1 //===-- X86Subtarget.cpp - X86 Subtarget Information ----------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file implements the X86 specific subclass of TargetSubtargetInfo.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "X86Subtarget.h"
15 #include "X86InstrInfo.h"
16 #include "X86TargetMachine.h"
17 #include "llvm/IR/Attributes.h"
18 #include "llvm/IR/Function.h"
19 #include "llvm/IR/GlobalValue.h"
20 #include "llvm/Support/CommandLine.h"
21 #include "llvm/Support/Debug.h"
22 #include "llvm/Support/ErrorHandling.h"
23 #include "llvm/Support/Host.h"
24 #include "llvm/Support/raw_ostream.h"
25 #include "llvm/Target/TargetMachine.h"
26 #include "llvm/Target/TargetOptions.h"
27 
28 #if defined(_MSC_VER)
29 #include <intrin.h>
30 #endif
31 
32 using namespace llvm;
33 
34 #define DEBUG_TYPE "subtarget"
35 
36 #define GET_SUBTARGETINFO_TARGET_DESC
37 #define GET_SUBTARGETINFO_CTOR
38 #include "X86GenSubtargetInfo.inc"
39 
40 // Temporary option to control early if-conversion for x86 while adding machine
41 // models.
42 static cl::opt<bool>
43 X86EarlyIfConv("x86-early-ifcvt", cl::Hidden,
44                cl::desc("Enable early if-conversion on X86"));
45 
46 
47 /// Classify a blockaddress reference for the current subtarget according to how
48 /// we should reference it in a non-pcrel context.
49 unsigned char X86Subtarget::ClassifyBlockAddressReference() const {
50   if (isPICStyleGOT())    // 32-bit ELF targets.
51     return X86II::MO_GOTOFF;
52 
53   if (isPICStyleStubPIC())   // Darwin/32 in PIC mode.
54     return X86II::MO_PIC_BASE_OFFSET;
55 
56   // Direct static reference to label.
57   return X86II::MO_NO_FLAG;
58 }
59 
60 /// Classify a global variable reference for the current subtarget according to
61 /// how we should reference it in a non-pcrel context.
62 unsigned char X86Subtarget::
63 ClassifyGlobalReference(const GlobalValue *GV, const TargetMachine &TM) const {
64   // DLLImport only exists on windows, it is implemented as a load from a
65   // DLLIMPORT stub.
66   if (GV->hasDLLImportStorageClass())
67     return X86II::MO_DLLIMPORT;
68 
69   bool isDef = GV->isStrongDefinitionForLinker();
70 
71   // X86-64 in PIC mode.
72   if (isPICStyleRIPRel()) {
73     // Large model never uses stubs.
74     if (TM.getCodeModel() == CodeModel::Large)
75       return X86II::MO_NO_FLAG;
76 
77     if (isTargetDarwin()) {
78       // If symbol visibility is hidden, the extra load is not needed if
79       // target is x86-64 or the symbol is definitely defined in the current
80       // translation unit.
81       if (GV->hasDefaultVisibility() && !isDef)
82         return X86II::MO_GOTPCREL;
83     } else if (!isTargetWin64()) {
84       assert(isTargetELF() && "Unknown rip-relative target");
85 
86       // Extra load is needed for all externally visible globals except with
87       // PIE as the definition of the global in an executable is not
88       // overridden.
89 
90       if (!GV->hasLocalLinkage() && GV->hasDefaultVisibility() &&
91           !isGlobalDefinedInPIE(GV, TM))
92         return X86II::MO_GOTPCREL;
93     }
94 
95     return X86II::MO_NO_FLAG;
96   }
97 
98   if (isPICStyleGOT()) {   // 32-bit ELF targets.
99     // Extra load is needed for all externally visible globals except with
100     // PIE as the definition of the global in an executable is not overridden.
101 
102     if (GV->hasLocalLinkage() || GV->hasHiddenVisibility() ||
103         isGlobalDefinedInPIE(GV, TM))
104       return X86II::MO_GOTOFF;
105     return X86II::MO_GOT;
106   }
107 
108   if (isPICStyleStubPIC()) {  // Darwin/32 in PIC mode.
109     // Determine whether we have a stub reference and/or whether the reference
110     // is relative to the PIC base or not.
111 
112     // If this is a strong reference to a definition, it is definitely not
113     // through a stub.
114     if (isDef)
115       return X86II::MO_PIC_BASE_OFFSET;
116 
117     // Unless we have a symbol with hidden visibility, we have to go through a
118     // normal $non_lazy_ptr stub because this symbol might be resolved late.
119     if (!GV->hasHiddenVisibility())  // Non-hidden $non_lazy_ptr reference.
120       return X86II::MO_DARWIN_NONLAZY_PIC_BASE;
121 
122     // If symbol visibility is hidden, we have a stub for common symbol
123     // references and external declarations.
124     if (GV->isDeclarationForLinker() || GV->hasCommonLinkage()) {
125       // Hidden $non_lazy_ptr reference.
126       return X86II::MO_DARWIN_HIDDEN_NONLAZY_PIC_BASE;
127     }
128 
129     // Otherwise, no stub.
130     return X86II::MO_PIC_BASE_OFFSET;
131   }
132 
133   if (isPICStyleStubNoDynamic()) {  // Darwin/32 in -mdynamic-no-pic mode.
134     // Determine whether we have a stub reference.
135 
136     // If this is a strong reference to a definition, it is definitely not
137     // through a stub.
138     if (isDef)
139       return X86II::MO_NO_FLAG;
140 
141     // Unless we have a symbol with hidden visibility, we have to go through a
142     // normal $non_lazy_ptr stub because this symbol might be resolved late.
143     if (!GV->hasHiddenVisibility())  // Non-hidden $non_lazy_ptr reference.
144       return X86II::MO_DARWIN_NONLAZY;
145 
146     // Otherwise, no stub.
147     return X86II::MO_NO_FLAG;
148   }
149 
150   // Direct static reference to global.
151   return X86II::MO_NO_FLAG;
152 }
153 
154 unsigned char X86Subtarget::classifyGlobalFunctionReference(
155     const GlobalValue *GV, const TargetMachine &TM) const {
156   // On ELF targets, in both X86-64 and X86-32 mode, direct calls to
157   // external symbols most go through the PLT in PIC mode.  If the symbol
158   // has hidden or protected visibility, or if it is static or local, then
159   // we don't need to use the PLT - we can directly call it.
160   // In PIE mode, calls to global functions don't need to go through PLT
161   if (isTargetELF() && TM.getRelocationModel() == Reloc::PIC_ &&
162       (!TM.Options.PositionIndependentExecutable ||
163        GV->isDeclarationForLinker()) &&
164       GV->hasDefaultVisibility() && !GV->hasLocalLinkage()) {
165     return X86II::MO_PLT;
166   } else if (isPICStyleStubAny() && !GV->isStrongDefinitionForLinker() &&
167              (!getTargetTriple().isMacOSX() ||
168               getTargetTriple().isMacOSXVersionLT(10, 5))) {
169     // PC-relative references to external symbols should go through $stub,
170     // unless we're building with the leopard linker or later, which
171     // automatically synthesizes these stubs.
172     return X86II::MO_DARWIN_STUB;
173   } else if (isPICStyleRIPRel() && isa<Function>(GV) &&
174              cast<Function>(GV)->hasFnAttribute(Attribute::NonLazyBind)) {
175     // If the function is marked as non-lazy, generate an indirect call
176     // which loads from the GOT directly. This avoids runtime overhead
177     // at the cost of eager binding (and one extra byte of encoding).
178     return X86II::MO_GOTPCREL;
179   }
180 
181   return X86II::MO_NO_FLAG;
182 }
183 
184 /// This function returns the name of a function which has an interface like
185 /// the non-standard bzero function, if such a function exists on the
186 /// current subtarget and it is considered preferable over memset with zero
187 /// passed as the second argument. Otherwise it returns null.
188 const char *X86Subtarget::getBZeroEntry() const {
189   // Darwin 10 has a __bzero entry point for this purpose.
190   if (getTargetTriple().isMacOSX() &&
191       !getTargetTriple().isMacOSXVersionLT(10, 6))
192     return "__bzero";
193 
194   return nullptr;
195 }
196 
197 bool X86Subtarget::hasSinCos() const {
198   return getTargetTriple().isMacOSX() &&
199     !getTargetTriple().isMacOSXVersionLT(10, 9) &&
200     is64Bit();
201 }
202 
203 /// Return true if the subtarget allows calls to immediate address.
204 bool X86Subtarget::IsLegalToCallImmediateAddr(const TargetMachine &TM) const {
205   // FIXME: I386 PE/COFF supports PC relative calls using IMAGE_REL_I386_REL32
206   // but WinCOFFObjectWriter::RecordRelocation cannot emit them.  Once it does,
207   // the following check for Win32 should be removed.
208   if (In64BitMode || isTargetWin32())
209     return false;
210   return isTargetELF() || TM.getRelocationModel() == Reloc::Static;
211 }
212 
213 void X86Subtarget::initSubtargetFeatures(StringRef CPU, StringRef FS) {
214   std::string CPUName = CPU;
215   if (CPUName.empty())
216     CPUName = "generic";
217 
218   // Make sure 64-bit features are available in 64-bit mode. (But make sure
219   // SSE2 can be turned off explicitly.)
220   std::string FullFS = FS;
221   if (In64BitMode) {
222     if (!FullFS.empty())
223       FullFS = "+64bit,+sse2," + FullFS;
224     else
225       FullFS = "+64bit,+sse2";
226   }
227 
228   // LAHF/SAHF are always supported in non-64-bit mode.
229   if (!In64BitMode) {
230     if (!FullFS.empty())
231       FullFS = "+sahf," + FullFS;
232     else
233       FullFS = "+sahf";
234   }
235 
236 
237   // Parse features string and set the CPU.
238   ParseSubtargetFeatures(CPUName, FullFS);
239 
240   // All CPUs that implement SSE4.2 or SSE4A support unaligned accesses of
241   // 16-bytes and under that are reasonably fast. These features were
242   // introduced with Intel's Nehalem/Silvermont and AMD's Family10h
243   // micro-architectures respectively.
244   if (hasSSE42() || hasSSE4A())
245     IsUAMem16Slow = false;
246 
247   InstrItins = getInstrItineraryForCPU(CPUName);
248 
249   // It's important to keep the MCSubtargetInfo feature bits in sync with
250   // target data structure which is shared with MC code emitter, etc.
251   if (In64BitMode)
252     ToggleFeature(X86::Mode64Bit);
253   else if (In32BitMode)
254     ToggleFeature(X86::Mode32Bit);
255   else if (In16BitMode)
256     ToggleFeature(X86::Mode16Bit);
257   else
258     llvm_unreachable("Not 16-bit, 32-bit or 64-bit mode!");
259 
260   DEBUG(dbgs() << "Subtarget features: SSELevel " << X86SSELevel
261                << ", 3DNowLevel " << X863DNowLevel
262                << ", 64bit " << HasX86_64 << "\n");
263   assert((!In64BitMode || HasX86_64) &&
264          "64-bit code requested on a subtarget that doesn't support it!");
265 
266   // Stack alignment is 16 bytes on Darwin, Linux and Solaris (both
267   // 32 and 64 bit) and for all 64-bit targets.
268   if (StackAlignOverride)
269     stackAlignment = StackAlignOverride;
270   else if (isTargetDarwin() || isTargetLinux() || isTargetSolaris() ||
271            In64BitMode)
272     stackAlignment = 16;
273 }
274 
275 void X86Subtarget::initializeEnvironment() {
276   X86SSELevel = NoSSE;
277   X863DNowLevel = NoThreeDNow;
278   HasX87 = false;
279   HasCMov = false;
280   HasX86_64 = false;
281   HasPOPCNT = false;
282   HasSSE4A = false;
283   HasAES = false;
284   HasFXSR = false;
285   HasXSAVE = false;
286   HasXSAVEOPT = false;
287   HasXSAVEC = false;
288   HasXSAVES = false;
289   HasPCLMUL = false;
290   HasFMA = false;
291   HasFMA4 = false;
292   HasXOP = false;
293   HasTBM = false;
294   HasMOVBE = false;
295   HasRDRAND = false;
296   HasF16C = false;
297   HasFSGSBase = false;
298   HasLZCNT = false;
299   HasBMI = false;
300   HasBMI2 = false;
301   HasVBMI = false;
302   HasIFMA = false;
303   HasRTM = false;
304   HasHLE = false;
305   HasERI = false;
306   HasCDI = false;
307   HasPFI = false;
308   HasDQI = false;
309   HasBWI = false;
310   HasVLX = false;
311   HasADX = false;
312   HasPKU = false;
313   HasSHA = false;
314   HasPRFCHW = false;
315   HasRDSEED = false;
316   HasLAHFSAHF = false;
317   HasMPX = false;
318   IsBTMemSlow = false;
319   IsSHLDSlow = false;
320   IsUAMem16Slow = false;
321   IsUAMem32Slow = false;
322   HasSSEUnalignedMem = false;
323   HasCmpxchg16b = false;
324   UseLeaForSP = false;
325   HasFastPartialYMMWrite = false;
326   HasSlowDivide32 = false;
327   HasSlowDivide64 = false;
328   PadShortFunctions = false;
329   CallRegIndirect = false;
330   LEAUsesAG = false;
331   SlowLEA = false;
332   SlowIncDec = false;
333   stackAlignment = 4;
334   // FIXME: this is a known good value for Yonah. How about others?
335   MaxInlineSizeThreshold = 128;
336   UseSoftFloat = false;
337 }
338 
339 X86Subtarget &X86Subtarget::initializeSubtargetDependencies(StringRef CPU,
340                                                             StringRef FS) {
341   initializeEnvironment();
342   initSubtargetFeatures(CPU, FS);
343   return *this;
344 }
345 
346 X86Subtarget::X86Subtarget(const Triple &TT, const std::string &CPU,
347                            const std::string &FS, const X86TargetMachine &TM,
348                            unsigned StackAlignOverride)
349     : X86GenSubtargetInfo(TT, CPU, FS), X86ProcFamily(Others),
350       PICStyle(PICStyles::None), TargetTriple(TT),
351       StackAlignOverride(StackAlignOverride),
352       In64BitMode(TargetTriple.getArch() == Triple::x86_64),
353       In32BitMode(TargetTriple.getArch() == Triple::x86 &&
354                   TargetTriple.getEnvironment() != Triple::CODE16),
355       In16BitMode(TargetTriple.getArch() == Triple::x86 &&
356                   TargetTriple.getEnvironment() == Triple::CODE16),
357       TSInfo(), InstrInfo(initializeSubtargetDependencies(CPU, FS)),
358       TLInfo(TM, *this), FrameLowering(*this, getStackAlignment()) {
359   // Determine the PICStyle based on the target selected.
360   if (TM.getRelocationModel() == Reloc::Static) {
361     // Unless we're in PIC or DynamicNoPIC mode, set the PIC style to None.
362     setPICStyle(PICStyles::None);
363   } else if (is64Bit()) {
364     // PIC in 64 bit mode is always rip-rel.
365     setPICStyle(PICStyles::RIPRel);
366   } else if (isTargetCOFF()) {
367     setPICStyle(PICStyles::None);
368   } else if (isTargetDarwin()) {
369     if (TM.getRelocationModel() == Reloc::PIC_)
370       setPICStyle(PICStyles::StubPIC);
371     else {
372       assert(TM.getRelocationModel() == Reloc::DynamicNoPIC);
373       setPICStyle(PICStyles::StubDynamicNoPIC);
374     }
375   } else if (isTargetELF()) {
376     setPICStyle(PICStyles::GOT);
377   }
378 }
379 
380 bool X86Subtarget::enableEarlyIfConversion() const {
381   return hasCMov() && X86EarlyIfConv;
382 }
383 
384