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
99 /// See 'ELF Handling for Thread-Local Storage' for more details.
100 /// SYMBOL_LABEL @TLSGD
103 /// MO_TLSLD - On a symbol operand this indicates that the immediate is
104 /// the offset of the GOT entry with the TLS index for the module that
105 /// contains the symbol. When this index is passed to a call to to
106 /// __tls_get_addr, the function will return the base address of the TLS
107 /// block for the symbol.
109 /// See 'ELF Handling for Thread-Local Storage' for more details.
110 /// SYMBOL_LABEL @TLSLD
113 /// MO_TLSLDM - On a symbol operand this indicates that the immediate is
114 /// the offset of the GOT entry with the TLS index for the module that
115 /// contains the symbol. When this index is passed to a call to to
116 /// ___tls_get_addr, the function will return the base address of the TLS
117 /// block for the symbol.
119 /// See 'ELF Handling for Thread-Local Storage' for more details.
120 /// SYMBOL_LABEL @TLSLDM
123 /// MO_GOTTPOFF - On a symbol operand this indicates that the immediate is
126 /// See 'ELF Handling for Thread-Local Storage' for more details.
127 /// SYMBOL_LABEL @GOTTPOFF
130 /// MO_INDNTPOFF - On a symbol operand this indicates that the immediate is
133 /// See 'ELF Handling for Thread-Local Storage' for more details.
134 /// SYMBOL_LABEL @INDNTPOFF
137 /// MO_TPOFF - On a symbol operand this indicates that the immediate is
140 /// See 'ELF Handling for Thread-Local Storage' for more details.
141 /// SYMBOL_LABEL @TPOFF
144 /// MO_DTPOFF - On a symbol operand this indicates that the immediate is
145 /// the offset of the GOT entry with the TLS offset of the symbol.
147 /// See 'ELF Handling for Thread-Local Storage' for more details.
148 /// SYMBOL_LABEL @DTPOFF
151 /// MO_NTPOFF - On a symbol operand this indicates that the immediate is
154 /// See 'ELF Handling for Thread-Local Storage' for more details.
155 /// SYMBOL_LABEL @NTPOFF
158 /// MO_GOTNTPOFF - On a symbol operand this indicates that the immediate is
161 /// See 'ELF Handling for Thread-Local Storage' for more details.
162 /// SYMBOL_LABEL @GOTNTPOFF
165 /// MO_DLLIMPORT - On a symbol operand "FOO", this indicates that the
166 /// reference is actually to the "__imp_FOO" symbol. This is used for
167 /// dllimport linkage on windows.
170 /// MO_DARWIN_STUB - On a symbol operand "FOO", this indicates that the
171 /// reference is actually to the "FOO$stub" symbol. This is used for calls
172 /// and jumps to external functions on Tiger and earlier.
175 /// MO_DARWIN_NONLAZY - On a symbol operand "FOO", this indicates that the
176 /// reference is actually to the "FOO$non_lazy_ptr" symbol, which is a
177 /// non-PIC-base-relative reference to a non-hidden dyld lazy pointer stub.
180 /// MO_DARWIN_NONLAZY_PIC_BASE - On a symbol operand "FOO", this indicates
181 /// that the reference is actually to "FOO$non_lazy_ptr - PICBASE", which is
182 /// a PIC-base-relative reference to a non-hidden dyld lazy pointer stub.
183 MO_DARWIN_NONLAZY_PIC_BASE,
185 /// MO_DARWIN_HIDDEN_NONLAZY_PIC_BASE - On a symbol operand "FOO", this
186 /// indicates that the reference is actually to "FOO$non_lazy_ptr -PICBASE",
187 /// which is a PIC-base-relative reference to a hidden dyld lazy pointer
189 MO_DARWIN_HIDDEN_NONLAZY_PIC_BASE,
191 /// MO_TLVP - On a symbol operand this indicates that the immediate is
194 /// This is the TLS offset for the Darwin TLS mechanism.
197 /// MO_TLVP_PIC_BASE - On a symbol operand this indicates that the immediate
198 /// is some TLS offset from the picbase.
200 /// This is the 32-bit TLS offset for Darwin TLS in PIC mode.
203 /// MO_SECREL - On a symbol operand this indicates that the immediate is
204 /// the offset from beginning of section.
206 /// This is the TLS offset for the COFF/Windows TLS mechanism.
211 //===------------------------------------------------------------------===//
212 // Instruction encodings. These are the standard/most common forms for X86
216 // PseudoFrm - This represents an instruction that is a pseudo instruction
217 // or one that has not been implemented yet. It is illegal to code generate
218 // it, but tolerated for intermediate implementation stages.
221 /// Raw - This form is for instructions that don't have any operands, so
222 /// they are just a fixed opcode value, like 'leave'.
225 /// AddRegFrm - This form is used for instructions like 'push r32' that have
226 /// their one register operand added to their opcode.
229 /// MRMDestReg - This form is used for instructions that use the Mod/RM byte
230 /// to specify a destination, which in this case is a register.
234 /// MRMDestMem - This form is used for instructions that use the Mod/RM byte
235 /// to specify a destination, which in this case is memory.
239 /// MRMSrcReg - This form is used for instructions that use the Mod/RM byte
240 /// to specify a source, which in this case is a register.
244 /// MRMSrcMem - This form is used for instructions that use the Mod/RM byte
245 /// to specify a source, which in this case is memory.
249 /// MRM[0-7][rm] - These forms are used to represent instructions that use
250 /// a Mod/RM byte, and use the middle field to hold extended opcode
251 /// information. In the intel manual these are represented as /0, /1, ...
254 // First, instructions that operate on a register r/m operand...
255 MRM0r = 16, MRM1r = 17, MRM2r = 18, MRM3r = 19, // Format /0 /1 /2 /3
256 MRM4r = 20, MRM5r = 21, MRM6r = 22, MRM7r = 23, // Format /4 /5 /6 /7
258 // Next, instructions that operate on a memory r/m operand...
259 MRM0m = 24, MRM1m = 25, MRM2m = 26, MRM3m = 27, // Format /0 /1 /2 /3
260 MRM4m = 28, MRM5m = 29, MRM6m = 30, MRM7m = 31, // Format /4 /5 /6 /7
262 // MRMInitReg - This form is used for instructions whose source and
263 // destinations are the same register.
266 //// MRM_XX - A mod/rm byte of exactly 0xXX.
267 MRM_C1 = 33, MRM_C2 = 34, MRM_C3 = 35, MRM_C4 = 36,
268 MRM_C8 = 37, MRM_C9 = 38, MRM_E8 = 39, MRM_F0 = 40,
269 MRM_F8 = 41, MRM_F9 = 42, MRM_D0 = 45, MRM_D1 = 46,
270 MRM_D4 = 47, MRM_D8 = 48, MRM_D9 = 49, MRM_DA = 50,
271 MRM_DB = 51, MRM_DC = 52, MRM_DD = 53, MRM_DE = 54,
274 /// RawFrmImm8 - This is used for the ENTER instruction, which has two
275 /// immediates, the first of which is a 16-bit immediate (specified by
276 /// the imm encoding) and the second is a 8-bit fixed value.
279 /// RawFrmImm16 - This is used for CALL FAR instructions, which have two
280 /// immediates, the first of which is a 16 or 32-bit immediate (specified by
281 /// the imm encoding) and the second is a 16-bit fixed value. In the AMD
282 /// manual, this operand is described as pntr16:32 and pntr16:16
287 //===------------------------------------------------------------------===//
290 // OpSize - Set if this instruction requires an operand size prefix (0x66),
291 // which most often indicates that the instruction operates on 16 bit data
292 // instead of 32 bit data.
295 // AsSize - Set if this instruction requires an operand size prefix (0x67),
296 // which most often indicates that the instruction address 16 bit address
297 // instead of 32 bit address (or 32 bit address in 64 bit mode).
300 //===------------------------------------------------------------------===//
301 // Op0Mask - There are several prefix bytes that are used to form two byte
302 // opcodes. These are currently 0x0F, 0xF3, and 0xD8-0xDF. This mask is
303 // used to obtain the setting of this field. If no bits in this field is
304 // set, there is no prefix byte for obtaining a multibyte opcode.
307 Op0Mask = 0x1F << Op0Shift,
309 // TB - TwoByte - Set if this instruction has a two byte opcode, which
310 // starts with a 0x0F byte before the real opcode.
313 // REP - The 0xF3 prefix byte indicating repetition of the following
317 // D8-DF - These escape opcodes are used by the floating point unit. These
318 // values must remain sequential.
319 D8 = 3 << Op0Shift, D9 = 4 << Op0Shift,
320 DA = 5 << Op0Shift, DB = 6 << Op0Shift,
321 DC = 7 << Op0Shift, DD = 8 << Op0Shift,
322 DE = 9 << Op0Shift, DF = 10 << Op0Shift,
324 // XS, XD - These prefix codes are for single and double precision scalar
325 // floating point operations performed in the SSE registers.
326 XD = 11 << Op0Shift, XS = 12 << Op0Shift,
328 // T8, TA, A6, A7 - Prefix after the 0x0F prefix.
329 T8 = 13 << Op0Shift, TA = 14 << Op0Shift,
330 A6 = 15 << Op0Shift, A7 = 16 << Op0Shift,
332 // T8XD - Prefix before and after 0x0F. Combination of T8 and XD.
333 T8XD = 17 << Op0Shift,
335 // T8XS - Prefix before and after 0x0F. Combination of T8 and XS.
336 T8XS = 18 << Op0Shift,
338 // TAXD - Prefix before and after 0x0F. Combination of TA and XD.
339 TAXD = 19 << Op0Shift,
341 // XOP8 - Prefix to include use of imm byte.
342 XOP8 = 20 << Op0Shift,
344 // XOP9 - Prefix to exclude use of imm byte.
345 XOP9 = 21 << Op0Shift,
347 //===------------------------------------------------------------------===//
348 // REX_W - REX prefixes are instruction prefixes used in 64-bit mode.
349 // They are used to specify GPRs and SSE registers, 64-bit operand size,
350 // etc. We only cares about REX.W and REX.R bits and only the former is
351 // statically determined.
353 REXShift = Op0Shift + 5,
354 REX_W = 1 << REXShift,
356 //===------------------------------------------------------------------===//
357 // This three-bit field describes the size of an immediate operand. Zero is
358 // unused so that we can tell if we forgot to set a value.
359 ImmShift = REXShift + 1,
360 ImmMask = 7 << ImmShift,
361 Imm8 = 1 << ImmShift,
362 Imm8PCRel = 2 << ImmShift,
363 Imm16 = 3 << ImmShift,
364 Imm16PCRel = 4 << ImmShift,
365 Imm32 = 5 << ImmShift,
366 Imm32PCRel = 6 << ImmShift,
367 Imm64 = 7 << ImmShift,
369 //===------------------------------------------------------------------===//
370 // FP Instruction Classification... Zero is non-fp instruction.
372 // FPTypeMask - Mask for all of the FP types...
373 FPTypeShift = ImmShift + 3,
374 FPTypeMask = 7 << FPTypeShift,
376 // NotFP - The default, set for instructions that do not use FP registers.
377 NotFP = 0 << FPTypeShift,
379 // ZeroArgFP - 0 arg FP instruction which implicitly pushes ST(0), f.e. fld0
380 ZeroArgFP = 1 << FPTypeShift,
382 // OneArgFP - 1 arg FP instructions which implicitly read ST(0), such as fst
383 OneArgFP = 2 << FPTypeShift,
385 // OneArgFPRW - 1 arg FP instruction which implicitly read ST(0) and write a
386 // result back to ST(0). For example, fcos, fsqrt, etc.
388 OneArgFPRW = 3 << FPTypeShift,
390 // TwoArgFP - 2 arg FP instructions which implicitly read ST(0), and an
391 // explicit argument, storing the result to either ST(0) or the implicit
392 // argument. For example: fadd, fsub, fmul, etc...
393 TwoArgFP = 4 << FPTypeShift,
395 // CompareFP - 2 arg FP instructions which implicitly read ST(0) and an
396 // explicit argument, but have no destination. Example: fucom, fucomi, ...
397 CompareFP = 5 << FPTypeShift,
399 // CondMovFP - "2 operand" floating point conditional move instructions.
400 CondMovFP = 6 << FPTypeShift,
402 // SpecialFP - Special instruction forms. Dispatch by opcode explicitly.
403 SpecialFP = 7 << FPTypeShift,
406 LOCKShift = FPTypeShift + 3,
407 LOCK = 1 << LOCKShift,
409 // Segment override prefixes. Currently we just need ability to address
410 // stuff in gs and fs segments.
411 SegOvrShift = LOCKShift + 1,
412 SegOvrMask = 3 << SegOvrShift,
413 FS = 1 << SegOvrShift,
414 GS = 2 << SegOvrShift,
416 // Execution domain for SSE instructions in bits 23, 24.
417 // 0 in bits 23-24 means normal, non-SSE instruction.
418 SSEDomainShift = SegOvrShift + 2,
420 OpcodeShift = SSEDomainShift + 2,
422 //===------------------------------------------------------------------===//
423 /// VEX - The opcode prefix used by AVX instructions
424 VEXShift = OpcodeShift + 8,
427 /// VEX_W - Has a opcode specific functionality, but is used in the same
428 /// way as REX_W is for regular SSE instructions.
431 /// VEX_4V - Used to specify an additional AVX/SSE register. Several 2
432 /// address instructions in SSE are represented as 3 address ones in AVX
433 /// and the additional register is encoded in VEX_VVVV prefix.
436 /// VEX_4VOp3 - Similar to VEX_4V, but used on instructions that encode
437 /// operand 3 with VEX.vvvv.
440 /// VEX_I8IMM - Specifies that the last register used in a AVX instruction,
441 /// must be encoded in the i8 immediate field. This usually happens in
442 /// instructions with 4 operands.
445 /// VEX_L - Stands for a bit in the VEX opcode prefix meaning the current
446 /// instruction uses 256-bit wide registers. This is usually auto detected
447 /// if a VR256 register is used, but some AVX instructions also have this
448 /// field marked when using a f256 memory references.
451 // VEX_LIG - Specifies that this instruction ignores the L-bit in the VEX
452 // prefix. Usually used for scalar instructions. Needed by disassembler.
455 /// Has3DNow0F0FOpcode - This flag indicates that the instruction uses the
456 /// wacky 0x0F 0x0F prefix for 3DNow! instructions. The manual documents
457 /// this as having a 0x0F prefix with a 0x0F opcode, and each instruction
458 /// storing a classifier in the imm8 field. To simplify our implementation,
459 /// we handle this by storeing the classifier in the opcode field and using
460 /// this flag to indicate that the encoder should do the wacky 3DNow! thing.
461 Has3DNow0F0FOpcode = 1U << 7,
463 /// MemOp4 - Used to indicate swapping of operand 3 and 4 to be encoded in
464 /// ModRM or I8IMM. This is used for FMA4 and XOP instructions.
467 /// XOP - Opcode prefix used by XOP instructions.
472 // getBaseOpcodeFor - This function returns the "base" X86 opcode for the
473 // specified machine instruction.
475 static inline unsigned char getBaseOpcodeFor(uint64_t TSFlags) {
476 return TSFlags >> X86II::OpcodeShift;
479 static inline bool hasImm(uint64_t TSFlags) {
480 return (TSFlags & X86II::ImmMask) != 0;
483 /// getSizeOfImm - Decode the "size of immediate" field from the TSFlags field
484 /// of the specified instruction.
485 static inline unsigned getSizeOfImm(uint64_t TSFlags) {
486 switch (TSFlags & X86II::ImmMask) {
487 default: llvm_unreachable("Unknown immediate size");
489 case X86II::Imm8PCRel: return 1;
491 case X86II::Imm16PCRel: return 2;
493 case X86II::Imm32PCRel: return 4;
494 case X86II::Imm64: return 8;
498 /// isImmPCRel - Return true if the immediate of the specified instruction's
499 /// TSFlags indicates that it is pc relative.
500 static inline unsigned isImmPCRel(uint64_t TSFlags) {
501 switch (TSFlags & X86II::ImmMask) {
502 default: llvm_unreachable("Unknown immediate size");
503 case X86II::Imm8PCRel:
504 case X86II::Imm16PCRel:
505 case X86II::Imm32PCRel:
515 /// getMemoryOperandNo - The function returns the MCInst operand # for the
516 /// first field of the memory operand. If the instruction doesn't have a
517 /// memory operand, this returns -1.
519 /// Note that this ignores tied operands. If there is a tied register which
520 /// is duplicated in the MCInst (e.g. "EAX = addl EAX, [mem]") it is only
521 /// counted as one operand.
523 static inline int getMemoryOperandNo(uint64_t TSFlags, unsigned Opcode) {
524 switch (TSFlags & X86II::FormMask) {
525 case X86II::MRMInitReg:
526 // FIXME: Remove this form.
528 default: llvm_unreachable("Unknown FormMask value in getMemoryOperandNo!");
531 case X86II::AddRegFrm:
532 case X86II::MRMDestReg:
533 case X86II::MRMSrcReg:
534 case X86II::RawFrmImm8:
535 case X86II::RawFrmImm16:
537 case X86II::MRMDestMem:
539 case X86II::MRMSrcMem: {
540 bool HasVEX_4V = (TSFlags >> X86II::VEXShift) & X86II::VEX_4V;
541 bool HasMemOp4 = (TSFlags >> X86II::VEXShift) & X86II::MemOp4;
542 unsigned FirstMemOp = 1;
544 ++FirstMemOp;// Skip the register source (which is encoded in VEX_VVVV).
546 ++FirstMemOp;// Skip the register source (which is encoded in I8IMM).
548 // FIXME: Maybe lea should have its own form? This is a horrible hack.
549 //if (Opcode == X86::LEA64r || Opcode == X86::LEA64_32r ||
550 // Opcode == X86::LEA16r || Opcode == X86::LEA32r)
553 case X86II::MRM0r: case X86II::MRM1r:
554 case X86II::MRM2r: case X86II::MRM3r:
555 case X86II::MRM4r: case X86II::MRM5r:
556 case X86II::MRM6r: case X86II::MRM7r:
558 case X86II::MRM0m: case X86II::MRM1m:
559 case X86II::MRM2m: case X86II::MRM3m:
560 case X86II::MRM4m: case X86II::MRM5m:
561 case X86II::MRM6m: case X86II::MRM7m: {
562 bool HasVEX_4V = (TSFlags >> X86II::VEXShift) & X86II::VEX_4V;
563 unsigned FirstMemOp = 0;
565 ++FirstMemOp;// Skip the register dest (which is encoded in VEX_VVVV).
568 case X86II::MRM_C1: case X86II::MRM_C2:
569 case X86II::MRM_C3: case X86II::MRM_C4:
570 case X86II::MRM_C8: case X86II::MRM_C9:
571 case X86II::MRM_E8: case X86II::MRM_F0:
572 case X86II::MRM_F8: case X86II::MRM_F9:
573 case X86II::MRM_D0: case X86II::MRM_D1:
574 case X86II::MRM_D4: case X86II::MRM_D8:
575 case X86II::MRM_D9: case X86II::MRM_DA:
576 case X86II::MRM_DB: case X86II::MRM_DC:
577 case X86II::MRM_DD: case X86II::MRM_DE:
583 /// isX86_64ExtendedReg - Is the MachineOperand a x86-64 extended (r8 or
584 /// higher) register? e.g. r8, xmm8, xmm13, etc.
585 static inline bool isX86_64ExtendedReg(unsigned RegNo) {
588 case X86::R8: case X86::R9: case X86::R10: case X86::R11:
589 case X86::R12: case X86::R13: case X86::R14: case X86::R15:
590 case X86::R8D: case X86::R9D: case X86::R10D: case X86::R11D:
591 case X86::R12D: case X86::R13D: case X86::R14D: case X86::R15D:
592 case X86::R8W: case X86::R9W: case X86::R10W: case X86::R11W:
593 case X86::R12W: case X86::R13W: case X86::R14W: case X86::R15W:
594 case X86::R8B: case X86::R9B: case X86::R10B: case X86::R11B:
595 case X86::R12B: case X86::R13B: case X86::R14B: case X86::R15B:
596 case X86::XMM8: case X86::XMM9: case X86::XMM10: case X86::XMM11:
597 case X86::XMM12: case X86::XMM13: case X86::XMM14: case X86::XMM15:
598 case X86::YMM8: case X86::YMM9: case X86::YMM10: case X86::YMM11:
599 case X86::YMM12: case X86::YMM13: case X86::YMM14: case X86::YMM15:
600 case X86::CR8: case X86::CR9: case X86::CR10: case X86::CR11:
601 case X86::CR12: case X86::CR13: case X86::CR14: case X86::CR15:
607 static inline bool isX86_64NonExtLowByteReg(unsigned reg) {
608 return (reg == X86::SPL || reg == X86::BPL ||
609 reg == X86::SIL || reg == X86::DIL);
613 } // end namespace llvm;