1 //===-- X86BaseInfo.h - Top level definitions for X86 -------- --*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
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..
15 //===----------------------------------------------------------------------===//
20 #include "X86MCTargetDesc.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.
43 } // end namespace X86;
46 /// X86II - This namespace holds all of the target specific flags that
47 /// instruction info tracks.
50 /// Target Operand Flag enum.
52 //===------------------------------------------------------------------===//
53 // X86 Specific MachineOperand flags.
57 /// MO_GOT_ABSOLUTE_ADDRESS - On a symbol operand, this represents a
59 /// SYMBOL_LABEL + [. - PICBASELABEL]
60 MO_GOT_ABSOLUTE_ADDRESS,
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
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.
70 /// See the X86-64 ELF ABI supplement for more details.
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.
77 /// See the X86-64 ELF ABI supplement for more details.
78 /// SYMBOL_LABEL @GOTOFF
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
85 /// See the X86-64 ELF ABI supplement for more details.
86 /// SYMBOL_LABEL @GOTPCREL
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.
92 /// See the X86-64 ELF ABI supplement for more details.
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.
101 /// See 'ELF Handling for Thread-Local Storage' for more details.
102 /// SYMBOL_LABEL @TLSGD
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 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.
111 /// See 'ELF Handling for Thread-Local Storage' for more details.
112 /// SYMBOL_LABEL @TLSLD
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 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.
121 /// See 'ELF Handling for Thread-Local Storage' for more details.
122 /// SYMBOL_LABEL @TLSLDM
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.
129 /// See 'ELF Handling for Thread-Local Storage' for more details.
130 /// SYMBOL_LABEL @GOTTPOFF
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
138 /// See 'ELF Handling for Thread-Local Storage' for more details.
139 /// SYMBOL_LABEL @INDNTPOFF
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.
146 /// See 'ELF Handling for Thread-Local Storage' for more details.
147 /// SYMBOL_LABEL @TPOFF
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.
154 /// See 'ELF Handling for Thread-Local Storage' for more details.
155 /// SYMBOL_LABEL @DTPOFF
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.
162 /// See 'ELF Handling for Thread-Local Storage' for more details.
163 /// SYMBOL_LABEL @NTPOFF
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.
170 /// See 'ELF Handling for Thread-Local Storage' for more details.
171 /// SYMBOL_LABEL @GOTNTPOFF
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.
179 /// MO_DARWIN_STUB - On a symbol operand "FOO", this indicates that the
180 /// reference is actually to the "FOO$stub" symbol. This is used for calls
181 /// and jumps to external functions on Tiger and earlier.
184 /// MO_DARWIN_NONLAZY - On a symbol operand "FOO", this indicates that the
185 /// reference is actually to the "FOO$non_lazy_ptr" symbol, which is a
186 /// non-PIC-base-relative reference to a non-hidden dyld lazy pointer stub.
189 /// MO_DARWIN_NONLAZY_PIC_BASE - On a symbol operand "FOO", this indicates
190 /// that the reference is actually to "FOO$non_lazy_ptr - PICBASE", which is
191 /// a PIC-base-relative reference to a non-hidden dyld lazy pointer stub.
192 MO_DARWIN_NONLAZY_PIC_BASE,
194 /// MO_DARWIN_HIDDEN_NONLAZY_PIC_BASE - On a symbol operand "FOO", this
195 /// indicates that the reference is actually to "FOO$non_lazy_ptr -PICBASE",
196 /// which is a PIC-base-relative reference to a hidden dyld lazy pointer
198 MO_DARWIN_HIDDEN_NONLAZY_PIC_BASE,
200 /// MO_TLVP - On a symbol operand this indicates that the immediate is
203 /// This is the TLS offset for the Darwin TLS mechanism.
206 /// MO_TLVP_PIC_BASE - On a symbol operand this indicates that the immediate
207 /// is some TLS offset from the picbase.
209 /// This is the 32-bit TLS offset for Darwin TLS in PIC mode.
212 /// MO_SECREL - On a symbol operand this indicates that the immediate is
213 /// the offset from beginning of section.
215 /// This is the TLS offset for the COFF/Windows TLS mechanism.
220 //===------------------------------------------------------------------===//
221 // Instruction encodings. These are the standard/most common forms for X86
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.
230 /// Raw - This form is for instructions that don't have any operands, so
231 /// they are just a fixed opcode value, like 'leave'.
234 /// AddRegFrm - This form is used for instructions like 'push r32' that have
235 /// their one register operand added to their opcode.
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.
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.
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.
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.
258 /// MRM[0-7][rm] - These forms are used to represent instructions that use
259 /// a Mod/RM byte, and use the middle field to hold extended opcode
260 /// information. In the intel manual these are represented as /0, /1, ...
263 // First, instructions that operate on a register r/m operand...
264 MRM0r = 16, MRM1r = 17, MRM2r = 18, MRM3r = 19, // Format /0 /1 /2 /3
265 MRM4r = 20, MRM5r = 21, MRM6r = 22, MRM7r = 23, // Format /4 /5 /6 /7
267 // Next, instructions that operate on a memory r/m operand...
268 MRM0m = 24, MRM1m = 25, MRM2m = 26, MRM3m = 27, // Format /0 /1 /2 /3
269 MRM4m = 28, MRM5m = 29, MRM6m = 30, MRM7m = 31, // Format /4 /5 /6 /7
271 // MRMInitReg - This form is used for instructions whose source and
272 // destinations are the same register.
275 //// MRM_XX - A mod/rm byte of exactly 0xXX.
276 MRM_C1 = 33, MRM_C2 = 34, MRM_C3 = 35, MRM_C4 = 36,
277 MRM_C8 = 37, MRM_C9 = 38, MRM_E8 = 39, MRM_F0 = 40,
278 MRM_F8 = 41, MRM_F9 = 42, MRM_D0 = 45, MRM_D1 = 46,
279 MRM_D4 = 47, MRM_D5 = 48, MRM_D8 = 49, MRM_D9 = 50,
280 MRM_DA = 51, MRM_DB = 52, MRM_DC = 53, MRM_DD = 54,
281 MRM_DE = 55, MRM_DF = 56,
283 /// RawFrmImm8 - This is used for the ENTER instruction, which has two
284 /// immediates, the first of which is a 16-bit immediate (specified by
285 /// the imm encoding) and the second is a 8-bit fixed value.
288 /// RawFrmImm16 - This is used for CALL FAR instructions, which have two
289 /// immediates, the first of which is a 16 or 32-bit immediate (specified by
290 /// the imm encoding) and the second is a 16-bit fixed value. In the AMD
291 /// manual, this operand is described as pntr16:32 and pntr16:16
296 //===------------------------------------------------------------------===//
299 // OpSize - Set if this instruction requires an operand size prefix (0x66),
300 // which most often indicates that the instruction operates on 16 bit data
301 // instead of 32 bit data.
304 // AsSize - Set if this instruction requires an operand size prefix (0x67),
305 // which most often indicates that the instruction address 16 bit address
306 // instead of 32 bit address (or 32 bit address in 64 bit mode).
309 //===------------------------------------------------------------------===//
310 // Op0Mask - There are several prefix bytes that are used to form two byte
311 // opcodes. These are currently 0x0F, 0xF3, and 0xD8-0xDF. This mask is
312 // used to obtain the setting of this field. If no bits in this field is
313 // set, there is no prefix byte for obtaining a multibyte opcode.
316 Op0Mask = 0x1F << Op0Shift,
318 // TB - TwoByte - Set if this instruction has a two byte opcode, which
319 // starts with a 0x0F byte before the real opcode.
322 // REP - The 0xF3 prefix byte indicating repetition of the following
326 // D8-DF - These escape opcodes are used by the floating point unit. These
327 // values must remain sequential.
328 D8 = 3 << Op0Shift, D9 = 4 << Op0Shift,
329 DA = 5 << Op0Shift, DB = 6 << Op0Shift,
330 DC = 7 << Op0Shift, DD = 8 << Op0Shift,
331 DE = 9 << Op0Shift, DF = 10 << Op0Shift,
333 // XS, XD - These prefix codes are for single and double precision scalar
334 // floating point operations performed in the SSE registers.
335 XD = 11 << Op0Shift, XS = 12 << Op0Shift,
337 // T8, TA, A6, A7 - Prefix after the 0x0F prefix.
338 T8 = 13 << Op0Shift, TA = 14 << Op0Shift,
339 A6 = 15 << Op0Shift, A7 = 16 << Op0Shift,
341 // T8XD - Prefix before and after 0x0F. Combination of T8 and XD.
342 T8XD = 17 << Op0Shift,
344 // T8XS - Prefix before and after 0x0F. Combination of T8 and XS.
345 T8XS = 18 << Op0Shift,
347 // TAXD - Prefix before and after 0x0F. Combination of TA and XD.
348 TAXD = 19 << Op0Shift,
350 // XOP8 - Prefix to include use of imm byte.
351 XOP8 = 20 << Op0Shift,
353 // XOP9 - Prefix to exclude use of imm byte.
354 XOP9 = 21 << Op0Shift,
356 //===------------------------------------------------------------------===//
357 // REX_W - REX prefixes are instruction prefixes used in 64-bit mode.
358 // They are used to specify GPRs and SSE registers, 64-bit operand size,
359 // etc. We only cares about REX.W and REX.R bits and only the former is
360 // statically determined.
362 REXShift = Op0Shift + 5,
363 REX_W = 1 << REXShift,
365 //===------------------------------------------------------------------===//
366 // This three-bit field describes the size of an immediate operand. Zero is
367 // unused so that we can tell if we forgot to set a value.
368 ImmShift = REXShift + 1,
369 ImmMask = 7 << ImmShift,
370 Imm8 = 1 << ImmShift,
371 Imm8PCRel = 2 << ImmShift,
372 Imm16 = 3 << ImmShift,
373 Imm16PCRel = 4 << ImmShift,
374 Imm32 = 5 << ImmShift,
375 Imm32PCRel = 6 << ImmShift,
376 Imm64 = 7 << ImmShift,
378 //===------------------------------------------------------------------===//
379 // FP Instruction Classification... Zero is non-fp instruction.
381 // FPTypeMask - Mask for all of the FP types...
382 FPTypeShift = ImmShift + 3,
383 FPTypeMask = 7 << FPTypeShift,
385 // NotFP - The default, set for instructions that do not use FP registers.
386 NotFP = 0 << FPTypeShift,
388 // ZeroArgFP - 0 arg FP instruction which implicitly pushes ST(0), f.e. fld0
389 ZeroArgFP = 1 << FPTypeShift,
391 // OneArgFP - 1 arg FP instructions which implicitly read ST(0), such as fst
392 OneArgFP = 2 << FPTypeShift,
394 // OneArgFPRW - 1 arg FP instruction which implicitly read ST(0) and write a
395 // result back to ST(0). For example, fcos, fsqrt, etc.
397 OneArgFPRW = 3 << FPTypeShift,
399 // TwoArgFP - 2 arg FP instructions which implicitly read ST(0), and an
400 // explicit argument, storing the result to either ST(0) or the implicit
401 // argument. For example: fadd, fsub, fmul, etc...
402 TwoArgFP = 4 << FPTypeShift,
404 // CompareFP - 2 arg FP instructions which implicitly read ST(0) and an
405 // explicit argument, but have no destination. Example: fucom, fucomi, ...
406 CompareFP = 5 << FPTypeShift,
408 // CondMovFP - "2 operand" floating point conditional move instructions.
409 CondMovFP = 6 << FPTypeShift,
411 // SpecialFP - Special instruction forms. Dispatch by opcode explicitly.
412 SpecialFP = 7 << FPTypeShift,
415 LOCKShift = FPTypeShift + 3,
416 LOCK = 1 << LOCKShift,
418 // Segment override prefixes. Currently we just need ability to address
419 // stuff in gs and fs segments.
420 SegOvrShift = LOCKShift + 1,
421 SegOvrMask = 3 << SegOvrShift,
422 FS = 1 << SegOvrShift,
423 GS = 2 << SegOvrShift,
425 // Execution domain for SSE instructions in bits 23, 24.
426 // 0 in bits 23-24 means normal, non-SSE instruction.
427 SSEDomainShift = SegOvrShift + 2,
429 OpcodeShift = SSEDomainShift + 2,
431 //===------------------------------------------------------------------===//
432 /// VEX - The opcode prefix used by AVX instructions
433 VEXShift = OpcodeShift + 8,
436 /// VEX_W - Has a opcode specific functionality, but is used in the same
437 /// way as REX_W is for regular SSE instructions.
440 /// VEX_4V - Used to specify an additional AVX/SSE register. Several 2
441 /// address instructions in SSE are represented as 3 address ones in AVX
442 /// and the additional register is encoded in VEX_VVVV prefix.
445 /// VEX_4VOp3 - Similar to VEX_4V, but used on instructions that encode
446 /// operand 3 with VEX.vvvv.
449 /// VEX_I8IMM - Specifies that the last register used in a AVX instruction,
450 /// must be encoded in the i8 immediate field. This usually happens in
451 /// instructions with 4 operands.
454 /// VEX_L - Stands for a bit in the VEX opcode prefix meaning the current
455 /// instruction uses 256-bit wide registers. This is usually auto detected
456 /// if a VR256 register is used, but some AVX instructions also have this
457 /// field marked when using a f256 memory references.
460 // VEX_LIG - Specifies that this instruction ignores the L-bit in the VEX
461 // prefix. Usually used for scalar instructions. Needed by disassembler.
464 /// Has3DNow0F0FOpcode - This flag indicates that the instruction uses the
465 /// wacky 0x0F 0x0F prefix for 3DNow! instructions. The manual documents
466 /// this as having a 0x0F prefix with a 0x0F opcode, and each instruction
467 /// storing a classifier in the imm8 field. To simplify our implementation,
468 /// we handle this by storeing the classifier in the opcode field and using
469 /// this flag to indicate that the encoder should do the wacky 3DNow! thing.
470 Has3DNow0F0FOpcode = 1U << 7,
472 /// MemOp4 - Used to indicate swapping of operand 3 and 4 to be encoded in
473 /// ModRM or I8IMM. This is used for FMA4 and XOP instructions.
476 /// XOP - Opcode prefix used by XOP instructions.
481 // getBaseOpcodeFor - This function returns the "base" X86 opcode for the
482 // specified machine instruction.
484 inline unsigned char getBaseOpcodeFor(uint64_t TSFlags) {
485 return TSFlags >> X86II::OpcodeShift;
488 inline bool hasImm(uint64_t TSFlags) {
489 return (TSFlags & X86II::ImmMask) != 0;
492 /// getSizeOfImm - Decode the "size of immediate" field from the TSFlags field
493 /// of the specified instruction.
494 inline unsigned getSizeOfImm(uint64_t TSFlags) {
495 switch (TSFlags & X86II::ImmMask) {
496 default: llvm_unreachable("Unknown immediate size");
498 case X86II::Imm8PCRel: return 1;
500 case X86II::Imm16PCRel: return 2;
502 case X86II::Imm32PCRel: return 4;
503 case X86II::Imm64: return 8;
507 /// isImmPCRel - Return true if the immediate of the specified instruction's
508 /// TSFlags indicates that it is pc relative.
509 inline unsigned isImmPCRel(uint64_t TSFlags) {
510 switch (TSFlags & X86II::ImmMask) {
511 default: llvm_unreachable("Unknown immediate size");
512 case X86II::Imm8PCRel:
513 case X86II::Imm16PCRel:
514 case X86II::Imm32PCRel:
524 /// getMemoryOperandNo - The function returns the MCInst operand # for the
525 /// first field of the memory operand. If the instruction doesn't have a
526 /// memory operand, this returns -1.
528 /// Note that this ignores tied operands. If there is a tied register which
529 /// is duplicated in the MCInst (e.g. "EAX = addl EAX, [mem]") it is only
530 /// counted as one operand.
532 inline int getMemoryOperandNo(uint64_t TSFlags, unsigned Opcode) {
533 switch (TSFlags & X86II::FormMask) {
534 case X86II::MRMInitReg:
535 // FIXME: Remove this form.
537 default: llvm_unreachable("Unknown FormMask value in getMemoryOperandNo!");
540 case X86II::AddRegFrm:
541 case X86II::MRMDestReg:
542 case X86II::MRMSrcReg:
543 case X86II::RawFrmImm8:
544 case X86II::RawFrmImm16:
546 case X86II::MRMDestMem:
548 case X86II::MRMSrcMem: {
549 bool HasVEX_4V = (TSFlags >> X86II::VEXShift) & X86II::VEX_4V;
550 bool HasMemOp4 = (TSFlags >> X86II::VEXShift) & X86II::MemOp4;
551 unsigned FirstMemOp = 1;
553 ++FirstMemOp;// Skip the register source (which is encoded in VEX_VVVV).
555 ++FirstMemOp;// Skip the register source (which is encoded in I8IMM).
557 // FIXME: Maybe lea should have its own form? This is a horrible hack.
558 //if (Opcode == X86::LEA64r || Opcode == X86::LEA64_32r ||
559 // Opcode == X86::LEA16r || Opcode == X86::LEA32r)
562 case X86II::MRM0r: case X86II::MRM1r:
563 case X86II::MRM2r: case X86II::MRM3r:
564 case X86II::MRM4r: case X86II::MRM5r:
565 case X86II::MRM6r: case X86II::MRM7r:
567 case X86II::MRM0m: case X86II::MRM1m:
568 case X86II::MRM2m: case X86II::MRM3m:
569 case X86II::MRM4m: case X86II::MRM5m:
570 case X86II::MRM6m: case X86II::MRM7m: {
571 bool HasVEX_4V = (TSFlags >> X86II::VEXShift) & X86II::VEX_4V;
572 unsigned FirstMemOp = 0;
574 ++FirstMemOp;// Skip the register dest (which is encoded in VEX_VVVV).
577 case X86II::MRM_C1: case X86II::MRM_C2:
578 case X86II::MRM_C3: case X86II::MRM_C4:
579 case X86II::MRM_C8: case X86II::MRM_C9:
580 case X86II::MRM_E8: case X86II::MRM_F0:
581 case X86II::MRM_F8: case X86II::MRM_F9:
582 case X86II::MRM_D0: case X86II::MRM_D1:
583 case X86II::MRM_D4: case X86II::MRM_D5:
584 case X86II::MRM_D8: case X86II::MRM_D9:
585 case X86II::MRM_DA: case X86II::MRM_DB:
586 case X86II::MRM_DC: case X86II::MRM_DD:
587 case X86II::MRM_DE: case X86II::MRM_DF:
592 /// isX86_64ExtendedReg - Is the MachineOperand a x86-64 extended (r8 or
593 /// higher) register? e.g. r8, xmm8, xmm13, etc.
594 inline bool isX86_64ExtendedReg(unsigned RegNo) {
597 case X86::R8: case X86::R9: case X86::R10: case X86::R11:
598 case X86::R12: case X86::R13: case X86::R14: case X86::R15:
599 case X86::R8D: case X86::R9D: case X86::R10D: case X86::R11D:
600 case X86::R12D: case X86::R13D: case X86::R14D: case X86::R15D:
601 case X86::R8W: case X86::R9W: case X86::R10W: case X86::R11W:
602 case X86::R12W: case X86::R13W: case X86::R14W: case X86::R15W:
603 case X86::R8B: case X86::R9B: case X86::R10B: case X86::R11B:
604 case X86::R12B: case X86::R13B: case X86::R14B: case X86::R15B:
605 case X86::XMM8: case X86::XMM9: case X86::XMM10: case X86::XMM11:
606 case X86::XMM12: case X86::XMM13: case X86::XMM14: case X86::XMM15:
607 case X86::YMM8: case X86::YMM9: case X86::YMM10: case X86::YMM11:
608 case X86::YMM12: case X86::YMM13: case X86::YMM14: case X86::YMM15:
609 case X86::CR8: case X86::CR9: case X86::CR10: case X86::CR11:
610 case X86::CR12: case X86::CR13: case X86::CR14: case X86::CR15:
616 inline bool isX86_64NonExtLowByteReg(unsigned reg) {
617 return (reg == X86::SPL || reg == X86::BPL ||
618 reg == X86::SIL || reg == X86::DIL);
622 } // end namespace llvm;