1 //===-- X86.td - Target definition file for the Intel X86 --*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This is a target description file for the Intel i386 architecture, referred
11 // to here as the "X86" architecture.
13 //===----------------------------------------------------------------------===//
15 // Get the target-independent interfaces which we are implementing...
17 include "llvm/Target/Target.td"
19 //===----------------------------------------------------------------------===//
20 // X86 Subtarget state
23 def Mode64Bit : SubtargetFeature<"64bit-mode", "In64BitMode", "true",
24 "64-bit mode (x86_64)">;
25 def Mode32Bit : SubtargetFeature<"32bit-mode", "In32BitMode", "true",
26 "32-bit mode (80386)">;
27 def Mode16Bit : SubtargetFeature<"16bit-mode", "In16BitMode", "true",
28 "16-bit mode (i8086)">;
30 //===----------------------------------------------------------------------===//
31 // X86 Subtarget features
32 //===----------------------------------------------------------------------===//
34 def FeatureCMOV : SubtargetFeature<"cmov","HasCMov", "true",
35 "Enable conditional move instructions">;
37 def FeaturePOPCNT : SubtargetFeature<"popcnt", "HasPOPCNT", "true",
38 "Support POPCNT instruction">;
41 def FeatureMMX : SubtargetFeature<"mmx","X86SSELevel", "MMX",
42 "Enable MMX instructions">;
43 def FeatureSSE1 : SubtargetFeature<"sse", "X86SSELevel", "SSE1",
44 "Enable SSE instructions",
45 // SSE codegen depends on cmovs, and all
46 // SSE1+ processors support them.
47 [FeatureMMX, FeatureCMOV]>;
48 def FeatureSSE2 : SubtargetFeature<"sse2", "X86SSELevel", "SSE2",
49 "Enable SSE2 instructions",
51 def FeatureSSE3 : SubtargetFeature<"sse3", "X86SSELevel", "SSE3",
52 "Enable SSE3 instructions",
54 def FeatureSSSE3 : SubtargetFeature<"ssse3", "X86SSELevel", "SSSE3",
55 "Enable SSSE3 instructions",
57 def FeatureSSE41 : SubtargetFeature<"sse4.1", "X86SSELevel", "SSE41",
58 "Enable SSE 4.1 instructions",
60 def FeatureSSE42 : SubtargetFeature<"sse4.2", "X86SSELevel", "SSE42",
61 "Enable SSE 4.2 instructions",
63 def Feature3DNow : SubtargetFeature<"3dnow", "X863DNowLevel", "ThreeDNow",
64 "Enable 3DNow! instructions",
66 def Feature3DNowA : SubtargetFeature<"3dnowa", "X863DNowLevel", "ThreeDNowA",
67 "Enable 3DNow! Athlon instructions",
69 // All x86-64 hardware has SSE2, but we don't mark SSE2 as an implied
70 // feature, because SSE2 can be disabled (e.g. for compiling OS kernels)
71 // without disabling 64-bit mode.
72 def Feature64Bit : SubtargetFeature<"64bit", "HasX86_64", "true",
73 "Support 64-bit instructions",
75 def FeatureCMPXCHG16B : SubtargetFeature<"cx16", "HasCmpxchg16b", "true",
76 "64-bit with cmpxchg16b",
78 def FeatureSlowBTMem : SubtargetFeature<"slow-bt-mem", "IsBTMemSlow", "true",
79 "Bit testing of memory is slow">;
80 def FeatureSlowSHLD : SubtargetFeature<"slow-shld", "IsSHLDSlow", "true",
81 "SHLD instruction is slow">;
82 // FIXME: This is a 16-byte (SSE/AVX) feature; we should rename it to make that
83 // explicit. Also, it seems this would be the default state for most chips
84 // going forward, so it would probably be better to negate the logic and
85 // match the 32-byte "slow mem" feature below.
86 def FeatureFastUAMem : SubtargetFeature<"fast-unaligned-mem",
87 "IsUAMemFast", "true",
88 "Fast unaligned memory access">;
89 def FeatureSlowUAMem32 : SubtargetFeature<"slow-unaligned-mem-32",
90 "IsUAMem32Slow", "true",
91 "Slow unaligned 32-byte memory access">;
92 def FeatureSSE4A : SubtargetFeature<"sse4a", "HasSSE4A", "true",
93 "Support SSE 4a instructions",
96 def FeatureAVX : SubtargetFeature<"avx", "X86SSELevel", "AVX",
97 "Enable AVX instructions",
99 def FeatureAVX2 : SubtargetFeature<"avx2", "X86SSELevel", "AVX2",
100 "Enable AVX2 instructions",
102 def FeatureAVX512 : SubtargetFeature<"avx512f", "X86SSELevel", "AVX512F",
103 "Enable AVX-512 instructions",
105 def FeatureERI : SubtargetFeature<"avx512er", "HasERI", "true",
106 "Enable AVX-512 Exponential and Reciprocal Instructions",
108 def FeatureCDI : SubtargetFeature<"avx512cd", "HasCDI", "true",
109 "Enable AVX-512 Conflict Detection Instructions",
111 def FeaturePFI : SubtargetFeature<"avx512pf", "HasPFI", "true",
112 "Enable AVX-512 PreFetch Instructions",
114 def FeatureDQI : SubtargetFeature<"avx512dq", "HasDQI", "true",
115 "Enable AVX-512 Doubleword and Quadword Instructions",
117 def FeatureBWI : SubtargetFeature<"avx512bw", "HasBWI", "true",
118 "Enable AVX-512 Byte and Word Instructions",
120 def FeatureVLX : SubtargetFeature<"avx512vl", "HasVLX", "true",
121 "Enable AVX-512 Vector Length eXtensions",
123 def FeaturePCLMUL : SubtargetFeature<"pclmul", "HasPCLMUL", "true",
124 "Enable packed carry-less multiplication instructions",
126 def FeatureFMA : SubtargetFeature<"fma", "HasFMA", "true",
127 "Enable three-operand fused multiple-add",
129 def FeatureFMA4 : SubtargetFeature<"fma4", "HasFMA4", "true",
130 "Enable four-operand fused multiple-add",
131 [FeatureAVX, FeatureSSE4A]>;
132 def FeatureXOP : SubtargetFeature<"xop", "HasXOP", "true",
133 "Enable XOP instructions",
135 def FeatureSSEUnalignedMem : SubtargetFeature<"sse-unaligned-mem",
136 "HasSSEUnalignedMem", "true",
137 "Allow unaligned memory operands with SSE instructions">;
138 def FeatureAES : SubtargetFeature<"aes", "HasAES", "true",
139 "Enable AES instructions",
141 def FeatureTBM : SubtargetFeature<"tbm", "HasTBM", "true",
142 "Enable TBM instructions">;
143 def FeatureMOVBE : SubtargetFeature<"movbe", "HasMOVBE", "true",
144 "Support MOVBE instruction">;
145 def FeatureRDRAND : SubtargetFeature<"rdrnd", "HasRDRAND", "true",
146 "Support RDRAND instruction">;
147 def FeatureF16C : SubtargetFeature<"f16c", "HasF16C", "true",
148 "Support 16-bit floating point conversion instructions",
150 def FeatureFSGSBase : SubtargetFeature<"fsgsbase", "HasFSGSBase", "true",
151 "Support FS/GS Base instructions">;
152 def FeatureLZCNT : SubtargetFeature<"lzcnt", "HasLZCNT", "true",
153 "Support LZCNT instruction">;
154 def FeatureBMI : SubtargetFeature<"bmi", "HasBMI", "true",
155 "Support BMI instructions">;
156 def FeatureBMI2 : SubtargetFeature<"bmi2", "HasBMI2", "true",
157 "Support BMI2 instructions">;
158 def FeatureRTM : SubtargetFeature<"rtm", "HasRTM", "true",
159 "Support RTM instructions">;
160 def FeatureHLE : SubtargetFeature<"hle", "HasHLE", "true",
162 def FeatureADX : SubtargetFeature<"adx", "HasADX", "true",
163 "Support ADX instructions">;
164 def FeatureSHA : SubtargetFeature<"sha", "HasSHA", "true",
165 "Enable SHA instructions",
167 def FeaturePRFCHW : SubtargetFeature<"prfchw", "HasPRFCHW", "true",
168 "Support PRFCHW instructions">;
169 def FeatureRDSEED : SubtargetFeature<"rdseed", "HasRDSEED", "true",
170 "Support RDSEED instruction">;
171 def FeatureLeaForSP : SubtargetFeature<"lea-sp", "UseLeaForSP", "true",
172 "Use LEA for adjusting the stack pointer">;
173 def FeatureSlowDivide32 : SubtargetFeature<"idivl-to-divb",
174 "HasSlowDivide32", "true",
175 "Use 8-bit divide for positive values less than 256">;
176 def FeatureSlowDivide64 : SubtargetFeature<"idivq-to-divw",
177 "HasSlowDivide64", "true",
178 "Use 16-bit divide for positive values less than 65536">;
179 def FeaturePadShortFunctions : SubtargetFeature<"pad-short-functions",
180 "PadShortFunctions", "true",
181 "Pad short functions">;
182 def FeatureCallRegIndirect : SubtargetFeature<"call-reg-indirect",
183 "CallRegIndirect", "true",
184 "Call register indirect">;
185 def FeatureLEAUsesAG : SubtargetFeature<"lea-uses-ag", "LEAUsesAG", "true",
186 "LEA instruction needs inputs at AG stage">;
187 def FeatureSlowLEA : SubtargetFeature<"slow-lea", "SlowLEA", "true",
188 "LEA instruction with certain arguments is slow">;
189 def FeatureSlowIncDec : SubtargetFeature<"slow-incdec", "SlowIncDec", "true",
190 "INC and DEC instructions are slower than ADD and SUB">;
192 : SubtargetFeature<"soft-float", "UseSoftFloat", "true",
193 "Use software floating point features.">;
195 //===----------------------------------------------------------------------===//
196 // X86 processors supported.
197 //===----------------------------------------------------------------------===//
199 include "X86Schedule.td"
201 def ProcIntelAtom : SubtargetFeature<"atom", "X86ProcFamily", "IntelAtom",
202 "Intel Atom processors">;
203 def ProcIntelSLM : SubtargetFeature<"slm", "X86ProcFamily", "IntelSLM",
204 "Intel Silvermont processors">;
206 class Proc<string Name, list<SubtargetFeature> Features>
207 : ProcessorModel<Name, GenericModel, Features>;
209 def : Proc<"generic", []>;
210 def : Proc<"i386", []>;
211 def : Proc<"i486", []>;
212 def : Proc<"i586", []>;
213 def : Proc<"pentium", []>;
214 def : Proc<"pentium-mmx", [FeatureMMX]>;
215 def : Proc<"i686", []>;
216 def : Proc<"pentiumpro", [FeatureCMOV]>;
217 def : Proc<"pentium2", [FeatureMMX, FeatureCMOV]>;
218 def : Proc<"pentium3", [FeatureSSE1]>;
219 def : Proc<"pentium3m", [FeatureSSE1, FeatureSlowBTMem]>;
220 def : Proc<"pentium-m", [FeatureSSE2, FeatureSlowBTMem]>;
221 def : Proc<"pentium4", [FeatureSSE2]>;
222 def : Proc<"pentium4m", [FeatureSSE2, FeatureSlowBTMem]>;
225 def : ProcessorModel<"yonah", SandyBridgeModel,
226 [FeatureSSE3, FeatureSlowBTMem]>;
229 def : Proc<"prescott", [FeatureSSE3, FeatureSlowBTMem]>;
230 def : Proc<"nocona", [FeatureSSE3, FeatureCMPXCHG16B, FeatureSlowBTMem]>;
232 // Intel Core 2 Solo/Duo.
233 def : ProcessorModel<"core2", SandyBridgeModel,
234 [FeatureSSSE3, FeatureCMPXCHG16B, FeatureSlowBTMem]>;
235 def : ProcessorModel<"penryn", SandyBridgeModel,
236 [FeatureSSE41, FeatureCMPXCHG16B, FeatureSlowBTMem]>;
239 class BonnellProc<string Name> : ProcessorModel<Name, AtomModel, [
248 FeatureCallRegIndirect,
250 FeaturePadShortFunctions
252 def : BonnellProc<"bonnell">;
253 def : BonnellProc<"atom">; // Pin the generic name to the baseline.
255 class SilvermontProc<string Name> : ProcessorModel<Name, SLMModel, [
264 FeatureCallRegIndirect,
271 def : SilvermontProc<"silvermont">;
272 def : SilvermontProc<"slm">; // Legacy alias.
274 // "Arrandale" along with corei3 and corei5
275 class NehalemProc<string Name> : ProcessorModel<Name, SandyBridgeModel, [
282 def : NehalemProc<"nehalem">;
283 def : NehalemProc<"corei7">;
285 // Westmere is a similar machine to nehalem with some additional features.
286 // Westmere is the corei3/i5/i7 path from nehalem to sandybridge
287 class WestmereProc<string Name> : ProcessorModel<Name, SandyBridgeModel, [
296 def : WestmereProc<"westmere">;
298 // SSE is not listed here since llvm treats AVX as a reimplementation of SSE,
299 // rather than a superset.
300 class SandyBridgeProc<string Name> : ProcessorModel<Name, SandyBridgeModel, [
309 def : SandyBridgeProc<"sandybridge">;
310 def : SandyBridgeProc<"corei7-avx">; // Legacy alias.
312 class IvyBridgeProc<string Name> : ProcessorModel<Name, SandyBridgeModel, [
324 def : IvyBridgeProc<"ivybridge">;
325 def : IvyBridgeProc<"core-avx-i">; // Legacy alias.
327 class HaswellProc<string Name> : ProcessorModel<Name, HaswellModel, [
346 def : HaswellProc<"haswell">;
347 def : HaswellProc<"core-avx2">; // Legacy alias.
349 class BroadwellProc<string Name> : ProcessorModel<Name, HaswellModel, [
370 def : BroadwellProc<"broadwell">;
372 // FIXME: define KNL model
373 class KnightsLandingProc<string Name> : ProcessorModel<Name, HaswellModel,
374 [FeatureAVX512, FeatureERI, FeatureCDI, FeaturePFI,
375 FeatureCMPXCHG16B, FeatureFastUAMem, FeaturePOPCNT,
376 FeatureAES, FeaturePCLMUL, FeatureRDRAND, FeatureF16C,
377 FeatureFSGSBase, FeatureMOVBE, FeatureLZCNT, FeatureBMI,
378 FeatureBMI2, FeatureFMA, FeatureRTM, FeatureHLE,
380 def : KnightsLandingProc<"knl">;
382 // FIXME: define SKX model
383 class SkylakeProc<string Name> : ProcessorModel<Name, HaswellModel,
384 [FeatureAVX512, FeatureCDI,
385 FeatureDQI, FeatureBWI, FeatureVLX,
386 FeatureCMPXCHG16B, FeatureFastUAMem, FeaturePOPCNT,
387 FeatureAES, FeaturePCLMUL, FeatureRDRAND, FeatureF16C,
388 FeatureFSGSBase, FeatureMOVBE, FeatureLZCNT, FeatureBMI,
389 FeatureBMI2, FeatureFMA, FeatureRTM, FeatureHLE,
391 def : SkylakeProc<"skylake">;
392 def : SkylakeProc<"skx">; // Legacy alias.
397 def : Proc<"k6", [FeatureMMX]>;
398 def : Proc<"k6-2", [Feature3DNow]>;
399 def : Proc<"k6-3", [Feature3DNow]>;
400 def : Proc<"athlon", [Feature3DNowA, FeatureSlowBTMem,
402 def : Proc<"athlon-tbird", [Feature3DNowA, FeatureSlowBTMem,
404 def : Proc<"athlon-4", [FeatureSSE1, Feature3DNowA, FeatureSlowBTMem,
406 def : Proc<"athlon-xp", [FeatureSSE1, Feature3DNowA, FeatureSlowBTMem,
408 def : Proc<"athlon-mp", [FeatureSSE1, Feature3DNowA, FeatureSlowBTMem,
410 def : Proc<"k8", [FeatureSSE2, Feature3DNowA, Feature64Bit,
411 FeatureSlowBTMem, FeatureSlowSHLD]>;
412 def : Proc<"opteron", [FeatureSSE2, Feature3DNowA, Feature64Bit,
413 FeatureSlowBTMem, FeatureSlowSHLD]>;
414 def : Proc<"athlon64", [FeatureSSE2, Feature3DNowA, Feature64Bit,
415 FeatureSlowBTMem, FeatureSlowSHLD]>;
416 def : Proc<"athlon-fx", [FeatureSSE2, Feature3DNowA, Feature64Bit,
417 FeatureSlowBTMem, FeatureSlowSHLD]>;
418 def : Proc<"k8-sse3", [FeatureSSE3, Feature3DNowA, FeatureCMPXCHG16B,
419 FeatureSlowBTMem, FeatureSlowSHLD]>;
420 def : Proc<"opteron-sse3", [FeatureSSE3, Feature3DNowA, FeatureCMPXCHG16B,
421 FeatureSlowBTMem, FeatureSlowSHLD]>;
422 def : Proc<"athlon64-sse3", [FeatureSSE3, Feature3DNowA, FeatureCMPXCHG16B,
423 FeatureSlowBTMem, FeatureSlowSHLD]>;
424 def : Proc<"amdfam10", [FeatureSSE4A,
425 Feature3DNowA, FeatureCMPXCHG16B, FeatureLZCNT,
426 FeaturePOPCNT, FeatureSlowBTMem,
428 def : Proc<"barcelona", [FeatureSSE4A,
429 Feature3DNowA, FeatureCMPXCHG16B, FeatureLZCNT,
430 FeaturePOPCNT, FeatureSlowBTMem,
433 def : Proc<"btver1", [FeatureSSSE3, FeatureSSE4A, FeatureCMPXCHG16B,
434 FeaturePRFCHW, FeatureLZCNT, FeaturePOPCNT,
438 def : ProcessorModel<"btver2", BtVer2Model,
439 [FeatureAVX, FeatureSSE4A, FeatureCMPXCHG16B,
440 FeaturePRFCHW, FeatureAES, FeaturePCLMUL,
441 FeatureBMI, FeatureF16C, FeatureMOVBE,
442 FeatureLZCNT, FeaturePOPCNT, FeatureFastUAMem,
445 // TODO: We should probably add 'FeatureFastUAMem' to all of the AMD chips.
448 def : Proc<"bdver1", [FeatureXOP, FeatureFMA4, FeatureCMPXCHG16B,
449 FeatureAES, FeaturePRFCHW, FeaturePCLMUL,
450 FeatureAVX, FeatureSSE4A, FeatureLZCNT,
451 FeaturePOPCNT, FeatureSlowSHLD]>;
453 def : Proc<"bdver2", [FeatureXOP, FeatureFMA4, FeatureCMPXCHG16B,
454 FeatureAES, FeaturePRFCHW, FeaturePCLMUL,
455 FeatureAVX, FeatureSSE4A, FeatureF16C,
456 FeatureLZCNT, FeaturePOPCNT, FeatureBMI,
457 FeatureTBM, FeatureFMA, FeatureSlowSHLD]>;
460 def : Proc<"bdver3", [FeatureXOP, FeatureFMA4, FeatureCMPXCHG16B,
461 FeatureAES, FeaturePRFCHW, FeaturePCLMUL,
462 FeatureAVX, FeatureSSE4A, FeatureF16C,
463 FeatureLZCNT, FeaturePOPCNT, FeatureBMI,
464 FeatureTBM, FeatureFMA, FeatureSlowSHLD,
468 def : Proc<"bdver4", [FeatureAVX2, FeatureXOP, FeatureFMA4,
469 FeatureCMPXCHG16B, FeatureAES, FeaturePRFCHW,
470 FeaturePCLMUL, FeatureF16C, FeatureLZCNT,
471 FeaturePOPCNT, FeatureBMI, FeatureBMI2,
472 FeatureTBM, FeatureFMA, FeatureSSE4A,
475 def : Proc<"geode", [Feature3DNowA]>;
477 def : Proc<"winchip-c6", [FeatureMMX]>;
478 def : Proc<"winchip2", [Feature3DNow]>;
479 def : Proc<"c3", [Feature3DNow]>;
480 def : Proc<"c3-2", [FeatureSSE1]>;
482 // We also provide a generic 64-bit specific x86 processor model which tries to
483 // be good for modern chips without enabling instruction set encodings past the
484 // basic SSE2 and 64-bit ones. It disables slow things from any mainstream and
485 // modern 64-bit x86 chip, and enables features that are generally beneficial.
487 // We currently use the Sandy Bridge model as the default scheduling model as
488 // we use it across Nehalem, Westmere, Sandy Bridge, and Ivy Bridge which
489 // covers a huge swath of x86 processors. If there are specific scheduling
490 // knobs which need to be tuned differently for AMD chips, we might consider
491 // forming a common base for them.
492 def : ProcessorModel<"x86-64", SandyBridgeModel,
493 [FeatureSSE2, Feature64Bit, FeatureSlowBTMem,
496 //===----------------------------------------------------------------------===//
497 // Register File Description
498 //===----------------------------------------------------------------------===//
500 include "X86RegisterInfo.td"
502 //===----------------------------------------------------------------------===//
503 // Instruction Descriptions
504 //===----------------------------------------------------------------------===//
506 include "X86InstrInfo.td"
508 def X86InstrInfo : InstrInfo;
510 //===----------------------------------------------------------------------===//
511 // Calling Conventions
512 //===----------------------------------------------------------------------===//
514 include "X86CallingConv.td"
517 //===----------------------------------------------------------------------===//
519 //===----------------------------------------------------------------------===//
521 def ATTAsmParser : AsmParser {
522 string AsmParserClassName = "AsmParser";
525 def ATTAsmParserVariant : AsmParserVariant {
531 // Discard comments in assembly strings.
532 string CommentDelimiter = "#";
534 // Recognize hard coded registers.
535 string RegisterPrefix = "%";
538 def IntelAsmParserVariant : AsmParserVariant {
542 string Name = "intel";
544 // Discard comments in assembly strings.
545 string CommentDelimiter = ";";
547 // Recognize hard coded registers.
548 string RegisterPrefix = "";
551 //===----------------------------------------------------------------------===//
553 //===----------------------------------------------------------------------===//
555 // The X86 target supports two different syntaxes for emitting machine code.
556 // This is controlled by the -x86-asm-syntax={att|intel}
557 def ATTAsmWriter : AsmWriter {
558 string AsmWriterClassName = "ATTInstPrinter";
561 def IntelAsmWriter : AsmWriter {
562 string AsmWriterClassName = "IntelInstPrinter";
567 // Information about the instructions...
568 let InstructionSet = X86InstrInfo;
569 let AssemblyParsers = [ATTAsmParser];
570 let AssemblyParserVariants = [ATTAsmParserVariant, IntelAsmParserVariant];
571 let AssemblyWriters = [ATTAsmWriter, IntelAsmWriter];