1 //===--- X86.cpp - Implement X86 target feature support -------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file implements X86 TargetInfo objects.
10 //
11 //===----------------------------------------------------------------------===//
12
13 #include "X86.h"
14 #include "clang/Basic/Builtins.h"
15 #include "clang/Basic/Diagnostic.h"
16 #include "clang/Basic/TargetBuiltins.h"
17 #include "llvm/ADT/StringExtras.h"
18 #include "llvm/ADT/StringRef.h"
19 #include "llvm/ADT/StringSwitch.h"
20 #include "llvm/TargetParser/X86TargetParser.h"
21 #include <optional>
22
23 namespace clang {
24 namespace targets {
25
26 static constexpr Builtin::Info BuiltinInfoX86[] = {
27 #define BUILTIN(ID, TYPE, ATTRS) \
28 {#ID, TYPE, ATTRS, nullptr, HeaderDesc::NO_HEADER, ALL_LANGUAGES},
29 #define TARGET_BUILTIN(ID, TYPE, ATTRS, FEATURE) \
30 {#ID, TYPE, ATTRS, FEATURE, HeaderDesc::NO_HEADER, ALL_LANGUAGES},
31 #define TARGET_HEADER_BUILTIN(ID, TYPE, ATTRS, HEADER, LANGS, FEATURE) \
32 {#ID, TYPE, ATTRS, FEATURE, HeaderDesc::HEADER, LANGS},
33 #include "clang/Basic/BuiltinsX86.def"
34
35 #define BUILTIN(ID, TYPE, ATTRS) \
36 {#ID, TYPE, ATTRS, nullptr, HeaderDesc::NO_HEADER, ALL_LANGUAGES},
37 #define TARGET_BUILTIN(ID, TYPE, ATTRS, FEATURE) \
38 {#ID, TYPE, ATTRS, FEATURE, HeaderDesc::NO_HEADER, ALL_LANGUAGES},
39 #define TARGET_HEADER_BUILTIN(ID, TYPE, ATTRS, HEADER, LANGS, FEATURE) \
40 {#ID, TYPE, ATTRS, FEATURE, HeaderDesc::HEADER, LANGS},
41 #include "clang/Basic/BuiltinsX86_64.def"
42 };
43
44 static const char *const GCCRegNames[] = {
45 "ax", "dx", "cx", "bx", "si", "di", "bp", "sp",
46 "st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)",
47 "argp", "flags", "fpcr", "fpsr", "dirflag", "frame", "xmm0", "xmm1",
48 "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7", "mm0", "mm1",
49 "mm2", "mm3", "mm4", "mm5", "mm6", "mm7", "r8", "r9",
50 "r10", "r11", "r12", "r13", "r14", "r15", "xmm8", "xmm9",
51 "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", "xmm15", "ymm0", "ymm1",
52 "ymm2", "ymm3", "ymm4", "ymm5", "ymm6", "ymm7", "ymm8", "ymm9",
53 "ymm10", "ymm11", "ymm12", "ymm13", "ymm14", "ymm15", "xmm16", "xmm17",
54 "xmm18", "xmm19", "xmm20", "xmm21", "xmm22", "xmm23", "xmm24", "xmm25",
55 "xmm26", "xmm27", "xmm28", "xmm29", "xmm30", "xmm31", "ymm16", "ymm17",
56 "ymm18", "ymm19", "ymm20", "ymm21", "ymm22", "ymm23", "ymm24", "ymm25",
57 "ymm26", "ymm27", "ymm28", "ymm29", "ymm30", "ymm31", "zmm0", "zmm1",
58 "zmm2", "zmm3", "zmm4", "zmm5", "zmm6", "zmm7", "zmm8", "zmm9",
59 "zmm10", "zmm11", "zmm12", "zmm13", "zmm14", "zmm15", "zmm16", "zmm17",
60 "zmm18", "zmm19", "zmm20", "zmm21", "zmm22", "zmm23", "zmm24", "zmm25",
61 "zmm26", "zmm27", "zmm28", "zmm29", "zmm30", "zmm31", "k0", "k1",
62 "k2", "k3", "k4", "k5", "k6", "k7",
63 "cr0", "cr2", "cr3", "cr4", "cr8",
64 "dr0", "dr1", "dr2", "dr3", "dr6", "dr7",
65 "bnd0", "bnd1", "bnd2", "bnd3",
66 "tmm0", "tmm1", "tmm2", "tmm3", "tmm4", "tmm5", "tmm6", "tmm7",
67 "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23",
68 "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31",
69 };
70
71 const TargetInfo::AddlRegName AddlRegNames[] = {
72 {{"al", "ah", "eax", "rax"}, 0},
73 {{"bl", "bh", "ebx", "rbx"}, 3},
74 {{"cl", "ch", "ecx", "rcx"}, 2},
75 {{"dl", "dh", "edx", "rdx"}, 1},
76 {{"esi", "rsi"}, 4},
77 {{"edi", "rdi"}, 5},
78 {{"esp", "rsp"}, 7},
79 {{"ebp", "rbp"}, 6},
80 {{"r8d", "r8w", "r8b"}, 38},
81 {{"r9d", "r9w", "r9b"}, 39},
82 {{"r10d", "r10w", "r10b"}, 40},
83 {{"r11d", "r11w", "r11b"}, 41},
84 {{"r12d", "r12w", "r12b"}, 42},
85 {{"r13d", "r13w", "r13b"}, 43},
86 {{"r14d", "r14w", "r14b"}, 44},
87 {{"r15d", "r15w", "r15b"}, 45},
88 {{"r16d", "r16w", "r16b"}, 165},
89 {{"r17d", "r17w", "r17b"}, 166},
90 {{"r18d", "r18w", "r18b"}, 167},
91 {{"r19d", "r19w", "r19b"}, 168},
92 {{"r20d", "r20w", "r20b"}, 169},
93 {{"r21d", "r21w", "r21b"}, 170},
94 {{"r22d", "r22w", "r22b"}, 171},
95 {{"r23d", "r23w", "r23b"}, 172},
96 {{"r24d", "r24w", "r24b"}, 173},
97 {{"r25d", "r25w", "r25b"}, 174},
98 {{"r26d", "r26w", "r26b"}, 175},
99 {{"r27d", "r27w", "r27b"}, 176},
100 {{"r28d", "r28w", "r28b"}, 177},
101 {{"r29d", "r29w", "r29b"}, 178},
102 {{"r30d", "r30w", "r30b"}, 179},
103 {{"r31d", "r31w", "r31b"}, 180},
104 };
105 } // namespace targets
106 } // namespace clang
107
108 using namespace clang;
109 using namespace clang::targets;
110
setFPMath(StringRef Name)111 bool X86TargetInfo::setFPMath(StringRef Name) {
112 if (Name == "387") {
113 FPMath = FP_387;
114 return true;
115 }
116 if (Name == "sse") {
117 FPMath = FP_SSE;
118 return true;
119 }
120 return false;
121 }
122
initFeatureMap(llvm::StringMap<bool> & Features,DiagnosticsEngine & Diags,StringRef CPU,const std::vector<std::string> & FeaturesVec) const123 bool X86TargetInfo::initFeatureMap(
124 llvm::StringMap<bool> &Features, DiagnosticsEngine &Diags, StringRef CPU,
125 const std::vector<std::string> &FeaturesVec) const {
126 // FIXME: This *really* should not be here.
127 // X86_64 always has SSE2.
128 if (getTriple().getArch() == llvm::Triple::x86_64)
129 setFeatureEnabled(Features, "sse2", true);
130
131 using namespace llvm::X86;
132
133 SmallVector<StringRef, 16> CPUFeatures;
134 getFeaturesForCPU(CPU, CPUFeatures);
135 for (auto &F : CPUFeatures)
136 setFeatureEnabled(Features, F, true);
137
138 std::vector<std::string> UpdatedFeaturesVec;
139 std::vector<std::string> UpdatedAVX10FeaturesVec;
140 enum { FE_NOSET = -1, FE_FALSE, FE_TRUE };
141 int HasEVEX512 = FE_NOSET;
142 bool HasAVX512F = Features.lookup("avx512f");
143 bool HasAVX10 = Features.lookup("avx10.1-256");
144 bool HasAVX10_512 = Features.lookup("avx10.1-512");
145 std::string LastAVX10;
146 std::string LastAVX512;
147 for (const auto &Feature : FeaturesVec) {
148 // Expand general-regs-only to -x86, -mmx and -sse
149 if (Feature == "+general-regs-only") {
150 UpdatedFeaturesVec.push_back("-x87");
151 UpdatedFeaturesVec.push_back("-mmx");
152 UpdatedFeaturesVec.push_back("-sse");
153 continue;
154 }
155
156 if (Feature.substr(1, 6) == "avx10.") {
157 if (Feature[0] == '+') {
158 HasAVX10 = true;
159 if (StringRef(Feature).ends_with("512"))
160 HasAVX10_512 = true;
161 LastAVX10 = Feature;
162 } else if (HasAVX10 && Feature == "-avx10.1-256") {
163 HasAVX10 = false;
164 HasAVX10_512 = false;
165 } else if (HasAVX10_512 && Feature == "-avx10.1-512") {
166 HasAVX10_512 = false;
167 }
168 // Postpone AVX10 features handling after AVX512 settled.
169 UpdatedAVX10FeaturesVec.push_back(Feature);
170 continue;
171 } else if (!HasAVX512F && StringRef(Feature).starts_with("+avx512")) {
172 HasAVX512F = true;
173 LastAVX512 = Feature;
174 } else if (HasAVX512F && Feature == "-avx512f") {
175 HasAVX512F = false;
176 } else if (HasEVEX512 != FE_TRUE && Feature == "+evex512") {
177 HasEVEX512 = FE_TRUE;
178 continue;
179 } else if (HasEVEX512 != FE_FALSE && Feature == "-evex512") {
180 HasEVEX512 = FE_FALSE;
181 continue;
182 }
183
184 UpdatedFeaturesVec.push_back(Feature);
185 }
186 llvm::append_range(UpdatedFeaturesVec, UpdatedAVX10FeaturesVec);
187 // HasEVEX512 is a three-states flag. We need to turn it into [+-]evex512
188 // according to other features.
189 if (HasAVX512F) {
190 UpdatedFeaturesVec.push_back(HasEVEX512 == FE_FALSE ? "-evex512"
191 : "+evex512");
192 if (HasAVX10 && !HasAVX10_512 && HasEVEX512 != FE_FALSE)
193 Diags.Report(diag::warn_invalid_feature_combination)
194 << LastAVX512 + " " + LastAVX10 + "; will be promoted to avx10.1-512";
195 } else if (HasAVX10) {
196 if (HasEVEX512 != FE_NOSET)
197 Diags.Report(diag::warn_invalid_feature_combination)
198 << LastAVX10 + (HasEVEX512 == FE_TRUE ? " +evex512" : " -evex512");
199 UpdatedFeaturesVec.push_back(HasAVX10_512 ? "+evex512" : "-evex512");
200 }
201
202 if (!TargetInfo::initFeatureMap(Features, Diags, CPU, UpdatedFeaturesVec))
203 return false;
204
205 // Can't do this earlier because we need to be able to explicitly enable
206 // or disable these features and the things that they depend upon.
207
208 // Enable popcnt if sse4.2 is enabled and popcnt is not explicitly disabled.
209 auto I = Features.find("sse4.2");
210 if (I != Features.end() && I->getValue() &&
211 !llvm::is_contained(UpdatedFeaturesVec, "-popcnt"))
212 Features["popcnt"] = true;
213
214 // Additionally, if SSE is enabled and mmx is not explicitly disabled,
215 // then enable MMX.
216 I = Features.find("sse");
217 if (I != Features.end() && I->getValue() &&
218 !llvm::is_contained(UpdatedFeaturesVec, "-mmx"))
219 Features["mmx"] = true;
220
221 // Enable xsave if avx is enabled and xsave is not explicitly disabled.
222 I = Features.find("avx");
223 if (I != Features.end() && I->getValue() &&
224 !llvm::is_contained(UpdatedFeaturesVec, "-xsave"))
225 Features["xsave"] = true;
226
227 // Enable CRC32 if SSE4.2 is enabled and CRC32 is not explicitly disabled.
228 I = Features.find("sse4.2");
229 if (I != Features.end() && I->getValue() &&
230 !llvm::is_contained(UpdatedFeaturesVec, "-crc32"))
231 Features["crc32"] = true;
232
233 return true;
234 }
235
setFeatureEnabled(llvm::StringMap<bool> & Features,StringRef Name,bool Enabled) const236 void X86TargetInfo::setFeatureEnabled(llvm::StringMap<bool> &Features,
237 StringRef Name, bool Enabled) const {
238 if (Name == "sse4") {
239 // We can get here via the __target__ attribute since that's not controlled
240 // via the -msse4/-mno-sse4 command line alias. Handle this the same way
241 // here - turn on the sse4.2 if enabled, turn off the sse4.1 level if
242 // disabled.
243 if (Enabled)
244 Name = "sse4.2";
245 else
246 Name = "sse4.1";
247 }
248
249 Features[Name] = Enabled;
250 llvm::X86::updateImpliedFeatures(Name, Enabled, Features);
251 }
252
253 /// handleTargetFeatures - Perform initialization based on the user
254 /// configured set of features.
handleTargetFeatures(std::vector<std::string> & Features,DiagnosticsEngine & Diags)255 bool X86TargetInfo::handleTargetFeatures(std::vector<std::string> &Features,
256 DiagnosticsEngine &Diags) {
257 for (const auto &Feature : Features) {
258 if (Feature[0] != '+')
259 continue;
260
261 if (Feature == "+mmx") {
262 HasMMX = true;
263 } else if (Feature == "+aes") {
264 HasAES = true;
265 } else if (Feature == "+vaes") {
266 HasVAES = true;
267 } else if (Feature == "+pclmul") {
268 HasPCLMUL = true;
269 } else if (Feature == "+vpclmulqdq") {
270 HasVPCLMULQDQ = true;
271 } else if (Feature == "+lzcnt") {
272 HasLZCNT = true;
273 } else if (Feature == "+rdrnd") {
274 HasRDRND = true;
275 } else if (Feature == "+fsgsbase") {
276 HasFSGSBASE = true;
277 } else if (Feature == "+bmi") {
278 HasBMI = true;
279 } else if (Feature == "+bmi2") {
280 HasBMI2 = true;
281 } else if (Feature == "+popcnt") {
282 HasPOPCNT = true;
283 } else if (Feature == "+rtm") {
284 HasRTM = true;
285 } else if (Feature == "+prfchw") {
286 HasPRFCHW = true;
287 } else if (Feature == "+rdseed") {
288 HasRDSEED = true;
289 } else if (Feature == "+adx") {
290 HasADX = true;
291 } else if (Feature == "+tbm") {
292 HasTBM = true;
293 } else if (Feature == "+lwp") {
294 HasLWP = true;
295 } else if (Feature == "+fma") {
296 HasFMA = true;
297 } else if (Feature == "+f16c") {
298 HasF16C = true;
299 } else if (Feature == "+gfni") {
300 HasGFNI = true;
301 } else if (Feature == "+evex512") {
302 HasEVEX512 = true;
303 } else if (Feature == "+avx10.1-256") {
304 HasAVX10_1 = true;
305 } else if (Feature == "+avx10.1-512") {
306 HasAVX10_1_512 = true;
307 } else if (Feature == "+avx512cd") {
308 HasAVX512CD = true;
309 } else if (Feature == "+avx512vpopcntdq") {
310 HasAVX512VPOPCNTDQ = true;
311 } else if (Feature == "+avx512vnni") {
312 HasAVX512VNNI = true;
313 } else if (Feature == "+avx512bf16") {
314 HasAVX512BF16 = true;
315 } else if (Feature == "+avx512fp16") {
316 HasAVX512FP16 = true;
317 HasLegalHalfType = true;
318 } else if (Feature == "+avx512dq") {
319 HasAVX512DQ = true;
320 } else if (Feature == "+avx512bitalg") {
321 HasAVX512BITALG = true;
322 } else if (Feature == "+avx512bw") {
323 HasAVX512BW = true;
324 } else if (Feature == "+avx512vl") {
325 HasAVX512VL = true;
326 } else if (Feature == "+avx512vbmi") {
327 HasAVX512VBMI = true;
328 } else if (Feature == "+avx512vbmi2") {
329 HasAVX512VBMI2 = true;
330 } else if (Feature == "+avx512ifma") {
331 HasAVX512IFMA = true;
332 } else if (Feature == "+avx512vp2intersect") {
333 HasAVX512VP2INTERSECT = true;
334 } else if (Feature == "+sha") {
335 HasSHA = true;
336 } else if (Feature == "+sha512") {
337 HasSHA512 = true;
338 } else if (Feature == "+shstk") {
339 HasSHSTK = true;
340 } else if (Feature == "+sm3") {
341 HasSM3 = true;
342 } else if (Feature == "+sm4") {
343 HasSM4 = true;
344 } else if (Feature == "+movbe") {
345 HasMOVBE = true;
346 } else if (Feature == "+sgx") {
347 HasSGX = true;
348 } else if (Feature == "+cx8") {
349 HasCX8 = true;
350 } else if (Feature == "+cx16") {
351 HasCX16 = true;
352 } else if (Feature == "+fxsr") {
353 HasFXSR = true;
354 } else if (Feature == "+xsave") {
355 HasXSAVE = true;
356 } else if (Feature == "+xsaveopt") {
357 HasXSAVEOPT = true;
358 } else if (Feature == "+xsavec") {
359 HasXSAVEC = true;
360 } else if (Feature == "+xsaves") {
361 HasXSAVES = true;
362 } else if (Feature == "+mwaitx") {
363 HasMWAITX = true;
364 } else if (Feature == "+pku") {
365 HasPKU = true;
366 } else if (Feature == "+clflushopt") {
367 HasCLFLUSHOPT = true;
368 } else if (Feature == "+clwb") {
369 HasCLWB = true;
370 } else if (Feature == "+wbnoinvd") {
371 HasWBNOINVD = true;
372 } else if (Feature == "+prefetchi") {
373 HasPREFETCHI = true;
374 } else if (Feature == "+clzero") {
375 HasCLZERO = true;
376 } else if (Feature == "+cldemote") {
377 HasCLDEMOTE = true;
378 } else if (Feature == "+rdpid") {
379 HasRDPID = true;
380 } else if (Feature == "+rdpru") {
381 HasRDPRU = true;
382 } else if (Feature == "+kl") {
383 HasKL = true;
384 } else if (Feature == "+widekl") {
385 HasWIDEKL = true;
386 } else if (Feature == "+retpoline-external-thunk") {
387 HasRetpolineExternalThunk = true;
388 } else if (Feature == "+sahf") {
389 HasLAHFSAHF = true;
390 } else if (Feature == "+waitpkg") {
391 HasWAITPKG = true;
392 } else if (Feature == "+movdiri") {
393 HasMOVDIRI = true;
394 } else if (Feature == "+movdir64b") {
395 HasMOVDIR64B = true;
396 } else if (Feature == "+pconfig") {
397 HasPCONFIG = true;
398 } else if (Feature == "+ptwrite") {
399 HasPTWRITE = true;
400 } else if (Feature == "+invpcid") {
401 HasINVPCID = true;
402 } else if (Feature == "+enqcmd") {
403 HasENQCMD = true;
404 } else if (Feature == "+hreset") {
405 HasHRESET = true;
406 } else if (Feature == "+amx-bf16") {
407 HasAMXBF16 = true;
408 } else if (Feature == "+amx-fp16") {
409 HasAMXFP16 = true;
410 } else if (Feature == "+amx-int8") {
411 HasAMXINT8 = true;
412 } else if (Feature == "+amx-tile") {
413 HasAMXTILE = true;
414 } else if (Feature == "+amx-complex") {
415 HasAMXCOMPLEX = true;
416 } else if (Feature == "+cmpccxadd") {
417 HasCMPCCXADD = true;
418 } else if (Feature == "+raoint") {
419 HasRAOINT = true;
420 } else if (Feature == "+avxifma") {
421 HasAVXIFMA = true;
422 } else if (Feature == "+avxneconvert") {
423 HasAVXNECONVERT= true;
424 } else if (Feature == "+avxvnni") {
425 HasAVXVNNI = true;
426 } else if (Feature == "+avxvnniint16") {
427 HasAVXVNNIINT16 = true;
428 } else if (Feature == "+avxvnniint8") {
429 HasAVXVNNIINT8 = true;
430 } else if (Feature == "+serialize") {
431 HasSERIALIZE = true;
432 } else if (Feature == "+tsxldtrk") {
433 HasTSXLDTRK = true;
434 } else if (Feature == "+uintr") {
435 HasUINTR = true;
436 } else if (Feature == "+usermsr") {
437 HasUSERMSR = true;
438 } else if (Feature == "+crc32") {
439 HasCRC32 = true;
440 } else if (Feature == "+x87") {
441 HasX87 = true;
442 } else if (Feature == "+fullbf16") {
443 HasFullBFloat16 = true;
444 } else if (Feature == "+egpr") {
445 HasEGPR = true;
446 } else if (Feature == "+inline-asm-use-gpr32") {
447 HasInlineAsmUseGPR32 = true;
448 } else if (Feature == "+push2pop2") {
449 HasPush2Pop2 = true;
450 } else if (Feature == "+ppx") {
451 HasPPX = true;
452 } else if (Feature == "+ndd") {
453 HasNDD = true;
454 } else if (Feature == "+ccmp") {
455 HasCCMP = true;
456 } else if (Feature == "+nf") {
457 HasNF = true;
458 } else if (Feature == "+cf") {
459 HasCF = true;
460 } else if (Feature == "+zu") {
461 HasZU = true;
462 } else if (Feature == "+branch-hint") {
463 HasBranchHint = true;
464 }
465
466 X86SSEEnum Level = llvm::StringSwitch<X86SSEEnum>(Feature)
467 .Case("+avx512f", AVX512F)
468 .Case("+avx2", AVX2)
469 .Case("+avx", AVX)
470 .Case("+sse4.2", SSE42)
471 .Case("+sse4.1", SSE41)
472 .Case("+ssse3", SSSE3)
473 .Case("+sse3", SSE3)
474 .Case("+sse2", SSE2)
475 .Case("+sse", SSE1)
476 .Default(NoSSE);
477 SSELevel = std::max(SSELevel, Level);
478
479 HasFloat16 = SSELevel >= SSE2;
480
481 // X86 target has bfloat16 emulation support in the backend, where
482 // bfloat16 is treated as a 32-bit float, arithmetic operations are
483 // performed in 32-bit, and the result is converted back to bfloat16.
484 // Truncation and extension between bfloat16 and 32-bit float are supported
485 // by the compiler-rt library. However, native bfloat16 support is currently
486 // not available in the X86 target. Hence, HasFullBFloat16 will be false
487 // until native bfloat16 support is available. HasFullBFloat16 is used to
488 // determine whether to automatically use excess floating point precision
489 // for bfloat16 arithmetic operations in the front-end.
490 HasBFloat16 = SSELevel >= SSE2;
491
492 XOPEnum XLevel = llvm::StringSwitch<XOPEnum>(Feature)
493 .Case("+xop", XOP)
494 .Case("+fma4", FMA4)
495 .Case("+sse4a", SSE4A)
496 .Default(NoXOP);
497 XOPLevel = std::max(XOPLevel, XLevel);
498 }
499
500 // LLVM doesn't have a separate switch for fpmath, so only accept it if it
501 // matches the selected sse level.
502 if ((FPMath == FP_SSE && SSELevel < SSE1) ||
503 (FPMath == FP_387 && SSELevel >= SSE1)) {
504 Diags.Report(diag::err_target_unsupported_fpmath)
505 << (FPMath == FP_SSE ? "sse" : "387");
506 return false;
507 }
508
509 // FIXME: We should allow long double type on 32-bits to match with GCC.
510 // This requires backend to be able to lower f80 without x87 first.
511 if (!HasX87 && LongDoubleFormat == &llvm::APFloat::x87DoubleExtended())
512 HasLongDouble = false;
513
514 return true;
515 }
516
517 /// X86TargetInfo::getTargetDefines - Return the set of the X86-specific macro
518 /// definitions for this particular subtarget.
getTargetDefines(const LangOptions & Opts,MacroBuilder & Builder) const519 void X86TargetInfo::getTargetDefines(const LangOptions &Opts,
520 MacroBuilder &Builder) const {
521 // Inline assembly supports X86 flag outputs.
522 Builder.defineMacro("__GCC_ASM_FLAG_OUTPUTS__");
523
524 std::string CodeModel = getTargetOpts().CodeModel;
525 if (CodeModel == "default")
526 CodeModel = "small";
527 Builder.defineMacro("__code_model_" + CodeModel + "__");
528
529 // Target identification.
530 if (getTriple().getArch() == llvm::Triple::x86_64) {
531 Builder.defineMacro("__amd64__");
532 Builder.defineMacro("__amd64");
533 Builder.defineMacro("__x86_64");
534 Builder.defineMacro("__x86_64__");
535 if (getTriple().getArchName() == "x86_64h") {
536 Builder.defineMacro("__x86_64h");
537 Builder.defineMacro("__x86_64h__");
538 }
539 } else {
540 DefineStd(Builder, "i386", Opts);
541 }
542
543 Builder.defineMacro("__SEG_GS");
544 Builder.defineMacro("__SEG_FS");
545 Builder.defineMacro("__seg_gs", "__attribute__((address_space(256)))");
546 Builder.defineMacro("__seg_fs", "__attribute__((address_space(257)))");
547
548 // Subtarget options.
549 // FIXME: We are hard-coding the tune parameters based on the CPU, but they
550 // truly should be based on -mtune options.
551 using namespace llvm::X86;
552 switch (CPU) {
553 case CK_None:
554 break;
555 case CK_i386:
556 // The rest are coming from the i386 define above.
557 Builder.defineMacro("__tune_i386__");
558 break;
559 case CK_i486:
560 case CK_WinChipC6:
561 case CK_WinChip2:
562 case CK_C3:
563 defineCPUMacros(Builder, "i486");
564 break;
565 case CK_PentiumMMX:
566 Builder.defineMacro("__pentium_mmx__");
567 Builder.defineMacro("__tune_pentium_mmx__");
568 [[fallthrough]];
569 case CK_i586:
570 case CK_Pentium:
571 defineCPUMacros(Builder, "i586");
572 defineCPUMacros(Builder, "pentium");
573 break;
574 case CK_Pentium3:
575 case CK_PentiumM:
576 Builder.defineMacro("__tune_pentium3__");
577 [[fallthrough]];
578 case CK_Pentium2:
579 case CK_C3_2:
580 Builder.defineMacro("__tune_pentium2__");
581 [[fallthrough]];
582 case CK_PentiumPro:
583 case CK_i686:
584 defineCPUMacros(Builder, "i686");
585 defineCPUMacros(Builder, "pentiumpro");
586 break;
587 case CK_Pentium4:
588 defineCPUMacros(Builder, "pentium4");
589 break;
590 case CK_Yonah:
591 case CK_Prescott:
592 case CK_Nocona:
593 defineCPUMacros(Builder, "nocona");
594 break;
595 case CK_Core2:
596 case CK_Penryn:
597 defineCPUMacros(Builder, "core2");
598 break;
599 case CK_Bonnell:
600 defineCPUMacros(Builder, "atom");
601 break;
602 case CK_Silvermont:
603 defineCPUMacros(Builder, "slm");
604 break;
605 case CK_Goldmont:
606 defineCPUMacros(Builder, "goldmont");
607 break;
608 case CK_GoldmontPlus:
609 defineCPUMacros(Builder, "goldmont_plus");
610 break;
611 case CK_Tremont:
612 defineCPUMacros(Builder, "tremont");
613 break;
614 // Gracemont and later atom-cores use P-core cpu macros.
615 case CK_Gracemont:
616 case CK_Nehalem:
617 case CK_Westmere:
618 case CK_SandyBridge:
619 case CK_IvyBridge:
620 case CK_Haswell:
621 case CK_Broadwell:
622 case CK_SkylakeClient:
623 case CK_SkylakeServer:
624 case CK_Cascadelake:
625 case CK_Cooperlake:
626 case CK_Cannonlake:
627 case CK_IcelakeClient:
628 case CK_Rocketlake:
629 case CK_IcelakeServer:
630 case CK_Tigerlake:
631 case CK_SapphireRapids:
632 case CK_Alderlake:
633 case CK_Raptorlake:
634 case CK_Meteorlake:
635 case CK_Arrowlake:
636 case CK_ArrowlakeS:
637 case CK_Lunarlake:
638 case CK_Pantherlake:
639 case CK_Sierraforest:
640 case CK_Grandridge:
641 case CK_Graniterapids:
642 case CK_GraniterapidsD:
643 case CK_Emeraldrapids:
644 case CK_Clearwaterforest:
645 // FIXME: Historically, we defined this legacy name, it would be nice to
646 // remove it at some point. We've never exposed fine-grained names for
647 // recent primary x86 CPUs, and we should keep it that way.
648 defineCPUMacros(Builder, "corei7");
649 break;
650 case CK_KNL:
651 defineCPUMacros(Builder, "knl");
652 break;
653 case CK_KNM:
654 break;
655 case CK_Lakemont:
656 defineCPUMacros(Builder, "i586", /*Tuning*/false);
657 defineCPUMacros(Builder, "pentium", /*Tuning*/false);
658 Builder.defineMacro("__tune_lakemont__");
659 break;
660 case CK_K6_2:
661 Builder.defineMacro("__k6_2__");
662 Builder.defineMacro("__tune_k6_2__");
663 [[fallthrough]];
664 case CK_K6_3:
665 if (CPU != CK_K6_2) { // In case of fallthrough
666 // FIXME: GCC may be enabling these in cases where some other k6
667 // architecture is specified but -m3dnow is explicitly provided. The
668 // exact semantics need to be determined and emulated here.
669 Builder.defineMacro("__k6_3__");
670 Builder.defineMacro("__tune_k6_3__");
671 }
672 [[fallthrough]];
673 case CK_K6:
674 defineCPUMacros(Builder, "k6");
675 break;
676 case CK_Athlon:
677 case CK_AthlonXP:
678 defineCPUMacros(Builder, "athlon");
679 if (SSELevel != NoSSE) {
680 Builder.defineMacro("__athlon_sse__");
681 Builder.defineMacro("__tune_athlon_sse__");
682 }
683 break;
684 case CK_K8:
685 case CK_K8SSE3:
686 case CK_x86_64:
687 defineCPUMacros(Builder, "k8");
688 break;
689 case CK_x86_64_v2:
690 case CK_x86_64_v3:
691 case CK_x86_64_v4:
692 break;
693 case CK_AMDFAM10:
694 defineCPUMacros(Builder, "amdfam10");
695 break;
696 case CK_BTVER1:
697 defineCPUMacros(Builder, "btver1");
698 break;
699 case CK_BTVER2:
700 defineCPUMacros(Builder, "btver2");
701 break;
702 case CK_BDVER1:
703 defineCPUMacros(Builder, "bdver1");
704 break;
705 case CK_BDVER2:
706 defineCPUMacros(Builder, "bdver2");
707 break;
708 case CK_BDVER3:
709 defineCPUMacros(Builder, "bdver3");
710 break;
711 case CK_BDVER4:
712 defineCPUMacros(Builder, "bdver4");
713 break;
714 case CK_ZNVER1:
715 defineCPUMacros(Builder, "znver1");
716 break;
717 case CK_ZNVER2:
718 defineCPUMacros(Builder, "znver2");
719 break;
720 case CK_ZNVER3:
721 defineCPUMacros(Builder, "znver3");
722 break;
723 case CK_ZNVER4:
724 defineCPUMacros(Builder, "znver4");
725 break;
726 case CK_ZNVER5:
727 defineCPUMacros(Builder, "znver5");
728 break;
729 case CK_Geode:
730 defineCPUMacros(Builder, "geode");
731 break;
732 }
733
734 // Target properties.
735 Builder.defineMacro("__REGISTER_PREFIX__", "");
736
737 // Define __NO_MATH_INLINES on linux/x86 so that we don't get inline
738 // functions in glibc header files that use FP Stack inline asm which the
739 // backend can't deal with (PR879).
740 Builder.defineMacro("__NO_MATH_INLINES");
741
742 if (HasAES)
743 Builder.defineMacro("__AES__");
744
745 if (HasVAES)
746 Builder.defineMacro("__VAES__");
747
748 if (HasPCLMUL)
749 Builder.defineMacro("__PCLMUL__");
750
751 if (HasVPCLMULQDQ)
752 Builder.defineMacro("__VPCLMULQDQ__");
753
754 // Note, in 32-bit mode, GCC does not define the macro if -mno-sahf. In LLVM,
755 // the feature flag only applies to 64-bit mode.
756 if (HasLAHFSAHF || getTriple().getArch() == llvm::Triple::x86)
757 Builder.defineMacro("__LAHF_SAHF__");
758
759 if (HasLZCNT)
760 Builder.defineMacro("__LZCNT__");
761
762 if (HasRDRND)
763 Builder.defineMacro("__RDRND__");
764
765 if (HasFSGSBASE)
766 Builder.defineMacro("__FSGSBASE__");
767
768 if (HasBMI)
769 Builder.defineMacro("__BMI__");
770
771 if (HasBMI2)
772 Builder.defineMacro("__BMI2__");
773
774 if (HasPOPCNT)
775 Builder.defineMacro("__POPCNT__");
776
777 if (HasRTM)
778 Builder.defineMacro("__RTM__");
779
780 if (HasPRFCHW)
781 Builder.defineMacro("__PRFCHW__");
782
783 if (HasRDSEED)
784 Builder.defineMacro("__RDSEED__");
785
786 if (HasADX)
787 Builder.defineMacro("__ADX__");
788
789 if (HasTBM)
790 Builder.defineMacro("__TBM__");
791
792 if (HasLWP)
793 Builder.defineMacro("__LWP__");
794
795 if (HasMWAITX)
796 Builder.defineMacro("__MWAITX__");
797
798 if (HasMOVBE)
799 Builder.defineMacro("__MOVBE__");
800
801 switch (XOPLevel) {
802 case XOP:
803 Builder.defineMacro("__XOP__");
804 [[fallthrough]];
805 case FMA4:
806 Builder.defineMacro("__FMA4__");
807 [[fallthrough]];
808 case SSE4A:
809 Builder.defineMacro("__SSE4A__");
810 [[fallthrough]];
811 case NoXOP:
812 break;
813 }
814
815 if (HasFMA)
816 Builder.defineMacro("__FMA__");
817
818 if (HasF16C)
819 Builder.defineMacro("__F16C__");
820
821 if (HasGFNI)
822 Builder.defineMacro("__GFNI__");
823
824 if (HasEVEX512)
825 Builder.defineMacro("__EVEX512__");
826 if (HasAVX10_1)
827 Builder.defineMacro("__AVX10_1__");
828 if (HasAVX10_1_512)
829 Builder.defineMacro("__AVX10_1_512__");
830 if (HasAVX512CD)
831 Builder.defineMacro("__AVX512CD__");
832 if (HasAVX512VPOPCNTDQ)
833 Builder.defineMacro("__AVX512VPOPCNTDQ__");
834 if (HasAVX512VNNI)
835 Builder.defineMacro("__AVX512VNNI__");
836 if (HasAVX512BF16)
837 Builder.defineMacro("__AVX512BF16__");
838 if (HasAVX512FP16)
839 Builder.defineMacro("__AVX512FP16__");
840 if (HasAVX512DQ)
841 Builder.defineMacro("__AVX512DQ__");
842 if (HasAVX512BITALG)
843 Builder.defineMacro("__AVX512BITALG__");
844 if (HasAVX512BW)
845 Builder.defineMacro("__AVX512BW__");
846 if (HasAVX512VL) {
847 Builder.defineMacro("__AVX512VL__");
848 Builder.defineMacro("__EVEX256__");
849 }
850 if (HasAVX512VBMI)
851 Builder.defineMacro("__AVX512VBMI__");
852 if (HasAVX512VBMI2)
853 Builder.defineMacro("__AVX512VBMI2__");
854 if (HasAVX512IFMA)
855 Builder.defineMacro("__AVX512IFMA__");
856 if (HasAVX512VP2INTERSECT)
857 Builder.defineMacro("__AVX512VP2INTERSECT__");
858 if (HasSHA)
859 Builder.defineMacro("__SHA__");
860 if (HasSHA512)
861 Builder.defineMacro("__SHA512__");
862
863 if (HasFXSR)
864 Builder.defineMacro("__FXSR__");
865 if (HasXSAVE)
866 Builder.defineMacro("__XSAVE__");
867 if (HasXSAVEOPT)
868 Builder.defineMacro("__XSAVEOPT__");
869 if (HasXSAVEC)
870 Builder.defineMacro("__XSAVEC__");
871 if (HasXSAVES)
872 Builder.defineMacro("__XSAVES__");
873 if (HasPKU)
874 Builder.defineMacro("__PKU__");
875 if (HasCLFLUSHOPT)
876 Builder.defineMacro("__CLFLUSHOPT__");
877 if (HasCLWB)
878 Builder.defineMacro("__CLWB__");
879 if (HasWBNOINVD)
880 Builder.defineMacro("__WBNOINVD__");
881 if (HasSHSTK)
882 Builder.defineMacro("__SHSTK__");
883 if (HasSGX)
884 Builder.defineMacro("__SGX__");
885 if (HasSM3)
886 Builder.defineMacro("__SM3__");
887 if (HasSM4)
888 Builder.defineMacro("__SM4__");
889 if (HasPREFETCHI)
890 Builder.defineMacro("__PREFETCHI__");
891 if (HasCLZERO)
892 Builder.defineMacro("__CLZERO__");
893 if (HasKL)
894 Builder.defineMacro("__KL__");
895 if (HasWIDEKL)
896 Builder.defineMacro("__WIDEKL__");
897 if (HasRDPID)
898 Builder.defineMacro("__RDPID__");
899 if (HasRDPRU)
900 Builder.defineMacro("__RDPRU__");
901 if (HasCLDEMOTE)
902 Builder.defineMacro("__CLDEMOTE__");
903 if (HasWAITPKG)
904 Builder.defineMacro("__WAITPKG__");
905 if (HasMOVDIRI)
906 Builder.defineMacro("__MOVDIRI__");
907 if (HasMOVDIR64B)
908 Builder.defineMacro("__MOVDIR64B__");
909 if (HasPCONFIG)
910 Builder.defineMacro("__PCONFIG__");
911 if (HasPTWRITE)
912 Builder.defineMacro("__PTWRITE__");
913 if (HasINVPCID)
914 Builder.defineMacro("__INVPCID__");
915 if (HasENQCMD)
916 Builder.defineMacro("__ENQCMD__");
917 if (HasHRESET)
918 Builder.defineMacro("__HRESET__");
919 if (HasAMXTILE)
920 Builder.defineMacro("__AMX_TILE__");
921 if (HasAMXINT8)
922 Builder.defineMacro("__AMX_INT8__");
923 if (HasAMXBF16)
924 Builder.defineMacro("__AMX_BF16__");
925 if (HasAMXFP16)
926 Builder.defineMacro("__AMX_FP16__");
927 if (HasAMXCOMPLEX)
928 Builder.defineMacro("__AMX_COMPLEX__");
929 if (HasCMPCCXADD)
930 Builder.defineMacro("__CMPCCXADD__");
931 if (HasRAOINT)
932 Builder.defineMacro("__RAOINT__");
933 if (HasAVXIFMA)
934 Builder.defineMacro("__AVXIFMA__");
935 if (HasAVXNECONVERT)
936 Builder.defineMacro("__AVXNECONVERT__");
937 if (HasAVXVNNI)
938 Builder.defineMacro("__AVXVNNI__");
939 if (HasAVXVNNIINT16)
940 Builder.defineMacro("__AVXVNNIINT16__");
941 if (HasAVXVNNIINT8)
942 Builder.defineMacro("__AVXVNNIINT8__");
943 if (HasSERIALIZE)
944 Builder.defineMacro("__SERIALIZE__");
945 if (HasTSXLDTRK)
946 Builder.defineMacro("__TSXLDTRK__");
947 if (HasUINTR)
948 Builder.defineMacro("__UINTR__");
949 if (HasUSERMSR)
950 Builder.defineMacro("__USERMSR__");
951 if (HasCRC32)
952 Builder.defineMacro("__CRC32__");
953 if (HasEGPR)
954 Builder.defineMacro("__EGPR__");
955 if (HasPush2Pop2)
956 Builder.defineMacro("__PUSH2POP2__");
957 if (HasPPX)
958 Builder.defineMacro("__PPX__");
959 if (HasNDD)
960 Builder.defineMacro("__NDD__");
961 if (HasCCMP)
962 Builder.defineMacro("__CCMP__");
963 if (HasNF)
964 Builder.defineMacro("__NF__");
965 if (HasCF)
966 Builder.defineMacro("__CF__");
967 if (HasZU)
968 Builder.defineMacro("__ZU__");
969 if (HasEGPR && HasPush2Pop2 && HasPPX && HasNDD && HasCCMP && HasNF &&
970 HasCF && HasZU)
971 Builder.defineMacro("__APX_F__");
972 if (HasEGPR && HasInlineAsmUseGPR32)
973 Builder.defineMacro("__APX_INLINE_ASM_USE_GPR32__");
974
975 // Each case falls through to the previous one here.
976 switch (SSELevel) {
977 case AVX512F:
978 Builder.defineMacro("__AVX512F__");
979 [[fallthrough]];
980 case AVX2:
981 Builder.defineMacro("__AVX2__");
982 [[fallthrough]];
983 case AVX:
984 Builder.defineMacro("__AVX__");
985 [[fallthrough]];
986 case SSE42:
987 Builder.defineMacro("__SSE4_2__");
988 [[fallthrough]];
989 case SSE41:
990 Builder.defineMacro("__SSE4_1__");
991 [[fallthrough]];
992 case SSSE3:
993 Builder.defineMacro("__SSSE3__");
994 [[fallthrough]];
995 case SSE3:
996 Builder.defineMacro("__SSE3__");
997 [[fallthrough]];
998 case SSE2:
999 Builder.defineMacro("__SSE2__");
1000 Builder.defineMacro("__SSE2_MATH__"); // -mfp-math=sse always implied.
1001 [[fallthrough]];
1002 case SSE1:
1003 Builder.defineMacro("__SSE__");
1004 Builder.defineMacro("__SSE_MATH__"); // -mfp-math=sse always implied.
1005 [[fallthrough]];
1006 case NoSSE:
1007 break;
1008 }
1009
1010 if (Opts.MicrosoftExt && getTriple().getArch() == llvm::Triple::x86) {
1011 switch (SSELevel) {
1012 case AVX512F:
1013 case AVX2:
1014 case AVX:
1015 case SSE42:
1016 case SSE41:
1017 case SSSE3:
1018 case SSE3:
1019 case SSE2:
1020 Builder.defineMacro("_M_IX86_FP", Twine(2));
1021 break;
1022 case SSE1:
1023 Builder.defineMacro("_M_IX86_FP", Twine(1));
1024 break;
1025 default:
1026 Builder.defineMacro("_M_IX86_FP", Twine(0));
1027 break;
1028 }
1029 }
1030
1031 // Each case falls through to the previous one here.
1032 if (HasMMX) {
1033 Builder.defineMacro("__MMX__");
1034 }
1035
1036 if (CPU >= CK_i486 || CPU == CK_None) {
1037 Builder.defineMacro("__GCC_HAVE_SYNC_COMPARE_AND_SWAP_1");
1038 Builder.defineMacro("__GCC_HAVE_SYNC_COMPARE_AND_SWAP_2");
1039 Builder.defineMacro("__GCC_HAVE_SYNC_COMPARE_AND_SWAP_4");
1040 }
1041 if (HasCX8)
1042 Builder.defineMacro("__GCC_HAVE_SYNC_COMPARE_AND_SWAP_8");
1043 if (HasCX16 && getTriple().getArch() == llvm::Triple::x86_64)
1044 Builder.defineMacro("__GCC_HAVE_SYNC_COMPARE_AND_SWAP_16");
1045
1046 if (HasFloat128)
1047 Builder.defineMacro("__SIZEOF_FLOAT128__", "16");
1048 }
1049
isValidFeatureName(StringRef Name) const1050 bool X86TargetInfo::isValidFeatureName(StringRef Name) const {
1051 return llvm::StringSwitch<bool>(Name)
1052 .Case("adx", true)
1053 .Case("aes", true)
1054 .Case("amx-bf16", true)
1055 .Case("amx-complex", true)
1056 .Case("amx-fp16", true)
1057 .Case("amx-int8", true)
1058 .Case("amx-tile", true)
1059 .Case("avx", true)
1060 .Case("avx10.1-256", true)
1061 .Case("avx10.1-512", true)
1062 .Case("avx2", true)
1063 .Case("avx512f", true)
1064 .Case("avx512cd", true)
1065 .Case("avx512vpopcntdq", true)
1066 .Case("avx512vnni", true)
1067 .Case("avx512bf16", true)
1068 .Case("avx512fp16", true)
1069 .Case("avx512dq", true)
1070 .Case("avx512bitalg", true)
1071 .Case("avx512bw", true)
1072 .Case("avx512vl", true)
1073 .Case("avx512vbmi", true)
1074 .Case("avx512vbmi2", true)
1075 .Case("avx512ifma", true)
1076 .Case("avx512vp2intersect", true)
1077 .Case("avxifma", true)
1078 .Case("avxneconvert", true)
1079 .Case("avxvnni", true)
1080 .Case("avxvnniint16", true)
1081 .Case("avxvnniint8", true)
1082 .Case("bmi", true)
1083 .Case("bmi2", true)
1084 .Case("cldemote", true)
1085 .Case("clflushopt", true)
1086 .Case("clwb", true)
1087 .Case("clzero", true)
1088 .Case("cmpccxadd", true)
1089 .Case("crc32", true)
1090 .Case("cx16", true)
1091 .Case("enqcmd", true)
1092 .Case("evex512", true)
1093 .Case("f16c", true)
1094 .Case("fma", true)
1095 .Case("fma4", true)
1096 .Case("fsgsbase", true)
1097 .Case("fxsr", true)
1098 .Case("general-regs-only", true)
1099 .Case("gfni", true)
1100 .Case("hreset", true)
1101 .Case("invpcid", true)
1102 .Case("kl", true)
1103 .Case("widekl", true)
1104 .Case("lwp", true)
1105 .Case("lzcnt", true)
1106 .Case("mmx", true)
1107 .Case("movbe", true)
1108 .Case("movdiri", true)
1109 .Case("movdir64b", true)
1110 .Case("mwaitx", true)
1111 .Case("pclmul", true)
1112 .Case("pconfig", true)
1113 .Case("pku", true)
1114 .Case("popcnt", true)
1115 .Case("prefetchi", true)
1116 .Case("prfchw", true)
1117 .Case("ptwrite", true)
1118 .Case("raoint", true)
1119 .Case("rdpid", true)
1120 .Case("rdpru", true)
1121 .Case("rdrnd", true)
1122 .Case("rdseed", true)
1123 .Case("rtm", true)
1124 .Case("sahf", true)
1125 .Case("serialize", true)
1126 .Case("sgx", true)
1127 .Case("sha", true)
1128 .Case("sha512", true)
1129 .Case("shstk", true)
1130 .Case("sm3", true)
1131 .Case("sm4", true)
1132 .Case("sse", true)
1133 .Case("sse2", true)
1134 .Case("sse3", true)
1135 .Case("ssse3", true)
1136 .Case("sse4", true)
1137 .Case("sse4.1", true)
1138 .Case("sse4.2", true)
1139 .Case("sse4a", true)
1140 .Case("tbm", true)
1141 .Case("tsxldtrk", true)
1142 .Case("uintr", true)
1143 .Case("usermsr", true)
1144 .Case("vaes", true)
1145 .Case("vpclmulqdq", true)
1146 .Case("wbnoinvd", true)
1147 .Case("waitpkg", true)
1148 .Case("x87", true)
1149 .Case("xop", true)
1150 .Case("xsave", true)
1151 .Case("xsavec", true)
1152 .Case("xsaves", true)
1153 .Case("xsaveopt", true)
1154 .Case("egpr", true)
1155 .Case("push2pop2", true)
1156 .Case("ppx", true)
1157 .Case("ndd", true)
1158 .Case("ccmp", true)
1159 .Case("nf", true)
1160 .Case("cf", true)
1161 .Case("zu", true)
1162 .Default(false);
1163 }
1164
hasFeature(StringRef Feature) const1165 bool X86TargetInfo::hasFeature(StringRef Feature) const {
1166 return llvm::StringSwitch<bool>(Feature)
1167 .Case("adx", HasADX)
1168 .Case("aes", HasAES)
1169 .Case("amx-bf16", HasAMXBF16)
1170 .Case("amx-complex", HasAMXCOMPLEX)
1171 .Case("amx-fp16", HasAMXFP16)
1172 .Case("amx-int8", HasAMXINT8)
1173 .Case("amx-tile", HasAMXTILE)
1174 .Case("avx", SSELevel >= AVX)
1175 .Case("avx10.1-256", HasAVX10_1)
1176 .Case("avx10.1-512", HasAVX10_1_512)
1177 .Case("avx2", SSELevel >= AVX2)
1178 .Case("avx512f", SSELevel >= AVX512F)
1179 .Case("avx512cd", HasAVX512CD)
1180 .Case("avx512vpopcntdq", HasAVX512VPOPCNTDQ)
1181 .Case("avx512vnni", HasAVX512VNNI)
1182 .Case("avx512bf16", HasAVX512BF16)
1183 .Case("avx512fp16", HasAVX512FP16)
1184 .Case("avx512dq", HasAVX512DQ)
1185 .Case("avx512bitalg", HasAVX512BITALG)
1186 .Case("avx512bw", HasAVX512BW)
1187 .Case("avx512vl", HasAVX512VL)
1188 .Case("avx512vbmi", HasAVX512VBMI)
1189 .Case("avx512vbmi2", HasAVX512VBMI2)
1190 .Case("avx512ifma", HasAVX512IFMA)
1191 .Case("avx512vp2intersect", HasAVX512VP2INTERSECT)
1192 .Case("avxifma", HasAVXIFMA)
1193 .Case("avxneconvert", HasAVXNECONVERT)
1194 .Case("avxvnni", HasAVXVNNI)
1195 .Case("avxvnniint16", HasAVXVNNIINT16)
1196 .Case("avxvnniint8", HasAVXVNNIINT8)
1197 .Case("bmi", HasBMI)
1198 .Case("bmi2", HasBMI2)
1199 .Case("cldemote", HasCLDEMOTE)
1200 .Case("clflushopt", HasCLFLUSHOPT)
1201 .Case("clwb", HasCLWB)
1202 .Case("clzero", HasCLZERO)
1203 .Case("cmpccxadd", HasCMPCCXADD)
1204 .Case("crc32", HasCRC32)
1205 .Case("cx8", HasCX8)
1206 .Case("cx16", HasCX16)
1207 .Case("enqcmd", HasENQCMD)
1208 .Case("evex512", HasEVEX512)
1209 .Case("f16c", HasF16C)
1210 .Case("fma", HasFMA)
1211 .Case("fma4", XOPLevel >= FMA4)
1212 .Case("fsgsbase", HasFSGSBASE)
1213 .Case("fxsr", HasFXSR)
1214 .Case("gfni", HasGFNI)
1215 .Case("hreset", HasHRESET)
1216 .Case("invpcid", HasINVPCID)
1217 .Case("kl", HasKL)
1218 .Case("widekl", HasWIDEKL)
1219 .Case("lwp", HasLWP)
1220 .Case("lzcnt", HasLZCNT)
1221 .Case("mmx", HasMMX)
1222 .Case("movbe", HasMOVBE)
1223 .Case("movdiri", HasMOVDIRI)
1224 .Case("movdir64b", HasMOVDIR64B)
1225 .Case("mwaitx", HasMWAITX)
1226 .Case("pclmul", HasPCLMUL)
1227 .Case("pconfig", HasPCONFIG)
1228 .Case("pku", HasPKU)
1229 .Case("popcnt", HasPOPCNT)
1230 .Case("prefetchi", HasPREFETCHI)
1231 .Case("prfchw", HasPRFCHW)
1232 .Case("ptwrite", HasPTWRITE)
1233 .Case("raoint", HasRAOINT)
1234 .Case("rdpid", HasRDPID)
1235 .Case("rdpru", HasRDPRU)
1236 .Case("rdrnd", HasRDRND)
1237 .Case("rdseed", HasRDSEED)
1238 .Case("retpoline-external-thunk", HasRetpolineExternalThunk)
1239 .Case("rtm", HasRTM)
1240 .Case("sahf", HasLAHFSAHF)
1241 .Case("serialize", HasSERIALIZE)
1242 .Case("sgx", HasSGX)
1243 .Case("sha", HasSHA)
1244 .Case("sha512", HasSHA512)
1245 .Case("shstk", HasSHSTK)
1246 .Case("sm3", HasSM3)
1247 .Case("sm4", HasSM4)
1248 .Case("sse", SSELevel >= SSE1)
1249 .Case("sse2", SSELevel >= SSE2)
1250 .Case("sse3", SSELevel >= SSE3)
1251 .Case("ssse3", SSELevel >= SSSE3)
1252 .Case("sse4.1", SSELevel >= SSE41)
1253 .Case("sse4.2", SSELevel >= SSE42)
1254 .Case("sse4a", XOPLevel >= SSE4A)
1255 .Case("tbm", HasTBM)
1256 .Case("tsxldtrk", HasTSXLDTRK)
1257 .Case("uintr", HasUINTR)
1258 .Case("usermsr", HasUSERMSR)
1259 .Case("vaes", HasVAES)
1260 .Case("vpclmulqdq", HasVPCLMULQDQ)
1261 .Case("wbnoinvd", HasWBNOINVD)
1262 .Case("waitpkg", HasWAITPKG)
1263 .Case("x86", true)
1264 .Case("x86_32", getTriple().getArch() == llvm::Triple::x86)
1265 .Case("x86_64", getTriple().getArch() == llvm::Triple::x86_64)
1266 .Case("x87", HasX87)
1267 .Case("xop", XOPLevel >= XOP)
1268 .Case("xsave", HasXSAVE)
1269 .Case("xsavec", HasXSAVEC)
1270 .Case("xsaves", HasXSAVES)
1271 .Case("xsaveopt", HasXSAVEOPT)
1272 .Case("fullbf16", HasFullBFloat16)
1273 .Case("egpr", HasEGPR)
1274 .Case("push2pop2", HasPush2Pop2)
1275 .Case("ppx", HasPPX)
1276 .Case("ndd", HasNDD)
1277 .Case("ccmp", HasCCMP)
1278 .Case("nf", HasNF)
1279 .Case("cf", HasCF)
1280 .Case("zu", HasZU)
1281 .Case("branch-hint", HasBranchHint)
1282 .Default(false);
1283 }
1284
1285 // We can't use a generic validation scheme for the features accepted here
1286 // versus subtarget features accepted in the target attribute because the
1287 // bitfield structure that's initialized in the runtime only supports the
1288 // below currently rather than the full range of subtarget features. (See
1289 // X86TargetInfo::hasFeature for a somewhat comprehensive list).
validateCpuSupports(StringRef FeatureStr) const1290 bool X86TargetInfo::validateCpuSupports(StringRef FeatureStr) const {
1291 return llvm::StringSwitch<bool>(FeatureStr)
1292 #define X86_FEATURE_COMPAT(ENUM, STR, PRIORITY) .Case(STR, true)
1293 #define X86_MICROARCH_LEVEL(ENUM, STR, PRIORITY) .Case(STR, true)
1294 #include "llvm/TargetParser/X86TargetParser.def"
1295 .Default(false);
1296 }
1297
getFeature(StringRef Name)1298 static llvm::X86::ProcessorFeatures getFeature(StringRef Name) {
1299 return llvm::StringSwitch<llvm::X86::ProcessorFeatures>(Name)
1300 #define X86_FEATURE_COMPAT(ENUM, STR, PRIORITY) \
1301 .Case(STR, llvm::X86::FEATURE_##ENUM)
1302
1303 #include "llvm/TargetParser/X86TargetParser.def"
1304 ;
1305 // Note, this function should only be used after ensuring the value is
1306 // correct, so it asserts if the value is out of range.
1307 }
1308
multiVersionSortPriority(StringRef Name) const1309 unsigned X86TargetInfo::multiVersionSortPriority(StringRef Name) const {
1310 // Valid CPUs have a 'key feature' that compares just better than its key
1311 // feature.
1312 using namespace llvm::X86;
1313 CPUKind Kind = parseArchX86(Name);
1314 if (Kind != CK_None) {
1315 ProcessorFeatures KeyFeature = getKeyFeature(Kind);
1316 return (getFeaturePriority(KeyFeature) << 1) + 1;
1317 }
1318
1319 // Now we know we have a feature, so get its priority and shift it a few so
1320 // that we have sufficient room for the CPUs (above).
1321 return getFeaturePriority(getFeature(Name)) << 1;
1322 }
1323
validateCPUSpecificCPUDispatch(StringRef Name) const1324 bool X86TargetInfo::validateCPUSpecificCPUDispatch(StringRef Name) const {
1325 return llvm::X86::validateCPUSpecificCPUDispatch(Name);
1326 }
1327
CPUSpecificManglingCharacter(StringRef Name) const1328 char X86TargetInfo::CPUSpecificManglingCharacter(StringRef Name) const {
1329 return llvm::X86::getCPUDispatchMangling(Name);
1330 }
1331
getCPUSpecificCPUDispatchFeatures(StringRef Name,llvm::SmallVectorImpl<StringRef> & Features) const1332 void X86TargetInfo::getCPUSpecificCPUDispatchFeatures(
1333 StringRef Name, llvm::SmallVectorImpl<StringRef> &Features) const {
1334 SmallVector<StringRef, 32> TargetCPUFeatures;
1335 llvm::X86::getFeaturesForCPU(Name, TargetCPUFeatures, true);
1336 for (auto &F : TargetCPUFeatures)
1337 Features.push_back(F);
1338 }
1339
1340 // We can't use a generic validation scheme for the cpus accepted here
1341 // versus subtarget cpus accepted in the target attribute because the
1342 // variables intitialized by the runtime only support the below currently
1343 // rather than the full range of cpus.
validateCpuIs(StringRef FeatureStr) const1344 bool X86TargetInfo::validateCpuIs(StringRef FeatureStr) const {
1345 return llvm::StringSwitch<bool>(FeatureStr)
1346 #define X86_VENDOR(ENUM, STRING) .Case(STRING, true)
1347 #define X86_CPU_TYPE_ALIAS(ENUM, ALIAS) .Case(ALIAS, true)
1348 #define X86_CPU_TYPE(ENUM, STR) .Case(STR, true)
1349 #define X86_CPU_SUBTYPE_ALIAS(ENUM, ALIAS) .Case(ALIAS, true)
1350 #define X86_CPU_SUBTYPE(ENUM, STR) .Case(STR, true)
1351 #include "llvm/TargetParser/X86TargetParser.def"
1352 .Default(false);
1353 }
1354
matchAsmCCConstraint(const char * Name)1355 static unsigned matchAsmCCConstraint(const char *Name) {
1356 auto RV = llvm::StringSwitch<unsigned>(Name)
1357 .Case("@cca", 4)
1358 .Case("@ccae", 5)
1359 .Case("@ccb", 4)
1360 .Case("@ccbe", 5)
1361 .Case("@ccc", 4)
1362 .Case("@cce", 4)
1363 .Case("@ccz", 4)
1364 .Case("@ccg", 4)
1365 .Case("@ccge", 5)
1366 .Case("@ccl", 4)
1367 .Case("@ccle", 5)
1368 .Case("@ccna", 5)
1369 .Case("@ccnae", 6)
1370 .Case("@ccnb", 5)
1371 .Case("@ccnbe", 6)
1372 .Case("@ccnc", 5)
1373 .Case("@ccne", 5)
1374 .Case("@ccnz", 5)
1375 .Case("@ccng", 5)
1376 .Case("@ccnge", 6)
1377 .Case("@ccnl", 5)
1378 .Case("@ccnle", 6)
1379 .Case("@ccno", 5)
1380 .Case("@ccnp", 5)
1381 .Case("@ccns", 5)
1382 .Case("@cco", 4)
1383 .Case("@ccp", 4)
1384 .Case("@ccs", 4)
1385 .Default(0);
1386 return RV;
1387 }
1388
validateAsmConstraint(const char * & Name,TargetInfo::ConstraintInfo & Info) const1389 bool X86TargetInfo::validateAsmConstraint(
1390 const char *&Name, TargetInfo::ConstraintInfo &Info) const {
1391 switch (*Name) {
1392 default:
1393 return false;
1394 // Constant constraints.
1395 case 'e': // 32-bit signed integer constant for use with sign-extending x86_64
1396 // instructions.
1397 case 'Z': // 32-bit unsigned integer constant for use with zero-extending
1398 // x86_64 instructions.
1399 case 's':
1400 Info.setRequiresImmediate();
1401 return true;
1402 case 'I':
1403 Info.setRequiresImmediate(0, 31);
1404 return true;
1405 case 'J':
1406 Info.setRequiresImmediate(0, 63);
1407 return true;
1408 case 'K':
1409 Info.setRequiresImmediate(-128, 127);
1410 return true;
1411 case 'L':
1412 Info.setRequiresImmediate({int(0xff), int(0xffff), int(0xffffffff)});
1413 return true;
1414 case 'M':
1415 Info.setRequiresImmediate(0, 3);
1416 return true;
1417 case 'N':
1418 Info.setRequiresImmediate(0, 255);
1419 return true;
1420 case 'O':
1421 Info.setRequiresImmediate(0, 127);
1422 return true;
1423 case 'W':
1424 switch (*++Name) {
1425 default:
1426 return false;
1427 case 's':
1428 Info.setAllowsRegister();
1429 return true;
1430 }
1431 // Register constraints.
1432 case 'Y': // 'Y' is the first character for several 2-character constraints.
1433 // Shift the pointer to the second character of the constraint.
1434 Name++;
1435 switch (*Name) {
1436 default:
1437 return false;
1438 case 'z': // First SSE register.
1439 case '2':
1440 case 't': // Any SSE register, when SSE2 is enabled.
1441 case 'i': // Any SSE register, when SSE2 and inter-unit moves enabled.
1442 case 'm': // Any MMX register, when inter-unit moves enabled.
1443 case 'k': // AVX512 arch mask registers: k1-k7.
1444 Info.setAllowsRegister();
1445 return true;
1446 }
1447 case 'f': // Any x87 floating point stack register.
1448 // Constraint 'f' cannot be used for output operands.
1449 if (Info.ConstraintStr[0] == '=')
1450 return false;
1451 Info.setAllowsRegister();
1452 return true;
1453 case 'a': // eax.
1454 case 'b': // ebx.
1455 case 'c': // ecx.
1456 case 'd': // edx.
1457 case 'S': // esi.
1458 case 'D': // edi.
1459 case 'A': // edx:eax.
1460 case 't': // Top of floating point stack.
1461 case 'u': // Second from top of floating point stack.
1462 case 'q': // Any register accessible as [r]l: a, b, c, and d.
1463 case 'y': // Any MMX register.
1464 case 'v': // Any {X,Y,Z}MM register (Arch & context dependent)
1465 case 'x': // Any SSE register.
1466 case 'k': // Any AVX512 mask register (same as Yk, additionally allows k0
1467 // for intermideate k reg operations).
1468 case 'Q': // Any register accessible as [r]h: a, b, c, and d.
1469 case 'R': // "Legacy" registers: ax, bx, cx, dx, di, si, sp, bp.
1470 case 'l': // "Index" registers: any general register that can be used as an
1471 // index in a base+index memory access.
1472 Info.setAllowsRegister();
1473 return true;
1474 // Floating point constant constraints.
1475 case 'C': // SSE floating point constant.
1476 case 'G': // x87 floating point constant.
1477 return true;
1478 case 'j':
1479 Name++;
1480 switch (*Name) {
1481 default:
1482 return false;
1483 case 'r':
1484 Info.setAllowsRegister();
1485 return true;
1486 case 'R':
1487 Info.setAllowsRegister();
1488 return true;
1489 }
1490 case '@':
1491 // CC condition changes.
1492 if (auto Len = matchAsmCCConstraint(Name)) {
1493 Name += Len - 1;
1494 Info.setAllowsRegister();
1495 return true;
1496 }
1497 return false;
1498 }
1499 }
1500
1501 // Below is based on the following information:
1502 // +------------------------------------+-------------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------+
1503 // | Processor Name | Cache Line Size (Bytes) | Source |
1504 // +------------------------------------+-------------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------+
1505 // | i386 | 64 | https://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-optimization-manual.pdf |
1506 // | i486 | 16 | "four doublewords" (doubleword = 32 bits, 4 bits * 32 bits = 16 bytes) https://en.wikichip.org/w/images/d/d3/i486_MICROPROCESSOR_HARDWARE_REFERENCE_MANUAL_%281990%29.pdf and http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.126.4216&rep=rep1&type=pdf (page 29) |
1507 // | i586/Pentium MMX | 32 | https://www.7-cpu.com/cpu/P-MMX.html |
1508 // | i686/Pentium | 32 | https://www.7-cpu.com/cpu/P6.html |
1509 // | Netburst/Pentium4 | 64 | https://www.7-cpu.com/cpu/P4-180.html |
1510 // | Atom | 64 | https://www.7-cpu.com/cpu/Atom.html |
1511 // | Westmere | 64 | https://en.wikichip.org/wiki/intel/microarchitectures/sandy_bridge_(client) "Cache Architecture" |
1512 // | Sandy Bridge | 64 | https://en.wikipedia.org/wiki/Sandy_Bridge and https://www.7-cpu.com/cpu/SandyBridge.html |
1513 // | Ivy Bridge | 64 | https://blog.stuffedcow.net/2013/01/ivb-cache-replacement/ and https://www.7-cpu.com/cpu/IvyBridge.html |
1514 // | Haswell | 64 | https://www.7-cpu.com/cpu/Haswell.html |
1515 // | Broadwell | 64 | https://www.7-cpu.com/cpu/Broadwell.html |
1516 // | Skylake (including skylake-avx512) | 64 | https://www.nas.nasa.gov/hecc/support/kb/skylake-processors_550.html "Cache Hierarchy" |
1517 // | Cascade Lake | 64 | https://www.nas.nasa.gov/hecc/support/kb/cascade-lake-processors_579.html "Cache Hierarchy" |
1518 // | Skylake | 64 | https://en.wikichip.org/wiki/intel/microarchitectures/kaby_lake "Memory Hierarchy" |
1519 // | Ice Lake | 64 | https://www.7-cpu.com/cpu/Ice_Lake.html |
1520 // | Knights Landing | 64 | https://software.intel.com/en-us/articles/intel-xeon-phi-processor-7200-family-memory-management-optimizations "The Intel® Xeon Phi™ Processor Architecture" |
1521 // | Knights Mill | 64 | https://software.intel.com/sites/default/files/managed/9e/bc/64-ia-32-architectures-optimization-manual.pdf?countrylabel=Colombia "2.5.5.2 L1 DCache " |
1522 // +------------------------------------+-------------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------+
getCPUCacheLineSize() const1523 std::optional<unsigned> X86TargetInfo::getCPUCacheLineSize() const {
1524 using namespace llvm::X86;
1525 switch (CPU) {
1526 // i386
1527 case CK_i386:
1528 // i486
1529 case CK_i486:
1530 case CK_WinChipC6:
1531 case CK_WinChip2:
1532 case CK_C3:
1533 // Lakemont
1534 case CK_Lakemont:
1535 return 16;
1536
1537 // i586
1538 case CK_i586:
1539 case CK_Pentium:
1540 case CK_PentiumMMX:
1541 // i686
1542 case CK_PentiumPro:
1543 case CK_i686:
1544 case CK_Pentium2:
1545 case CK_Pentium3:
1546 case CK_PentiumM:
1547 case CK_C3_2:
1548 // K6
1549 case CK_K6:
1550 case CK_K6_2:
1551 case CK_K6_3:
1552 // Geode
1553 case CK_Geode:
1554 return 32;
1555
1556 // Netburst
1557 case CK_Pentium4:
1558 case CK_Prescott:
1559 case CK_Nocona:
1560 // Atom
1561 case CK_Bonnell:
1562 case CK_Silvermont:
1563 case CK_Goldmont:
1564 case CK_GoldmontPlus:
1565 case CK_Tremont:
1566 case CK_Gracemont:
1567
1568 case CK_Westmere:
1569 case CK_SandyBridge:
1570 case CK_IvyBridge:
1571 case CK_Haswell:
1572 case CK_Broadwell:
1573 case CK_SkylakeClient:
1574 case CK_SkylakeServer:
1575 case CK_Cascadelake:
1576 case CK_Nehalem:
1577 case CK_Cooperlake:
1578 case CK_Cannonlake:
1579 case CK_Tigerlake:
1580 case CK_SapphireRapids:
1581 case CK_IcelakeClient:
1582 case CK_Rocketlake:
1583 case CK_IcelakeServer:
1584 case CK_Alderlake:
1585 case CK_Raptorlake:
1586 case CK_Meteorlake:
1587 case CK_Arrowlake:
1588 case CK_ArrowlakeS:
1589 case CK_Lunarlake:
1590 case CK_Pantherlake:
1591 case CK_Sierraforest:
1592 case CK_Grandridge:
1593 case CK_Graniterapids:
1594 case CK_GraniterapidsD:
1595 case CK_Emeraldrapids:
1596 case CK_Clearwaterforest:
1597 case CK_KNL:
1598 case CK_KNM:
1599 // K7
1600 case CK_Athlon:
1601 case CK_AthlonXP:
1602 // K8
1603 case CK_K8:
1604 case CK_K8SSE3:
1605 case CK_AMDFAM10:
1606 // Bobcat
1607 case CK_BTVER1:
1608 case CK_BTVER2:
1609 // Bulldozer
1610 case CK_BDVER1:
1611 case CK_BDVER2:
1612 case CK_BDVER3:
1613 case CK_BDVER4:
1614 // Zen
1615 case CK_ZNVER1:
1616 case CK_ZNVER2:
1617 case CK_ZNVER3:
1618 case CK_ZNVER4:
1619 case CK_ZNVER5:
1620 // Deprecated
1621 case CK_x86_64:
1622 case CK_x86_64_v2:
1623 case CK_x86_64_v3:
1624 case CK_x86_64_v4:
1625 case CK_Yonah:
1626 case CK_Penryn:
1627 case CK_Core2:
1628 return 64;
1629
1630 // The following currently have unknown cache line sizes (but they are probably all 64):
1631 // Core
1632 case CK_None:
1633 return std::nullopt;
1634 }
1635 llvm_unreachable("Unknown CPU kind");
1636 }
1637
validateOutputSize(const llvm::StringMap<bool> & FeatureMap,StringRef Constraint,unsigned Size) const1638 bool X86TargetInfo::validateOutputSize(const llvm::StringMap<bool> &FeatureMap,
1639 StringRef Constraint,
1640 unsigned Size) const {
1641 // Strip off constraint modifiers.
1642 Constraint = Constraint.ltrim("=+&");
1643
1644 return validateOperandSize(FeatureMap, Constraint, Size);
1645 }
1646
validateInputSize(const llvm::StringMap<bool> & FeatureMap,StringRef Constraint,unsigned Size) const1647 bool X86TargetInfo::validateInputSize(const llvm::StringMap<bool> &FeatureMap,
1648 StringRef Constraint,
1649 unsigned Size) const {
1650 return validateOperandSize(FeatureMap, Constraint, Size);
1651 }
1652
validateOperandSize(const llvm::StringMap<bool> & FeatureMap,StringRef Constraint,unsigned Size) const1653 bool X86TargetInfo::validateOperandSize(const llvm::StringMap<bool> &FeatureMap,
1654 StringRef Constraint,
1655 unsigned Size) const {
1656 switch (Constraint[0]) {
1657 default:
1658 break;
1659 case 'k':
1660 // Registers k0-k7 (AVX512) size limit is 64 bit.
1661 case 'y':
1662 return Size <= 64;
1663 case 'f':
1664 case 't':
1665 case 'u':
1666 return Size <= 128;
1667 case 'Y':
1668 // 'Y' is the first character for several 2-character constraints.
1669 switch (Constraint[1]) {
1670 default:
1671 return false;
1672 case 'm':
1673 // 'Ym' is synonymous with 'y'.
1674 case 'k':
1675 return Size <= 64;
1676 case 'z':
1677 // XMM0/YMM/ZMM0
1678 if (hasFeatureEnabled(FeatureMap, "avx512f") &&
1679 hasFeatureEnabled(FeatureMap, "evex512"))
1680 // ZMM0 can be used if target supports AVX512F and EVEX512 is set.
1681 return Size <= 512U;
1682 else if (hasFeatureEnabled(FeatureMap, "avx"))
1683 // YMM0 can be used if target supports AVX.
1684 return Size <= 256U;
1685 else if (hasFeatureEnabled(FeatureMap, "sse"))
1686 return Size <= 128U;
1687 return false;
1688 case 'i':
1689 case 't':
1690 case '2':
1691 // 'Yi','Yt','Y2' are synonymous with 'x' when SSE2 is enabled.
1692 if (SSELevel < SSE2)
1693 return false;
1694 break;
1695 }
1696 break;
1697 case 'v':
1698 case 'x':
1699 if (hasFeatureEnabled(FeatureMap, "avx512f") &&
1700 hasFeatureEnabled(FeatureMap, "evex512"))
1701 // 512-bit zmm registers can be used if target supports AVX512F and
1702 // EVEX512 is set.
1703 return Size <= 512U;
1704 else if (hasFeatureEnabled(FeatureMap, "avx"))
1705 // 256-bit ymm registers can be used if target supports AVX.
1706 return Size <= 256U;
1707 return Size <= 128U;
1708
1709 }
1710
1711 return true;
1712 }
1713
convertConstraint(const char * & Constraint) const1714 std::string X86TargetInfo::convertConstraint(const char *&Constraint) const {
1715 switch (*Constraint) {
1716 case '@':
1717 if (auto Len = matchAsmCCConstraint(Constraint)) {
1718 std::string Converted = "{" + std::string(Constraint, Len) + "}";
1719 Constraint += Len - 1;
1720 return Converted;
1721 }
1722 return std::string(1, *Constraint);
1723 case 'a':
1724 return std::string("{ax}");
1725 case 'b':
1726 return std::string("{bx}");
1727 case 'c':
1728 return std::string("{cx}");
1729 case 'd':
1730 return std::string("{dx}");
1731 case 'S':
1732 return std::string("{si}");
1733 case 'D':
1734 return std::string("{di}");
1735 case 'p': // Keep 'p' constraint (address).
1736 return std::string("p");
1737 case 't': // top of floating point stack.
1738 return std::string("{st}");
1739 case 'u': // second from top of floating point stack.
1740 return std::string("{st(1)}"); // second from top of floating point stack.
1741 case 'W':
1742 assert(Constraint[1] == 's');
1743 return '^' + std::string(Constraint++, 2);
1744 case 'Y':
1745 switch (Constraint[1]) {
1746 default:
1747 // Break from inner switch and fall through (copy single char),
1748 // continue parsing after copying the current constraint into
1749 // the return string.
1750 break;
1751 case 'k':
1752 case 'm':
1753 case 'i':
1754 case 't':
1755 case 'z':
1756 case '2':
1757 // "^" hints llvm that this is a 2 letter constraint.
1758 // "Constraint++" is used to promote the string iterator
1759 // to the next constraint.
1760 return std::string("^") + std::string(Constraint++, 2);
1761 }
1762 [[fallthrough]];
1763 case 'j':
1764 switch (Constraint[1]) {
1765 default:
1766 // Break from inner switch and fall through (copy single char),
1767 // continue parsing after copying the current constraint into
1768 // the return string.
1769 break;
1770 case 'r':
1771 case 'R':
1772 // "^" hints llvm that this is a 2 letter constraint.
1773 // "Constraint++" is used to promote the string iterator
1774 // to the next constraint.
1775 return std::string("^") + std::string(Constraint++, 2);
1776 }
1777 [[fallthrough]];
1778 default:
1779 return std::string(1, *Constraint);
1780 }
1781 }
1782
fillValidCPUList(SmallVectorImpl<StringRef> & Values) const1783 void X86TargetInfo::fillValidCPUList(SmallVectorImpl<StringRef> &Values) const {
1784 bool Only64Bit = getTriple().getArch() != llvm::Triple::x86;
1785 llvm::X86::fillValidCPUArchList(Values, Only64Bit);
1786 }
1787
fillValidTuneCPUList(SmallVectorImpl<StringRef> & Values) const1788 void X86TargetInfo::fillValidTuneCPUList(SmallVectorImpl<StringRef> &Values) const {
1789 llvm::X86::fillValidTuneCPUList(Values);
1790 }
1791
getGCCRegNames() const1792 ArrayRef<const char *> X86TargetInfo::getGCCRegNames() const {
1793 return llvm::ArrayRef(GCCRegNames);
1794 }
1795
getGCCAddlRegNames() const1796 ArrayRef<TargetInfo::AddlRegName> X86TargetInfo::getGCCAddlRegNames() const {
1797 return llvm::ArrayRef(AddlRegNames);
1798 }
1799
getTargetBuiltins() const1800 ArrayRef<Builtin::Info> X86_32TargetInfo::getTargetBuiltins() const {
1801 return llvm::ArrayRef(BuiltinInfoX86, clang::X86::LastX86CommonBuiltin -
1802 Builtin::FirstTSBuiltin + 1);
1803 }
1804
getTargetBuiltins() const1805 ArrayRef<Builtin::Info> X86_64TargetInfo::getTargetBuiltins() const {
1806 return llvm::ArrayRef(BuiltinInfoX86,
1807 X86::LastTSBuiltin - Builtin::FirstTSBuiltin);
1808 }
1809