1 //===-- X86Subtarget.h - Define Subtarget for the X86 ----------*- C++ -*--===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file declares the X86 specific subclass of TargetSubtargetInfo.
12 //===----------------------------------------------------------------------===//
14 #ifndef X86SUBTARGET_H
15 #define X86SUBTARGET_H
17 #include "llvm/ADT/Triple.h"
18 #include "llvm/IR/CallingConv.h"
19 #include "llvm/Target/TargetSubtargetInfo.h"
22 #define GET_SUBTARGETINFO_HEADER
23 #include "X86GenSubtargetInfo.inc"
30 /// PICStyles - The X86 backend supports a number of different styles of PIC.
34 StubPIC, // Used on i386-darwin in -fPIC mode.
35 StubDynamicNoPIC, // Used on i386-darwin in -mdynamic-no-pic mode.
36 GOT, // Used on many 32-bit unices in -fPIC mode.
37 RIPRel, // Used on X86-64 when not in -static mode.
38 None // Set when in -static mode (not PIC or DynamicNoPIC mode).
42 class X86Subtarget final : public X86GenSubtargetInfo {
45 NoMMXSSE, MMX, SSE1, SSE2, SSE3, SSSE3, SSE41, SSE42, AVX, AVX2, AVX512F
49 NoThreeDNow, ThreeDNow, ThreeDNowA
52 enum X86ProcFamilyEnum {
53 Others, IntelAtom, IntelSLM
56 /// X86ProcFamily - X86 processor family: Intel Atom, and others
57 X86ProcFamilyEnum X86ProcFamily;
59 /// PICStyle - Which PIC style to use
61 PICStyles::Style PICStyle;
63 /// X86SSELevel - MMX, SSE1, SSE2, SSE3, SSSE3, SSE41, SSE42, or
65 X86SSEEnum X86SSELevel;
67 /// X863DNowLevel - 3DNow or 3DNow Athlon, or none supported.
69 X863DNowEnum X863DNowLevel;
71 /// HasCMov - True if this processor has conditional move instructions
72 /// (generally pentium pro+).
75 /// HasX86_64 - True if the processor supports X86-64 instructions.
79 /// HasPOPCNT - True if the processor supports POPCNT.
82 /// HasSSE4A - True if the processor supports SSE4A instructions.
85 /// HasAES - Target has AES instructions
88 /// HasPCLMUL - Target has carry-less multiplication
91 /// HasFMA - Target has 3-operand fused multiply-add
94 /// HasFMA4 - Target has 4-operand fused multiply-add
97 /// HasXOP - Target has XOP instructions
100 /// HasTBM - Target has TBM instructions.
103 /// HasMOVBE - True if the processor has the MOVBE instruction.
106 /// HasRDRAND - True if the processor has the RDRAND instruction.
109 /// HasF16C - Processor has 16-bit floating point conversion instructions.
112 /// HasFSGSBase - Processor has FS/GS base insturctions.
115 /// HasLZCNT - Processor has LZCNT instruction.
118 /// HasBMI - Processor has BMI1 instructions.
121 /// HasBMI2 - Processor has BMI2 instructions.
124 /// HasRTM - Processor has RTM instructions.
127 /// HasHLE - Processor has HLE.
130 /// HasADX - Processor has ADX instructions.
133 /// HasSHA - Processor has SHA instructions.
136 /// HasPRFCHW - Processor has PRFCHW instructions.
139 /// HasRDSEED - Processor has RDSEED instructions.
142 /// IsBTMemSlow - True if BT (bit test) of memory instructions are slow.
145 /// IsSHLDSlow - True if SHLD instructions are slow.
148 /// IsUAMemFast - True if unaligned memory access is fast.
151 /// HasVectorUAMem - True if SIMD operations can have unaligned memory
152 /// operands. This may require setting a feature bit in the processor.
155 /// HasCmpxchg16b - True if this processor has the CMPXCHG16B instruction;
156 /// this is true for most x86-64 chips, but not the first AMD chips.
159 /// UseLeaForSP - True if the LEA instruction should be used for adjusting
160 /// the stack pointer. This is an optimization for Intel Atom processors.
163 /// HasSlowDivide - True if smaller divides are significantly faster than
164 /// full divides and should be used when possible.
167 /// PostRAScheduler - True if using post-register-allocation scheduler.
168 bool PostRAScheduler;
170 /// PadShortFunctions - True if the short functions should be padded to prevent
171 /// a stall when returning too early.
172 bool PadShortFunctions;
174 /// CallRegIndirect - True if the Calls with memory reference should be converted
175 /// to a register-based indirect call.
176 bool CallRegIndirect;
177 /// LEAUsesAG - True if the LEA instruction inputs have to be ready at
178 /// address generation (AG) time.
181 /// Processor has AVX-512 PreFetch Instructions
184 /// Processor has AVX-512 Exponential and Reciprocal Instructions
187 /// Processor has AVX-512 Conflict Detection Instructions
190 /// stackAlignment - The minimum alignment known to hold of the stack frame on
191 /// entry to the function and which must be maintained by every function.
192 unsigned stackAlignment;
194 /// Max. memset / memcpy size that is turned into rep/movs, rep/stos ops.
196 unsigned MaxInlineSizeThreshold;
198 /// TargetTriple - What processor and OS we're targeting.
201 /// Instruction itineraries for scheduling
202 InstrItineraryData InstrItins;
205 /// StackAlignOverride - Override the stack alignment.
206 unsigned StackAlignOverride;
208 /// In64BitMode - True if compiling for 64-bit, false for 16-bit or 32-bit.
211 /// In32BitMode - True if compiling for 32-bit, false for 16-bit or 64-bit.
214 /// In16BitMode - True if compiling for 16-bit, false for 32-bit or 64-bit.
218 /// This constructor initializes the data members to match that
219 /// of the specified triple.
221 X86Subtarget(const std::string &TT, const std::string &CPU,
222 const std::string &FS,
223 unsigned StackAlignOverride);
225 /// getStackAlignment - Returns the minimum alignment known to hold of the
226 /// stack frame on entry to the function and which must be maintained by every
227 /// function for this subtarget.
228 unsigned getStackAlignment() const { return stackAlignment; }
230 /// getMaxInlineSizeThreshold - Returns the maximum memset / memcpy size
231 /// that still makes it profitable to inline the call.
232 unsigned getMaxInlineSizeThreshold() const { return MaxInlineSizeThreshold; }
234 /// ParseSubtargetFeatures - Parses features string setting specified
235 /// subtarget options. Definition of function is auto generated by tblgen.
236 void ParseSubtargetFeatures(StringRef CPU, StringRef FS);
238 /// \brief Reset the features for the X86 target.
239 void resetSubtargetFeatures(const MachineFunction *MF) override;
241 void initializeEnvironment();
242 void resetSubtargetFeatures(StringRef CPU, StringRef FS);
244 /// Is this x86_64? (disregarding specific ABI / programming model)
245 bool is64Bit() const {
249 bool is32Bit() const {
253 bool is16Bit() const {
257 /// Is this x86_64 with the ILP32 programming model (x32 ABI)?
258 bool isTarget64BitILP32() const {
259 return In64BitMode && (TargetTriple.getEnvironment() == Triple::GNUX32 ||
260 TargetTriple.getOS() == Triple::NaCl);
263 /// Is this x86_64 with the LP64 programming model (standard AMD64, no x32)?
264 bool isTarget64BitLP64() const {
265 return In64BitMode && (TargetTriple.getEnvironment() != Triple::GNUX32);
268 PICStyles::Style getPICStyle() const { return PICStyle; }
269 void setPICStyle(PICStyles::Style Style) { PICStyle = Style; }
271 bool hasCMov() const { return HasCMov; }
272 bool hasMMX() const { return X86SSELevel >= MMX; }
273 bool hasSSE1() const { return X86SSELevel >= SSE1; }
274 bool hasSSE2() const { return X86SSELevel >= SSE2; }
275 bool hasSSE3() const { return X86SSELevel >= SSE3; }
276 bool hasSSSE3() const { return X86SSELevel >= SSSE3; }
277 bool hasSSE41() const { return X86SSELevel >= SSE41; }
278 bool hasSSE42() const { return X86SSELevel >= SSE42; }
279 bool hasAVX() const { return X86SSELevel >= AVX; }
280 bool hasAVX2() const { return X86SSELevel >= AVX2; }
281 bool hasAVX512() const { return X86SSELevel >= AVX512F; }
282 bool hasFp256() const { return hasAVX(); }
283 bool hasInt256() const { return hasAVX2(); }
284 bool hasSSE4A() const { return HasSSE4A; }
285 bool has3DNow() const { return X863DNowLevel >= ThreeDNow; }
286 bool has3DNowA() const { return X863DNowLevel >= ThreeDNowA; }
287 bool hasPOPCNT() const { return HasPOPCNT; }
288 bool hasAES() const { return HasAES; }
289 bool hasPCLMUL() const { return HasPCLMUL; }
290 bool hasFMA() const { return HasFMA; }
291 // FIXME: Favor FMA when both are enabled. Is this the right thing to do?
292 bool hasFMA4() const { return HasFMA4 && !HasFMA; }
293 bool hasXOP() const { return HasXOP; }
294 bool hasTBM() const { return HasTBM; }
295 bool hasMOVBE() const { return HasMOVBE; }
296 bool hasRDRAND() const { return HasRDRAND; }
297 bool hasF16C() const { return HasF16C; }
298 bool hasFSGSBase() const { return HasFSGSBase; }
299 bool hasLZCNT() const { return HasLZCNT; }
300 bool hasBMI() const { return HasBMI; }
301 bool hasBMI2() const { return HasBMI2; }
302 bool hasRTM() const { return HasRTM; }
303 bool hasHLE() const { return HasHLE; }
304 bool hasADX() const { return HasADX; }
305 bool hasSHA() const { return HasSHA; }
306 bool hasPRFCHW() const { return HasPRFCHW; }
307 bool hasRDSEED() const { return HasRDSEED; }
308 bool isBTMemSlow() const { return IsBTMemSlow; }
309 bool isSHLDSlow() const { return IsSHLDSlow; }
310 bool isUnalignedMemAccessFast() const { return IsUAMemFast; }
311 bool hasVectorUAMem() const { return HasVectorUAMem; }
312 bool hasCmpxchg16b() const { return HasCmpxchg16b; }
313 bool useLeaForSP() const { return UseLeaForSP; }
314 bool hasSlowDivide() const { return HasSlowDivide; }
315 bool padShortFunctions() const { return PadShortFunctions; }
316 bool callRegIndirect() const { return CallRegIndirect; }
317 bool LEAusesAG() const { return LEAUsesAG; }
318 bool hasCDI() const { return HasCDI; }
319 bool hasPFI() const { return HasPFI; }
320 bool hasERI() const { return HasERI; }
322 bool isAtom() const { return X86ProcFamily == IntelAtom; }
324 const Triple &getTargetTriple() const { return TargetTriple; }
326 bool isTargetDarwin() const { return TargetTriple.isOSDarwin(); }
327 bool isTargetFreeBSD() const {
328 return TargetTriple.getOS() == Triple::FreeBSD;
330 bool isTargetSolaris() const {
331 return TargetTriple.getOS() == Triple::Solaris;
334 bool isTargetELF() const { return TargetTriple.isOSBinFormatELF(); }
335 bool isTargetCOFF() const { return TargetTriple.isOSBinFormatCOFF(); }
336 bool isTargetMacho() const { return TargetTriple.isOSBinFormatMachO(); }
338 bool isTargetLinux() const { return TargetTriple.isOSLinux(); }
339 bool isTargetNaCl() const { return TargetTriple.isOSNaCl(); }
340 bool isTargetNaCl32() const { return isTargetNaCl() && !is64Bit(); }
341 bool isTargetNaCl64() const { return isTargetNaCl() && is64Bit(); }
343 bool isTargetWindowsMSVC() const {
344 return TargetTriple.isWindowsMSVCEnvironment();
347 bool isTargetKnownWindowsMSVC() const {
348 return TargetTriple.isKnownWindowsMSVCEnvironment();
351 bool isTargetWindowsCygwin() const {
352 return TargetTriple.isWindowsCygwinEnvironment();
355 bool isTargetWindowsGNU() const {
356 return TargetTriple.isWindowsGNUEnvironment();
359 bool isTargetCygMing() const { return TargetTriple.isOSCygMing(); }
361 bool isOSWindows() const { return TargetTriple.isOSWindows(); }
363 bool isTargetWin64() const {
364 return In64BitMode && TargetTriple.isOSWindows();
367 bool isTargetWin32() const {
368 return !In64BitMode && (isTargetCygMing() || isTargetKnownWindowsMSVC());
371 bool isPICStyleSet() const { return PICStyle != PICStyles::None; }
372 bool isPICStyleGOT() const { return PICStyle == PICStyles::GOT; }
373 bool isPICStyleRIPRel() const { return PICStyle == PICStyles::RIPRel; }
375 bool isPICStyleStubPIC() const {
376 return PICStyle == PICStyles::StubPIC;
379 bool isPICStyleStubNoDynamic() const {
380 return PICStyle == PICStyles::StubDynamicNoPIC;
382 bool isPICStyleStubAny() const {
383 return PICStyle == PICStyles::StubDynamicNoPIC ||
384 PICStyle == PICStyles::StubPIC;
387 bool isCallingConvWin64(CallingConv::ID CC) const {
388 return (isTargetWin64() && CC != CallingConv::X86_64_SysV) ||
389 CC == CallingConv::X86_64_Win64;
392 /// ClassifyGlobalReference - Classify a global variable reference for the
393 /// current subtarget according to how we should reference it in a non-pcrel
395 unsigned char ClassifyGlobalReference(const GlobalValue *GV,
396 const TargetMachine &TM)const;
398 /// ClassifyBlockAddressReference - Classify a blockaddress reference for the
399 /// current subtarget according to how we should reference it in a non-pcrel
401 unsigned char ClassifyBlockAddressReference() const;
403 /// IsLegalToCallImmediateAddr - Return true if the subtarget allows calls
404 /// to immediate address.
405 bool IsLegalToCallImmediateAddr(const TargetMachine &TM) const;
407 /// This function returns the name of a function which has an interface
408 /// like the non-standard bzero function, if such a function exists on
409 /// the current subtarget and it is considered prefereable over
410 /// memset with zero passed as the second argument. Otherwise it
412 const char *getBZeroEntry() const;
414 /// This function returns true if the target has sincos() routine in its
415 /// compiler runtime or math libraries.
416 bool hasSinCos() const;
418 /// Enable the MachineScheduler pass for all X86 subtargets.
419 bool enableMachineScheduler() const override { return true; }
421 /// enablePostRAScheduler - run for Atom optimization.
422 bool enablePostRAScheduler(CodeGenOpt::Level OptLevel,
423 TargetSubtargetInfo::AntiDepBreakMode& Mode,
424 RegClassVector& CriticalPathRCs) const override;
426 bool postRAScheduler() const { return PostRAScheduler; }
428 /// getInstrItins = Return the instruction itineraries based on the
429 /// subtarget selection.
430 const InstrItineraryData &getInstrItineraryData() const { return InstrItins; }
433 } // End llvm namespace