1 //===-- X86BaseInfo.h - Top level definitions for X86 -------- --*- C++ -*-===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This file contains small standalone helper functions and enum definitions for 10 // the X86 target useful for the compiler back-end and the MC libraries. 11 // As such, it deliberately does not include references to LLVM core 12 // code gen types, passes, etc.. 13 // 14 //===----------------------------------------------------------------------===// 15 16 #ifndef LLVM_LIB_TARGET_X86_MCTARGETDESC_X86BASEINFO_H 17 #define LLVM_LIB_TARGET_X86_MCTARGETDESC_X86BASEINFO_H 18 19 #include "X86MCTargetDesc.h" 20 #include "llvm/MC/MCInstrDesc.h" 21 #include "llvm/Support/DataTypes.h" 22 #include "llvm/Support/ErrorHandling.h" 23 24 namespace llvm { 25 26 namespace X86 { 27 // Enums for memory operand decoding. Each memory operand is represented with 28 // a 5 operand sequence in the form: 29 // [BaseReg, ScaleAmt, IndexReg, Disp, Segment] 30 // These enums help decode this. 31 enum { 32 AddrBaseReg = 0, 33 AddrScaleAmt = 1, 34 AddrIndexReg = 2, 35 AddrDisp = 3, 36 37 /// AddrSegmentReg - The operand # of the segment in the memory operand. 38 AddrSegmentReg = 4, 39 40 /// AddrNumOperands - Total number of operands in a memory reference. 41 AddrNumOperands = 5 42 }; 43 44 /// AVX512 static rounding constants. These need to match the values in 45 /// avx512fintrin.h. 46 enum STATIC_ROUNDING { 47 TO_NEAREST_INT = 0, 48 TO_NEG_INF = 1, 49 TO_POS_INF = 2, 50 TO_ZERO = 3, 51 CUR_DIRECTION = 4, 52 NO_EXC = 8 53 }; 54 55 /// The constants to describe instr prefixes if there are 56 enum IPREFIXES { 57 IP_NO_PREFIX = 0, 58 IP_HAS_OP_SIZE = 1, 59 IP_HAS_AD_SIZE = 2, 60 IP_HAS_REPEAT_NE = 4, 61 IP_HAS_REPEAT = 8, 62 IP_HAS_LOCK = 16, 63 IP_HAS_NOTRACK = 32, 64 IP_USE_VEX3 = 64, 65 }; 66 67 enum OperandType : unsigned { 68 /// AVX512 embedded rounding control. This should only have values 0-3. 69 OPERAND_ROUNDING_CONTROL = MCOI::OPERAND_FIRST_TARGET, 70 OPERAND_COND_CODE, 71 }; 72 73 // X86 specific condition code. These correspond to X86_*_COND in 74 // X86InstrInfo.td. They must be kept in synch. 75 enum CondCode { 76 COND_O = 0, 77 COND_NO = 1, 78 COND_B = 2, 79 COND_AE = 3, 80 COND_E = 4, 81 COND_NE = 5, 82 COND_BE = 6, 83 COND_A = 7, 84 COND_S = 8, 85 COND_NS = 9, 86 COND_P = 10, 87 COND_NP = 11, 88 COND_L = 12, 89 COND_GE = 13, 90 COND_LE = 14, 91 COND_G = 15, 92 LAST_VALID_COND = COND_G, 93 94 // Artificial condition codes. These are used by AnalyzeBranch 95 // to indicate a block terminated with two conditional branches that together 96 // form a compound condition. They occur in code using FCMP_OEQ or FCMP_UNE, 97 // which can't be represented on x86 with a single condition. These 98 // are never used in MachineInstrs and are inverses of one another. 99 COND_NE_OR_P, 100 COND_E_AND_NP, 101 102 COND_INVALID 103 }; 104 } // end namespace X86; 105 106 /// X86II - This namespace holds all of the target specific flags that 107 /// instruction info tracks. 108 /// 109 namespace X86II { 110 /// Target Operand Flag enum. 111 enum TOF { 112 //===------------------------------------------------------------------===// 113 // X86 Specific MachineOperand flags. 114 115 MO_NO_FLAG, 116 117 /// MO_GOT_ABSOLUTE_ADDRESS - On a symbol operand, this represents a 118 /// relocation of: 119 /// SYMBOL_LABEL + [. - PICBASELABEL] 120 MO_GOT_ABSOLUTE_ADDRESS, 121 122 /// MO_PIC_BASE_OFFSET - On a symbol operand this indicates that the 123 /// immediate should get the value of the symbol minus the PIC base label: 124 /// SYMBOL_LABEL - PICBASELABEL 125 MO_PIC_BASE_OFFSET, 126 127 /// MO_GOT - On a symbol operand this indicates that the immediate is the 128 /// offset to the GOT entry for the symbol name from the base of the GOT. 129 /// 130 /// See the X86-64 ELF ABI supplement for more details. 131 /// SYMBOL_LABEL @GOT 132 MO_GOT, 133 134 /// MO_GOTOFF - On a symbol operand this indicates that the immediate is 135 /// the offset to the location of the symbol name from the base of the GOT. 136 /// 137 /// See the X86-64 ELF ABI supplement for more details. 138 /// SYMBOL_LABEL @GOTOFF 139 MO_GOTOFF, 140 141 /// MO_GOTPCREL - On a symbol operand this indicates that the immediate is 142 /// offset to the GOT entry for the symbol name from the current code 143 /// location. 144 /// 145 /// See the X86-64 ELF ABI supplement for more details. 146 /// SYMBOL_LABEL @GOTPCREL 147 MO_GOTPCREL, 148 149 /// MO_PLT - On a symbol operand this indicates that the immediate is 150 /// offset to the PLT entry of symbol name from the current code location. 151 /// 152 /// See the X86-64 ELF ABI supplement for more details. 153 /// SYMBOL_LABEL @PLT 154 MO_PLT, 155 156 /// MO_TLSGD - On a symbol operand this indicates that the immediate is 157 /// the offset of the GOT entry with the TLS index structure that contains 158 /// the module number and variable offset for the symbol. Used in the 159 /// general dynamic TLS access model. 160 /// 161 /// See 'ELF Handling for Thread-Local Storage' for more details. 162 /// SYMBOL_LABEL @TLSGD 163 MO_TLSGD, 164 165 /// MO_TLSLD - On a symbol operand this indicates that the immediate is 166 /// the offset of the GOT entry with the TLS index for the module that 167 /// contains the symbol. When this index is passed to a call to 168 /// __tls_get_addr, the function will return the base address of the TLS 169 /// block for the symbol. Used in the x86-64 local dynamic TLS access model. 170 /// 171 /// See 'ELF Handling for Thread-Local Storage' for more details. 172 /// SYMBOL_LABEL @TLSLD 173 MO_TLSLD, 174 175 /// MO_TLSLDM - On a symbol operand this indicates that the immediate is 176 /// the offset of the GOT entry with the TLS index for the module that 177 /// contains the symbol. When this index is passed to a call to 178 /// ___tls_get_addr, the function will return the base address of the TLS 179 /// block for the symbol. Used in the IA32 local dynamic TLS access model. 180 /// 181 /// See 'ELF Handling for Thread-Local Storage' for more details. 182 /// SYMBOL_LABEL @TLSLDM 183 MO_TLSLDM, 184 185 /// MO_GOTTPOFF - On a symbol operand this indicates that the immediate is 186 /// the offset of the GOT entry with the thread-pointer offset for the 187 /// symbol. Used in the x86-64 initial exec TLS access model. 188 /// 189 /// See 'ELF Handling for Thread-Local Storage' for more details. 190 /// SYMBOL_LABEL @GOTTPOFF 191 MO_GOTTPOFF, 192 193 /// MO_INDNTPOFF - On a symbol operand this indicates that the immediate is 194 /// the absolute address of the GOT entry with the negative thread-pointer 195 /// offset for the symbol. Used in the non-PIC IA32 initial exec TLS access 196 /// model. 197 /// 198 /// See 'ELF Handling for Thread-Local Storage' for more details. 199 /// SYMBOL_LABEL @INDNTPOFF 200 MO_INDNTPOFF, 201 202 /// MO_TPOFF - On a symbol operand this indicates that the immediate is 203 /// the thread-pointer offset for the symbol. Used in the x86-64 local 204 /// exec TLS access model. 205 /// 206 /// See 'ELF Handling for Thread-Local Storage' for more details. 207 /// SYMBOL_LABEL @TPOFF 208 MO_TPOFF, 209 210 /// MO_DTPOFF - On a symbol operand this indicates that the immediate is 211 /// the offset of the GOT entry with the TLS offset of the symbol. Used 212 /// in the local dynamic TLS access model. 213 /// 214 /// See 'ELF Handling for Thread-Local Storage' for more details. 215 /// SYMBOL_LABEL @DTPOFF 216 MO_DTPOFF, 217 218 /// MO_NTPOFF - On a symbol operand this indicates that the immediate is 219 /// the negative thread-pointer offset for the symbol. Used in the IA32 220 /// local exec TLS access model. 221 /// 222 /// See 'ELF Handling for Thread-Local Storage' for more details. 223 /// SYMBOL_LABEL @NTPOFF 224 MO_NTPOFF, 225 226 /// MO_GOTNTPOFF - On a symbol operand this indicates that the immediate is 227 /// the offset of the GOT entry with the negative thread-pointer offset for 228 /// the symbol. Used in the PIC IA32 initial exec TLS access model. 229 /// 230 /// See 'ELF Handling for Thread-Local Storage' for more details. 231 /// SYMBOL_LABEL @GOTNTPOFF 232 MO_GOTNTPOFF, 233 234 /// MO_DLLIMPORT - On a symbol operand "FOO", this indicates that the 235 /// reference is actually to the "__imp_FOO" symbol. This is used for 236 /// dllimport linkage on windows. 237 MO_DLLIMPORT, 238 239 /// MO_DARWIN_NONLAZY - On a symbol operand "FOO", this indicates that the 240 /// reference is actually to the "FOO$non_lazy_ptr" symbol, which is a 241 /// non-PIC-base-relative reference to a non-hidden dyld lazy pointer stub. 242 MO_DARWIN_NONLAZY, 243 244 /// MO_DARWIN_NONLAZY_PIC_BASE - On a symbol operand "FOO", this indicates 245 /// that the reference is actually to "FOO$non_lazy_ptr - PICBASE", which is 246 /// a PIC-base-relative reference to a non-hidden dyld lazy pointer stub. 247 MO_DARWIN_NONLAZY_PIC_BASE, 248 249 /// MO_TLVP - On a symbol operand this indicates that the immediate is 250 /// some TLS offset. 251 /// 252 /// This is the TLS offset for the Darwin TLS mechanism. 253 MO_TLVP, 254 255 /// MO_TLVP_PIC_BASE - On a symbol operand this indicates that the immediate 256 /// is some TLS offset from the picbase. 257 /// 258 /// This is the 32-bit TLS offset for Darwin TLS in PIC mode. 259 MO_TLVP_PIC_BASE, 260 261 /// MO_SECREL - On a symbol operand this indicates that the immediate is 262 /// the offset from beginning of section. 263 /// 264 /// This is the TLS offset for the COFF/Windows TLS mechanism. 265 MO_SECREL, 266 267 /// MO_ABS8 - On a symbol operand this indicates that the symbol is known 268 /// to be an absolute symbol in range [0,128), so we can use the @ABS8 269 /// symbol modifier. 270 MO_ABS8, 271 272 /// MO_COFFSTUB - On a symbol operand "FOO", this indicates that the 273 /// reference is actually to the ".refptr.FOO" symbol. This is used for 274 /// stub symbols on windows. 275 MO_COFFSTUB, 276 }; 277 278 enum : uint64_t { 279 //===------------------------------------------------------------------===// 280 // Instruction encodings. These are the standard/most common forms for X86 281 // instructions. 282 // 283 284 // PseudoFrm - This represents an instruction that is a pseudo instruction 285 // or one that has not been implemented yet. It is illegal to code generate 286 // it, but tolerated for intermediate implementation stages. 287 Pseudo = 0, 288 289 /// Raw - This form is for instructions that don't have any operands, so 290 /// they are just a fixed opcode value, like 'leave'. 291 RawFrm = 1, 292 293 /// AddRegFrm - This form is used for instructions like 'push r32' that have 294 /// their one register operand added to their opcode. 295 AddRegFrm = 2, 296 297 /// RawFrmMemOffs - This form is for instructions that store an absolute 298 /// memory offset as an immediate with a possible segment override. 299 RawFrmMemOffs = 3, 300 301 /// RawFrmSrc - This form is for instructions that use the source index 302 /// register SI/ESI/RSI with a possible segment override. 303 RawFrmSrc = 4, 304 305 /// RawFrmDst - This form is for instructions that use the destination index 306 /// register DI/EDI/RDI. 307 RawFrmDst = 5, 308 309 /// RawFrmDstSrc - This form is for instructions that use the source index 310 /// register SI/ESI/RSI with a possible segment override, and also the 311 /// destination index register DI/EDI/RDI. 312 RawFrmDstSrc = 6, 313 314 /// RawFrmImm8 - This is used for the ENTER instruction, which has two 315 /// immediates, the first of which is a 16-bit immediate (specified by 316 /// the imm encoding) and the second is a 8-bit fixed value. 317 RawFrmImm8 = 7, 318 319 /// RawFrmImm16 - This is used for CALL FAR instructions, which have two 320 /// immediates, the first of which is a 16 or 32-bit immediate (specified by 321 /// the imm encoding) and the second is a 16-bit fixed value. In the AMD 322 /// manual, this operand is described as pntr16:32 and pntr16:16 323 RawFrmImm16 = 8, 324 325 /// AddCCFrm - This form is used for Jcc that encode the condition code 326 /// in the lower 4 bits of the opcode. 327 AddCCFrm = 9, 328 329 /// MRM[0-7][rm] - These forms are used to represent instructions that use 330 /// a Mod/RM byte, and use the middle field to hold extended opcode 331 /// information. In the intel manual these are represented as /0, /1, ... 332 /// 333 334 /// MRMDestMem - This form is used for instructions that use the Mod/RM byte 335 /// to specify a destination, which in this case is memory. 336 /// 337 MRMDestMem = 32, 338 339 /// MRMSrcMem - This form is used for instructions that use the Mod/RM byte 340 /// to specify a source, which in this case is memory. 341 /// 342 MRMSrcMem = 33, 343 344 /// MRMSrcMem4VOp3 - This form is used for instructions that encode 345 /// operand 3 with VEX.VVVV and load from memory. 346 /// 347 MRMSrcMem4VOp3 = 34, 348 349 /// MRMSrcMemOp4 - This form is used for instructions that use the Mod/RM 350 /// byte to specify the fourth source, which in this case is memory. 351 /// 352 MRMSrcMemOp4 = 35, 353 354 /// MRMSrcMemCC - This form is used for instructions that use the Mod/RM 355 /// byte to specify the operands and also encodes a condition code. 356 /// 357 MRMSrcMemCC = 36, 358 359 /// MRMXm - This form is used for instructions that use the Mod/RM byte 360 /// to specify a memory source, but doesn't use the middle field. And has 361 /// a condition code. 362 /// 363 MRMXmCC = 38, 364 365 /// MRMXm - This form is used for instructions that use the Mod/RM byte 366 /// to specify a memory source, but doesn't use the middle field. 367 /// 368 MRMXm = 39, 369 370 // Next, instructions that operate on a memory r/m operand... 371 MRM0m = 40, MRM1m = 41, MRM2m = 42, MRM3m = 43, // Format /0 /1 /2 /3 372 MRM4m = 44, MRM5m = 45, MRM6m = 46, MRM7m = 47, // Format /4 /5 /6 /7 373 374 /// MRMDestReg - This form is used for instructions that use the Mod/RM byte 375 /// to specify a destination, which in this case is a register. 376 /// 377 MRMDestReg = 48, 378 379 /// MRMSrcReg - This form is used for instructions that use the Mod/RM byte 380 /// to specify a source, which in this case is a register. 381 /// 382 MRMSrcReg = 49, 383 384 /// MRMSrcReg4VOp3 - This form is used for instructions that encode 385 /// operand 3 with VEX.VVVV and do not load from memory. 386 /// 387 MRMSrcReg4VOp3 = 50, 388 389 /// MRMSrcRegOp4 - This form is used for instructions that use the Mod/RM 390 /// byte to specify the fourth source, which in this case is a register. 391 /// 392 MRMSrcRegOp4 = 51, 393 394 /// MRMSrcRegCC - This form is used for instructions that use the Mod/RM 395 /// byte to specify the operands and also encodes a condition code 396 /// 397 MRMSrcRegCC = 52, 398 399 /// MRMXCCr - This form is used for instructions that use the Mod/RM byte 400 /// to specify a register source, but doesn't use the middle field. And has 401 /// a condition code. 402 /// 403 MRMXrCC = 54, 404 405 /// MRMXr - This form is used for instructions that use the Mod/RM byte 406 /// to specify a register source, but doesn't use the middle field. 407 /// 408 MRMXr = 55, 409 410 // Instructions that operate on a register r/m operand... 411 MRM0r = 56, MRM1r = 57, MRM2r = 58, MRM3r = 59, // Format /0 /1 /2 /3 412 MRM4r = 60, MRM5r = 61, MRM6r = 62, MRM7r = 63, // Format /4 /5 /6 /7 413 414 /// MRM_XX - A mod/rm byte of exactly 0xXX. 415 MRM_C0 = 64, MRM_C1 = 65, MRM_C2 = 66, MRM_C3 = 67, 416 MRM_C4 = 68, MRM_C5 = 69, MRM_C6 = 70, MRM_C7 = 71, 417 MRM_C8 = 72, MRM_C9 = 73, MRM_CA = 74, MRM_CB = 75, 418 MRM_CC = 76, MRM_CD = 77, MRM_CE = 78, MRM_CF = 79, 419 MRM_D0 = 80, MRM_D1 = 81, MRM_D2 = 82, MRM_D3 = 83, 420 MRM_D4 = 84, MRM_D5 = 85, MRM_D6 = 86, MRM_D7 = 87, 421 MRM_D8 = 88, MRM_D9 = 89, MRM_DA = 90, MRM_DB = 91, 422 MRM_DC = 92, MRM_DD = 93, MRM_DE = 94, MRM_DF = 95, 423 MRM_E0 = 96, MRM_E1 = 97, MRM_E2 = 98, MRM_E3 = 99, 424 MRM_E4 = 100, MRM_E5 = 101, MRM_E6 = 102, MRM_E7 = 103, 425 MRM_E8 = 104, MRM_E9 = 105, MRM_EA = 106, MRM_EB = 107, 426 MRM_EC = 108, MRM_ED = 109, MRM_EE = 110, MRM_EF = 111, 427 MRM_F0 = 112, MRM_F1 = 113, MRM_F2 = 114, MRM_F3 = 115, 428 MRM_F4 = 116, MRM_F5 = 117, MRM_F6 = 118, MRM_F7 = 119, 429 MRM_F8 = 120, MRM_F9 = 121, MRM_FA = 122, MRM_FB = 123, 430 MRM_FC = 124, MRM_FD = 125, MRM_FE = 126, MRM_FF = 127, 431 432 FormMask = 127, 433 434 //===------------------------------------------------------------------===// 435 // Actual flags... 436 437 // OpSize - OpSizeFixed implies instruction never needs a 0x66 prefix. 438 // OpSize16 means this is a 16-bit instruction and needs 0x66 prefix in 439 // 32-bit mode. OpSize32 means this is a 32-bit instruction needs a 0x66 440 // prefix in 16-bit mode. 441 OpSizeShift = 7, 442 OpSizeMask = 0x3 << OpSizeShift, 443 444 OpSizeFixed = 0 << OpSizeShift, 445 OpSize16 = 1 << OpSizeShift, 446 OpSize32 = 2 << OpSizeShift, 447 448 // AsSize - AdSizeX implies this instruction determines its need of 0x67 449 // prefix from a normal ModRM memory operand. The other types indicate that 450 // an operand is encoded with a specific width and a prefix is needed if 451 // it differs from the current mode. 452 AdSizeShift = OpSizeShift + 2, 453 AdSizeMask = 0x3 << AdSizeShift, 454 455 AdSizeX = 0 << AdSizeShift, 456 AdSize16 = 1 << AdSizeShift, 457 AdSize32 = 2 << AdSizeShift, 458 AdSize64 = 3 << AdSizeShift, 459 460 //===------------------------------------------------------------------===// 461 // OpPrefix - There are several prefix bytes that are used as opcode 462 // extensions. These are 0x66, 0xF3, and 0xF2. If this field is 0 there is 463 // no prefix. 464 // 465 OpPrefixShift = AdSizeShift + 2, 466 OpPrefixMask = 0x3 << OpPrefixShift, 467 468 // PD - Prefix code for packed double precision vector floating point 469 // operations performed in the SSE registers. 470 PD = 1 << OpPrefixShift, 471 472 // XS, XD - These prefix codes are for single and double precision scalar 473 // floating point operations performed in the SSE registers. 474 XS = 2 << OpPrefixShift, XD = 3 << OpPrefixShift, 475 476 //===------------------------------------------------------------------===// 477 // OpMap - This field determines which opcode map this instruction 478 // belongs to. i.e. one-byte, two-byte, 0x0f 0x38, 0x0f 0x3a, etc. 479 // 480 OpMapShift = OpPrefixShift + 2, 481 OpMapMask = 0x7 << OpMapShift, 482 483 // OB - OneByte - Set if this instruction has a one byte opcode. 484 OB = 0 << OpMapShift, 485 486 // TB - TwoByte - Set if this instruction has a two byte opcode, which 487 // starts with a 0x0F byte before the real opcode. 488 TB = 1 << OpMapShift, 489 490 // T8, TA - Prefix after the 0x0F prefix. 491 T8 = 2 << OpMapShift, TA = 3 << OpMapShift, 492 493 // XOP8 - Prefix to include use of imm byte. 494 XOP8 = 4 << OpMapShift, 495 496 // XOP9 - Prefix to exclude use of imm byte. 497 XOP9 = 5 << OpMapShift, 498 499 // XOPA - Prefix to encode 0xA in VEX.MMMM of XOP instructions. 500 XOPA = 6 << OpMapShift, 501 502 /// ThreeDNow - This indicates that the instruction uses the 503 /// wacky 0x0F 0x0F prefix for 3DNow! instructions. The manual documents 504 /// this as having a 0x0F prefix with a 0x0F opcode, and each instruction 505 /// storing a classifier in the imm8 field. To simplify our implementation, 506 /// we handle this by storeing the classifier in the opcode field and using 507 /// this flag to indicate that the encoder should do the wacky 3DNow! thing. 508 ThreeDNow = 7 << OpMapShift, 509 510 //===------------------------------------------------------------------===// 511 // REX_W - REX prefixes are instruction prefixes used in 64-bit mode. 512 // They are used to specify GPRs and SSE registers, 64-bit operand size, 513 // etc. We only cares about REX.W and REX.R bits and only the former is 514 // statically determined. 515 // 516 REXShift = OpMapShift + 3, 517 REX_W = 1 << REXShift, 518 519 //===------------------------------------------------------------------===// 520 // This three-bit field describes the size of an immediate operand. Zero is 521 // unused so that we can tell if we forgot to set a value. 522 ImmShift = REXShift + 1, 523 ImmMask = 15 << ImmShift, 524 Imm8 = 1 << ImmShift, 525 Imm8PCRel = 2 << ImmShift, 526 Imm8Reg = 3 << ImmShift, 527 Imm16 = 4 << ImmShift, 528 Imm16PCRel = 5 << ImmShift, 529 Imm32 = 6 << ImmShift, 530 Imm32PCRel = 7 << ImmShift, 531 Imm32S = 8 << ImmShift, 532 Imm64 = 9 << ImmShift, 533 534 //===------------------------------------------------------------------===// 535 // FP Instruction Classification... Zero is non-fp instruction. 536 537 // FPTypeMask - Mask for all of the FP types... 538 FPTypeShift = ImmShift + 4, 539 FPTypeMask = 7 << FPTypeShift, 540 541 // NotFP - The default, set for instructions that do not use FP registers. 542 NotFP = 0 << FPTypeShift, 543 544 // ZeroArgFP - 0 arg FP instruction which implicitly pushes ST(0), f.e. fld0 545 ZeroArgFP = 1 << FPTypeShift, 546 547 // OneArgFP - 1 arg FP instructions which implicitly read ST(0), such as fst 548 OneArgFP = 2 << FPTypeShift, 549 550 // OneArgFPRW - 1 arg FP instruction which implicitly read ST(0) and write a 551 // result back to ST(0). For example, fcos, fsqrt, etc. 552 // 553 OneArgFPRW = 3 << FPTypeShift, 554 555 // TwoArgFP - 2 arg FP instructions which implicitly read ST(0), and an 556 // explicit argument, storing the result to either ST(0) or the implicit 557 // argument. For example: fadd, fsub, fmul, etc... 558 TwoArgFP = 4 << FPTypeShift, 559 560 // CompareFP - 2 arg FP instructions which implicitly read ST(0) and an 561 // explicit argument, but have no destination. Example: fucom, fucomi, ... 562 CompareFP = 5 << FPTypeShift, 563 564 // CondMovFP - "2 operand" floating point conditional move instructions. 565 CondMovFP = 6 << FPTypeShift, 566 567 // SpecialFP - Special instruction forms. Dispatch by opcode explicitly. 568 SpecialFP = 7 << FPTypeShift, 569 570 // Lock prefix 571 LOCKShift = FPTypeShift + 3, 572 LOCK = 1 << LOCKShift, 573 574 // REP prefix 575 REPShift = LOCKShift + 1, 576 REP = 1 << REPShift, 577 578 // Execution domain for SSE instructions. 579 // 0 means normal, non-SSE instruction. 580 SSEDomainShift = REPShift + 1, 581 582 // Encoding 583 EncodingShift = SSEDomainShift + 2, 584 EncodingMask = 0x3 << EncodingShift, 585 586 // VEX - encoding using 0xC4/0xC5 587 VEX = 1 << EncodingShift, 588 589 /// XOP - Opcode prefix used by XOP instructions. 590 XOP = 2 << EncodingShift, 591 592 // VEX_EVEX - Specifies that this instruction use EVEX form which provides 593 // syntax support up to 32 512-bit register operands and up to 7 16-bit 594 // mask operands as well as source operand data swizzling/memory operand 595 // conversion, eviction hint, and rounding mode. 596 EVEX = 3 << EncodingShift, 597 598 // Opcode 599 OpcodeShift = EncodingShift + 2, 600 601 /// VEX_W - Has a opcode specific functionality, but is used in the same 602 /// way as REX_W is for regular SSE instructions. 603 VEX_WShift = OpcodeShift + 8, 604 VEX_W = 1ULL << VEX_WShift, 605 606 /// VEX_4V - Used to specify an additional AVX/SSE register. Several 2 607 /// address instructions in SSE are represented as 3 address ones in AVX 608 /// and the additional register is encoded in VEX_VVVV prefix. 609 VEX_4VShift = VEX_WShift + 1, 610 VEX_4V = 1ULL << VEX_4VShift, 611 612 /// VEX_L - Stands for a bit in the VEX opcode prefix meaning the current 613 /// instruction uses 256-bit wide registers. This is usually auto detected 614 /// if a VR256 register is used, but some AVX instructions also have this 615 /// field marked when using a f256 memory references. 616 VEX_LShift = VEX_4VShift + 1, 617 VEX_L = 1ULL << VEX_LShift, 618 619 // EVEX_K - Set if this instruction requires masking 620 EVEX_KShift = VEX_LShift + 1, 621 EVEX_K = 1ULL << EVEX_KShift, 622 623 // EVEX_Z - Set if this instruction has EVEX.Z field set. 624 EVEX_ZShift = EVEX_KShift + 1, 625 EVEX_Z = 1ULL << EVEX_ZShift, 626 627 // EVEX_L2 - Set if this instruction has EVEX.L' field set. 628 EVEX_L2Shift = EVEX_ZShift + 1, 629 EVEX_L2 = 1ULL << EVEX_L2Shift, 630 631 // EVEX_B - Set if this instruction has EVEX.B field set. 632 EVEX_BShift = EVEX_L2Shift + 1, 633 EVEX_B = 1ULL << EVEX_BShift, 634 635 // The scaling factor for the AVX512's 8-bit compressed displacement. 636 CD8_Scale_Shift = EVEX_BShift + 1, 637 CD8_Scale_Mask = 127ULL << CD8_Scale_Shift, 638 639 /// Explicitly specified rounding control 640 EVEX_RCShift = CD8_Scale_Shift + 7, 641 EVEX_RC = 1ULL << EVEX_RCShift, 642 643 // NOTRACK prefix 644 NoTrackShift = EVEX_RCShift + 1, 645 NOTRACK = 1ULL << NoTrackShift 646 }; 647 648 // getBaseOpcodeFor - This function returns the "base" X86 opcode for the 649 // specified machine instruction. 650 // 651 inline uint8_t getBaseOpcodeFor(uint64_t TSFlags) { 652 return TSFlags >> X86II::OpcodeShift; 653 } 654 655 inline bool hasImm(uint64_t TSFlags) { 656 return (TSFlags & X86II::ImmMask) != 0; 657 } 658 659 /// getSizeOfImm - Decode the "size of immediate" field from the TSFlags field 660 /// of the specified instruction. 661 inline unsigned getSizeOfImm(uint64_t TSFlags) { 662 switch (TSFlags & X86II::ImmMask) { 663 default: llvm_unreachable("Unknown immediate size"); 664 case X86II::Imm8: 665 case X86II::Imm8PCRel: 666 case X86II::Imm8Reg: return 1; 667 case X86II::Imm16: 668 case X86II::Imm16PCRel: return 2; 669 case X86II::Imm32: 670 case X86II::Imm32S: 671 case X86II::Imm32PCRel: return 4; 672 case X86II::Imm64: return 8; 673 } 674 } 675 676 /// isImmPCRel - Return true if the immediate of the specified instruction's 677 /// TSFlags indicates that it is pc relative. 678 inline unsigned isImmPCRel(uint64_t TSFlags) { 679 switch (TSFlags & X86II::ImmMask) { 680 default: llvm_unreachable("Unknown immediate size"); 681 case X86II::Imm8PCRel: 682 case X86II::Imm16PCRel: 683 case X86II::Imm32PCRel: 684 return true; 685 case X86II::Imm8: 686 case X86II::Imm8Reg: 687 case X86II::Imm16: 688 case X86II::Imm32: 689 case X86II::Imm32S: 690 case X86II::Imm64: 691 return false; 692 } 693 } 694 695 /// isImmSigned - Return true if the immediate of the specified instruction's 696 /// TSFlags indicates that it is signed. 697 inline unsigned isImmSigned(uint64_t TSFlags) { 698 switch (TSFlags & X86II::ImmMask) { 699 default: llvm_unreachable("Unknown immediate signedness"); 700 case X86II::Imm32S: 701 return true; 702 case X86II::Imm8: 703 case X86II::Imm8PCRel: 704 case X86II::Imm8Reg: 705 case X86II::Imm16: 706 case X86II::Imm16PCRel: 707 case X86II::Imm32: 708 case X86II::Imm32PCRel: 709 case X86II::Imm64: 710 return false; 711 } 712 } 713 714 /// getOperandBias - compute whether all of the def operands are repeated 715 /// in the uses and therefore should be skipped. 716 /// This determines the start of the unique operand list. We need to determine 717 /// if all of the defs have a corresponding tied operand in the uses. 718 /// Unfortunately, the tied operand information is encoded in the uses not 719 /// the defs so we have to use some heuristics to find which operands to 720 /// query. 721 inline unsigned getOperandBias(const MCInstrDesc& Desc) { 722 unsigned NumDefs = Desc.getNumDefs(); 723 unsigned NumOps = Desc.getNumOperands(); 724 switch (NumDefs) { 725 default: llvm_unreachable("Unexpected number of defs"); 726 case 0: 727 return 0; 728 case 1: 729 // Common two addr case. 730 if (NumOps > 1 && Desc.getOperandConstraint(1, MCOI::TIED_TO) == 0) 731 return 1; 732 // Check for AVX-512 scatter which has a TIED_TO in the second to last 733 // operand. 734 if (NumOps == 8 && 735 Desc.getOperandConstraint(6, MCOI::TIED_TO) == 0) 736 return 1; 737 return 0; 738 case 2: 739 // XCHG/XADD have two destinations and two sources. 740 if (NumOps >= 4 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0 && 741 Desc.getOperandConstraint(3, MCOI::TIED_TO) == 1) 742 return 2; 743 // Check for gather. AVX-512 has the second tied operand early. AVX2 744 // has it as the last op. 745 if (NumOps == 9 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0 && 746 (Desc.getOperandConstraint(3, MCOI::TIED_TO) == 1 || 747 Desc.getOperandConstraint(8, MCOI::TIED_TO) == 1)) 748 return 2; 749 return 0; 750 } 751 } 752 753 /// getMemoryOperandNo - The function returns the MCInst operand # for the 754 /// first field of the memory operand. If the instruction doesn't have a 755 /// memory operand, this returns -1. 756 /// 757 /// Note that this ignores tied operands. If there is a tied register which 758 /// is duplicated in the MCInst (e.g. "EAX = addl EAX, [mem]") it is only 759 /// counted as one operand. 760 /// 761 inline int getMemoryOperandNo(uint64_t TSFlags) { 762 bool HasVEX_4V = TSFlags & X86II::VEX_4V; 763 bool HasEVEX_K = TSFlags & X86II::EVEX_K; 764 765 switch (TSFlags & X86II::FormMask) { 766 default: llvm_unreachable("Unknown FormMask value in getMemoryOperandNo!"); 767 case X86II::Pseudo: 768 case X86II::RawFrm: 769 case X86II::AddRegFrm: 770 case X86II::RawFrmImm8: 771 case X86II::RawFrmImm16: 772 case X86II::RawFrmMemOffs: 773 case X86II::RawFrmSrc: 774 case X86II::RawFrmDst: 775 case X86II::RawFrmDstSrc: 776 case X86II::AddCCFrm: 777 return -1; 778 case X86II::MRMDestMem: 779 return 0; 780 case X86II::MRMSrcMem: 781 // Start from 1, skip any registers encoded in VEX_VVVV or I8IMM, or a 782 // mask register. 783 return 1 + HasVEX_4V + HasEVEX_K; 784 case X86II::MRMSrcMem4VOp3: 785 // Skip registers encoded in reg. 786 return 1 + HasEVEX_K; 787 case X86II::MRMSrcMemOp4: 788 // Skip registers encoded in reg, VEX_VVVV, and I8IMM. 789 return 3; 790 case X86II::MRMSrcMemCC: 791 // Start from 1, skip any registers encoded in VEX_VVVV or I8IMM, or a 792 // mask register. 793 return 1; 794 case X86II::MRMDestReg: 795 case X86II::MRMSrcReg: 796 case X86II::MRMSrcReg4VOp3: 797 case X86II::MRMSrcRegOp4: 798 case X86II::MRMSrcRegCC: 799 case X86II::MRMXrCC: 800 case X86II::MRMXr: 801 case X86II::MRM0r: case X86II::MRM1r: 802 case X86II::MRM2r: case X86II::MRM3r: 803 case X86II::MRM4r: case X86II::MRM5r: 804 case X86II::MRM6r: case X86II::MRM7r: 805 return -1; 806 case X86II::MRMXmCC: 807 case X86II::MRMXm: 808 case X86II::MRM0m: case X86II::MRM1m: 809 case X86II::MRM2m: case X86II::MRM3m: 810 case X86II::MRM4m: case X86II::MRM5m: 811 case X86II::MRM6m: case X86II::MRM7m: 812 // Start from 0, skip registers encoded in VEX_VVVV or a mask register. 813 return 0 + HasVEX_4V + HasEVEX_K; 814 case X86II::MRM_C0: case X86II::MRM_C1: case X86II::MRM_C2: 815 case X86II::MRM_C3: case X86II::MRM_C4: case X86II::MRM_C5: 816 case X86II::MRM_C6: case X86II::MRM_C7: case X86II::MRM_C8: 817 case X86II::MRM_C9: case X86II::MRM_CA: case X86II::MRM_CB: 818 case X86II::MRM_CC: case X86II::MRM_CD: case X86II::MRM_CE: 819 case X86II::MRM_CF: case X86II::MRM_D0: case X86II::MRM_D1: 820 case X86II::MRM_D2: case X86II::MRM_D3: case X86II::MRM_D4: 821 case X86II::MRM_D5: case X86II::MRM_D6: case X86II::MRM_D7: 822 case X86II::MRM_D8: case X86II::MRM_D9: case X86II::MRM_DA: 823 case X86II::MRM_DB: case X86II::MRM_DC: case X86II::MRM_DD: 824 case X86II::MRM_DE: case X86II::MRM_DF: case X86II::MRM_E0: 825 case X86II::MRM_E1: case X86II::MRM_E2: case X86II::MRM_E3: 826 case X86II::MRM_E4: case X86II::MRM_E5: case X86II::MRM_E6: 827 case X86II::MRM_E7: case X86II::MRM_E8: case X86II::MRM_E9: 828 case X86II::MRM_EA: case X86II::MRM_EB: case X86II::MRM_EC: 829 case X86II::MRM_ED: case X86II::MRM_EE: case X86II::MRM_EF: 830 case X86II::MRM_F0: case X86II::MRM_F1: case X86II::MRM_F2: 831 case X86II::MRM_F3: case X86II::MRM_F4: case X86II::MRM_F5: 832 case X86II::MRM_F6: case X86II::MRM_F7: case X86II::MRM_F8: 833 case X86II::MRM_F9: case X86II::MRM_FA: case X86II::MRM_FB: 834 case X86II::MRM_FC: case X86II::MRM_FD: case X86II::MRM_FE: 835 case X86II::MRM_FF: 836 return -1; 837 } 838 } 839 840 /// isX86_64ExtendedReg - Is the MachineOperand a x86-64 extended (r8 or 841 /// higher) register? e.g. r8, xmm8, xmm13, etc. 842 inline bool isX86_64ExtendedReg(unsigned RegNo) { 843 if ((RegNo >= X86::XMM8 && RegNo <= X86::XMM31) || 844 (RegNo >= X86::YMM8 && RegNo <= X86::YMM31) || 845 (RegNo >= X86::ZMM8 && RegNo <= X86::ZMM31)) 846 return true; 847 848 switch (RegNo) { 849 default: break; 850 case X86::R8: case X86::R9: case X86::R10: case X86::R11: 851 case X86::R12: case X86::R13: case X86::R14: case X86::R15: 852 case X86::R8D: case X86::R9D: case X86::R10D: case X86::R11D: 853 case X86::R12D: case X86::R13D: case X86::R14D: case X86::R15D: 854 case X86::R8W: case X86::R9W: case X86::R10W: case X86::R11W: 855 case X86::R12W: case X86::R13W: case X86::R14W: case X86::R15W: 856 case X86::R8B: case X86::R9B: case X86::R10B: case X86::R11B: 857 case X86::R12B: case X86::R13B: case X86::R14B: case X86::R15B: 858 case X86::CR8: case X86::CR9: case X86::CR10: case X86::CR11: 859 case X86::CR12: case X86::CR13: case X86::CR14: case X86::CR15: 860 case X86::DR8: case X86::DR9: case X86::DR10: case X86::DR11: 861 case X86::DR12: case X86::DR13: case X86::DR14: case X86::DR15: 862 return true; 863 } 864 return false; 865 } 866 867 /// is32ExtendedReg - Is the MemoryOperand a 32 extended (zmm16 or higher) 868 /// registers? e.g. zmm21, etc. 869 static inline bool is32ExtendedReg(unsigned RegNo) { 870 return ((RegNo >= X86::XMM16 && RegNo <= X86::XMM31) || 871 (RegNo >= X86::YMM16 && RegNo <= X86::YMM31) || 872 (RegNo >= X86::ZMM16 && RegNo <= X86::ZMM31)); 873 } 874 875 876 inline bool isX86_64NonExtLowByteReg(unsigned reg) { 877 return (reg == X86::SPL || reg == X86::BPL || 878 reg == X86::SIL || reg == X86::DIL); 879 } 880 881 /// isKMasked - Is this a masked instruction. 882 inline bool isKMasked(uint64_t TSFlags) { 883 return (TSFlags & X86II::EVEX_K) != 0; 884 } 885 886 /// isKMergedMasked - Is this a merge masked instruction. 887 inline bool isKMergeMasked(uint64_t TSFlags) { 888 return isKMasked(TSFlags) && (TSFlags & X86II::EVEX_Z) == 0; 889 } 890 } 891 892 } // end namespace llvm; 893 894 #endif 895