
| 6802 | case Dis_StopHere: |
| 6803 | break; |
| 6804 | default: |
| 6805 | vassert(0); |
| 6806 | } |
| 6807 | DIP("\n"); |
| 6808 | } else { |
| 6809 | /* All decode failures end up here. */ |
| 6810 | if (sigill_diag_IN) { |
| 6811 | Int i, j; |
| 6812 | UChar buf[64]; |
| 6813 | UInt insn |
| 6814 | = getUIntLittleEndianly( (UChar*)&guest_code_IN[delta_IN] ); |
| 6815 | vex_bzero(buf, sizeof(buf)); |
| 6816 | for (i = j = 0; i < 32; i++) { |
| 6817 | if (i > 0) { |
| 6818 | if ((i & 7) == 0) buf[j++] = ' '; |
| 6819 | else if ((i & 3) == 0) buf[j++] = '\''; |
| 6820 | } |
| 6821 | buf[j++] = (insn & (1<<(31-i))) ? '1' : '0'; |
| 6822 | } |
| 6823 | vex_printf("disInstr(arm64): unhandled instruction 0x%08x\n", insn); |
| 6824 | vex_printf("disInstr(arm64): %s\n", buf); |
| 6825 | } |
| 6826 | |
| 6827 | /* Tell the dispatcher that this insn cannot be decoded, and so |
| 6828 | has not been executed, and (is currently) the next to be |
| 6829 | executed. PC should be up-to-date since it is made so at the |
| 6830 | start of each insn, but nevertheless be paranoid and update |
| 6831 | it again right now. */ |
| 6832 | putPC( mkU64(guest_PC_curr_instr) ); |
| 6833 | dres.whatNext = Dis_StopHere; |
| 6834 | dres.len = 0; |
| 6835 | dres.continueAt = 0; |
| 6836 | dres.jk_StopHere = Ijk_NoDecode; |
| 6837 | } |
| 6838 | return dres; |
| 6839 | } |
| 6840 | |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 6841 | //////////////////////////////////////////////////////////////////////// |
| 6842 | //////////////////////////////////////////////////////////////////////// |
| 6843 | |
| 6844 | /* Spare code for doing reference implementations of various 128-bit |
| 6845 | SIMD interleaves/deinterleaves/concatenation ops. For 64-bit |
| 6846 | equivalents see the end of guest_arm_toIR.c. */ |
| 6847 | |
| 6848 | //////////////////////////////////////////////////////////////// |
| 6849 | // 64x2 operations |
| 6850 | // |
| 6851 | static IRExpr* mk_CatEvenLanes64x2 ( IRTemp a10, IRTemp b10 ) |
| 6852 | { |
| 6853 | // returns a0 b0 |
| 6854 | return binop(Iop_64HLtoV128, unop(Iop_V128to64, mkexpr(a10)), |
| 6855 | unop(Iop_V128to64, mkexpr(b10))); |
| 6856 | } |
| 6857 | |
| 6858 | static IRExpr* mk_CatOddLanes64x2 ( IRTemp a10, IRTemp b10 ) |
| 6859 | { |
| 6860 | // returns a1 b1 |
| 6861 | return binop(Iop_64HLtoV128, unop(Iop_V128HIto64, mkexpr(a10)), |
| 6862 | unop(Iop_V128HIto64, mkexpr(b10))); |
| 6863 | } |
| 6864 | |
| 6865 | |
| 6866 | //////////////////////////////////////////////////////////////// |
| 6867 | // 32x4 operations |
| 6868 | // |
| 6869 | |
| 6870 | // Split a 128 bit value into 4 32 bit ones, in 64-bit IRTemps with |
| 6871 | // the top halves guaranteed to be zero. |
| 6872 | static void breakV128to32s ( IRTemp* out3, IRTemp* out2, IRTemp* out1, |
| 6873 | IRTemp* out0, IRTemp v128 ) |
| 6874 | { |
| 6875 | if (out3) *out3 = newTemp(Ity_I64); |
| 6876 | if (out2) *out2 = newTemp(Ity_I64); |
| 6877 | if (out1) *out1 = newTemp(Ity_I64); |
| 6878 | if (out0) *out0 = newTemp(Ity_I64); |
| 6879 | IRTemp hi64 = newTemp(Ity_I64); |
| 6880 | IRTemp lo64 = newTemp(Ity_I64); |
| 6881 | assign(hi64, unop(Iop_V128HIto64, mkexpr(v128)) ); |
| 6882 | assign(lo64, unop(Iop_V128to64, mkexpr(v128)) ); |
| 6883 | if (out3) assign(*out3, binop(Iop_Shr64, mkexpr(hi64), mkU8(32))); |
| 6884 | if (out2) assign(*out2, binop(Iop_And64, mkexpr(hi64), mkU64(0xFFFFFFFF))); |
| 6885 | if (out1) assign(*out1, binop(Iop_Shr64, mkexpr(lo64), mkU8(32))); |
| 6886 | if (out0) assign(*out0, binop(Iop_And64, mkexpr(lo64), mkU64(0xFFFFFFFF))); |
| 6887 | } |
| 6888 | |
| 6889 | // Make a V128 bit value from 4 32 bit ones, each of which is in a 64 bit |
| 6890 | // IRTemp. |
| 6891 | static IRTemp mkV128from32s ( IRTemp in3, IRTemp in2, IRTemp in1, IRTemp in0 ) |
| 6892 | { |
| 6893 | IRTemp hi64 = newTemp(Ity_I64); |
| 6894 | IRTemp lo64 = newTemp(Ity_I64); |
| 6895 | assign(hi64, |
| 6896 | binop(Iop_Or64, |
| 6897 | binop(Iop_Shl64, mkexpr(in3), mkU8(32)), |
| 6898 | binop(Iop_And64, mkexpr(in2), mkU64(0xFFFFFFFF)))); |
| 6899 | assign(lo64, |
| 6900 | binop(Iop_Or64, |
| 6901 | binop(Iop_Shl64, mkexpr(in1), mkU8(32)), |
| 6902 | binop(Iop_And64, mkexpr(in0), mkU64(0xFFFFFFFF)))); |
| 6903 | IRTemp res = newTemp(Ity_V128); |
| 6904 | assign(res, binop(Iop_64HLtoV128, mkexpr(hi64), mkexpr(lo64))); |
| 6905 | return res; |
| 6906 | } |
| 6907 | |
| 6908 | static IRExpr* mk_CatEvenLanes32x4 ( IRTemp a3210, IRTemp b3210 ) |
| 6909 | { |
| 6910 | // returns a2 a0 b2 b0 |
| 6911 | IRTemp a2, a0, b2, b0; |
| 6912 | breakV128to32s(NULL, &a2, NULL, &a0, a3210); |
| 6913 | breakV128to32s(NULL, &b2, NULL, &b0, b3210); |
| 6914 | return mkexpr(mkV128from32s(a2, a0, b2, b0)); |
| 6915 | } |
| 6916 | |
| 6917 | static IRExpr* mk_CatOddLanes32x4 ( IRTemp a3210, IRTemp b3210 ) |
| 6918 | { |
| 6919 | // returns a3 a1 b3 b1 |
| 6920 | IRTemp a3, a1, b3, b1; |
| 6921 | breakV128to32s(&a3, NULL, &a1, NULL, a3210); |
| 6922 | breakV128to32s(&b3, NULL, &b1, NULL, b3210); |
| 6923 | return mkexpr(mkV128from32s(a3, a1, b3, b1)); |
| 6924 | } |
| 6925 | |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 6926 | static IRExpr* mk_InterleaveLO32x4 ( IRTemp a3210, IRTemp b3210 ) |
| 6927 | { |
| 6928 | // returns a1 b1 a0 b0 |
| 6929 | IRTemp a1, a0, b1, b0; |
| 6930 | breakV128to32s(NULL, NULL, &a1, &a0, a3210); |
| 6931 | breakV128to32s(NULL, NULL, &b1, &b0, b3210); |
| 6932 | return mkexpr(mkV128from32s(a1, b1, a0, b0)); |
| 6933 | } |
| 6934 | |
| 6935 | static IRExpr* mk_InterleaveHI32x4 ( IRTemp a3210, IRTemp b3210 ) |
| 6936 | { |
| 6937 | // returns a3 b3 a2 b2 |
| 6938 | IRTemp a3, a2, b3, b2; |
| 6939 | breakV128to32s(&a3, &a2, NULL, NULL, a3210); |
| 6940 | breakV128to32s(&b3, &b2, NULL, NULL, b3210); |
| 6941 | return mkexpr(mkV128from32s(a3, b3, a2, b2)); |
| 6942 | } |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 6943 | |
| 6944 | //////////////////////////////////////////////////////////////// |
| 6945 | // 16x8 operations |
| 6946 | // |
| 6947 | |
| 6948 | static void breakV128to16s ( IRTemp* out7, IRTemp* out6, IRTemp* out5, |
| 6949 | IRTemp* out4, IRTemp* out3, IRTemp* out2, |
| 6950 | IRTemp* out1,IRTemp* out0, IRTemp v128 ) |
| 6951 | { |
| 6952 | if (out7) *out7 = newTemp(Ity_I64); |
| 6953 | if (out6) *out6 = newTemp(Ity_I64); |
| 6954 | if (out5) *out5 = newTemp(Ity_I64); |
| 6955 | if (out4) *out4 = newTemp(Ity_I64); |
| 6956 | if (out3) *out3 = newTemp(Ity_I64); |
| 6957 | if (out2) *out2 = newTemp(Ity_I64); |
| 6958 | if (out1) *out1 = newTemp(Ity_I64); |
| 6959 | if (out0) *out0 = newTemp(Ity_I64); |
| 6960 | IRTemp hi64 = newTemp(Ity_I64); |
| 6961 | IRTemp lo64 = newTemp(Ity_I64); |
| 6962 | assign(hi64, unop(Iop_V128HIto64, mkexpr(v128)) ); |
| 6963 | assign(lo64, unop(Iop_V128to64, mkexpr(v128)) ); |
| 6964 | if (out7) |
| 6965 | assign(*out7, binop(Iop_And64, |
| 6966 | binop(Iop_Shr64, mkexpr(hi64), mkU8(48)), |
| 6967 | mkU64(0xFFFF))); |
| 6968 | if (out6) |
| 6969 | assign(*out6, binop(Iop_And64, |
| 6970 | binop(Iop_Shr64, mkexpr(hi64), mkU8(32)), |
| 6971 | mkU64(0xFFFF))); |
| 6972 | if (out5) |
| 6973 | assign(*out5, binop(Iop_And64, |
| 6974 | binop(Iop_Shr64, mkexpr(hi64), mkU8(16)), |
| 6975 | mkU64(0xFFFF))); |
| 6976 | if (out4) |
| 6977 | assign(*out4, binop(Iop_And64, mkexpr(hi64), mkU64(0xFFFF))); |
| 6978 | if (out3) |
| 6979 | assign(*out3, binop(Iop_And64, |
| 6980 | binop(Iop_Shr64, mkexpr(lo64), mkU8(48)), |
| 6981 | mkU64(0xFFFF))); |
| 6982 | if (out2) |
| 6983 | assign(*out2, binop(Iop_And64, |
| 6984 | binop(Iop_Shr64, mkexpr(lo64), mkU8(32)), |
| 6985 | mkU64(0xFFFF))); |
| 6986 | if (out1) |
| 6987 | assign(*out1, binop(Iop_And64, |
| 6988 | binop(Iop_Shr64, mkexpr(lo64), mkU8(16)), |
| 6989 | mkU64(0xFFFF))); |
| 6990 | if (out0) |
| 6991 | assign(*out0, binop(Iop_And64, mkexpr(lo64), mkU64(0xFFFF))); |
| 6992 | } |
| 6993 | |
| 6994 | static IRTemp mkV128from16s ( IRTemp in7, IRTemp in6, IRTemp in5, IRTemp in4, |
| 6995 | IRTemp in3, IRTemp in2, IRTemp in1, IRTemp in0 ) |
| 6996 | { |
| 6997 | IRTemp hi64 = newTemp(Ity_I64); |
| 6998 | IRTemp lo64 = newTemp(Ity_I64); |
| 6999 | assign(hi64, |
| 7000 | binop(Iop_Or64, |
| 7001 | binop(Iop_Or64, |
| 7002 | binop(Iop_Shl64, |
| 7003 | binop(Iop_And64, mkexpr(in7), mkU64(0xFFFF)), |
| 7004 | mkU8(48)), |
| 7005 | binop(Iop_Shl64, |
| 7006 | binop(Iop_And64, mkexpr(in6), mkU64(0xFFFF)), |
| 7007 | mkU8(32))), |
| 7008 | binop(Iop_Or64, |
| 7009 | binop(Iop_Shl64, |
| 7010 | binop(Iop_And64, mkexpr(in5), mkU64(0xFFFF)), |
| 7011 | mkU8(16)), |
| 7012 | binop(Iop_And64, |
| 7013 | mkexpr(in4), mkU64(0xFFFF))))); |
| 7014 | assign(lo64, |
| 7015 | binop(Iop_Or64, |
| 7016 | binop(Iop_Or64, |
| 7017 | binop(Iop_Shl64, |
| 7018 | binop(Iop_And64, mkexpr(in3), mkU64(0xFFFF)), |
| 7019 | mkU8(48)), |
| 7020 | binop(Iop_Shl64, |
| 7021 | binop(Iop_And64, mkexpr(in2), mkU64(0xFFFF)), |
| 7022 | mkU8(32))), |
| 7023 | binop(Iop_Or64, |
| 7024 | binop(Iop_Shl64, |
| 7025 | binop(Iop_And64, mkexpr(in1), mkU64(0xFFFF)), |
| 7026 | mkU8(16)), |
| 7027 | binop(Iop_And64, |
| 7028 | mkexpr(in0), mkU64(0xFFFF))))); |
| 7029 | IRTemp res = newTemp(Ity_V128); |
| 7030 | assign(res, binop(Iop_64HLtoV128, mkexpr(hi64), mkexpr(lo64))); |
| 7031 | return res; |
| 7032 | } |
| 7033 | |
| 7034 | static IRExpr* mk_CatEvenLanes16x8 ( IRTemp a76543210, IRTemp b76543210 ) |
| 7035 | { |
| 7036 | // returns a6 a4 a2 a0 b6 b4 b2 b0 |
| 7037 | IRTemp a6, a4, a2, a0, b6, b4, b2, b0; |
| 7038 | breakV128to16s(NULL, &a6, NULL, &a4, NULL, &a2, NULL, &a0, a76543210); |
| 7039 | breakV128to16s(NULL, &b6, NULL, &b4, NULL, &b2, NULL, &b0, b76543210); |
| 7040 | return mkexpr(mkV128from16s(a6, a4, a2, a0, b6, b4, b2, b0)); |
| 7041 | } |
| 7042 | |
| 7043 | static IRExpr* mk_CatOddLanes16x8 ( IRTemp a76543210, IRTemp b76543210 ) |
| 7044 | { |
| 7045 | // returns a7 a5 a3 a1 b7 b5 b3 b1 |
| 7046 | IRTemp a7, a5, a3, a1, b7, b5, b3, b1; |
| 7047 | breakV128to16s(&a7, NULL, &a5, NULL, &a3, NULL, &a1, NULL, a76543210); |
| 7048 | breakV128to16s(&b7, NULL, &b5, NULL, &b3, NULL, &b1, NULL, b76543210); |
| 7049 | return mkexpr(mkV128from16s(a7, a5, a3, a1, b7, b5, b3, b1)); |
| 7050 | } |
| 7051 | |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 7052 | static IRExpr* mk_InterleaveLO16x8 ( IRTemp a76543210, IRTemp b76543210 ) |
| 7053 | { |
| 7054 | // returns a3 b3 a2 b2 a1 b1 a0 b0 |
| 7055 | IRTemp a3, b3, a2, b2, a1, a0, b1, b0; |
| 7056 | breakV128to16s(NULL, NULL, NULL, NULL, &a3, &a2, &a1, &a0, a76543210); |
| 7057 | breakV128to16s(NULL, NULL, NULL, NULL, &b3, &b2, &b1, &b0, b76543210); |
| 7058 | return mkexpr(mkV128from16s(a3, b3, a2, b2, a1, b1, a0, b0)); |
| 7059 | } |
| 7060 | |
| 7061 | static IRExpr* mk_InterleaveHI16x8 ( IRTemp a76543210, IRTemp b76543210 ) |
| 7062 | { |
| 7063 | // returns a7 b7 a6 b6 a5 b5 a4 b4 |
| 7064 | IRTemp a7, b7, a6, b6, a5, b5, a4, b4; |
| 7065 | breakV128to16s(&a7, &a6, &a5, &a4, NULL, NULL, NULL, NULL, a76543210); |
| 7066 | breakV128to16s(&b7, &b6, &b5, &b4, NULL, NULL, NULL, NULL, b76543210); |
| 7067 | return mkexpr(mkV128from16s(a7, b7, a6, b6, a5, b5, a4, b4)); |
| 7068 | } |
| 7069 | |
sewardj | fab0914 | 2014-02-10 10:28:13 +0000 | [diff] [blame] | 7070 | //////////////////////////////////////////////////////////////// |
| 7071 | // 8x16 operations |
| 7072 | // |
| 7073 | |
| 7074 | static void breakV128to8s ( IRTemp* outF, IRTemp* outE, IRTemp* outD, |
| 7075 | IRTemp* outC, IRTemp* outB, IRTemp* outA, |
| 7076 | IRTemp* out9, IRTemp* out8, |
| 7077 | IRTemp* out7, IRTemp* out6, IRTemp* out5, |
| 7078 | IRTemp* out4, IRTemp* out3, IRTemp* out2, |
| 7079 | IRTemp* out1,IRTemp* out0, IRTemp v128 ) |
| 7080 | { |
| 7081 | if (outF) *outF = newTemp(Ity_I64); |
| 7082 | if (outE) *outE = newTemp(Ity_I64); |
| 7083 | if (outD) *outD = newTemp(Ity_I64); |
| 7084 | if (outC) *outC = newTemp(Ity_I64); |
| 7085 | if (outB) *outB = newTemp(Ity_I64); |
| 7086 | if (outA) *outA = newTemp(Ity_I64); |
| 7087 | if (out9) *out9 = newTemp(Ity_I64); |
| 7088 | if (out8) *out8 = newTemp(Ity_I64); |
| 7089 | if (out7) *out7 = newTemp(Ity_I64); |
| 7090 | if (out6) *out6 = newTemp(Ity_I64); |
| 7091 | if (out5) *out5 = newTemp(Ity_I64); |
| 7092 | if (out4) *out4 = newTemp(Ity_I64); |
| 7093 | if (out3) *out3 = newTemp(Ity_I64); |
| 7094 | if (out2) *out2 = newTemp(Ity_I64); |
| 7095 | if (out1) *out1 = newTemp(Ity_I64); |
| 7096 | if (out0) *out0 = newTemp(Ity_I64); |
| 7097 | IRTemp hi64 = newTemp(Ity_I64); |
| 7098 | IRTemp lo64 = newTemp(Ity_I64); |
| 7099 | assign(hi64, unop(Iop_V128HIto64, mkexpr(v128)) ); |
| 7100 | assign(lo64, unop(Iop_V128to64, mkexpr(v128)) ); |
| 7101 | if (outF) |
| 7102 | assign(*outF, binop(Iop_And64, |
| 7103 | binop(Iop_Shr64, mkexpr(hi64), mkU8(56)), |
| 7104 | mkU64(0xFF))); |
| 7105 | if (outE) |
| 7106 | assign(*outE, binop(Iop_And64, |
| 7107 | binop(Iop_Shr64, mkexpr(hi64), mkU8(48)), |
| 7108 | mkU64(0xFF))); |
| 7109 | if (outD) |
| 7110 | assign(*outD, binop(Iop_And64, |
| 7111 | binop(Iop_Shr64, mkexpr(hi64), mkU8(40)), |
| 7112 | mkU64(0xFF))); |
| 7113 | if (outC) |
| 7114 | assign(*outC, binop(Iop_And64, |
| 7115 | binop(Iop_Shr64, mkexpr(hi64), mkU8(32)), |
| 7116 | mkU64(0xFF))); |
| 7117 | if (outB) |
| 7118 | assign(*outB, binop(Iop_And64, |
| 7119 | binop(Iop_Shr64, mkexpr(hi64), mkU8(24)), |
| 7120 | mkU64(0xFF))); |
| 7121 | if (outA) |
| 7122 | assign(*outA, binop(Iop_And64, |
| 7123 | binop(Iop_Shr64, mkexpr(hi64), mkU8(16)), |
| 7124 | mkU64(0xFF))); |
| 7125 | if (out9) |
| 7126 | assign(*out9, binop(Iop_And64, |
| 7127 | binop(Iop_Shr64, mkexpr(hi64), mkU8(8)), |
| 7128 | mkU64(0xFF))); |
| 7129 | if (out8) |
| 7130 | assign(*out8, binop(Iop_And64, |
| 7131 | binop(Iop_Shr64, mkexpr(hi64), mkU8(0)), |
| 7132 | mkU64(0xFF))); |
| 7133 | if (out7) |
| 7134 | assign(*out7, binop(Iop_And64, |
| 7135 | binop(Iop_Shr64, mkexpr(lo64), mkU8(56)), |
| 7136 | mkU64(0xFF))); |
| 7137 | if (out6) |
| 7138 | assign(*out6, binop(Iop_And64, |
| 7139 | binop(Iop_Shr64, mkexpr(lo64), mkU8(48)), |
| 7140 | mkU64(0xFF))); |
| 7141 | if (out5) |
| 7142 | assign(*out5, binop(Iop_And64, |
| 7143 | binop(Iop_Shr64, mkexpr(lo64), mkU8(40)), |
| 7144 | mkU64(0xFF))); |
| 7145 | if (out4) |
| 7146 | assign(*out4, binop(Iop_And64, |
| 7147 | binop(Iop_Shr64, mkexpr(lo64), mkU8(32)), |
| 7148 | mkU64(0xFF))); |
| 7149 | if (out3) |
| 7150 | assign(*out3, binop(Iop_And64, |
| 7151 | binop(Iop_Shr64, mkexpr(lo64), mkU8(24)), |
| 7152 | mkU64(0xFF))); |
| 7153 | if (out2) |
| 7154 | assign(*out2, binop(Iop_And64, |
| 7155 | binop(Iop_Shr64, mkexpr(lo64), mkU8(16)), |
| 7156 | mkU64(0xFF))); |
| 7157 | if (out1) |
| 7158 | assign(*out1, binop(Iop_And64, |
| 7159 | binop(Iop_Shr64, mkexpr(lo64), mkU8(8)), |
| 7160 | mkU64(0xFF))); |
| 7161 | if (out0) |
| 7162 | assign(*out0, binop(Iop_And64, |
| 7163 | binop(Iop_Shr64, mkexpr(lo64), mkU8(0)), |
| 7164 | mkU64(0xFF))); |
| 7165 | } |
| 7166 | |
| 7167 | static IRTemp mkV128from8s ( IRTemp inF, IRTemp inE, IRTemp inD, IRTemp inC, |
| 7168 | IRTemp inB, IRTemp inA, IRTemp in9, IRTemp in8, |
| 7169 | IRTemp in7, IRTemp in6, IRTemp in5, IRTemp in4, |
| 7170 | IRTemp in3, IRTemp in2, IRTemp in1, IRTemp in0 ) |
| 7171 | { |
| 7172 | IRTemp vFE = newTemp(Ity_I64); |
| 7173 | IRTemp vDC = newTemp(Ity_I64); |
| 7174 | IRTemp vBA = newTemp(Ity_I64); |
| 7175 | IRTemp v98 = newTemp(Ity_I64); |
| 7176 | IRTemp v76 = newTemp(Ity_I64); |
| 7177 | IRTemp v54 = newTemp(Ity_I64); |
| 7178 | IRTemp v32 = newTemp(Ity_I64); |
| 7179 | IRTemp v10 = newTemp(Ity_I64); |
| 7180 | assign(vFE, binop(Iop_Or64, |
| 7181 | binop(Iop_Shl64, |
| 7182 | binop(Iop_And64, mkexpr(inF), mkU64(0xFF)), mkU8(8)), |
| 7183 | binop(Iop_And64, mkexpr(inE), mkU64(0xFF)))); |
| 7184 | assign(vDC, binop(Iop_Or64, |
| 7185 | binop(Iop_Shl64, |
| 7186 | binop(Iop_And64, mkexpr(inD), mkU64(0xFF)), mkU8(8)), |
| 7187 | binop(Iop_And64, mkexpr(inC), mkU64(0xFF)))); |
| 7188 | assign(vBA, binop(Iop_Or64, |
| 7189 | binop(Iop_Shl64, |
| 7190 | binop(Iop_And64, mkexpr(inB), mkU64(0xFF)), mkU8(8)), |
| 7191 | binop(Iop_And64, mkexpr(inA), mkU64(0xFF)))); |
| 7192 | assign(v98, binop(Iop_Or64, |
| 7193 | binop(Iop_Shl64, |
| 7194 | binop(Iop_And64, mkexpr(in9), mkU64(0xFF)), mkU8(8)), |
| 7195 | binop(Iop_And64, mkexpr(in8), mkU64(0xFF)))); |
| 7196 | assign(v76, binop(Iop_Or64, |
| 7197 | binop(Iop_Shl64, |
| 7198 | binop(Iop_And64, mkexpr(in7), mkU64(0xFF)), mkU8(8)), |
| 7199 | binop(Iop_And64, mkexpr(in6), mkU64(0xFF)))); |
| 7200 | assign(v54, binop(Iop_Or64, |
| 7201 | binop(Iop_Shl64, |
| 7202 | binop(Iop_And64, mkexpr(in5), mkU64(0xFF)), mkU8(8)), |
| 7203 | binop(Iop_And64, mkexpr(in4), mkU64(0xFF)))); |
| 7204 | assign(v32, binop(Iop_Or64, |
| 7205 | binop(Iop_Shl64, |
| 7206 | binop(Iop_And64, mkexpr(in3), mkU64(0xFF)), mkU8(8)), |
| 7207 | binop(Iop_And64, mkexpr(in2), mkU64(0xFF)))); |
| 7208 | assign(v10, binop(Iop_Or64, |
| 7209 | binop(Iop_Shl64, |
| 7210 | binop(Iop_And64, mkexpr(in1), mkU64(0xFF)), mkU8(8)), |
| 7211 | binop(Iop_And64, mkexpr(in0), mkU64(0xFF)))); |
| 7212 | return mkV128from16s(vFE, vDC, vBA, v98, v76, v54, v32, v10); |
| 7213 | } |
| 7214 | |
| 7215 | static IRExpr* mk_CatEvenLanes8x16 ( IRTemp aFEDCBA9876543210, |
| 7216 | IRTemp bFEDCBA9876543210 ) |
| 7217 | { |
| 7218 | // returns aE aC aA a8 a6 a4 a2 a0 bE bC bA b8 b6 b4 b2 b0 |
| 7219 | IRTemp aE, aC, aA, a8, a6, a4, a2, a0, bE, bC, bA, b8, b6, b4, b2, b0; |
| 7220 | breakV128to8s(NULL, &aE, NULL, &aC, NULL, &aA, NULL, &a8, |
| 7221 | NULL, &a6, NULL, &a4, NULL, &a2, NULL, &a0, |
| 7222 | aFEDCBA9876543210); |
| 7223 | breakV128to8s(NULL, &bE, NULL, &bC, NULL, &bA, NULL, &b8, |
| 7224 | NULL, &b6, NULL, &b4, NULL, &b2, NULL, &b0, |
| 7225 | bFEDCBA9876543210); |
| 7226 | return mkexpr(mkV128from8s(aE, aC, aA, a8, a6, a4, a2, a0, |
| 7227 | bE, bC, bA, b8, b6, b4, b2, b0)); |
| 7228 | } |
| 7229 | |
| 7230 | static IRExpr* mk_CatOddLanes8x16 ( IRTemp aFEDCBA9876543210, |
| 7231 | IRTemp bFEDCBA9876543210 ) |
| 7232 | { |
| 7233 | // returns aF aD aB a9 a7 a5 a3 a1 bF bD bB b9 b7 b5 b3 b1 |
| 7234 | IRTemp aF, aD, aB, a9, a7, a5, a3, a1, bF, bD, bB, b9, b7, b5, b3, b1; |
| 7235 | breakV128to8s(&aF, NULL, &aD, NULL, &aB, NULL, &a9, NULL, |
| 7236 | &a7, NULL, &a5, NULL, &a3, NULL, &a1, NULL, |
| 7237 | aFEDCBA9876543210); |
| 7238 | |
| 7239 | breakV128to8s(&bF, NULL, &bD, NULL, &bB, NULL, &b9, NULL, |
| 7240 | &b7, NULL, &b5, NULL, &b3, NULL, &b1, NULL, |
| 7241 | aFEDCBA9876543210); |
| 7242 | |
| 7243 | return mkexpr(mkV128from8s(aF, aD, aB, a9, a7, a5, a3, a1, |
| 7244 | bF, bD, bB, b9, b7, b5, b3, b1)); |
| 7245 | } |
| 7246 | |
sewardj | e520bb3 | 2014-02-17 11:00:53 +0000 | [diff] [blame] | 7247 | static IRExpr* mk_InterleaveLO8x16 ( IRTemp aFEDCBA9876543210, |
| 7248 | IRTemp bFEDCBA9876543210 ) |
| 7249 | { |
| 7250 | // returns a7 b7 a6 b6 a5 b5 a4 b4 a3 b3 a2 b2 a1 b1 a0 b0 |
| 7251 | IRTemp a7, b7, a6, b6, a5, b5, a4, b4, a3, b3, a2, b2, a1, b1, a0, b0; |
| 7252 | breakV128to8s(NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, |
| 7253 | &a7, &a6, &a5, &a4, &a3, &a2, &a1, &a0, |
| 7254 | aFEDCBA9876543210); |
| 7255 | breakV128to8s(NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, |
| 7256 | &b7, &b6, &b5, &b4, &b3, &b2, &b1, &b0, |
| 7257 | bFEDCBA9876543210); |
| 7258 | return mkexpr(mkV128from8s(a7, b7, a6, b6, a5, b5, a4, b4, |
| 7259 | a3, b3, a2, b2, a1, b1, a0, b0)); |
| 7260 | } |
| 7261 | |
| 7262 | static IRExpr* mk_InterleaveHI8x16 ( IRTemp aFEDCBA9876543210, |
| 7263 | IRTemp bFEDCBA9876543210 ) |
| 7264 | { |
| 7265 | // returns aF bF aE bE aD bD aC bC aB bB aA bA a9 b9 a8 b8 |
| 7266 | IRTemp aF, bF, aE, bE, aD, bD, aC, bC, aB, bB, aA, bA, a9, b9, a8, b8; |
| 7267 | breakV128to8s(&aF, &aE, &aD, &aC, &aB, &aA, &a9, &a8, |
| 7268 | NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, |
| 7269 | aFEDCBA9876543210); |
| 7270 | breakV128to8s(&bF, &bE, &bD, &bC, &bB, &bA, &b9, &b8, |
| 7271 | NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, |
| 7272 | bFEDCBA9876543210); |
| 7273 | return mkexpr(mkV128from8s(aF, bF, aE, bE, aD, bD, aC, bC, |
| 7274 | aB, bB, aA, bA, a9, b9, a8, b8)); |
| 7275 | } |
sewardj | ecde697 | 2014-02-05 11:01:19 +0000 | [diff] [blame] | 7276 | |
sewardj | bbcf188 | 2014-01-12 12:49:10 +0000 | [diff] [blame] | 7277 | /*--------------------------------------------------------------------*/ |
| 7278 | /*--- end guest_arm64_toIR.c ---*/ |
| 7279 | /*--------------------------------------------------------------------*/ |