blob: b690de2d3fb2d482129253ca6aefc5dbfd93ad35 [file] [log] [blame]
Evan Cheng7e763d82011-07-25 18:43:53 +00001//===-- 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"
Chandler Carruth8a8cd2b2014-01-07 11:48:04 +000021#include "llvm/MC/MCInstrInfo.h"
Evan Cheng7e763d82011-07-25 18:43:53 +000022#include "llvm/Support/DataTypes.h"
Craig Topper4ed72782012-02-05 05:38:58 +000023#include "llvm/Support/ErrorHandling.h"
Evan Cheng7e763d82011-07-25 18:43:53 +000024
25namespace llvm {
26
27namespace 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;
Evan Cheng7e763d82011-07-25 18:43:53 +000045
46/// X86II - This namespace holds all of the target specific flags that
47/// instruction info tracks.
48///
49namespace 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
Hans Wennborg09610f32012-06-04 09:55:36 +000097 /// 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.
Evan Cheng7e763d82011-07-25 18:43:53 +0000100 ///
101 /// See 'ELF Handling for Thread-Local Storage' for more details.
102 /// SYMBOL_LABEL @TLSGD
103 MO_TLSGD,
104
Hans Wennborg789acfb2012-06-01 16:27:21 +0000105 /// 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
Hans Wennborg5deecd92013-01-29 14:05:57 +0000107 /// contains the symbol. When this index is passed to a call to
Hans Wennborg789acfb2012-06-01 16:27:21 +0000108 /// __tls_get_addr, the function will return the base address of the TLS
Hans Wennborg09610f32012-06-04 09:55:36 +0000109 /// block for the symbol. Used in the x86-64 local dynamic TLS access model.
Hans Wennborg789acfb2012-06-01 16:27:21 +0000110 ///
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
Hans Wennborg5deecd92013-01-29 14:05:57 +0000117 /// contains the symbol. When this index is passed to a call to
Hans Wennborg789acfb2012-06-01 16:27:21 +0000118 /// ___tls_get_addr, the function will return the base address of the TLS
Hans Wennborg09610f32012-06-04 09:55:36 +0000119 /// block for the symbol. Used in the IA32 local dynamic TLS access model.
Hans Wennborg789acfb2012-06-01 16:27:21 +0000120 ///
121 /// See 'ELF Handling for Thread-Local Storage' for more details.
122 /// SYMBOL_LABEL @TLSLDM
123 MO_TLSLDM,
124
Evan Cheng7e763d82011-07-25 18:43:53 +0000125 /// MO_GOTTPOFF - On a symbol operand this indicates that the immediate is
Hans Wennborg09610f32012-06-04 09:55:36 +0000126 /// 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.
Evan Cheng7e763d82011-07-25 18:43:53 +0000128 ///
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
Hans Wennborg09610f32012-06-04 09:55:36 +0000134 /// 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.
Evan Cheng7e763d82011-07-25 18:43:53 +0000137 ///
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
Hans Wennborg09610f32012-06-04 09:55:36 +0000143 /// the thread-pointer offset for the symbol. Used in the x86-64 local
144 /// exec TLS access model.
Evan Cheng7e763d82011-07-25 18:43:53 +0000145 ///
146 /// See 'ELF Handling for Thread-Local Storage' for more details.
147 /// SYMBOL_LABEL @TPOFF
148 MO_TPOFF,
149
Hans Wennborg789acfb2012-06-01 16:27:21 +0000150 /// MO_DTPOFF - On a symbol operand this indicates that the immediate is
Hans Wennborg09610f32012-06-04 09:55:36 +0000151 /// the offset of the GOT entry with the TLS offset of the symbol. Used
152 /// in the local dynamic TLS access model.
Hans Wennborg789acfb2012-06-01 16:27:21 +0000153 ///
154 /// See 'ELF Handling for Thread-Local Storage' for more details.
155 /// SYMBOL_LABEL @DTPOFF
156 MO_DTPOFF,
157
Evan Cheng7e763d82011-07-25 18:43:53 +0000158 /// MO_NTPOFF - On a symbol operand this indicates that the immediate is
Hans Wennborg09610f32012-06-04 09:55:36 +0000159 /// the negative thread-pointer offset for the symbol. Used in the IA32
160 /// local exec TLS access model.
Evan Cheng7e763d82011-07-25 18:43:53 +0000161 ///
162 /// See 'ELF Handling for Thread-Local Storage' for more details.
163 /// SYMBOL_LABEL @NTPOFF
164 MO_NTPOFF,
165
Hans Wennborgf9d0e442012-05-11 10:11:01 +0000166 /// MO_GOTNTPOFF - On a symbol operand this indicates that the immediate is
Hans Wennborg09610f32012-06-04 09:55:36 +0000167 /// 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.
Hans Wennborgf9d0e442012-05-11 10:11:01 +0000169 ///
170 /// See 'ELF Handling for Thread-Local Storage' for more details.
171 /// SYMBOL_LABEL @GOTNTPOFF
172 MO_GOTNTPOFF,
173
Evan Cheng7e763d82011-07-25 18:43:53 +0000174 /// 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.
Anton Korobeynikovc6b40172012-02-11 17:26:53 +0000210 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
Evan Cheng7e763d82011-07-25 18:43:53 +0000217 };
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
Craig Topper35da3d12014-01-16 07:36:58 +0000258 /// 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
David Woodhouse2ef8d9c2014-01-22 15:08:08 +0000262 /// 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
David Woodhouseb33c2ef2014-01-22 15:08:21 +0000266 /// RawFrmDst - This form is for instructions that use the destination index
267 /// register DI/EDI/ESI.
268 RawFrmDst = 9,
269
David Woodhouse9bbf7ca2014-01-22 15:08:36 +0000270 /// 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
Evan Cheng7e763d82011-07-25 18:43:53 +0000275 /// MRM[0-7][rm] - These forms are used to represent instructions that use
276 /// a Mod/RM byte, and use the middle field to hold extended opcode
277 /// information. In the intel manual these are represented as /0, /1, ...
278 ///
279
280 // First, instructions that operate on a register r/m operand...
281 MRM0r = 16, MRM1r = 17, MRM2r = 18, MRM3r = 19, // Format /0 /1 /2 /3
282 MRM4r = 20, MRM5r = 21, MRM6r = 22, MRM7r = 23, // Format /4 /5 /6 /7
283
284 // Next, instructions that operate on a memory r/m operand...
285 MRM0m = 24, MRM1m = 25, MRM2m = 26, MRM3m = 27, // Format /0 /1 /2 /3
286 MRM4m = 28, MRM5m = 29, MRM6m = 30, MRM7m = 31, // Format /4 /5 /6 /7
287
Craig Toppered7aa462012-02-18 08:19:49 +0000288 //// MRM_XX - A mod/rm byte of exactly 0xXX.
289 MRM_C1 = 33, MRM_C2 = 34, MRM_C3 = 35, MRM_C4 = 36,
Michael Liao95d944032013-04-11 04:52:28 +0000290 MRM_C8 = 37, MRM_C9 = 38, MRM_CA = 39, MRM_CB = 40,
291 MRM_E8 = 41, MRM_F0 = 42, MRM_F8 = 45, MRM_F9 = 46,
292 MRM_D0 = 47, MRM_D1 = 48, MRM_D4 = 49, MRM_D5 = 50,
293 MRM_D6 = 51, MRM_D8 = 52, MRM_D9 = 53, MRM_DA = 54,
294 MRM_DB = 55, MRM_DC = 56, MRM_DD = 57, MRM_DE = 58,
295 MRM_DF = 59,
Evan Cheng7e763d82011-07-25 18:43:53 +0000296
297 /// RawFrmImm8 - This is used for the ENTER instruction, which has two
298 /// immediates, the first of which is a 16-bit immediate (specified by
299 /// the imm encoding) and the second is a 8-bit fixed value.
300 RawFrmImm8 = 43,
301
302 /// RawFrmImm16 - This is used for CALL FAR instructions, which have two
303 /// immediates, the first of which is a 16 or 32-bit immediate (specified by
304 /// the imm encoding) and the second is a 16-bit fixed value. In the AMD
305 /// manual, this operand is described as pntr16:32 and pntr16:16
306 RawFrmImm16 = 44,
307
308 FormMask = 63,
309
310 //===------------------------------------------------------------------===//
311 // Actual flags...
312
313 // OpSize - Set if this instruction requires an operand size prefix (0x66),
314 // which most often indicates that the instruction operates on 16 bit data
Craig Topper7ceb54a2014-01-06 06:02:58 +0000315 // instead of 32 bit data. OpSize16 in 16 bit mode indicates that the
316 // instruction operates on 32 bit data instead of 16 bit data.
Evan Cheng7e763d82011-07-25 18:43:53 +0000317 OpSize = 1 << 6,
Craig Topper7ceb54a2014-01-06 06:02:58 +0000318 OpSize16 = 1 << 7,
Evan Cheng7e763d82011-07-25 18:43:53 +0000319
320 // AsSize - Set if this instruction requires an operand size prefix (0x67),
321 // which most often indicates that the instruction address 16 bit address
322 // instead of 32 bit address (or 32 bit address in 64 bit mode).
Craig Topper7ceb54a2014-01-06 06:02:58 +0000323 AdSize = 1 << 8,
Evan Cheng7e763d82011-07-25 18:43:53 +0000324
325 //===------------------------------------------------------------------===//
326 // Op0Mask - There are several prefix bytes that are used to form two byte
Craig Topperec688662014-01-31 07:00:55 +0000327 // opcodes. This mask is used to obtain the setting of this field. If no
328 // bits in this field is set, there is no prefix byte for obtaining a
329 // multibyte opcode.
Evan Cheng7e763d82011-07-25 18:43:53 +0000330 //
Craig Topper7ceb54a2014-01-06 06:02:58 +0000331 Op0Shift = 9,
Evan Cheng7e763d82011-07-25 18:43:53 +0000332 Op0Mask = 0x1F << Op0Shift,
333
334 // TB - TwoByte - Set if this instruction has a two byte opcode, which
335 // starts with a 0x0F byte before the real opcode.
336 TB = 1 << Op0Shift,
337
Evan Cheng7e763d82011-07-25 18:43:53 +0000338 // D8-DF - These escape opcodes are used by the floating point unit. These
339 // values must remain sequential.
340 D8 = 3 << Op0Shift, D9 = 4 << Op0Shift,
341 DA = 5 << Op0Shift, DB = 6 << Op0Shift,
342 DC = 7 << Op0Shift, DD = 8 << Op0Shift,
343 DE = 9 << Op0Shift, DF = 10 << Op0Shift,
344
345 // XS, XD - These prefix codes are for single and double precision scalar
346 // floating point operations performed in the SSE registers.
347 XD = 11 << Op0Shift, XS = 12 << Op0Shift,
348
349 // T8, TA, A6, A7 - Prefix after the 0x0F prefix.
350 T8 = 13 << Op0Shift, TA = 14 << Op0Shift,
351 A6 = 15 << Op0Shift, A7 = 16 << Op0Shift,
352
Craig Topper96fa5972011-10-16 16:50:08 +0000353 // T8XD - Prefix before and after 0x0F. Combination of T8 and XD.
354 T8XD = 17 << Op0Shift,
355
356 // T8XS - Prefix before and after 0x0F. Combination of T8 and XS.
357 T8XS = 18 << Op0Shift,
Evan Cheng7e763d82011-07-25 18:43:53 +0000358
Craig Topper980d5982011-10-23 07:34:00 +0000359 // TAXD - Prefix before and after 0x0F. Combination of TA and XD.
360 TAXD = 19 << Op0Shift,
361
Jan Sjödin6dd24882011-12-12 19:12:26 +0000362 // XOP8 - Prefix to include use of imm byte.
363 XOP8 = 20 << Op0Shift,
364
365 // XOP9 - Prefix to exclude use of imm byte.
366 XOP9 = 21 << Op0Shift,
367
Yunzhong Gaob8bbcbf2013-09-27 18:38:42 +0000368 // XOPA - Prefix to encode 0xA in VEX.MMMM of XOP instructions.
369 XOPA = 22 << Op0Shift,
370
Craig Topperae11aed2014-01-14 07:41:20 +0000371 // PD - Prefix code for packed double precision vector floating point
372 // operations performed in the SSE registers.
373 PD = 23 << Op0Shift,
374
375 // T8PD - Prefix before and after 0x0F. Combination of T8 and PD.
376 T8PD = 24 << Op0Shift,
377
378 // TAPD - Prefix before and after 0x0F. Combination of TA and PD.
379 TAPD = 25 << Op0Shift,
380
Evan Cheng7e763d82011-07-25 18:43:53 +0000381 //===------------------------------------------------------------------===//
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 = Op0Shift + 5,
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,
David Woodhouse0b6c9492014-01-30 22:20:41 +0000394 ImmMask = 15 << ImmShift,
Evan Cheng7e763d82011-07-25 18:43:53 +0000395 Imm8 = 1 << ImmShift,
396 Imm8PCRel = 2 << ImmShift,
397 Imm16 = 3 << ImmShift,
398 Imm16PCRel = 4 << ImmShift,
399 Imm32 = 5 << ImmShift,
400 Imm32PCRel = 6 << ImmShift,
David Woodhouse0b6c9492014-01-30 22:20:41 +0000401 Imm32S = 7 << ImmShift,
402 Imm64 = 8 << ImmShift,
Evan Cheng7e763d82011-07-25 18:43:53 +0000403
404 //===------------------------------------------------------------------===//
405 // FP Instruction Classification... Zero is non-fp instruction.
406
407 // FPTypeMask - Mask for all of the FP types...
David Woodhouse0b6c9492014-01-30 22:20:41 +0000408 FPTypeShift = ImmShift + 4,
Evan Cheng7e763d82011-07-25 18:43:53 +0000409 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
Craig Topperec688662014-01-31 07:00:55 +0000444 // 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,
Evan Cheng7e763d82011-07-25 18:43:53 +0000451
452 OpcodeShift = SSEDomainShift + 2,
453
454 //===------------------------------------------------------------------===//
455 /// VEX - The opcode prefix used by AVX instructions
456 VEXShift = OpcodeShift + 8,
457 VEX = 1U << 0,
458
459 /// VEX_W - Has a opcode specific functionality, but is used in the same
460 /// way as REX_W is for regular SSE instructions.
461 VEX_W = 1U << 1,
462
463 /// VEX_4V - Used to specify an additional AVX/SSE register. Several 2
464 /// address instructions in SSE are represented as 3 address ones in AVX
465 /// and the additional register is encoded in VEX_VVVV prefix.
466 VEX_4V = 1U << 2,
467
Craig Topperaea148c2011-10-16 07:55:05 +0000468 /// VEX_4VOp3 - Similar to VEX_4V, but used on instructions that encode
469 /// operand 3 with VEX.vvvv.
470 VEX_4VOp3 = 1U << 3,
471
Evan Cheng7e763d82011-07-25 18:43:53 +0000472 /// VEX_I8IMM - Specifies that the last register used in a AVX instruction,
473 /// must be encoded in the i8 immediate field. This usually happens in
474 /// instructions with 4 operands.
Craig Topperaea148c2011-10-16 07:55:05 +0000475 VEX_I8IMM = 1U << 4,
Evan Cheng7e763d82011-07-25 18:43:53 +0000476
477 /// VEX_L - Stands for a bit in the VEX opcode prefix meaning the current
478 /// instruction uses 256-bit wide registers. This is usually auto detected
479 /// if a VR256 register is used, but some AVX instructions also have this
480 /// field marked when using a f256 memory references.
Craig Topperaea148c2011-10-16 07:55:05 +0000481 VEX_L = 1U << 5,
Evan Cheng7e763d82011-07-25 18:43:53 +0000482
Craig Topperf18c8962011-10-04 06:30:42 +0000483 // VEX_LIG - Specifies that this instruction ignores the L-bit in the VEX
484 // prefix. Usually used for scalar instructions. Needed by disassembler.
Craig Topperaea148c2011-10-16 07:55:05 +0000485 VEX_LIG = 1U << 6,
Craig Topperf18c8962011-10-04 06:30:42 +0000486
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000487 // TODO: we should combine VEX_L and VEX_LIG together to form a 2-bit field
488 // with following encoding:
489 // - 00 V128
490 // - 01 V256
491 // - 10 V512
492 // - 11 LIG (but, in insn encoding, leave VEX.L and EVEX.L in zeros.
493 // this will save 1 tsflag bit
494
495 // VEX_EVEX - Specifies that this instruction use EVEX form which provides
496 // syntax support up to 32 512-bit register operands and up to 7 16-bit
497 // mask operands as well as source operand data swizzling/memory operand
498 // conversion, eviction hint, and rounding mode.
499 EVEX = 1U << 7,
500
501 // EVEX_K - Set if this instruction requires masking
502 EVEX_K = 1U << 8,
503
504 // EVEX_Z - Set if this instruction has EVEX.Z field set.
505 EVEX_Z = 1U << 9,
506
507 // EVEX_L2 - Set if this instruction has EVEX.L' field set.
508 EVEX_L2 = 1U << 10,
509
510 // EVEX_B - Set if this instruction has EVEX.B field set.
511 EVEX_B = 1U << 11,
512
513 // EVEX_CD8E - compressed disp8 form, element-size
514 EVEX_CD8EShift = VEXShift + 12,
515 EVEX_CD8EMask = 3,
516
517 // EVEX_CD8V - compressed disp8 form, vector-width
518 EVEX_CD8VShift = EVEX_CD8EShift + 2,
519 EVEX_CD8VMask = 7,
520
Evan Cheng7e763d82011-07-25 18:43:53 +0000521 /// Has3DNow0F0FOpcode - This flag indicates that the instruction uses the
522 /// wacky 0x0F 0x0F prefix for 3DNow! instructions. The manual documents
523 /// this as having a 0x0F prefix with a 0x0F opcode, and each instruction
524 /// storing a classifier in the imm8 field. To simplify our implementation,
525 /// we handle this by storeing the classifier in the opcode field and using
526 /// this flag to indicate that the encoder should do the wacky 3DNow! thing.
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000527 Has3DNow0F0FOpcode = 1U << 17,
Bruno Cardoso Lopes0f9a1f52011-11-25 19:33:42 +0000528
Craig Toppercd93de92011-12-30 04:48:54 +0000529 /// MemOp4 - Used to indicate swapping of operand 3 and 4 to be encoded in
530 /// ModRM or I8IMM. This is used for FMA4 and XOP instructions.
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000531 MemOp4 = 1U << 18,
Jan Sjödin6dd24882011-12-12 19:12:26 +0000532
533 /// XOP - Opcode prefix used by XOP instructions.
Elena Demikhovskyb19c9dc2014-01-13 12:55:03 +0000534 XOP = 1U << 19,
Jan Sjödin6dd24882011-12-12 19:12:26 +0000535
Elena Demikhovskyb19c9dc2014-01-13 12:55:03 +0000536 /// Explicitly specified rounding control
537 EVEX_RC = 1U << 20
Evan Cheng7e763d82011-07-25 18:43:53 +0000538 };
539
540 // getBaseOpcodeFor - This function returns the "base" X86 opcode for the
541 // specified machine instruction.
542 //
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000543 inline unsigned char getBaseOpcodeFor(uint64_t TSFlags) {
Evan Cheng7e763d82011-07-25 18:43:53 +0000544 return TSFlags >> X86II::OpcodeShift;
545 }
546
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000547 inline bool hasImm(uint64_t TSFlags) {
Evan Cheng7e763d82011-07-25 18:43:53 +0000548 return (TSFlags & X86II::ImmMask) != 0;
549 }
550
551 /// getSizeOfImm - Decode the "size of immediate" field from the TSFlags field
552 /// of the specified instruction.
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000553 inline unsigned getSizeOfImm(uint64_t TSFlags) {
Evan Cheng7e763d82011-07-25 18:43:53 +0000554 switch (TSFlags & X86II::ImmMask) {
Craig Topper4ed72782012-02-05 05:38:58 +0000555 default: llvm_unreachable("Unknown immediate size");
Evan Cheng7e763d82011-07-25 18:43:53 +0000556 case X86II::Imm8:
557 case X86II::Imm8PCRel: return 1;
558 case X86II::Imm16:
559 case X86II::Imm16PCRel: return 2;
560 case X86II::Imm32:
David Woodhouse0b6c9492014-01-30 22:20:41 +0000561 case X86II::Imm32S:
Evan Cheng7e763d82011-07-25 18:43:53 +0000562 case X86II::Imm32PCRel: return 4;
563 case X86II::Imm64: return 8;
564 }
565 }
566
567 /// isImmPCRel - Return true if the immediate of the specified instruction's
568 /// TSFlags indicates that it is pc relative.
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000569 inline unsigned isImmPCRel(uint64_t TSFlags) {
Evan Cheng7e763d82011-07-25 18:43:53 +0000570 switch (TSFlags & X86II::ImmMask) {
Craig Topper4ed72782012-02-05 05:38:58 +0000571 default: llvm_unreachable("Unknown immediate size");
Evan Cheng7e763d82011-07-25 18:43:53 +0000572 case X86II::Imm8PCRel:
573 case X86II::Imm16PCRel:
574 case X86II::Imm32PCRel:
575 return true;
576 case X86II::Imm8:
577 case X86II::Imm16:
578 case X86II::Imm32:
David Woodhouse0b6c9492014-01-30 22:20:41 +0000579 case X86II::Imm32S:
580 case X86II::Imm64:
581 return false;
582 }
583 }
584
585 /// isImmSigned - Return true if the immediate of the specified instruction's
586 /// TSFlags indicates that it is signed.
587 inline unsigned isImmSigned(uint64_t TSFlags) {
588 switch (TSFlags & X86II::ImmMask) {
589 default: llvm_unreachable("Unknown immediate signedness");
590 case X86II::Imm32S:
591 return true;
592 case X86II::Imm8:
593 case X86II::Imm8PCRel:
594 case X86II::Imm16:
595 case X86II::Imm16PCRel:
596 case X86II::Imm32:
597 case X86II::Imm32PCRel:
Evan Cheng7e763d82011-07-25 18:43:53 +0000598 case X86II::Imm64:
599 return false;
600 }
601 }
602
Preston Gurdddf96b52013-04-10 20:11:59 +0000603 /// getOperandBias - compute any additional adjustment needed to
604 /// the offset to the start of the memory operand
605 /// in this instruction.
606 /// If this is a two-address instruction,skip one of the register operands.
607 /// FIXME: This should be handled during MCInst lowering.
608 inline int getOperandBias(const MCInstrDesc& Desc)
609 {
610 unsigned NumOps = Desc.getNumOperands();
611 unsigned CurOp = 0;
612 if (NumOps > 1 && Desc.getOperandConstraint(1, MCOI::TIED_TO) == 0)
613 ++CurOp;
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000614 else if (NumOps > 3 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0 &&
615 Desc.getOperandConstraint(3, MCOI::TIED_TO) == 1)
616 // Special case for AVX-512 GATHER with 2 TIED_TO operands
617 // Skip the first 2 operands: dst, mask_wb
618 CurOp += 2;
619 else if (NumOps > 3 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0 &&
620 Desc.getOperandConstraint(NumOps - 1, MCOI::TIED_TO) == 1)
Preston Gurdddf96b52013-04-10 20:11:59 +0000621 // Special case for GATHER with 2 TIED_TO operands
622 // Skip the first 2 operands: dst, mask_wb
623 CurOp += 2;
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000624 else if (NumOps > 2 && Desc.getOperandConstraint(NumOps - 2, MCOI::TIED_TO) == 0)
625 // SCATTER
626 ++CurOp;
Preston Gurdddf96b52013-04-10 20:11:59 +0000627 return CurOp;
628 }
629
Evan Cheng7e763d82011-07-25 18:43:53 +0000630 /// getMemoryOperandNo - The function returns the MCInst operand # for the
631 /// first field of the memory operand. If the instruction doesn't have a
632 /// memory operand, this returns -1.
633 ///
634 /// Note that this ignores tied operands. If there is a tied register which
635 /// is duplicated in the MCInst (e.g. "EAX = addl EAX, [mem]") it is only
636 /// counted as one operand.
637 ///
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000638 inline int getMemoryOperandNo(uint64_t TSFlags, unsigned Opcode) {
Evan Cheng7e763d82011-07-25 18:43:53 +0000639 switch (TSFlags & X86II::FormMask) {
Craig Topper4ed72782012-02-05 05:38:58 +0000640 default: llvm_unreachable("Unknown FormMask value in getMemoryOperandNo!");
Evan Cheng7e763d82011-07-25 18:43:53 +0000641 case X86II::Pseudo:
642 case X86II::RawFrm:
643 case X86II::AddRegFrm:
644 case X86II::MRMDestReg:
645 case X86II::MRMSrcReg:
646 case X86II::RawFrmImm8:
647 case X86II::RawFrmImm16:
Craig Topper35da3d12014-01-16 07:36:58 +0000648 case X86II::RawFrmMemOffs:
David Woodhouse2ef8d9c2014-01-22 15:08:08 +0000649 case X86II::RawFrmSrc:
David Woodhouseb33c2ef2014-01-22 15:08:21 +0000650 case X86II::RawFrmDst:
David Woodhouse9bbf7ca2014-01-22 15:08:36 +0000651 case X86II::RawFrmDstSrc:
Evan Cheng7e763d82011-07-25 18:43:53 +0000652 return -1;
653 case X86II::MRMDestMem:
654 return 0;
655 case X86II::MRMSrcMem: {
656 bool HasVEX_4V = (TSFlags >> X86II::VEXShift) & X86II::VEX_4V;
Craig Toppercd93de92011-12-30 04:48:54 +0000657 bool HasMemOp4 = (TSFlags >> X86II::VEXShift) & X86II::MemOp4;
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000658 bool HasEVEX = (TSFlags >> X86II::VEXShift) & X86II::EVEX;
659 bool HasEVEX_K = HasEVEX && ((TSFlags >> X86II::VEXShift) & X86II::EVEX_K);
Evan Cheng7e763d82011-07-25 18:43:53 +0000660 unsigned FirstMemOp = 1;
Craig Topperaea148c2011-10-16 07:55:05 +0000661 if (HasVEX_4V)
Evan Cheng7e763d82011-07-25 18:43:53 +0000662 ++FirstMemOp;// Skip the register source (which is encoded in VEX_VVVV).
Craig Toppercd93de92011-12-30 04:48:54 +0000663 if (HasMemOp4)
Bruno Cardoso Lopes0f9a1f52011-11-25 19:33:42 +0000664 ++FirstMemOp;// Skip the register source (which is encoded in I8IMM).
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000665 if (HasEVEX_K)
666 ++FirstMemOp;// Skip the mask register
Evan Cheng7e763d82011-07-25 18:43:53 +0000667 // FIXME: Maybe lea should have its own form? This is a horrible hack.
668 //if (Opcode == X86::LEA64r || Opcode == X86::LEA64_32r ||
669 // Opcode == X86::LEA16r || Opcode == X86::LEA32r)
670 return FirstMemOp;
671 }
672 case X86II::MRM0r: case X86II::MRM1r:
673 case X86II::MRM2r: case X86II::MRM3r:
674 case X86II::MRM4r: case X86II::MRM5r:
675 case X86II::MRM6r: case X86II::MRM7r:
676 return -1;
677 case X86II::MRM0m: case X86II::MRM1m:
678 case X86II::MRM2m: case X86II::MRM3m:
679 case X86II::MRM4m: case X86II::MRM5m:
Craig Topper27ad1252011-10-15 20:46:47 +0000680 case X86II::MRM6m: case X86II::MRM7m: {
681 bool HasVEX_4V = (TSFlags >> X86II::VEXShift) & X86II::VEX_4V;
682 unsigned FirstMemOp = 0;
683 if (HasVEX_4V)
684 ++FirstMemOp;// Skip the register dest (which is encoded in VEX_VVVV).
685 return FirstMemOp;
686 }
Dave Zarzycki656e8512013-03-25 18:59:43 +0000687 case X86II::MRM_C1: case X86II::MRM_C2: case X86II::MRM_C3:
688 case X86II::MRM_C4: case X86II::MRM_C8: case X86II::MRM_C9:
Michael Liao95d944032013-04-11 04:52:28 +0000689 case X86II::MRM_CA: case X86II::MRM_CB: case X86II::MRM_E8:
690 case X86II::MRM_F0: case X86II::MRM_F8: case X86II::MRM_F9:
691 case X86II::MRM_D0: case X86II::MRM_D1: case X86II::MRM_D4:
692 case X86II::MRM_D5: case X86II::MRM_D6: case X86II::MRM_D8:
693 case X86II::MRM_D9: case X86II::MRM_DA: case X86II::MRM_DB:
694 case X86II::MRM_DC: case X86II::MRM_DD: case X86II::MRM_DE:
695 case X86II::MRM_DF:
Evan Cheng7e763d82011-07-25 18:43:53 +0000696 return -1;
697 }
698 }
699
700 /// isX86_64ExtendedReg - Is the MachineOperand a x86-64 extended (r8 or
701 /// higher) register? e.g. r8, xmm8, xmm13, etc.
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000702 inline bool isX86_64ExtendedReg(unsigned RegNo) {
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000703 if ((RegNo > X86::XMM7 && RegNo <= X86::XMM15) ||
704 (RegNo > X86::XMM23 && RegNo <= X86::XMM31) ||
705 (RegNo > X86::YMM7 && RegNo <= X86::YMM15) ||
706 (RegNo > X86::YMM23 && RegNo <= X86::YMM31) ||
707 (RegNo > X86::ZMM7 && RegNo <= X86::ZMM15) ||
708 (RegNo > X86::ZMM23 && RegNo <= X86::ZMM31))
709 return true;
710
Evan Cheng7e763d82011-07-25 18:43:53 +0000711 switch (RegNo) {
712 default: break;
713 case X86::R8: case X86::R9: case X86::R10: case X86::R11:
714 case X86::R12: case X86::R13: case X86::R14: case X86::R15:
715 case X86::R8D: case X86::R9D: case X86::R10D: case X86::R11D:
716 case X86::R12D: case X86::R13D: case X86::R14D: case X86::R15D:
717 case X86::R8W: case X86::R9W: case X86::R10W: case X86::R11W:
718 case X86::R12W: case X86::R13W: case X86::R14W: case X86::R15W:
719 case X86::R8B: case X86::R9B: case X86::R10B: case X86::R11B:
720 case X86::R12B: case X86::R13B: case X86::R14B: case X86::R15B:
Evan Cheng7e763d82011-07-25 18:43:53 +0000721 case X86::CR8: case X86::CR9: case X86::CR10: case X86::CR11:
722 case X86::CR12: case X86::CR13: case X86::CR14: case X86::CR15:
723 return true;
724 }
725 return false;
726 }
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000727
728 /// is32ExtendedReg - Is the MemoryOperand a 32 extended (zmm16 or higher)
729 /// registers? e.g. zmm21, etc.
730 static inline bool is32ExtendedReg(unsigned RegNo) {
731 return ((RegNo > X86::XMM15 && RegNo <= X86::XMM31) ||
732 (RegNo > X86::YMM15 && RegNo <= X86::YMM31) ||
733 (RegNo > X86::ZMM15 && RegNo <= X86::ZMM31));
734 }
735
Evan Cheng7e763d82011-07-25 18:43:53 +0000736
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000737 inline bool isX86_64NonExtLowByteReg(unsigned reg) {
Evan Cheng7e763d82011-07-25 18:43:53 +0000738 return (reg == X86::SPL || reg == X86::BPL ||
739 reg == X86::SIL || reg == X86::DIL);
740 }
741}
742
743} // end namespace llvm;
744
745#endif