blob: d4698bf492adfd38ca3fc2d92865a9672132c212 [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
Benjamin Kramera7c40ef2014-08-13 16:26:38 +000017#ifndef LLVM_LIB_TARGET_X86_MCTARGETDESC_X86BASEINFO_H
18#define LLVM_LIB_TARGET_X86_MCTARGETDESC_X86BASEINFO_H
Evan Cheng7e763d82011-07-25 18:43:53 +000019
20#include "X86MCTargetDesc.h"
Craig Topperc6d4efa2014-03-19 06:53:25 +000021#include "llvm/MC/MCInstrDesc.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
Craig Topperf655cdd2014-11-11 07:32:32 +0000219 enum : uint64_t {
Evan Cheng7e763d82011-07-25 18:43:53 +0000220 //===------------------------------------------------------------------===//
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
Craig Topper2fb696b2014-02-19 06:59:13 +0000275 /// RawFrmImm8 - This is used for the ENTER instruction, which has two
276 /// immediates, the first of which is a 16-bit immediate (specified by
277 /// the imm encoding) and the second is a 8-bit fixed value.
278 RawFrmImm8 = 11,
279
280 /// RawFrmImm16 - This is used for CALL FAR instructions, which have two
281 /// immediates, the first of which is a 16 or 32-bit immediate (specified by
282 /// the imm encoding) and the second is a 16-bit fixed value. In the AMD
283 /// manual, this operand is described as pntr16:32 and pntr16:16
284 RawFrmImm16 = 12,
285
Craig Toppera0869dc2014-02-10 06:55:41 +0000286 /// MRMX[rm] - The forms are used to represent instructions that use a
287 /// Mod/RM byte, and don't use the middle field for anything.
288 MRMXr = 14, MRMXm = 15,
289
Evan Cheng7e763d82011-07-25 18:43:53 +0000290 /// MRM[0-7][rm] - These forms are used to represent instructions that use
291 /// a Mod/RM byte, and use the middle field to hold extended opcode
292 /// information. In the intel manual these are represented as /0, /1, ...
293 ///
294
295 // First, instructions that operate on a register r/m operand...
296 MRM0r = 16, MRM1r = 17, MRM2r = 18, MRM3r = 19, // Format /0 /1 /2 /3
297 MRM4r = 20, MRM5r = 21, MRM6r = 22, MRM7r = 23, // Format /4 /5 /6 /7
298
299 // Next, instructions that operate on a memory r/m operand...
300 MRM0m = 24, MRM1m = 25, MRM2m = 26, MRM3m = 27, // Format /0 /1 /2 /3
301 MRM4m = 28, MRM5m = 29, MRM6m = 30, MRM7m = 31, // Format /4 /5 /6 /7
302
Craig Toppered7aa462012-02-18 08:19:49 +0000303 //// MRM_XX - A mod/rm byte of exactly 0xXX.
Craig Topper0d1fd552014-02-19 05:34:21 +0000304 MRM_C0 = 32, MRM_C1 = 33, MRM_C2 = 34, MRM_C3 = 35,
Craig Toppera3776de2015-02-15 04:16:44 +0000305 MRM_C4 = 36, MRM_C5 = 37, MRM_C6 = 38, MRM_C7 = 39,
306 MRM_C8 = 40, MRM_C9 = 41, MRM_CA = 42, MRM_CB = 43,
307 MRM_CC = 44, MRM_CD = 45, MRM_CE = 46, MRM_CF = 47,
308 MRM_D0 = 48, MRM_D1 = 49, MRM_D2 = 50, MRM_D3 = 51,
309 MRM_D4 = 52, MRM_D5 = 53, MRM_D6 = 54, MRM_D7 = 55,
310 MRM_D8 = 56, MRM_D9 = 57, MRM_DA = 58, MRM_DB = 59,
311 MRM_DC = 60, MRM_DD = 61, MRM_DE = 62, MRM_DF = 63,
312 MRM_E0 = 64, MRM_E1 = 65, MRM_E2 = 66, MRM_E3 = 67,
313 MRM_E4 = 68, MRM_E5 = 69, MRM_E6 = 70, MRM_E7 = 71,
314 MRM_E8 = 72, MRM_E9 = 73, MRM_EA = 74, MRM_EB = 75,
315 MRM_EC = 76, MRM_ED = 77, MRM_EE = 78, MRM_EF = 79,
316 MRM_F0 = 80, MRM_F1 = 81, MRM_F2 = 82, MRM_F3 = 83,
317 MRM_F4 = 84, MRM_F5 = 85, MRM_F6 = 86, MRM_F7 = 87,
318 MRM_F8 = 88, MRM_F9 = 89, MRM_FA = 90, MRM_FB = 91,
319 MRM_FC = 92, MRM_FD = 93, MRM_FE = 94, MRM_FF = 95,
Evan Cheng7e763d82011-07-25 18:43:53 +0000320
Craig Topper56f0ed812014-02-19 08:25:02 +0000321 FormMask = 127,
Evan Cheng7e763d82011-07-25 18:43:53 +0000322
323 //===------------------------------------------------------------------===//
324 // Actual flags...
325
Craig Topperfa6298a2014-02-02 09:25:09 +0000326 // OpSize - OpSizeFixed implies instruction never needs a 0x66 prefix.
327 // OpSize16 means this is a 16-bit instruction and needs 0x66 prefix in
328 // 32-bit mode. OpSize32 means this is a 32-bit instruction needs a 0x66
329 // prefix in 16-bit mode.
Craig Topper56f0ed812014-02-19 08:25:02 +0000330 OpSizeShift = 7,
Craig Topperfa6298a2014-02-02 09:25:09 +0000331 OpSizeMask = 0x3 << OpSizeShift,
332
Craig Topperb86338f2014-12-24 06:05:22 +0000333 OpSizeFixed = 0 << OpSizeShift,
334 OpSize16 = 1 << OpSizeShift,
335 OpSize32 = 2 << OpSizeShift,
Evan Cheng7e763d82011-07-25 18:43:53 +0000336
Craig Topperb86338f2014-12-24 06:05:22 +0000337 // AsSize - AdSizeX implies this instruction determines its need of 0x67
338 // prefix from a normal ModRM memory operand. The other types indicate that
339 // an operand is encoded with a specific width and a prefix is needed if
340 // it differs from the current mode.
Craig Topper56f0ed812014-02-19 08:25:02 +0000341 AdSizeShift = OpSizeShift + 2,
Craig Topperb86338f2014-12-24 06:05:22 +0000342 AdSizeMask = 0x3 << AdSizeShift,
343
344 AdSizeX = 1 << AdSizeShift,
345 AdSize16 = 1 << AdSizeShift,
346 AdSize32 = 2 << AdSizeShift,
347 AdSize64 = 3 << AdSizeShift,
Evan Cheng7e763d82011-07-25 18:43:53 +0000348
349 //===------------------------------------------------------------------===//
Craig Topper10243c82014-01-31 08:47:06 +0000350 // OpPrefix - There are several prefix bytes that are used as opcode
351 // extensions. These are 0x66, 0xF3, and 0xF2. If this field is 0 there is
352 // no prefix.
Evan Cheng7e763d82011-07-25 18:43:53 +0000353 //
Craig Topperb86338f2014-12-24 06:05:22 +0000354 OpPrefixShift = AdSizeShift + 2,
Craig Topper5ccb6172014-02-18 00:21:49 +0000355 OpPrefixMask = 0x7 << OpPrefixShift,
Yunzhong Gaob8bbcbf2013-09-27 18:38:42 +0000356
Craig Topper5ccb6172014-02-18 00:21:49 +0000357 // PS, PD - Prefix code for packed single and double precision vector
358 // floating point operations performed in the SSE registers.
359 PS = 1 << OpPrefixShift, PD = 2 << OpPrefixShift,
Craig Topperae11aed2014-01-14 07:41:20 +0000360
Craig Topper10243c82014-01-31 08:47:06 +0000361 // XS, XD - These prefix codes are for single and double precision scalar
362 // floating point operations performed in the SSE registers.
Craig Topper5ccb6172014-02-18 00:21:49 +0000363 XS = 3 << OpPrefixShift, XD = 4 << OpPrefixShift,
Craig Topperae11aed2014-01-14 07:41:20 +0000364
Craig Topper10243c82014-01-31 08:47:06 +0000365 //===------------------------------------------------------------------===//
366 // OpMap - This field determines which opcode map this instruction
367 // belongs to. i.e. one-byte, two-byte, 0x0f 0x38, 0x0f 0x3a, etc.
368 //
Craig Topper5ccb6172014-02-18 00:21:49 +0000369 OpMapShift = OpPrefixShift + 3,
Craig Topper56f0ed812014-02-19 08:25:02 +0000370 OpMapMask = 0x7 << OpMapShift,
Craig Topper10243c82014-01-31 08:47:06 +0000371
372 // OB - OneByte - Set if this instruction has a one byte opcode.
373 OB = 0 << OpMapShift,
374
375 // TB - TwoByte - Set if this instruction has a two byte opcode, which
376 // starts with a 0x0F byte before the real opcode.
377 TB = 1 << OpMapShift,
378
379 // T8, TA - Prefix after the 0x0F prefix.
380 T8 = 2 << OpMapShift, TA = 3 << OpMapShift,
381
382 // XOP8 - Prefix to include use of imm byte.
383 XOP8 = 4 << OpMapShift,
384
385 // XOP9 - Prefix to exclude use of imm byte.
386 XOP9 = 5 << OpMapShift,
387
388 // XOPA - Prefix to encode 0xA in VEX.MMMM of XOP instructions.
389 XOPA = 6 << OpMapShift,
390
Evan Cheng7e763d82011-07-25 18:43:53 +0000391 //===------------------------------------------------------------------===//
392 // REX_W - REX prefixes are instruction prefixes used in 64-bit mode.
393 // They are used to specify GPRs and SSE registers, 64-bit operand size,
394 // etc. We only cares about REX.W and REX.R bits and only the former is
395 // statically determined.
396 //
Craig Topper56f0ed812014-02-19 08:25:02 +0000397 REXShift = OpMapShift + 3,
Evan Cheng7e763d82011-07-25 18:43:53 +0000398 REX_W = 1 << REXShift,
399
400 //===------------------------------------------------------------------===//
401 // This three-bit field describes the size of an immediate operand. Zero is
402 // unused so that we can tell if we forgot to set a value.
403 ImmShift = REXShift + 1,
David Woodhouse0b6c9492014-01-30 22:20:41 +0000404 ImmMask = 15 << ImmShift,
Evan Cheng7e763d82011-07-25 18:43:53 +0000405 Imm8 = 1 << ImmShift,
406 Imm8PCRel = 2 << ImmShift,
407 Imm16 = 3 << ImmShift,
408 Imm16PCRel = 4 << ImmShift,
409 Imm32 = 5 << ImmShift,
410 Imm32PCRel = 6 << ImmShift,
David Woodhouse0b6c9492014-01-30 22:20:41 +0000411 Imm32S = 7 << ImmShift,
412 Imm64 = 8 << ImmShift,
Evan Cheng7e763d82011-07-25 18:43:53 +0000413
414 //===------------------------------------------------------------------===//
415 // FP Instruction Classification... Zero is non-fp instruction.
416
417 // FPTypeMask - Mask for all of the FP types...
David Woodhouse0b6c9492014-01-30 22:20:41 +0000418 FPTypeShift = ImmShift + 4,
Evan Cheng7e763d82011-07-25 18:43:53 +0000419 FPTypeMask = 7 << FPTypeShift,
420
421 // NotFP - The default, set for instructions that do not use FP registers.
422 NotFP = 0 << FPTypeShift,
423
424 // ZeroArgFP - 0 arg FP instruction which implicitly pushes ST(0), f.e. fld0
425 ZeroArgFP = 1 << FPTypeShift,
426
427 // OneArgFP - 1 arg FP instructions which implicitly read ST(0), such as fst
428 OneArgFP = 2 << FPTypeShift,
429
430 // OneArgFPRW - 1 arg FP instruction which implicitly read ST(0) and write a
431 // result back to ST(0). For example, fcos, fsqrt, etc.
432 //
433 OneArgFPRW = 3 << FPTypeShift,
434
435 // TwoArgFP - 2 arg FP instructions which implicitly read ST(0), and an
436 // explicit argument, storing the result to either ST(0) or the implicit
437 // argument. For example: fadd, fsub, fmul, etc...
438 TwoArgFP = 4 << FPTypeShift,
439
440 // CompareFP - 2 arg FP instructions which implicitly read ST(0) and an
441 // explicit argument, but have no destination. Example: fucom, fucomi, ...
442 CompareFP = 5 << FPTypeShift,
443
444 // CondMovFP - "2 operand" floating point conditional move instructions.
445 CondMovFP = 6 << FPTypeShift,
446
447 // SpecialFP - Special instruction forms. Dispatch by opcode explicitly.
448 SpecialFP = 7 << FPTypeShift,
449
450 // Lock prefix
451 LOCKShift = FPTypeShift + 3,
452 LOCK = 1 << LOCKShift,
453
Craig Topperec688662014-01-31 07:00:55 +0000454 // REP prefix
455 REPShift = LOCKShift + 1,
456 REP = 1 << REPShift,
457
458 // Execution domain for SSE instructions.
459 // 0 means normal, non-SSE instruction.
460 SSEDomainShift = REPShift + 1,
Evan Cheng7e763d82011-07-25 18:43:53 +0000461
Craig Topperd402df32014-02-02 07:08:01 +0000462 // Encoding
463 EncodingShift = SSEDomainShift + 2,
464 EncodingMask = 0x3 << EncodingShift,
465
466 // VEX - encoding using 0xC4/0xC5
Craig Topperf655cdd2014-11-11 07:32:32 +0000467 VEX = 1 << EncodingShift,
Craig Topperd402df32014-02-02 07:08:01 +0000468
469 /// XOP - Opcode prefix used by XOP instructions.
Craig Topperf655cdd2014-11-11 07:32:32 +0000470 XOP = 2 << EncodingShift,
Craig Topperd402df32014-02-02 07:08:01 +0000471
472 // VEX_EVEX - Specifies that this instruction use EVEX form which provides
473 // syntax support up to 32 512-bit register operands and up to 7 16-bit
474 // mask operands as well as source operand data swizzling/memory operand
475 // conversion, eviction hint, and rounding mode.
Craig Topperf655cdd2014-11-11 07:32:32 +0000476 EVEX = 3 << EncodingShift,
Craig Topperd402df32014-02-02 07:08:01 +0000477
478 // Opcode
479 OpcodeShift = EncodingShift + 2,
Evan Cheng7e763d82011-07-25 18:43:53 +0000480
Evan Cheng7e763d82011-07-25 18:43:53 +0000481 /// VEX_W - Has a opcode specific functionality, but is used in the same
482 /// way as REX_W is for regular SSE instructions.
Craig Topperf655cdd2014-11-11 07:32:32 +0000483 VEX_WShift = OpcodeShift + 8,
484 VEX_W = 1ULL << VEX_WShift,
Evan Cheng7e763d82011-07-25 18:43:53 +0000485
486 /// VEX_4V - Used to specify an additional AVX/SSE register. Several 2
487 /// address instructions in SSE are represented as 3 address ones in AVX
488 /// and the additional register is encoded in VEX_VVVV prefix.
Craig Topperf655cdd2014-11-11 07:32:32 +0000489 VEX_4VShift = VEX_WShift + 1,
490 VEX_4V = 1ULL << VEX_4VShift,
Evan Cheng7e763d82011-07-25 18:43:53 +0000491
Craig Topperaea148c2011-10-16 07:55:05 +0000492 /// VEX_4VOp3 - Similar to VEX_4V, but used on instructions that encode
493 /// operand 3 with VEX.vvvv.
Craig Topperf655cdd2014-11-11 07:32:32 +0000494 VEX_4VOp3Shift = VEX_4VShift + 1,
495 VEX_4VOp3 = 1ULL << VEX_4VOp3Shift,
Craig Topperaea148c2011-10-16 07:55:05 +0000496
Evan Cheng7e763d82011-07-25 18:43:53 +0000497 /// VEX_I8IMM - Specifies that the last register used in a AVX instruction,
498 /// must be encoded in the i8 immediate field. This usually happens in
499 /// instructions with 4 operands.
Craig Topperf655cdd2014-11-11 07:32:32 +0000500 VEX_I8IMMShift = VEX_4VOp3Shift + 1,
501 VEX_I8IMM = 1ULL << VEX_I8IMMShift,
Evan Cheng7e763d82011-07-25 18:43:53 +0000502
503 /// VEX_L - Stands for a bit in the VEX opcode prefix meaning the current
504 /// instruction uses 256-bit wide registers. This is usually auto detected
505 /// if a VR256 register is used, but some AVX instructions also have this
506 /// field marked when using a f256 memory references.
Craig Topperf655cdd2014-11-11 07:32:32 +0000507 VEX_LShift = VEX_I8IMMShift + 1,
508 VEX_L = 1ULL << VEX_LShift,
Evan Cheng7e763d82011-07-25 18:43:53 +0000509
Craig Topperf18c8962011-10-04 06:30:42 +0000510 // VEX_LIG - Specifies that this instruction ignores the L-bit in the VEX
511 // prefix. Usually used for scalar instructions. Needed by disassembler.
Craig Topperf655cdd2014-11-11 07:32:32 +0000512 VEX_LIGShift = VEX_LShift + 1,
513 VEX_LIG = 1ULL << VEX_LIGShift,
Craig Topperf18c8962011-10-04 06:30:42 +0000514
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000515 // TODO: we should combine VEX_L and VEX_LIG together to form a 2-bit field
516 // with following encoding:
517 // - 00 V128
518 // - 01 V256
519 // - 10 V512
520 // - 11 LIG (but, in insn encoding, leave VEX.L and EVEX.L in zeros.
521 // this will save 1 tsflag bit
522
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000523 // EVEX_K - Set if this instruction requires masking
Craig Topperf655cdd2014-11-11 07:32:32 +0000524 EVEX_KShift = VEX_LIGShift + 1,
525 EVEX_K = 1ULL << EVEX_KShift,
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000526
527 // EVEX_Z - Set if this instruction has EVEX.Z field set.
Craig Topperf655cdd2014-11-11 07:32:32 +0000528 EVEX_ZShift = EVEX_KShift + 1,
529 EVEX_Z = 1ULL << EVEX_ZShift,
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000530
531 // EVEX_L2 - Set if this instruction has EVEX.L' field set.
Craig Topperf655cdd2014-11-11 07:32:32 +0000532 EVEX_L2Shift = EVEX_ZShift + 1,
533 EVEX_L2 = 1ULL << EVEX_L2Shift,
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000534
535 // EVEX_B - Set if this instruction has EVEX.B field set.
Craig Topperf655cdd2014-11-11 07:32:32 +0000536 EVEX_BShift = EVEX_L2Shift + 1,
537 EVEX_B = 1ULL << EVEX_BShift,
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000538
Adam Nemet54adb0f2014-07-17 17:04:50 +0000539 // The scaling factor for the AVX512's 8-bit compressed displacement.
Craig Topperf655cdd2014-11-11 07:32:32 +0000540 CD8_Scale_Shift = EVEX_BShift + 1,
541 CD8_Scale_Mask = 127ULL << CD8_Scale_Shift,
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000542
Evan Cheng7e763d82011-07-25 18:43:53 +0000543 /// Has3DNow0F0FOpcode - This flag indicates that the instruction uses the
544 /// wacky 0x0F 0x0F prefix for 3DNow! instructions. The manual documents
545 /// this as having a 0x0F prefix with a 0x0F opcode, and each instruction
546 /// storing a classifier in the imm8 field. To simplify our implementation,
547 /// we handle this by storeing the classifier in the opcode field and using
548 /// this flag to indicate that the encoder should do the wacky 3DNow! thing.
Adam Nemet54adb0f2014-07-17 17:04:50 +0000549 Has3DNow0F0FOpcodeShift = CD8_Scale_Shift + 7,
Craig Topperf655cdd2014-11-11 07:32:32 +0000550 Has3DNow0F0FOpcode = 1ULL << Has3DNow0F0FOpcodeShift,
Bruno Cardoso Lopes0f9a1f52011-11-25 19:33:42 +0000551
Craig Toppercd93de92011-12-30 04:48:54 +0000552 /// MemOp4 - Used to indicate swapping of operand 3 and 4 to be encoded in
553 /// ModRM or I8IMM. This is used for FMA4 and XOP instructions.
Adam Nemetcf7c9052014-07-14 23:18:39 +0000554 MemOp4Shift = Has3DNow0F0FOpcodeShift + 1,
Craig Topperf655cdd2014-11-11 07:32:32 +0000555 MemOp4 = 1ULL << MemOp4Shift,
Jan Sjödin6dd24882011-12-12 19:12:26 +0000556
Elena Demikhovskyb19c9dc2014-01-13 12:55:03 +0000557 /// Explicitly specified rounding control
Adam Nemetcf7c9052014-07-14 23:18:39 +0000558 EVEX_RCShift = MemOp4Shift + 1,
Craig Topperf655cdd2014-11-11 07:32:32 +0000559 EVEX_RC = 1ULL << EVEX_RCShift
Evan Cheng7e763d82011-07-25 18:43:53 +0000560 };
561
562 // getBaseOpcodeFor - This function returns the "base" X86 opcode for the
563 // specified machine instruction.
564 //
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000565 inline unsigned char getBaseOpcodeFor(uint64_t TSFlags) {
Evan Cheng7e763d82011-07-25 18:43:53 +0000566 return TSFlags >> X86II::OpcodeShift;
567 }
568
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000569 inline bool hasImm(uint64_t TSFlags) {
Evan Cheng7e763d82011-07-25 18:43:53 +0000570 return (TSFlags & X86II::ImmMask) != 0;
571 }
572
573 /// getSizeOfImm - Decode the "size of immediate" field from the TSFlags field
574 /// of the specified instruction.
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000575 inline unsigned getSizeOfImm(uint64_t TSFlags) {
Evan Cheng7e763d82011-07-25 18:43:53 +0000576 switch (TSFlags & X86II::ImmMask) {
Craig Topper4ed72782012-02-05 05:38:58 +0000577 default: llvm_unreachable("Unknown immediate size");
Evan Cheng7e763d82011-07-25 18:43:53 +0000578 case X86II::Imm8:
579 case X86II::Imm8PCRel: return 1;
580 case X86II::Imm16:
581 case X86II::Imm16PCRel: return 2;
582 case X86II::Imm32:
David Woodhouse0b6c9492014-01-30 22:20:41 +0000583 case X86II::Imm32S:
Evan Cheng7e763d82011-07-25 18:43:53 +0000584 case X86II::Imm32PCRel: return 4;
585 case X86II::Imm64: return 8;
586 }
587 }
588
589 /// isImmPCRel - Return true if the immediate of the specified instruction's
590 /// TSFlags indicates that it is pc relative.
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000591 inline unsigned isImmPCRel(uint64_t TSFlags) {
Evan Cheng7e763d82011-07-25 18:43:53 +0000592 switch (TSFlags & X86II::ImmMask) {
Craig Topper4ed72782012-02-05 05:38:58 +0000593 default: llvm_unreachable("Unknown immediate size");
Evan Cheng7e763d82011-07-25 18:43:53 +0000594 case X86II::Imm8PCRel:
595 case X86II::Imm16PCRel:
596 case X86II::Imm32PCRel:
597 return true;
598 case X86II::Imm8:
599 case X86II::Imm16:
600 case X86II::Imm32:
David Woodhouse0b6c9492014-01-30 22:20:41 +0000601 case X86II::Imm32S:
602 case X86II::Imm64:
603 return false;
604 }
605 }
606
607 /// isImmSigned - Return true if the immediate of the specified instruction's
608 /// TSFlags indicates that it is signed.
609 inline unsigned isImmSigned(uint64_t TSFlags) {
610 switch (TSFlags & X86II::ImmMask) {
611 default: llvm_unreachable("Unknown immediate signedness");
612 case X86II::Imm32S:
613 return true;
614 case X86II::Imm8:
615 case X86II::Imm8PCRel:
616 case X86II::Imm16:
617 case X86II::Imm16PCRel:
618 case X86II::Imm32:
619 case X86II::Imm32PCRel:
Evan Cheng7e763d82011-07-25 18:43:53 +0000620 case X86II::Imm64:
621 return false;
622 }
623 }
624
Preston Gurdddf96b52013-04-10 20:11:59 +0000625 /// getOperandBias - compute any additional adjustment needed to
626 /// the offset to the start of the memory operand
627 /// in this instruction.
628 /// If this is a two-address instruction,skip one of the register operands.
629 /// FIXME: This should be handled during MCInst lowering.
630 inline int getOperandBias(const MCInstrDesc& Desc)
631 {
632 unsigned NumOps = Desc.getNumOperands();
633 unsigned CurOp = 0;
634 if (NumOps > 1 && Desc.getOperandConstraint(1, MCOI::TIED_TO) == 0)
635 ++CurOp;
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000636 else if (NumOps > 3 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0 &&
637 Desc.getOperandConstraint(3, MCOI::TIED_TO) == 1)
638 // Special case for AVX-512 GATHER with 2 TIED_TO operands
639 // Skip the first 2 operands: dst, mask_wb
640 CurOp += 2;
641 else if (NumOps > 3 && Desc.getOperandConstraint(2, MCOI::TIED_TO) == 0 &&
642 Desc.getOperandConstraint(NumOps - 1, MCOI::TIED_TO) == 1)
Preston Gurdddf96b52013-04-10 20:11:59 +0000643 // Special case for GATHER with 2 TIED_TO operands
644 // Skip the first 2 operands: dst, mask_wb
645 CurOp += 2;
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000646 else if (NumOps > 2 && Desc.getOperandConstraint(NumOps - 2, MCOI::TIED_TO) == 0)
647 // SCATTER
648 ++CurOp;
Preston Gurdddf96b52013-04-10 20:11:59 +0000649 return CurOp;
650 }
651
Evan Cheng7e763d82011-07-25 18:43:53 +0000652 /// getMemoryOperandNo - The function returns the MCInst operand # for the
653 /// first field of the memory operand. If the instruction doesn't have a
654 /// memory operand, this returns -1.
655 ///
656 /// Note that this ignores tied operands. If there is a tied register which
657 /// is duplicated in the MCInst (e.g. "EAX = addl EAX, [mem]") it is only
658 /// counted as one operand.
659 ///
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000660 inline int getMemoryOperandNo(uint64_t TSFlags, unsigned Opcode) {
Craig Topperf655cdd2014-11-11 07:32:32 +0000661 bool HasVEX_4V = TSFlags & X86II::VEX_4V;
662 bool HasMemOp4 = TSFlags & X86II::MemOp4;
663 bool HasEVEX_K = TSFlags & X86II::EVEX_K;
664
Evan Cheng7e763d82011-07-25 18:43:53 +0000665 switch (TSFlags & X86II::FormMask) {
Craig Topper4ed72782012-02-05 05:38:58 +0000666 default: llvm_unreachable("Unknown FormMask value in getMemoryOperandNo!");
Evan Cheng7e763d82011-07-25 18:43:53 +0000667 case X86II::Pseudo:
668 case X86II::RawFrm:
669 case X86II::AddRegFrm:
670 case X86II::MRMDestReg:
671 case X86II::MRMSrcReg:
672 case X86II::RawFrmImm8:
673 case X86II::RawFrmImm16:
Craig Topper35da3d12014-01-16 07:36:58 +0000674 case X86II::RawFrmMemOffs:
David Woodhouse2ef8d9c2014-01-22 15:08:08 +0000675 case X86II::RawFrmSrc:
David Woodhouseb33c2ef2014-01-22 15:08:21 +0000676 case X86II::RawFrmDst:
David Woodhouse9bbf7ca2014-01-22 15:08:36 +0000677 case X86II::RawFrmDstSrc:
Evan Cheng7e763d82011-07-25 18:43:53 +0000678 return -1;
679 case X86II::MRMDestMem:
680 return 0;
Craig Topper3dcdde22015-01-05 08:19:10 +0000681 case X86II::MRMSrcMem:
682 // Start from 1, skip any registers encoded in VEX_VVVV or I8IMM, or a
683 // mask register.
684 return 1 + HasVEX_4V + HasMemOp4 + HasEVEX_K;
Craig Toppera0869dc2014-02-10 06:55:41 +0000685 case X86II::MRMXr:
Evan Cheng7e763d82011-07-25 18:43:53 +0000686 case X86II::MRM0r: case X86II::MRM1r:
687 case X86II::MRM2r: case X86II::MRM3r:
688 case X86II::MRM4r: case X86II::MRM5r:
689 case X86II::MRM6r: case X86II::MRM7r:
690 return -1;
Craig Toppera0869dc2014-02-10 06:55:41 +0000691 case X86II::MRMXm:
Evan Cheng7e763d82011-07-25 18:43:53 +0000692 case X86II::MRM0m: case X86II::MRM1m:
693 case X86II::MRM2m: case X86II::MRM3m:
694 case X86II::MRM4m: case X86II::MRM5m:
Craig Topper3dcdde22015-01-05 08:19:10 +0000695 case X86II::MRM6m: case X86II::MRM7m:
696 // Start from 0, skip registers encoded in VEX_VVVV or a mask register.
697 return 0 + HasVEX_4V + HasEVEX_K;
Craig Topper0d1fd552014-02-19 05:34:21 +0000698 case X86II::MRM_C0: case X86II::MRM_C1: case X86II::MRM_C2:
699 case X86II::MRM_C3: case X86II::MRM_C4: case X86II::MRM_C8:
700 case X86II::MRM_C9: case X86II::MRM_CA: case X86II::MRM_CB:
Kevin Enderby0d928a12014-07-31 23:57:38 +0000701 case X86II::MRM_CF: case X86II::MRM_D0: case X86II::MRM_D1:
702 case X86II::MRM_D4: case X86II::MRM_D5: case X86II::MRM_D6:
703 case X86II::MRM_D7: case X86II::MRM_D8: case X86II::MRM_D9:
704 case X86II::MRM_DA: case X86II::MRM_DB: case X86II::MRM_DC:
705 case X86II::MRM_DD: case X86II::MRM_DE: case X86II::MRM_DF:
706 case X86II::MRM_E0: case X86II::MRM_E1: case X86II::MRM_E2:
707 case X86II::MRM_E3: case X86II::MRM_E4: case X86II::MRM_E5:
708 case X86II::MRM_E8: case X86II::MRM_E9: case X86II::MRM_EA:
709 case X86II::MRM_EB: case X86II::MRM_EC: case X86II::MRM_ED:
710 case X86II::MRM_EE: case X86II::MRM_F0: case X86II::MRM_F1:
711 case X86II::MRM_F2: case X86II::MRM_F3: case X86II::MRM_F4:
712 case X86II::MRM_F5: case X86II::MRM_F6: case X86II::MRM_F7:
713 case X86II::MRM_F8: case X86II::MRM_F9: case X86II::MRM_FA:
714 case X86II::MRM_FB: case X86II::MRM_FC: case X86II::MRM_FD:
715 case X86II::MRM_FE: case X86II::MRM_FF:
Evan Cheng7e763d82011-07-25 18:43:53 +0000716 return -1;
717 }
718 }
719
720 /// isX86_64ExtendedReg - Is the MachineOperand a x86-64 extended (r8 or
721 /// higher) register? e.g. r8, xmm8, xmm13, etc.
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000722 inline bool isX86_64ExtendedReg(unsigned RegNo) {
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000723 if ((RegNo > X86::XMM7 && RegNo <= X86::XMM15) ||
724 (RegNo > X86::XMM23 && RegNo <= X86::XMM31) ||
725 (RegNo > X86::YMM7 && RegNo <= X86::YMM15) ||
726 (RegNo > X86::YMM23 && RegNo <= X86::YMM31) ||
727 (RegNo > X86::ZMM7 && RegNo <= X86::ZMM15) ||
728 (RegNo > X86::ZMM23 && RegNo <= X86::ZMM31))
729 return true;
730
Evan Cheng7e763d82011-07-25 18:43:53 +0000731 switch (RegNo) {
732 default: break;
733 case X86::R8: case X86::R9: case X86::R10: case X86::R11:
734 case X86::R12: case X86::R13: case X86::R14: case X86::R15:
735 case X86::R8D: case X86::R9D: case X86::R10D: case X86::R11D:
736 case X86::R12D: case X86::R13D: case X86::R14D: case X86::R15D:
737 case X86::R8W: case X86::R9W: case X86::R10W: case X86::R11W:
738 case X86::R12W: case X86::R13W: case X86::R14W: case X86::R15W:
739 case X86::R8B: case X86::R9B: case X86::R10B: case X86::R11B:
740 case X86::R12B: case X86::R13B: case X86::R14B: case X86::R15B:
Evan Cheng7e763d82011-07-25 18:43:53 +0000741 case X86::CR8: case X86::CR9: case X86::CR10: case X86::CR11:
742 case X86::CR12: case X86::CR13: case X86::CR14: case X86::CR15:
743 return true;
744 }
745 return false;
746 }
Elena Demikhovsky003e7d72013-07-28 08:28:38 +0000747
748 /// is32ExtendedReg - Is the MemoryOperand a 32 extended (zmm16 or higher)
749 /// registers? e.g. zmm21, etc.
750 static inline bool is32ExtendedReg(unsigned RegNo) {
751 return ((RegNo > X86::XMM15 && RegNo <= X86::XMM31) ||
752 (RegNo > X86::YMM15 && RegNo <= X86::YMM31) ||
753 (RegNo > X86::ZMM15 && RegNo <= X86::ZMM31));
754 }
755
Michael Liao5bf95782014-12-04 05:20:33 +0000756
Chandler Carruth5c0997f2012-06-20 08:39:33 +0000757 inline bool isX86_64NonExtLowByteReg(unsigned reg) {
Evan Cheng7e763d82011-07-25 18:43:53 +0000758 return (reg == X86::SPL || reg == X86::BPL ||
759 reg == X86::SIL || reg == X86::DIL);
760 }
761}
762
763} // end namespace llvm;
764
765#endif