1 //===-- X86BaseInfo.h - Top level definitions for X86 -------- --*- C++ -*-===//
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
7 //===----------------------------------------------------------------------===//
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..
14 //===----------------------------------------------------------------------===//
16 #ifndef LLVM_LIB_TARGET_X86_MCTARGETDESC_X86BASEINFO_H
17 #define LLVM_LIB_TARGET_X86_MCTARGETDESC_X86BASEINFO_H
19 #include "X86MCTargetDesc.h"
20 #include "llvm/MC/MCInstrDesc.h"
21 #include "llvm/Support/DataTypes.h"
22 #include "llvm/Support/ErrorHandling.h"
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.
37 /// AddrSegmentReg - The operand # of the segment in the memory operand.
40 /// AddrNumOperands - Total number of operands in a memory reference.
44 /// AVX512 static rounding constants. These need to match the values in
46 enum STATIC_ROUNDING {
55 /// The constants to describe instr prefixes if there are
67 enum OperandType : unsigned {
68 /// AVX512 embedded rounding control. This should only have values 0-3.
69 OPERAND_ROUNDING_CONTROL = MCOI::OPERAND_FIRST_TARGET,
73 // X86 specific condition code. These correspond to X86_*_COND in
74 // X86InstrInfo.td. They must be kept in synch.
92 LAST_VALID_COND = COND_G,
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.
105 // The classification for the first instruction in macro fusion.
106 enum class FirstMacroFusionInstKind {
117 // Not valid as a first macro fusion instruction
121 enum class SecondMacroFusionInstKind {
122 // JA, JB and variants.
124 // JE, JL, JG and variants.
126 // JS, JP, JO and variants
128 // Not a fusible jump.
132 /// \returns the type of the first instruction in macro-fusion.
133 inline FirstMacroFusionInstKind
134 classifyFirstOpcodeInMacroFusion(unsigned Opcode) {
137 return FirstMacroFusionInstKind::Invalid;
149 case X86::TEST64ri32:
155 return FirstMacroFusionInstKind::Test;
161 case X86::AND16rr_REV:
167 case X86::AND32rr_REV:
173 case X86::AND64rr_REV:
179 case X86::AND8rr_REV:
180 return FirstMacroFusionInstKind::And;
188 case X86::CMP16rr_REV:
195 case X86::CMP32rr_REV:
202 case X86::CMP64rr_REV:
209 case X86::CMP8rr_REV:
210 return FirstMacroFusionInstKind::Cmp;
217 case X86::ADD16rr_REV:
223 case X86::ADD32rr_REV:
229 case X86::ADD64rr_REV:
235 case X86::ADD8rr_REV:
242 case X86::SUB16rr_REV:
248 case X86::SUB32rr_REV:
254 case X86::SUB64rr_REV:
260 case X86::SUB8rr_REV:
261 return FirstMacroFusionInstKind::AddSub;
264 case X86::INC16r_alt:
266 case X86::INC32r_alt:
271 case X86::DEC16r_alt:
273 case X86::DEC32r_alt:
276 return FirstMacroFusionInstKind::IncDec;
280 /// \returns the type of the second instruction in macro-fusion.
281 inline SecondMacroFusionInstKind
282 classifySecondCondCodeInMacroFusion(X86::CondCode CC) {
283 if (CC == X86::COND_INVALID)
284 return SecondMacroFusionInstKind::Invalid;
288 return SecondMacroFusionInstKind::Invalid;
301 return SecondMacroFusionInstKind::ELG;
310 return SecondMacroFusionInstKind::AB;
323 return SecondMacroFusionInstKind::SPO;
327 /// \param FirstKind kind of the first instruction in macro fusion.
328 /// \param SecondKind kind of the second instruction in macro fusion.
330 /// \returns true if the two instruction can be macro fused.
331 inline bool isMacroFused(FirstMacroFusionInstKind FirstKind,
332 SecondMacroFusionInstKind SecondKind) {
334 case X86::FirstMacroFusionInstKind::Test:
335 case X86::FirstMacroFusionInstKind::And:
337 case X86::FirstMacroFusionInstKind::Cmp:
338 case X86::FirstMacroFusionInstKind::AddSub:
339 return SecondKind == X86::SecondMacroFusionInstKind::AB ||
340 SecondKind == X86::SecondMacroFusionInstKind::ELG;
341 case X86::FirstMacroFusionInstKind::IncDec:
342 return SecondKind == X86::SecondMacroFusionInstKind::ELG;
343 case X86::FirstMacroFusionInstKind::Invalid:
346 llvm_unreachable("unknown fusion type");
349 /// Defines the possible values of the branch boundary alignment mask.
350 enum AlignBranchBoundaryKind : uint8_t {
352 AlignBranchFused = 1U << 0,
353 AlignBranchJcc = 1U << 1,
354 AlignBranchJmp = 1U << 2,
355 AlignBranchCall = 1U << 3,
356 AlignBranchRet = 1U << 4,
357 AlignBranchIndirect = 1U << 5
359 } // end namespace X86;
361 /// X86II - This namespace holds all of the target specific flags that
362 /// instruction info tracks.
365 /// Target Operand Flag enum.
367 //===------------------------------------------------------------------===//
368 // X86 Specific MachineOperand flags.
372 /// MO_GOT_ABSOLUTE_ADDRESS - On a symbol operand, this represents a
374 /// SYMBOL_LABEL + [. - PICBASELABEL]
375 MO_GOT_ABSOLUTE_ADDRESS,
377 /// MO_PIC_BASE_OFFSET - On a symbol operand this indicates that the
378 /// immediate should get the value of the symbol minus the PIC base label:
379 /// SYMBOL_LABEL - PICBASELABEL
382 /// MO_GOT - On a symbol operand this indicates that the immediate is the
383 /// offset to the GOT entry for the symbol name from the base of the GOT.
385 /// See the X86-64 ELF ABI supplement for more details.
386 /// SYMBOL_LABEL @GOT
389 /// MO_GOTOFF - On a symbol operand this indicates that the immediate is
390 /// the offset to the location of the symbol name from the base of the GOT.
392 /// See the X86-64 ELF ABI supplement for more details.
393 /// SYMBOL_LABEL @GOTOFF
396 /// MO_GOTPCREL - On a symbol operand this indicates that the immediate is
397 /// offset to the GOT entry for the symbol name from the current code
400 /// See the X86-64 ELF ABI supplement for more details.
401 /// SYMBOL_LABEL @GOTPCREL
404 /// MO_PLT - On a symbol operand this indicates that the immediate is
405 /// offset to the PLT entry of symbol name from the current code location.
407 /// See the X86-64 ELF ABI supplement for more details.
408 /// SYMBOL_LABEL @PLT
411 /// MO_TLSGD - On a symbol operand this indicates that the immediate is
412 /// the offset of the GOT entry with the TLS index structure that contains
413 /// the module number and variable offset for the symbol. Used in the
414 /// general dynamic TLS access model.
416 /// See 'ELF Handling for Thread-Local Storage' for more details.
417 /// SYMBOL_LABEL @TLSGD
420 /// MO_TLSLD - On a symbol operand this indicates that the immediate is
421 /// the offset of the GOT entry with the TLS index for the module that
422 /// contains the symbol. When this index is passed to a call to
423 /// __tls_get_addr, the function will return the base address of the TLS
424 /// block for the symbol. Used in the x86-64 local dynamic TLS access model.
426 /// See 'ELF Handling for Thread-Local Storage' for more details.
427 /// SYMBOL_LABEL @TLSLD
430 /// MO_TLSLDM - On a symbol operand this indicates that the immediate is
431 /// the offset of the GOT entry with the TLS index for the module that
432 /// contains the symbol. When this index is passed to a call to
433 /// ___tls_get_addr, the function will return the base address of the TLS
434 /// block for the symbol. Used in the IA32 local dynamic TLS access model.
436 /// See 'ELF Handling for Thread-Local Storage' for more details.
437 /// SYMBOL_LABEL @TLSLDM
440 /// MO_GOTTPOFF - On a symbol operand this indicates that the immediate is
441 /// the offset of the GOT entry with the thread-pointer offset for the
442 /// symbol. Used in the x86-64 initial exec TLS access model.
444 /// See 'ELF Handling for Thread-Local Storage' for more details.
445 /// SYMBOL_LABEL @GOTTPOFF
448 /// MO_INDNTPOFF - On a symbol operand this indicates that the immediate is
449 /// the absolute address of the GOT entry with the negative thread-pointer
450 /// offset for the symbol. Used in the non-PIC IA32 initial exec TLS access
453 /// See 'ELF Handling for Thread-Local Storage' for more details.
454 /// SYMBOL_LABEL @INDNTPOFF
457 /// MO_TPOFF - On a symbol operand this indicates that the immediate is
458 /// the thread-pointer offset for the symbol. Used in the x86-64 local
459 /// exec TLS access model.
461 /// See 'ELF Handling for Thread-Local Storage' for more details.
462 /// SYMBOL_LABEL @TPOFF
465 /// MO_DTPOFF - On a symbol operand this indicates that the immediate is
466 /// the offset of the GOT entry with the TLS offset of the symbol. Used
467 /// in the local dynamic TLS access model.
469 /// See 'ELF Handling for Thread-Local Storage' for more details.
470 /// SYMBOL_LABEL @DTPOFF
473 /// MO_NTPOFF - On a symbol operand this indicates that the immediate is
474 /// the negative thread-pointer offset for the symbol. Used in the IA32
475 /// local exec TLS access model.
477 /// See 'ELF Handling for Thread-Local Storage' for more details.
478 /// SYMBOL_LABEL @NTPOFF
481 /// MO_GOTNTPOFF - On a symbol operand this indicates that the immediate is
482 /// the offset of the GOT entry with the negative thread-pointer offset for
483 /// the symbol. Used in the PIC IA32 initial exec TLS access model.
485 /// See 'ELF Handling for Thread-Local Storage' for more details.
486 /// SYMBOL_LABEL @GOTNTPOFF
489 /// MO_DLLIMPORT - On a symbol operand "FOO", this indicates that the
490 /// reference is actually to the "__imp_FOO" symbol. This is used for
491 /// dllimport linkage on windows.
494 /// MO_DARWIN_NONLAZY - On a symbol operand "FOO", this indicates that the
495 /// reference is actually to the "FOO$non_lazy_ptr" symbol, which is a
496 /// non-PIC-base-relative reference to a non-hidden dyld lazy pointer stub.
499 /// MO_DARWIN_NONLAZY_PIC_BASE - On a symbol operand "FOO", this indicates
500 /// that the reference is actually to "FOO$non_lazy_ptr - PICBASE", which is
501 /// a PIC-base-relative reference to a non-hidden dyld lazy pointer stub.
502 MO_DARWIN_NONLAZY_PIC_BASE,
504 /// MO_TLVP - On a symbol operand this indicates that the immediate is
507 /// This is the TLS offset for the Darwin TLS mechanism.
510 /// MO_TLVP_PIC_BASE - On a symbol operand this indicates that the immediate
511 /// is some TLS offset from the picbase.
513 /// This is the 32-bit TLS offset for Darwin TLS in PIC mode.
516 /// MO_SECREL - On a symbol operand this indicates that the immediate is
517 /// the offset from beginning of section.
519 /// This is the TLS offset for the COFF/Windows TLS mechanism.
522 /// MO_ABS8 - On a symbol operand this indicates that the symbol is known
523 /// to be an absolute symbol in range [0,128), so we can use the @ABS8
527 /// MO_COFFSTUB - On a symbol operand "FOO", this indicates that the
528 /// reference is actually to the ".refptr.FOO" symbol. This is used for
529 /// stub symbols on windows.
534 //===------------------------------------------------------------------===//
535 // Instruction encodings. These are the standard/most common forms for X86
539 // PseudoFrm - This represents an instruction that is a pseudo instruction
540 // or one that has not been implemented yet. It is illegal to code generate
541 // it, but tolerated for intermediate implementation stages.
544 /// Raw - This form is for instructions that don't have any operands, so
545 /// they are just a fixed opcode value, like 'leave'.
548 /// AddRegFrm - This form is used for instructions like 'push r32' that have
549 /// their one register operand added to their opcode.
552 /// RawFrmMemOffs - This form is for instructions that store an absolute
553 /// memory offset as an immediate with a possible segment override.
556 /// RawFrmSrc - This form is for instructions that use the source index
557 /// register SI/ESI/RSI with a possible segment override.
560 /// RawFrmDst - This form is for instructions that use the destination index
561 /// register DI/EDI/RDI.
564 /// RawFrmDstSrc - This form is for instructions that use the source index
565 /// register SI/ESI/RSI with a possible segment override, and also the
566 /// destination index register DI/EDI/RDI.
569 /// RawFrmImm8 - This is used for the ENTER instruction, which has two
570 /// immediates, the first of which is a 16-bit immediate (specified by
571 /// the imm encoding) and the second is a 8-bit fixed value.
574 /// RawFrmImm16 - This is used for CALL FAR instructions, which have two
575 /// immediates, the first of which is a 16 or 32-bit immediate (specified by
576 /// the imm encoding) and the second is a 16-bit fixed value. In the AMD
577 /// manual, this operand is described as pntr16:32 and pntr16:16
580 /// AddCCFrm - This form is used for Jcc that encode the condition code
581 /// in the lower 4 bits of the opcode.
584 /// MRM[0-7][rm] - These forms are used to represent instructions that use
585 /// a Mod/RM byte, and use the middle field to hold extended opcode
586 /// information. In the intel manual these are represented as /0, /1, ...
589 /// MRMDestMem - This form is used for instructions that use the Mod/RM byte
590 /// to specify a destination, which in this case is memory.
594 /// MRMSrcMem - This form is used for instructions that use the Mod/RM byte
595 /// to specify a source, which in this case is memory.
599 /// MRMSrcMem4VOp3 - This form is used for instructions that encode
600 /// operand 3 with VEX.VVVV and load from memory.
604 /// MRMSrcMemOp4 - This form is used for instructions that use the Mod/RM
605 /// byte to specify the fourth source, which in this case is memory.
609 /// MRMSrcMemCC - This form is used for instructions that use the Mod/RM
610 /// byte to specify the operands and also encodes a condition code.
614 /// MRMXm - This form is used for instructions that use the Mod/RM byte
615 /// to specify a memory source, but doesn't use the middle field. And has
616 /// a condition code.
620 /// MRMXm - This form is used for instructions that use the Mod/RM byte
621 /// to specify a memory source, but doesn't use the middle field.
625 // Next, instructions that operate on a memory r/m operand...
626 MRM0m = 40, MRM1m = 41, MRM2m = 42, MRM3m = 43, // Format /0 /1 /2 /3
627 MRM4m = 44, MRM5m = 45, MRM6m = 46, MRM7m = 47, // Format /4 /5 /6 /7
629 /// MRMDestReg - This form is used for instructions that use the Mod/RM byte
630 /// to specify a destination, which in this case is a register.
634 /// MRMSrcReg - This form is used for instructions that use the Mod/RM byte
635 /// to specify a source, which in this case is a register.
639 /// MRMSrcReg4VOp3 - This form is used for instructions that encode
640 /// operand 3 with VEX.VVVV and do not load from memory.
644 /// MRMSrcRegOp4 - This form is used for instructions that use the Mod/RM
645 /// byte to specify the fourth source, which in this case is a register.
649 /// MRMSrcRegCC - This form is used for instructions that use the Mod/RM
650 /// byte to specify the operands and also encodes a condition code
654 /// MRMXCCr - This form is used for instructions that use the Mod/RM byte
655 /// to specify a register source, but doesn't use the middle field. And has
656 /// a condition code.
660 /// MRMXr - This form is used for instructions that use the Mod/RM byte
661 /// to specify a register source, but doesn't use the middle field.
665 // Instructions that operate on a register r/m operand...
666 MRM0r = 56, MRM1r = 57, MRM2r = 58, MRM3r = 59, // Format /0 /1 /2 /3
667 MRM4r = 60, MRM5r = 61, MRM6r = 62, MRM7r = 63, // Format /4 /5 /6 /7
669 /// MRM_XX - A mod/rm byte of exactly 0xXX.
670 MRM_C0 = 64, MRM_C1 = 65, MRM_C2 = 66, MRM_C3 = 67,
671 MRM_C4 = 68, MRM_C5 = 69, MRM_C6 = 70, MRM_C7 = 71,
672 MRM_C8 = 72, MRM_C9 = 73, MRM_CA = 74, MRM_CB = 75,
673 MRM_CC = 76, MRM_CD = 77, MRM_CE = 78, MRM_CF = 79,
674 MRM_D0 = 80, MRM_D1 = 81, MRM_D2 = 82, MRM_D3 = 83,
675 MRM_D4 = 84, MRM_D5 = 85, MRM_D6 = 86, MRM_D7 = 87,
676 MRM_D8 = 88, MRM_D9 = 89, MRM_DA = 90, MRM_DB = 91,
677 MRM_DC = 92, MRM_DD = 93, MRM_DE = 94, MRM_DF = 95,
678 MRM_E0 = 96, MRM_E1 = 97, MRM_E2 = 98, MRM_E3 = 99,
679 MRM_E4 = 100, MRM_E5 = 101, MRM_E6 = 102, MRM_E7 = 103,
680 MRM_E8 = 104, MRM_E9 = 105, MRM_EA = 106, MRM_EB = 107,
681 MRM_EC = 108, MRM_ED = 109, MRM_EE = 110, MRM_EF = 111,
682 MRM_F0 = 112, MRM_F1 = 113, MRM_F2 = 114, MRM_F3 = 115,
683 MRM_F4 = 116, MRM_F5 = 117, MRM_F6 = 118, MRM_F7 = 119,
684 MRM_F8 = 120, MRM_F9 = 121, MRM_FA = 122, MRM_FB = 123,
685 MRM_FC = 124, MRM_FD = 125, MRM_FE = 126, MRM_FF = 127,
689 //===------------------------------------------------------------------===//
692 // OpSize - OpSizeFixed implies instruction never needs a 0x66 prefix.
693 // OpSize16 means this is a 16-bit instruction and needs 0x66 prefix in
694 // 32-bit mode. OpSize32 means this is a 32-bit instruction needs a 0x66
695 // prefix in 16-bit mode.
697 OpSizeMask = 0x3 << OpSizeShift,
699 OpSizeFixed = 0 << OpSizeShift,
700 OpSize16 = 1 << OpSizeShift,
701 OpSize32 = 2 << OpSizeShift,
703 // AsSize - AdSizeX implies this instruction determines its need of 0x67
704 // prefix from a normal ModRM memory operand. The other types indicate that
705 // an operand is encoded with a specific width and a prefix is needed if
706 // it differs from the current mode.
707 AdSizeShift = OpSizeShift + 2,
708 AdSizeMask = 0x3 << AdSizeShift,
710 AdSizeX = 0 << AdSizeShift,
711 AdSize16 = 1 << AdSizeShift,
712 AdSize32 = 2 << AdSizeShift,
713 AdSize64 = 3 << AdSizeShift,
715 //===------------------------------------------------------------------===//
716 // OpPrefix - There are several prefix bytes that are used as opcode
717 // extensions. These are 0x66, 0xF3, and 0xF2. If this field is 0 there is
720 OpPrefixShift = AdSizeShift + 2,
721 OpPrefixMask = 0x3 << OpPrefixShift,
723 // PD - Prefix code for packed double precision vector floating point
724 // operations performed in the SSE registers.
725 PD = 1 << OpPrefixShift,
727 // XS, XD - These prefix codes are for single and double precision scalar
728 // floating point operations performed in the SSE registers.
729 XS = 2 << OpPrefixShift, XD = 3 << OpPrefixShift,
731 //===------------------------------------------------------------------===//
732 // OpMap - This field determines which opcode map this instruction
733 // belongs to. i.e. one-byte, two-byte, 0x0f 0x38, 0x0f 0x3a, etc.
735 OpMapShift = OpPrefixShift + 2,
736 OpMapMask = 0x7 << OpMapShift,
738 // OB - OneByte - Set if this instruction has a one byte opcode.
739 OB = 0 << OpMapShift,
741 // TB - TwoByte - Set if this instruction has a two byte opcode, which
742 // starts with a 0x0F byte before the real opcode.
743 TB = 1 << OpMapShift,
745 // T8, TA - Prefix after the 0x0F prefix.
746 T8 = 2 << OpMapShift, TA = 3 << OpMapShift,
748 // XOP8 - Prefix to include use of imm byte.
749 XOP8 = 4 << OpMapShift,
751 // XOP9 - Prefix to exclude use of imm byte.
752 XOP9 = 5 << OpMapShift,
754 // XOPA - Prefix to encode 0xA in VEX.MMMM of XOP instructions.
755 XOPA = 6 << OpMapShift,
757 /// ThreeDNow - This indicates that the instruction uses the
758 /// wacky 0x0F 0x0F prefix for 3DNow! instructions. The manual documents
759 /// this as having a 0x0F prefix with a 0x0F opcode, and each instruction
760 /// storing a classifier in the imm8 field. To simplify our implementation,
761 /// we handle this by storeing the classifier in the opcode field and using
762 /// this flag to indicate that the encoder should do the wacky 3DNow! thing.
763 ThreeDNow = 7 << OpMapShift,
765 //===------------------------------------------------------------------===//
766 // REX_W - REX prefixes are instruction prefixes used in 64-bit mode.
767 // They are used to specify GPRs and SSE registers, 64-bit operand size,
768 // etc. We only cares about REX.W and REX.R bits and only the former is
769 // statically determined.
771 REXShift = OpMapShift + 3,
772 REX_W = 1 << REXShift,
774 //===------------------------------------------------------------------===//
775 // This three-bit field describes the size of an immediate operand. Zero is
776 // unused so that we can tell if we forgot to set a value.
777 ImmShift = REXShift + 1,
778 ImmMask = 15 << ImmShift,
779 Imm8 = 1 << ImmShift,
780 Imm8PCRel = 2 << ImmShift,
781 Imm8Reg = 3 << ImmShift,
782 Imm16 = 4 << ImmShift,
783 Imm16PCRel = 5 << ImmShift,
784 Imm32 = 6 << ImmShift,
785 Imm32PCRel = 7 << ImmShift,
786 Imm32S = 8 << ImmShift,
787 Imm64 = 9 << ImmShift,
789 //===------------------------------------------------------------------===//
790 // FP Instruction Classification... Zero is non-fp instruction.
792 // FPTypeMask - Mask for all of the FP types...
793 FPTypeShift = ImmShift + 4,
794 FPTypeMask = 7 << FPTypeShift,
796 // NotFP - The default, set for instructions that do not use FP registers.
797 NotFP = 0 << FPTypeShift,
799 // ZeroArgFP - 0 arg FP instruction which implicitly pushes ST(0), f.e. fld0
800 ZeroArgFP = 1 << FPTypeShift,
802 // OneArgFP - 1 arg FP instructions which implicitly read ST(0), such as fst
803 OneArgFP = 2 << FPTypeShift,
805 // OneArgFPRW - 1 arg FP instruction which implicitly read ST(0) and write a
806 // result back to ST(0). For example, fcos, fsqrt, etc.
808 OneArgFPRW = 3 << FPTypeShift,
810 // TwoArgFP - 2 arg FP instructions which implicitly read ST(0), and an
811 // explicit argument, storing the result to either ST(0) or the implicit
812 // argument. For example: fadd, fsub, fmul, etc...
813 TwoArgFP = 4 << FPTypeShift,
815 // CompareFP - 2 arg FP instructions which implicitly read ST(0) and an
816 // explicit argument, but have no destination. Example: fucom, fucomi, ...
817 CompareFP = 5 << FPTypeShift,
819 // CondMovFP - "2 operand" floating point conditional move instructions.
820 CondMovFP = 6 << FPTypeShift,
822 // SpecialFP - Special instruction forms. Dispatch by opcode explicitly.
823 SpecialFP = 7 << FPTypeShift,
826 LOCKShift = FPTypeShift + 3,
827 LOCK = 1 << LOCKShift,
830 REPShift = LOCKShift + 1,
833 // Execution domain for SSE instructions.
834 // 0 means normal, non-SSE instruction.
835 SSEDomainShift = REPShift + 1,
838 EncodingShift = SSEDomainShift + 2,
839 EncodingMask = 0x3 << EncodingShift,
841 // VEX - encoding using 0xC4/0xC5
842 VEX = 1 << EncodingShift,
844 /// XOP - Opcode prefix used by XOP instructions.
845 XOP = 2 << EncodingShift,
847 // VEX_EVEX - Specifies that this instruction use EVEX form which provides
848 // syntax support up to 32 512-bit register operands and up to 7 16-bit
849 // mask operands as well as source operand data swizzling/memory operand
850 // conversion, eviction hint, and rounding mode.
851 EVEX = 3 << EncodingShift,
854 OpcodeShift = EncodingShift + 2,
856 /// VEX_W - Has a opcode specific functionality, but is used in the same
857 /// way as REX_W is for regular SSE instructions.
858 VEX_WShift = OpcodeShift + 8,
859 VEX_W = 1ULL << VEX_WShift,
861 /// VEX_4V - Used to specify an additional AVX/SSE register. Several 2
862 /// address instructions in SSE are represented as 3 address ones in AVX
863 /// and the additional register is encoded in VEX_VVVV prefix.
864 VEX_4VShift = VEX_WShift + 1,
865 VEX_4V = 1ULL << VEX_4VShift,
867 /// VEX_L - Stands for a bit in the VEX opcode prefix meaning the current
868 /// instruction uses 256-bit wide registers. This is usually auto detected
869 /// if a VR256 register is used, but some AVX instructions also have this
870 /// field marked when using a f256 memory references.
871 VEX_LShift = VEX_4VShift + 1,
872 VEX_L = 1ULL << VEX_LShift,
874 // EVEX_K - Set if this instruction requires masking
875 EVEX_KShift = VEX_LShift + 1,
876 EVEX_K = 1ULL << EVEX_KShift,
878 // EVEX_Z - Set if this instruction has EVEX.Z field set.
879 EVEX_ZShift = EVEX_KShift + 1,
880 EVEX_Z = 1ULL << EVEX_ZShift,
882 // EVEX_L2 - Set if this instruction has EVEX.L' field set.
883 EVEX_L2Shift = EVEX_ZShift + 1,
884 EVEX_L2 = 1ULL << EVEX_L2Shift,
886 // EVEX_B - Set if this instruction has EVEX.B field set.
887 EVEX_BShift = EVEX_L2Shift + 1,
888 EVEX_B = 1ULL << EVEX_BShift,
890 // The scaling factor for the AVX512's 8-bit compressed displacement.
891 CD8_Scale_Shift = EVEX_BShift + 1,
892 CD8_Scale_Mask = 127ULL << CD8_Scale_Shift,
894 /// Explicitly specified rounding control
895 EVEX_RCShift = CD8_Scale_Shift + 7,
896 EVEX_RC = 1ULL << EVEX_RCShift,
899 NoTrackShift = EVEX_RCShift + 1,
900 NOTRACK = 1ULL << NoTrackShift
903 /// \returns the "base" X86 opcode for the specified machine
905 inline uint8_t getBaseOpcodeFor(uint64_t TSFlags) {
906 return TSFlags >> X86II::OpcodeShift;
909 inline bool hasImm(uint64_t TSFlags) {
910 return (TSFlags & X86II::ImmMask) != 0;
913 /// Decode the "size of immediate" field from the TSFlags field of the
914 /// specified instruction.
915 inline unsigned getSizeOfImm(uint64_t TSFlags) {
916 switch (TSFlags & X86II::ImmMask) {
917 default: llvm_unreachable("Unknown immediate size");
919 case X86II::Imm8PCRel:
920 case X86II::Imm8Reg: return 1;
922 case X86II::Imm16PCRel: return 2;
925 case X86II::Imm32PCRel: return 4;
926 case X86II::Imm64: return 8;
930 /// \returns true if the immediate of the specified instruction's TSFlags
931 /// indicates that it is pc relative.
932 inline bool isImmPCRel(uint64_t TSFlags) {
933 switch (TSFlags & X86II::ImmMask) {
934 default: llvm_unreachable("Unknown immediate size");
935 case X86II::Imm8PCRel:
936 case X86II::Imm16PCRel:
937 case X86II::Imm32PCRel:
949 /// \returns true if the immediate of the specified instruction's
950 /// TSFlags indicates that it is signed.
951 inline bool isImmSigned(uint64_t TSFlags) {
952 switch (TSFlags & X86II::ImmMask) {
953 default: llvm_unreachable("Unknown immediate signedness");
957 case X86II::Imm8PCRel:
960 case X86II::Imm16PCRel:
962 case X86II::Imm32PCRel:
968 /// Compute whether all of the def operands are repeated in the uses and
969 /// therefore should be skipped.
970 /// This determines the start of the unique operand list. We need to determine
971 /// if all of the defs have a corresponding tied operand in the uses.
972 /// Unfortunately, the tied operand information is encoded in the uses not
973 /// the defs so we have to use some heuristics to find which operands to
975 inline unsigned getOperandBias(const MCInstrDesc& Desc) {
976 unsigned NumDefs = Desc.getNumDefs();
977 unsigned NumOps = Desc.getNumOperands();
979 default: llvm_unreachable("Unexpected number of defs");
983 // Common two addr case.
984 if (NumOps > 1 && Desc.getOperandConstraint(1, MCOI::TIED_TO) == 0)
986 // Check for AVX-512 scatter which has a TIED_TO in the second to last
989 Desc.getOperandConstraint(6, MCOI::TIED_TO) == 0)
993 // XCHG/XADD have two destinations and two sources.
994 if (NumOps >= 4 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0 &&
995 Desc.getOperandConstraint(3, MCOI::TIED_TO) == 1)
997 // Check for gather. AVX-512 has the second tied operand early. AVX2
998 // has it as the last op.
999 if (NumOps == 9 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0 &&
1000 (Desc.getOperandConstraint(3, MCOI::TIED_TO) == 1 ||
1001 Desc.getOperandConstraint(8, MCOI::TIED_TO) == 1))
1007 /// The function returns the MCInst operand # for the first field of the
1008 /// memory operand. If the instruction doesn't have a
1009 /// memory operand, this returns -1.
1011 /// Note that this ignores tied operands. If there is a tied register which
1012 /// is duplicated in the MCInst (e.g. "EAX = addl EAX, [mem]") it is only
1013 /// counted as one operand.
1015 inline int getMemoryOperandNo(uint64_t TSFlags) {
1016 bool HasVEX_4V = TSFlags & X86II::VEX_4V;
1017 bool HasEVEX_K = TSFlags & X86II::EVEX_K;
1019 switch (TSFlags & X86II::FormMask) {
1020 default: llvm_unreachable("Unknown FormMask value in getMemoryOperandNo!");
1023 case X86II::AddRegFrm:
1024 case X86II::RawFrmImm8:
1025 case X86II::RawFrmImm16:
1026 case X86II::RawFrmMemOffs:
1027 case X86II::RawFrmSrc:
1028 case X86II::RawFrmDst:
1029 case X86II::RawFrmDstSrc:
1030 case X86II::AddCCFrm:
1032 case X86II::MRMDestMem:
1034 case X86II::MRMSrcMem:
1035 // Start from 1, skip any registers encoded in VEX_VVVV or I8IMM, or a
1037 return 1 + HasVEX_4V + HasEVEX_K;
1038 case X86II::MRMSrcMem4VOp3:
1039 // Skip registers encoded in reg.
1040 return 1 + HasEVEX_K;
1041 case X86II::MRMSrcMemOp4:
1042 // Skip registers encoded in reg, VEX_VVVV, and I8IMM.
1044 case X86II::MRMSrcMemCC:
1045 // Start from 1, skip any registers encoded in VEX_VVVV or I8IMM, or a
1048 case X86II::MRMDestReg:
1049 case X86II::MRMSrcReg:
1050 case X86II::MRMSrcReg4VOp3:
1051 case X86II::MRMSrcRegOp4:
1052 case X86II::MRMSrcRegCC:
1053 case X86II::MRMXrCC:
1055 case X86II::MRM0r: case X86II::MRM1r:
1056 case X86II::MRM2r: case X86II::MRM3r:
1057 case X86II::MRM4r: case X86II::MRM5r:
1058 case X86II::MRM6r: case X86II::MRM7r:
1060 case X86II::MRMXmCC:
1062 case X86II::MRM0m: case X86II::MRM1m:
1063 case X86II::MRM2m: case X86II::MRM3m:
1064 case X86II::MRM4m: case X86II::MRM5m:
1065 case X86II::MRM6m: case X86II::MRM7m:
1066 // Start from 0, skip registers encoded in VEX_VVVV or a mask register.
1067 return 0 + HasVEX_4V + HasEVEX_K;
1068 case X86II::MRM_C0: case X86II::MRM_C1: case X86II::MRM_C2:
1069 case X86II::MRM_C3: case X86II::MRM_C4: case X86II::MRM_C5:
1070 case X86II::MRM_C6: case X86II::MRM_C7: case X86II::MRM_C8:
1071 case X86II::MRM_C9: case X86II::MRM_CA: case X86II::MRM_CB:
1072 case X86II::MRM_CC: case X86II::MRM_CD: case X86II::MRM_CE:
1073 case X86II::MRM_CF: case X86II::MRM_D0: case X86II::MRM_D1:
1074 case X86II::MRM_D2: case X86II::MRM_D3: case X86II::MRM_D4:
1075 case X86II::MRM_D5: case X86II::MRM_D6: case X86II::MRM_D7:
1076 case X86II::MRM_D8: case X86II::MRM_D9: case X86II::MRM_DA:
1077 case X86II::MRM_DB: case X86II::MRM_DC: case X86II::MRM_DD:
1078 case X86II::MRM_DE: case X86II::MRM_DF: case X86II::MRM_E0:
1079 case X86II::MRM_E1: case X86II::MRM_E2: case X86II::MRM_E3:
1080 case X86II::MRM_E4: case X86II::MRM_E5: case X86II::MRM_E6:
1081 case X86II::MRM_E7: case X86II::MRM_E8: case X86II::MRM_E9:
1082 case X86II::MRM_EA: case X86II::MRM_EB: case X86II::MRM_EC:
1083 case X86II::MRM_ED: case X86II::MRM_EE: case X86II::MRM_EF:
1084 case X86II::MRM_F0: case X86II::MRM_F1: case X86II::MRM_F2:
1085 case X86II::MRM_F3: case X86II::MRM_F4: case X86II::MRM_F5:
1086 case X86II::MRM_F6: case X86II::MRM_F7: case X86II::MRM_F8:
1087 case X86II::MRM_F9: case X86II::MRM_FA: case X86II::MRM_FB:
1088 case X86II::MRM_FC: case X86II::MRM_FD: case X86II::MRM_FE:
1094 /// \returns true if the MachineOperand is a x86-64 extended (r8 or
1095 /// higher) register, e.g. r8, xmm8, xmm13, etc.
1096 inline bool isX86_64ExtendedReg(unsigned RegNo) {
1097 if ((RegNo >= X86::XMM8 && RegNo <= X86::XMM31) ||
1098 (RegNo >= X86::YMM8 && RegNo <= X86::YMM31) ||
1099 (RegNo >= X86::ZMM8 && RegNo <= X86::ZMM31))
1104 case X86::R8: case X86::R9: case X86::R10: case X86::R11:
1105 case X86::R12: case X86::R13: case X86::R14: case X86::R15:
1106 case X86::R8D: case X86::R9D: case X86::R10D: case X86::R11D:
1107 case X86::R12D: case X86::R13D: case X86::R14D: case X86::R15D:
1108 case X86::R8W: case X86::R9W: case X86::R10W: case X86::R11W:
1109 case X86::R12W: case X86::R13W: case X86::R14W: case X86::R15W:
1110 case X86::R8B: case X86::R9B: case X86::R10B: case X86::R11B:
1111 case X86::R12B: case X86::R13B: case X86::R14B: case X86::R15B:
1112 case X86::CR8: case X86::CR9: case X86::CR10: case X86::CR11:
1113 case X86::CR12: case X86::CR13: case X86::CR14: case X86::CR15:
1114 case X86::DR8: case X86::DR9: case X86::DR10: case X86::DR11:
1115 case X86::DR12: case X86::DR13: case X86::DR14: case X86::DR15:
1121 /// \returns true if the MemoryOperand is a 32 extended (zmm16 or higher)
1122 /// registers, e.g. zmm21, etc.
1123 static inline bool is32ExtendedReg(unsigned RegNo) {
1124 return ((RegNo >= X86::XMM16 && RegNo <= X86::XMM31) ||
1125 (RegNo >= X86::YMM16 && RegNo <= X86::YMM31) ||
1126 (RegNo >= X86::ZMM16 && RegNo <= X86::ZMM31));
1130 inline bool isX86_64NonExtLowByteReg(unsigned reg) {
1131 return (reg == X86::SPL || reg == X86::BPL ||
1132 reg == X86::SIL || reg == X86::DIL);
1135 /// \returns true if this is a masked instruction.
1136 inline bool isKMasked(uint64_t TSFlags) {
1137 return (TSFlags & X86II::EVEX_K) != 0;
1140 /// \returns true if this is a merge masked instruction.
1141 inline bool isKMergeMasked(uint64_t TSFlags) {
1142 return isKMasked(TSFlags) && (TSFlags & X86II::EVEX_Z) == 0;
1146 } // end namespace llvm;