1 //===-- X86BaseInfo.h - Top level definitions for X86 -------- --*- C++ -*-===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file contains small standalone helper functions and enum definitions for 11 // the X86 target useful for the compiler back-end and the MC libraries. 12 // As such, it deliberately does not include references to LLVM core 13 // code gen types, passes, etc.. 14 // 15 //===----------------------------------------------------------------------===// 16 17 #ifndef X86BASEINFO_H 18 #define X86BASEINFO_H 19 20 #include "X86MCTargetDesc.h" 21 #include "llvm/MC/MCInstrDesc.h" 22 #include "llvm/Support/DataTypes.h" 23 #include "llvm/Support/ErrorHandling.h" 24 25 namespace llvm { 26 27 namespace X86 { 28 // Enums for memory operand decoding. Each memory operand is represented with 29 // a 5 operand sequence in the form: 30 // [BaseReg, ScaleAmt, IndexReg, Disp, Segment] 31 // These enums help decode this. 32 enum { 33 AddrBaseReg = 0, 34 AddrScaleAmt = 1, 35 AddrIndexReg = 2, 36 AddrDisp = 3, 37 38 /// AddrSegmentReg - The operand # of the segment in the memory operand. 39 AddrSegmentReg = 4, 40 41 /// AddrNumOperands - Total number of operands in a memory reference. 42 AddrNumOperands = 5 43 }; 44 } // end namespace X86; 45 46 /// X86II - This namespace holds all of the target specific flags that 47 /// instruction info tracks. 48 /// 49 namespace X86II { 50 /// Target Operand Flag enum. 51 enum TOF { 52 //===------------------------------------------------------------------===// 53 // X86 Specific MachineOperand flags. 54 55 MO_NO_FLAG, 56 57 /// MO_GOT_ABSOLUTE_ADDRESS - On a symbol operand, this represents a 58 /// relocation of: 59 /// SYMBOL_LABEL + [. - PICBASELABEL] 60 MO_GOT_ABSOLUTE_ADDRESS, 61 62 /// MO_PIC_BASE_OFFSET - On a symbol operand this indicates that the 63 /// immediate should get the value of the symbol minus the PIC base label: 64 /// SYMBOL_LABEL - PICBASELABEL 65 MO_PIC_BASE_OFFSET, 66 67 /// MO_GOT - On a symbol operand this indicates that the immediate is the 68 /// offset to the GOT entry for the symbol name from the base of the GOT. 69 /// 70 /// See the X86-64 ELF ABI supplement for more details. 71 /// SYMBOL_LABEL @GOT 72 MO_GOT, 73 74 /// MO_GOTOFF - On a symbol operand this indicates that the immediate is 75 /// the offset to the location of the symbol name from the base of the GOT. 76 /// 77 /// See the X86-64 ELF ABI supplement for more details. 78 /// SYMBOL_LABEL @GOTOFF 79 MO_GOTOFF, 80 81 /// MO_GOTPCREL - On a symbol operand this indicates that the immediate is 82 /// offset to the GOT entry for the symbol name from the current code 83 /// location. 84 /// 85 /// See the X86-64 ELF ABI supplement for more details. 86 /// SYMBOL_LABEL @GOTPCREL 87 MO_GOTPCREL, 88 89 /// MO_PLT - On a symbol operand this indicates that the immediate is 90 /// offset to the PLT entry of symbol name from the current code location. 91 /// 92 /// See the X86-64 ELF ABI supplement for more details. 93 /// SYMBOL_LABEL @PLT 94 MO_PLT, 95 96 /// MO_TLSGD - On a symbol operand this indicates that the immediate is 97 /// the offset of the GOT entry with the TLS index structure that contains 98 /// the module number and variable offset for the symbol. Used in the 99 /// general dynamic TLS access model. 100 /// 101 /// See 'ELF Handling for Thread-Local Storage' for more details. 102 /// SYMBOL_LABEL @TLSGD 103 MO_TLSGD, 104 105 /// MO_TLSLD - On a symbol operand this indicates that the immediate is 106 /// the offset of the GOT entry with the TLS index for the module that 107 /// contains the symbol. When this index is passed to a call to 108 /// __tls_get_addr, the function will return the base address of the TLS 109 /// block for the symbol. Used in the x86-64 local dynamic TLS access model. 110 /// 111 /// See 'ELF Handling for Thread-Local Storage' for more details. 112 /// SYMBOL_LABEL @TLSLD 113 MO_TLSLD, 114 115 /// MO_TLSLDM - On a symbol operand this indicates that the immediate is 116 /// the offset of the GOT entry with the TLS index for the module that 117 /// contains the symbol. When this index is passed to a call to 118 /// ___tls_get_addr, the function will return the base address of the TLS 119 /// block for the symbol. Used in the IA32 local dynamic TLS access model. 120 /// 121 /// See 'ELF Handling for Thread-Local Storage' for more details. 122 /// SYMBOL_LABEL @TLSLDM 123 MO_TLSLDM, 124 125 /// MO_GOTTPOFF - On a symbol operand this indicates that the immediate is 126 /// the offset of the GOT entry with the thread-pointer offset for the 127 /// symbol. Used in the x86-64 initial exec TLS access model. 128 /// 129 /// See 'ELF Handling for Thread-Local Storage' for more details. 130 /// SYMBOL_LABEL @GOTTPOFF 131 MO_GOTTPOFF, 132 133 /// MO_INDNTPOFF - On a symbol operand this indicates that the immediate is 134 /// the absolute address of the GOT entry with the negative thread-pointer 135 /// offset for the symbol. Used in the non-PIC IA32 initial exec TLS access 136 /// model. 137 /// 138 /// See 'ELF Handling for Thread-Local Storage' for more details. 139 /// SYMBOL_LABEL @INDNTPOFF 140 MO_INDNTPOFF, 141 142 /// MO_TPOFF - On a symbol operand this indicates that the immediate is 143 /// the thread-pointer offset for the symbol. Used in the x86-64 local 144 /// exec TLS access model. 145 /// 146 /// See 'ELF Handling for Thread-Local Storage' for more details. 147 /// SYMBOL_LABEL @TPOFF 148 MO_TPOFF, 149 150 /// MO_DTPOFF - On a symbol operand this indicates that the immediate is 151 /// the offset of the GOT entry with the TLS offset of the symbol. Used 152 /// in the local dynamic TLS access model. 153 /// 154 /// See 'ELF Handling for Thread-Local Storage' for more details. 155 /// SYMBOL_LABEL @DTPOFF 156 MO_DTPOFF, 157 158 /// MO_NTPOFF - On a symbol operand this indicates that the immediate is 159 /// the negative thread-pointer offset for the symbol. Used in the IA32 160 /// local exec TLS access model. 161 /// 162 /// See 'ELF Handling for Thread-Local Storage' for more details. 163 /// SYMBOL_LABEL @NTPOFF 164 MO_NTPOFF, 165 166 /// MO_GOTNTPOFF - On a symbol operand this indicates that the immediate is 167 /// the offset of the GOT entry with the negative thread-pointer offset for 168 /// the symbol. Used in the PIC IA32 initial exec TLS access model. 169 /// 170 /// See 'ELF Handling for Thread-Local Storage' for more details. 171 /// SYMBOL_LABEL @GOTNTPOFF 172 MO_GOTNTPOFF, 173 174 /// MO_DLLIMPORT - On a symbol operand "FOO", this indicates that the 175 /// reference is actually to the "__imp_FOO" symbol. This is used for 176 /// dllimport linkage on windows. 177 MO_DLLIMPORT, 178 179 /// MO_DARWIN_STUB - On a symbol operand "FOO", this indicates that the 180 /// reference is actually to the "FOO$stub" symbol. This is used for calls 181 /// and jumps to external functions on Tiger and earlier. 182 MO_DARWIN_STUB, 183 184 /// MO_DARWIN_NONLAZY - On a symbol operand "FOO", this indicates that the 185 /// reference is actually to the "FOO$non_lazy_ptr" symbol, which is a 186 /// non-PIC-base-relative reference to a non-hidden dyld lazy pointer stub. 187 MO_DARWIN_NONLAZY, 188 189 /// MO_DARWIN_NONLAZY_PIC_BASE - On a symbol operand "FOO", this indicates 190 /// that the reference is actually to "FOO$non_lazy_ptr - PICBASE", which is 191 /// a PIC-base-relative reference to a non-hidden dyld lazy pointer stub. 192 MO_DARWIN_NONLAZY_PIC_BASE, 193 194 /// MO_DARWIN_HIDDEN_NONLAZY_PIC_BASE - On a symbol operand "FOO", this 195 /// indicates that the reference is actually to "FOO$non_lazy_ptr -PICBASE", 196 /// which is a PIC-base-relative reference to a hidden dyld lazy pointer 197 /// stub. 198 MO_DARWIN_HIDDEN_NONLAZY_PIC_BASE, 199 200 /// MO_TLVP - On a symbol operand this indicates that the immediate is 201 /// some TLS offset. 202 /// 203 /// This is the TLS offset for the Darwin TLS mechanism. 204 MO_TLVP, 205 206 /// MO_TLVP_PIC_BASE - On a symbol operand this indicates that the immediate 207 /// is some TLS offset from the picbase. 208 /// 209 /// This is the 32-bit TLS offset for Darwin TLS in PIC mode. 210 MO_TLVP_PIC_BASE, 211 212 /// MO_SECREL - On a symbol operand this indicates that the immediate is 213 /// the offset from beginning of section. 214 /// 215 /// This is the TLS offset for the COFF/Windows TLS mechanism. 216 MO_SECREL 217 }; 218 219 enum { 220 //===------------------------------------------------------------------===// 221 // Instruction encodings. These are the standard/most common forms for X86 222 // instructions. 223 // 224 225 // PseudoFrm - This represents an instruction that is a pseudo instruction 226 // or one that has not been implemented yet. It is illegal to code generate 227 // it, but tolerated for intermediate implementation stages. 228 Pseudo = 0, 229 230 /// Raw - This form is for instructions that don't have any operands, so 231 /// they are just a fixed opcode value, like 'leave'. 232 RawFrm = 1, 233 234 /// AddRegFrm - This form is used for instructions like 'push r32' that have 235 /// their one register operand added to their opcode. 236 AddRegFrm = 2, 237 238 /// MRMDestReg - This form is used for instructions that use the Mod/RM byte 239 /// to specify a destination, which in this case is a register. 240 /// 241 MRMDestReg = 3, 242 243 /// MRMDestMem - This form is used for instructions that use the Mod/RM byte 244 /// to specify a destination, which in this case is memory. 245 /// 246 MRMDestMem = 4, 247 248 /// MRMSrcReg - This form is used for instructions that use the Mod/RM byte 249 /// to specify a source, which in this case is a register. 250 /// 251 MRMSrcReg = 5, 252 253 /// MRMSrcMem - This form is used for instructions that use the Mod/RM byte 254 /// to specify a source, which in this case is memory. 255 /// 256 MRMSrcMem = 6, 257 258 /// RawFrmMemOffs - This form is for instructions that store an absolute 259 /// memory offset as an immediate with a possible segment override. 260 RawFrmMemOffs = 7, 261 262 /// RawFrmSrc - This form is for instructions that use the source index 263 /// register SI/ESI/RSI with a possible segment override. 264 RawFrmSrc = 8, 265 266 /// RawFrmDst - This form is for instructions that use the destination index 267 /// register DI/EDI/ESI. 268 RawFrmDst = 9, 269 270 /// RawFrmSrc - This form is for instructions that use the the source index 271 /// register SI/ESI/ERI with a possible segment override, and also the 272 /// destination index register DI/ESI/RDI. 273 RawFrmDstSrc = 10, 274 275 /// RawFrmImm8 - This is used for the ENTER instruction, which has two 276 /// immediates, the first of which is a 16-bit immediate (specified by 277 /// the imm encoding) and the second is a 8-bit fixed value. 278 RawFrmImm8 = 11, 279 280 /// RawFrmImm16 - This is used for CALL FAR instructions, which have two 281 /// immediates, the first of which is a 16 or 32-bit immediate (specified by 282 /// the imm encoding) and the second is a 16-bit fixed value. In the AMD 283 /// manual, this operand is described as pntr16:32 and pntr16:16 284 RawFrmImm16 = 12, 285 286 /// MRMX[rm] - The forms are used to represent instructions that use a 287 /// Mod/RM byte, and don't use the middle field for anything. 288 MRMXr = 14, MRMXm = 15, 289 290 /// MRM[0-7][rm] - These forms are used to represent instructions that use 291 /// a Mod/RM byte, and use the middle field to hold extended opcode 292 /// information. In the intel manual these are represented as /0, /1, ... 293 /// 294 295 // First, instructions that operate on a register r/m operand... 296 MRM0r = 16, MRM1r = 17, MRM2r = 18, MRM3r = 19, // Format /0 /1 /2 /3 297 MRM4r = 20, MRM5r = 21, MRM6r = 22, MRM7r = 23, // Format /4 /5 /6 /7 298 299 // Next, instructions that operate on a memory r/m operand... 300 MRM0m = 24, MRM1m = 25, MRM2m = 26, MRM3m = 27, // Format /0 /1 /2 /3 301 MRM4m = 28, MRM5m = 29, MRM6m = 30, MRM7m = 31, // Format /4 /5 /6 /7 302 303 //// MRM_XX - A mod/rm byte of exactly 0xXX. 304 MRM_C0 = 32, MRM_C1 = 33, MRM_C2 = 34, MRM_C3 = 35, 305 MRM_C4 = 36, MRM_C8 = 37, MRM_C9 = 38, MRM_CA = 39, 306 MRM_CB = 40, MRM_D0 = 41, MRM_D1 = 42, MRM_D4 = 43, 307 MRM_D5 = 44, MRM_D6 = 45, MRM_D8 = 46, MRM_D9 = 47, 308 MRM_DA = 48, MRM_DB = 49, MRM_DC = 50, MRM_DD = 51, 309 MRM_DE = 52, MRM_DF = 53, MRM_E0 = 54, MRM_E1 = 55, 310 MRM_E2 = 56, MRM_E3 = 57, MRM_E4 = 58, MRM_E5 = 59, 311 MRM_E8 = 60, MRM_E9 = 61, MRM_EA = 62, MRM_EB = 63, 312 MRM_EC = 64, MRM_ED = 65, MRM_EE = 66, MRM_F0 = 67, 313 MRM_F1 = 68, MRM_F2 = 69, MRM_F3 = 70, MRM_F4 = 71, 314 MRM_F5 = 72, MRM_F6 = 73, MRM_F7 = 74, MRM_F8 = 75, 315 MRM_F9 = 76, MRM_FA = 77, MRM_FB = 78, MRM_FC = 79, 316 MRM_FD = 80, MRM_FE = 81, MRM_FF = 82, 317 318 FormMask = 127, 319 320 //===------------------------------------------------------------------===// 321 // Actual flags... 322 323 // OpSize - OpSizeFixed implies instruction never needs a 0x66 prefix. 324 // OpSize16 means this is a 16-bit instruction and needs 0x66 prefix in 325 // 32-bit mode. OpSize32 means this is a 32-bit instruction needs a 0x66 326 // prefix in 16-bit mode. 327 OpSizeShift = 7, 328 OpSizeMask = 0x3 << OpSizeShift, 329 330 OpSize16 = 1, 331 OpSize32 = 2, 332 333 // AsSize - Set if this instruction requires an operand size prefix (0x67), 334 // which most often indicates that the instruction address 16 bit address 335 // instead of 32 bit address (or 32 bit address in 64 bit mode). 336 AdSizeShift = OpSizeShift + 2, 337 AdSize = 1 << AdSizeShift, 338 339 //===------------------------------------------------------------------===// 340 // OpPrefix - There are several prefix bytes that are used as opcode 341 // extensions. These are 0x66, 0xF3, and 0xF2. If this field is 0 there is 342 // no prefix. 343 // 344 OpPrefixShift = AdSizeShift + 1, 345 OpPrefixMask = 0x7 << OpPrefixShift, 346 347 // PS, PD - Prefix code for packed single and double precision vector 348 // floating point operations performed in the SSE registers. 349 PS = 1 << OpPrefixShift, PD = 2 << OpPrefixShift, 350 351 // XS, XD - These prefix codes are for single and double precision scalar 352 // floating point operations performed in the SSE registers. 353 XS = 3 << OpPrefixShift, XD = 4 << OpPrefixShift, 354 355 //===------------------------------------------------------------------===// 356 // OpMap - This field determines which opcode map this instruction 357 // belongs to. i.e. one-byte, two-byte, 0x0f 0x38, 0x0f 0x3a, etc. 358 // 359 OpMapShift = OpPrefixShift + 3, 360 OpMapMask = 0x7 << OpMapShift, 361 362 // OB - OneByte - Set if this instruction has a one byte opcode. 363 OB = 0 << OpMapShift, 364 365 // TB - TwoByte - Set if this instruction has a two byte opcode, which 366 // starts with a 0x0F byte before the real opcode. 367 TB = 1 << OpMapShift, 368 369 // T8, TA - Prefix after the 0x0F prefix. 370 T8 = 2 << OpMapShift, TA = 3 << OpMapShift, 371 372 // XOP8 - Prefix to include use of imm byte. 373 XOP8 = 4 << OpMapShift, 374 375 // XOP9 - Prefix to exclude use of imm byte. 376 XOP9 = 5 << OpMapShift, 377 378 // XOPA - Prefix to encode 0xA in VEX.MMMM of XOP instructions. 379 XOPA = 6 << OpMapShift, 380 381 //===------------------------------------------------------------------===// 382 // REX_W - REX prefixes are instruction prefixes used in 64-bit mode. 383 // They are used to specify GPRs and SSE registers, 64-bit operand size, 384 // etc. We only cares about REX.W and REX.R bits and only the former is 385 // statically determined. 386 // 387 REXShift = OpMapShift + 3, 388 REX_W = 1 << REXShift, 389 390 //===------------------------------------------------------------------===// 391 // This three-bit field describes the size of an immediate operand. Zero is 392 // unused so that we can tell if we forgot to set a value. 393 ImmShift = REXShift + 1, 394 ImmMask = 15 << ImmShift, 395 Imm8 = 1 << ImmShift, 396 Imm8PCRel = 2 << ImmShift, 397 Imm16 = 3 << ImmShift, 398 Imm16PCRel = 4 << ImmShift, 399 Imm32 = 5 << ImmShift, 400 Imm32PCRel = 6 << ImmShift, 401 Imm32S = 7 << ImmShift, 402 Imm64 = 8 << ImmShift, 403 404 //===------------------------------------------------------------------===// 405 // FP Instruction Classification... Zero is non-fp instruction. 406 407 // FPTypeMask - Mask for all of the FP types... 408 FPTypeShift = ImmShift + 4, 409 FPTypeMask = 7 << FPTypeShift, 410 411 // NotFP - The default, set for instructions that do not use FP registers. 412 NotFP = 0 << FPTypeShift, 413 414 // ZeroArgFP - 0 arg FP instruction which implicitly pushes ST(0), f.e. fld0 415 ZeroArgFP = 1 << FPTypeShift, 416 417 // OneArgFP - 1 arg FP instructions which implicitly read ST(0), such as fst 418 OneArgFP = 2 << FPTypeShift, 419 420 // OneArgFPRW - 1 arg FP instruction which implicitly read ST(0) and write a 421 // result back to ST(0). For example, fcos, fsqrt, etc. 422 // 423 OneArgFPRW = 3 << FPTypeShift, 424 425 // TwoArgFP - 2 arg FP instructions which implicitly read ST(0), and an 426 // explicit argument, storing the result to either ST(0) or the implicit 427 // argument. For example: fadd, fsub, fmul, etc... 428 TwoArgFP = 4 << FPTypeShift, 429 430 // CompareFP - 2 arg FP instructions which implicitly read ST(0) and an 431 // explicit argument, but have no destination. Example: fucom, fucomi, ... 432 CompareFP = 5 << FPTypeShift, 433 434 // CondMovFP - "2 operand" floating point conditional move instructions. 435 CondMovFP = 6 << FPTypeShift, 436 437 // SpecialFP - Special instruction forms. Dispatch by opcode explicitly. 438 SpecialFP = 7 << FPTypeShift, 439 440 // Lock prefix 441 LOCKShift = FPTypeShift + 3, 442 LOCK = 1 << LOCKShift, 443 444 // REP prefix 445 REPShift = LOCKShift + 1, 446 REP = 1 << REPShift, 447 448 // Execution domain for SSE instructions. 449 // 0 means normal, non-SSE instruction. 450 SSEDomainShift = REPShift + 1, 451 452 // Encoding 453 EncodingShift = SSEDomainShift + 2, 454 EncodingMask = 0x3 << EncodingShift, 455 456 // VEX - encoding using 0xC4/0xC5 457 VEX = 1, 458 459 /// XOP - Opcode prefix used by XOP instructions. 460 XOP = 2, 461 462 // VEX_EVEX - Specifies that this instruction use EVEX form which provides 463 // syntax support up to 32 512-bit register operands and up to 7 16-bit 464 // mask operands as well as source operand data swizzling/memory operand 465 // conversion, eviction hint, and rounding mode. 466 EVEX = 3, 467 468 // Opcode 469 OpcodeShift = EncodingShift + 2, 470 471 //===------------------------------------------------------------------===// 472 /// VEX - The opcode prefix used by AVX instructions 473 VEXShift = OpcodeShift + 8, 474 475 /// VEX_W - Has a opcode specific functionality, but is used in the same 476 /// way as REX_W is for regular SSE instructions. 477 VEX_W = 1U << 0, 478 479 /// VEX_4V - Used to specify an additional AVX/SSE register. Several 2 480 /// address instructions in SSE are represented as 3 address ones in AVX 481 /// and the additional register is encoded in VEX_VVVV prefix. 482 VEX_4V = 1U << 1, 483 484 /// VEX_4VOp3 - Similar to VEX_4V, but used on instructions that encode 485 /// operand 3 with VEX.vvvv. 486 VEX_4VOp3 = 1U << 2, 487 488 /// VEX_I8IMM - Specifies that the last register used in a AVX instruction, 489 /// must be encoded in the i8 immediate field. This usually happens in 490 /// instructions with 4 operands. 491 VEX_I8IMM = 1U << 3, 492 493 /// VEX_L - Stands for a bit in the VEX opcode prefix meaning the current 494 /// instruction uses 256-bit wide registers. This is usually auto detected 495 /// if a VR256 register is used, but some AVX instructions also have this 496 /// field marked when using a f256 memory references. 497 VEX_L = 1U << 4, 498 499 // VEX_LIG - Specifies that this instruction ignores the L-bit in the VEX 500 // prefix. Usually used for scalar instructions. Needed by disassembler. 501 VEX_LIG = 1U << 5, 502 503 // TODO: we should combine VEX_L and VEX_LIG together to form a 2-bit field 504 // with following encoding: 505 // - 00 V128 506 // - 01 V256 507 // - 10 V512 508 // - 11 LIG (but, in insn encoding, leave VEX.L and EVEX.L in zeros. 509 // this will save 1 tsflag bit 510 511 // EVEX_K - Set if this instruction requires masking 512 EVEX_K = 1U << 6, 513 514 // EVEX_Z - Set if this instruction has EVEX.Z field set. 515 EVEX_Z = 1U << 7, 516 517 // EVEX_L2 - Set if this instruction has EVEX.L' field set. 518 EVEX_L2 = 1U << 8, 519 520 // EVEX_B - Set if this instruction has EVEX.B field set. 521 EVEX_B = 1U << 9, 522 523 // EVEX_CD8E - compressed disp8 form, element-size 524 EVEX_CD8EShift = VEXShift + 10, 525 EVEX_CD8EMask = 3, 526 527 // EVEX_CD8V - compressed disp8 form, vector-width 528 EVEX_CD8VShift = EVEX_CD8EShift + 2, 529 EVEX_CD8VMask = 7, 530 531 /// Has3DNow0F0FOpcode - This flag indicates that the instruction uses the 532 /// wacky 0x0F 0x0F prefix for 3DNow! instructions. The manual documents 533 /// this as having a 0x0F prefix with a 0x0F opcode, and each instruction 534 /// storing a classifier in the imm8 field. To simplify our implementation, 535 /// we handle this by storeing the classifier in the opcode field and using 536 /// this flag to indicate that the encoder should do the wacky 3DNow! thing. 537 Has3DNow0F0FOpcode = 1U << 15, 538 539 /// MemOp4 - Used to indicate swapping of operand 3 and 4 to be encoded in 540 /// ModRM or I8IMM. This is used for FMA4 and XOP instructions. 541 MemOp4 = 1U << 16, 542 543 /// Explicitly specified rounding control 544 EVEX_RC = 1U << 17 545 }; 546 547 // getBaseOpcodeFor - This function returns the "base" X86 opcode for the 548 // specified machine instruction. 549 // getBaseOpcodeFor(uint64_t TSFlags)550 inline unsigned char getBaseOpcodeFor(uint64_t TSFlags) { 551 return TSFlags >> X86II::OpcodeShift; 552 } 553 hasImm(uint64_t TSFlags)554 inline bool hasImm(uint64_t TSFlags) { 555 return (TSFlags & X86II::ImmMask) != 0; 556 } 557 558 /// getSizeOfImm - Decode the "size of immediate" field from the TSFlags field 559 /// of the specified instruction. getSizeOfImm(uint64_t TSFlags)560 inline unsigned getSizeOfImm(uint64_t TSFlags) { 561 switch (TSFlags & X86II::ImmMask) { 562 default: llvm_unreachable("Unknown immediate size"); 563 case X86II::Imm8: 564 case X86II::Imm8PCRel: return 1; 565 case X86II::Imm16: 566 case X86II::Imm16PCRel: return 2; 567 case X86II::Imm32: 568 case X86II::Imm32S: 569 case X86II::Imm32PCRel: return 4; 570 case X86II::Imm64: return 8; 571 } 572 } 573 574 /// isImmPCRel - Return true if the immediate of the specified instruction's 575 /// TSFlags indicates that it is pc relative. isImmPCRel(uint64_t TSFlags)576 inline unsigned isImmPCRel(uint64_t TSFlags) { 577 switch (TSFlags & X86II::ImmMask) { 578 default: llvm_unreachable("Unknown immediate size"); 579 case X86II::Imm8PCRel: 580 case X86II::Imm16PCRel: 581 case X86II::Imm32PCRel: 582 return true; 583 case X86II::Imm8: 584 case X86II::Imm16: 585 case X86II::Imm32: 586 case X86II::Imm32S: 587 case X86II::Imm64: 588 return false; 589 } 590 } 591 592 /// isImmSigned - Return true if the immediate of the specified instruction's 593 /// TSFlags indicates that it is signed. isImmSigned(uint64_t TSFlags)594 inline unsigned isImmSigned(uint64_t TSFlags) { 595 switch (TSFlags & X86II::ImmMask) { 596 default: llvm_unreachable("Unknown immediate signedness"); 597 case X86II::Imm32S: 598 return true; 599 case X86II::Imm8: 600 case X86II::Imm8PCRel: 601 case X86II::Imm16: 602 case X86II::Imm16PCRel: 603 case X86II::Imm32: 604 case X86II::Imm32PCRel: 605 case X86II::Imm64: 606 return false; 607 } 608 } 609 610 /// getOperandBias - compute any additional adjustment needed to 611 /// the offset to the start of the memory operand 612 /// in this instruction. 613 /// If this is a two-address instruction,skip one of the register operands. 614 /// FIXME: This should be handled during MCInst lowering. getOperandBias(const MCInstrDesc & Desc)615 inline int getOperandBias(const MCInstrDesc& Desc) 616 { 617 unsigned NumOps = Desc.getNumOperands(); 618 unsigned CurOp = 0; 619 if (NumOps > 1 && Desc.getOperandConstraint(1, MCOI::TIED_TO) == 0) 620 ++CurOp; 621 else if (NumOps > 3 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0 && 622 Desc.getOperandConstraint(3, MCOI::TIED_TO) == 1) 623 // Special case for AVX-512 GATHER with 2 TIED_TO operands 624 // Skip the first 2 operands: dst, mask_wb 625 CurOp += 2; 626 else if (NumOps > 3 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0 && 627 Desc.getOperandConstraint(NumOps - 1, MCOI::TIED_TO) == 1) 628 // Special case for GATHER with 2 TIED_TO operands 629 // Skip the first 2 operands: dst, mask_wb 630 CurOp += 2; 631 else if (NumOps > 2 && Desc.getOperandConstraint(NumOps - 2, MCOI::TIED_TO) == 0) 632 // SCATTER 633 ++CurOp; 634 return CurOp; 635 } 636 637 /// getMemoryOperandNo - The function returns the MCInst operand # for the 638 /// first field of the memory operand. If the instruction doesn't have a 639 /// memory operand, this returns -1. 640 /// 641 /// Note that this ignores tied operands. If there is a tied register which 642 /// is duplicated in the MCInst (e.g. "EAX = addl EAX, [mem]") it is only 643 /// counted as one operand. 644 /// getMemoryOperandNo(uint64_t TSFlags,unsigned Opcode)645 inline int getMemoryOperandNo(uint64_t TSFlags, unsigned Opcode) { 646 bool HasVEX_4V = (TSFlags >> X86II::VEXShift) & X86II::VEX_4V; 647 bool HasMemOp4 = (TSFlags >> X86II::VEXShift) & X86II::MemOp4; 648 bool HasEVEX_K = ((TSFlags >> X86II::VEXShift) & X86II::EVEX_K); 649 650 switch (TSFlags & X86II::FormMask) { 651 default: llvm_unreachable("Unknown FormMask value in getMemoryOperandNo!"); 652 case X86II::Pseudo: 653 case X86II::RawFrm: 654 case X86II::AddRegFrm: 655 case X86II::MRMDestReg: 656 case X86II::MRMSrcReg: 657 case X86II::RawFrmImm8: 658 case X86II::RawFrmImm16: 659 case X86II::RawFrmMemOffs: 660 case X86II::RawFrmSrc: 661 case X86II::RawFrmDst: 662 case X86II::RawFrmDstSrc: 663 return -1; 664 case X86II::MRMDestMem: 665 return 0; 666 case X86II::MRMSrcMem: { 667 unsigned FirstMemOp = 1; 668 if (HasVEX_4V) 669 ++FirstMemOp;// Skip the register source (which is encoded in VEX_VVVV). 670 if (HasMemOp4) 671 ++FirstMemOp;// Skip the register source (which is encoded in I8IMM). 672 if (HasEVEX_K) 673 ++FirstMemOp;// Skip the mask register 674 // FIXME: Maybe lea should have its own form? This is a horrible hack. 675 //if (Opcode == X86::LEA64r || Opcode == X86::LEA64_32r || 676 // Opcode == X86::LEA16r || Opcode == X86::LEA32r) 677 return FirstMemOp; 678 } 679 case X86II::MRMXr: 680 case X86II::MRM0r: case X86II::MRM1r: 681 case X86II::MRM2r: case X86II::MRM3r: 682 case X86II::MRM4r: case X86II::MRM5r: 683 case X86II::MRM6r: case X86II::MRM7r: 684 return -1; 685 case X86II::MRMXm: 686 case X86II::MRM0m: case X86II::MRM1m: 687 case X86II::MRM2m: case X86II::MRM3m: 688 case X86II::MRM4m: case X86II::MRM5m: 689 case X86II::MRM6m: case X86II::MRM7m: { 690 bool HasVEX_4V = (TSFlags >> X86II::VEXShift) & X86II::VEX_4V; 691 unsigned FirstMemOp = 0; 692 if (HasVEX_4V) 693 ++FirstMemOp;// Skip the register dest (which is encoded in VEX_VVVV). 694 if (HasEVEX_K) 695 ++FirstMemOp;// Skip the mask register 696 return FirstMemOp; 697 } 698 case X86II::MRM_C0: case X86II::MRM_C1: case X86II::MRM_C2: 699 case X86II::MRM_C3: case X86II::MRM_C4: case X86II::MRM_C8: 700 case X86II::MRM_C9: case X86II::MRM_CA: case X86II::MRM_CB: 701 case X86II::MRM_D0: case X86II::MRM_D1: case X86II::MRM_D4: 702 case X86II::MRM_D5: case X86II::MRM_D6: case X86II::MRM_D8: 703 case X86II::MRM_D9: case X86II::MRM_DA: case X86II::MRM_DB: 704 case X86II::MRM_DC: case X86II::MRM_DD: case X86II::MRM_DE: 705 case X86II::MRM_DF: case X86II::MRM_E0: case X86II::MRM_E1: 706 case X86II::MRM_E2: case X86II::MRM_E3: case X86II::MRM_E4: 707 case X86II::MRM_E5: case X86II::MRM_E8: case X86II::MRM_E9: 708 case X86II::MRM_EA: case X86II::MRM_EB: case X86II::MRM_EC: 709 case X86II::MRM_ED: case X86II::MRM_EE: case X86II::MRM_F0: 710 case X86II::MRM_F1: case X86II::MRM_F2: case X86II::MRM_F3: 711 case X86II::MRM_F4: case X86II::MRM_F5: case X86II::MRM_F6: 712 case X86II::MRM_F7: case X86II::MRM_F8: case X86II::MRM_F9: 713 case X86II::MRM_FA: case X86II::MRM_FB: case X86II::MRM_FC: 714 case X86II::MRM_FD: case X86II::MRM_FE: case X86II::MRM_FF: 715 return -1; 716 } 717 } 718 719 /// isX86_64ExtendedReg - Is the MachineOperand a x86-64 extended (r8 or 720 /// higher) register? e.g. r8, xmm8, xmm13, etc. isX86_64ExtendedReg(unsigned RegNo)721 inline bool isX86_64ExtendedReg(unsigned RegNo) { 722 if ((RegNo > X86::XMM7 && RegNo <= X86::XMM15) || 723 (RegNo > X86::XMM23 && RegNo <= X86::XMM31) || 724 (RegNo > X86::YMM7 && RegNo <= X86::YMM15) || 725 (RegNo > X86::YMM23 && RegNo <= X86::YMM31) || 726 (RegNo > X86::ZMM7 && RegNo <= X86::ZMM15) || 727 (RegNo > X86::ZMM23 && RegNo <= X86::ZMM31)) 728 return true; 729 730 switch (RegNo) { 731 default: break; 732 case X86::R8: case X86::R9: case X86::R10: case X86::R11: 733 case X86::R12: case X86::R13: case X86::R14: case X86::R15: 734 case X86::R8D: case X86::R9D: case X86::R10D: case X86::R11D: 735 case X86::R12D: case X86::R13D: case X86::R14D: case X86::R15D: 736 case X86::R8W: case X86::R9W: case X86::R10W: case X86::R11W: 737 case X86::R12W: case X86::R13W: case X86::R14W: case X86::R15W: 738 case X86::R8B: case X86::R9B: case X86::R10B: case X86::R11B: 739 case X86::R12B: case X86::R13B: case X86::R14B: case X86::R15B: 740 case X86::CR8: case X86::CR9: case X86::CR10: case X86::CR11: 741 case X86::CR12: case X86::CR13: case X86::CR14: case X86::CR15: 742 return true; 743 } 744 return false; 745 } 746 747 /// is32ExtendedReg - Is the MemoryOperand a 32 extended (zmm16 or higher) 748 /// registers? e.g. zmm21, etc. is32ExtendedReg(unsigned RegNo)749 static inline bool is32ExtendedReg(unsigned RegNo) { 750 return ((RegNo > X86::XMM15 && RegNo <= X86::XMM31) || 751 (RegNo > X86::YMM15 && RegNo <= X86::YMM31) || 752 (RegNo > X86::ZMM15 && RegNo <= X86::ZMM31)); 753 } 754 755 isX86_64NonExtLowByteReg(unsigned reg)756 inline bool isX86_64NonExtLowByteReg(unsigned reg) { 757 return (reg == X86::SPL || reg == X86::BPL || 758 reg == X86::SIL || reg == X86::DIL); 759 } 760 } 761 762 } // end namespace llvm; 763 764 #endif 765