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       !isGlobalDefinedInPIE(GV, TM) &&
163       GV->hasDefaultVisibility() && !GV->hasLocalLinkage()) {
164     return X86II::MO_PLT;
165   } else if (isPICStyleStubAny() && !GV->isStrongDefinitionForLinker() &&
166              (!getTargetTriple().isMacOSX() ||
167               getTargetTriple().isMacOSXVersionLT(10, 5))) {
168     // PC-relative references to external symbols should go through $stub,
169     // unless we're building with the leopard linker or later, which
170     // automatically synthesizes these stubs.
171     return X86II::MO_DARWIN_STUB;
172   } else if (isPICStyleRIPRel() && isa<Function>(GV) &&
173              cast<Function>(GV)->hasFnAttribute(Attribute::NonLazyBind)) {
174     // If the function is marked as non-lazy, generate an indirect call
175     // which loads from the GOT directly. This avoids runtime overhead
176     // at the cost of eager binding (and one extra byte of encoding).
177     return X86II::MO_GOTPCREL;
178   }
179 
180   return X86II::MO_NO_FLAG;
181 }
182 
183 /// This function returns the name of a function which has an interface like
184 /// the non-standard bzero function, if such a function exists on the
185 /// current subtarget and it is considered preferable over memset with zero
186 /// passed as the second argument. Otherwise it returns null.
187 const char *X86Subtarget::getBZeroEntry() const {
188   // Darwin 10 has a __bzero entry point for this purpose.
189   if (getTargetTriple().isMacOSX() &&
190       !getTargetTriple().isMacOSXVersionLT(10, 6))
191     return "__bzero";
192 
193   return nullptr;
194 }
195 
196 bool X86Subtarget::hasSinCos() const {
197   return getTargetTriple().isMacOSX() &&
198     !getTargetTriple().isMacOSXVersionLT(10, 9) &&
199     is64Bit();
200 }
201 
202 /// Return true if the subtarget allows calls to immediate address.
203 bool X86Subtarget::IsLegalToCallImmediateAddr(const TargetMachine &TM) const {
204   // FIXME: I386 PE/COFF supports PC relative calls using IMAGE_REL_I386_REL32
205   // but WinCOFFObjectWriter::RecordRelocation cannot emit them.  Once it does,
206   // the following check for Win32 should be removed.
207   if (In64BitMode || isTargetWin32())
208     return false;
209   return isTargetELF() || TM.getRelocationModel() == Reloc::Static;
210 }
211 
212 void X86Subtarget::initSubtargetFeatures(StringRef CPU, StringRef FS) {
213   std::string CPUName = CPU;
214   if (CPUName.empty())
215     CPUName = "generic";
216 
217   // Make sure 64-bit features are available in 64-bit mode. (But make sure
218   // SSE2 can be turned off explicitly.)
219   std::string FullFS = FS;
220   if (In64BitMode) {
221     if (!FullFS.empty())
222       FullFS = "+64bit,+sse2," + FullFS;
223     else
224       FullFS = "+64bit,+sse2";
225   }
226 
227   // LAHF/SAHF are always supported in non-64-bit mode.
228   if (!In64BitMode) {
229     if (!FullFS.empty())
230       FullFS = "+sahf," + FullFS;
231     else
232       FullFS = "+sahf";
233   }
234 
235 
236   // Parse features string and set the CPU.
237   ParseSubtargetFeatures(CPUName, FullFS);
238 
239   // All CPUs that implement SSE4.2 or SSE4A support unaligned accesses of
240   // 16-bytes and under that are reasonably fast. These features were
241   // introduced with Intel's Nehalem/Silvermont and AMD's Family10h
242   // micro-architectures respectively.
243   if (hasSSE42() || hasSSE4A())
244     IsUAMem16Slow = false;
245 
246   InstrItins = getInstrItineraryForCPU(CPUName);
247 
248   // It's important to keep the MCSubtargetInfo feature bits in sync with
249   // target data structure which is shared with MC code emitter, etc.
250   if (In64BitMode)
251     ToggleFeature(X86::Mode64Bit);
252   else if (In32BitMode)
253     ToggleFeature(X86::Mode32Bit);
254   else if (In16BitMode)
255     ToggleFeature(X86::Mode16Bit);
256   else
257     llvm_unreachable("Not 16-bit, 32-bit or 64-bit mode!");
258 
259   DEBUG(dbgs() << "Subtarget features: SSELevel " << X86SSELevel
260                << ", 3DNowLevel " << X863DNowLevel
261                << ", 64bit " << HasX86_64 << "\n");
262   assert((!In64BitMode || HasX86_64) &&
263          "64-bit code requested on a subtarget that doesn't support it!");
264 
265   // Stack alignment is 16 bytes on Darwin, Linux and Solaris (both
266   // 32 and 64 bit) and for all 64-bit targets.
267   if (StackAlignOverride)
268     stackAlignment = StackAlignOverride;
269   else if (isTargetDarwin() || isTargetLinux() || isTargetSolaris() ||
270            In64BitMode)
271     stackAlignment = 16;
272 }
273 
274 void X86Subtarget::initializeEnvironment() {
275   X86SSELevel = NoSSE;
276   X863DNowLevel = NoThreeDNow;
277   HasX87 = false;
278   HasCMov = false;
279   HasX86_64 = false;
280   HasPOPCNT = false;
281   HasSSE4A = false;
282   HasAES = false;
283   HasFXSR = false;
284   HasXSAVE = false;
285   HasXSAVEOPT = false;
286   HasXSAVEC = false;
287   HasXSAVES = false;
288   HasPCLMUL = false;
289   HasFMA = false;
290   HasFMA4 = false;
291   HasXOP = false;
292   HasTBM = false;
293   HasMOVBE = false;
294   HasRDRAND = false;
295   HasF16C = false;
296   HasFSGSBase = false;
297   HasLZCNT = false;
298   HasBMI = false;
299   HasBMI2 = false;
300   HasVBMI = false;
301   HasIFMA = false;
302   HasRTM = false;
303   HasHLE = false;
304   HasERI = false;
305   HasCDI = false;
306   HasPFI = false;
307   HasDQI = false;
308   HasBWI = false;
309   HasVLX = false;
310   HasADX = false;
311   HasPKU = false;
312   HasSHA = false;
313   HasPRFCHW = false;
314   HasRDSEED = false;
315   HasLAHFSAHF = false;
316   HasMPX = false;
317   IsBTMemSlow = false;
318   IsSHLDSlow = false;
319   IsUAMem16Slow = false;
320   IsUAMem32Slow = false;
321   HasSSEUnalignedMem = false;
322   HasCmpxchg16b = false;
323   UseLeaForSP = false;
324   HasFastPartialYMMWrite = false;
325   HasSlowDivide32 = false;
326   HasSlowDivide64 = false;
327   PadShortFunctions = false;
328   CallRegIndirect = false;
329   LEAUsesAG = false;
330   SlowLEA = false;
331   SlowIncDec = false;
332   stackAlignment = 4;
333   // FIXME: this is a known good value for Yonah. How about others?
334   MaxInlineSizeThreshold = 128;
335   UseSoftFloat = false;
336 }
337 
338 X86Subtarget &X86Subtarget::initializeSubtargetDependencies(StringRef CPU,
339                                                             StringRef FS) {
340   initializeEnvironment();
341   initSubtargetFeatures(CPU, FS);
342   return *this;
343 }
344 
345 X86Subtarget::X86Subtarget(const Triple &TT, const std::string &CPU,
346                            const std::string &FS, const X86TargetMachine &TM,
347                            unsigned StackAlignOverride)
348     : X86GenSubtargetInfo(TT, CPU, FS), X86ProcFamily(Others),
349       PICStyle(PICStyles::None), TargetTriple(TT),
350       StackAlignOverride(StackAlignOverride),
351       In64BitMode(TargetTriple.getArch() == Triple::x86_64),
352       In32BitMode(TargetTriple.getArch() == Triple::x86 &&
353                   TargetTriple.getEnvironment() != Triple::CODE16),
354       In16BitMode(TargetTriple.getArch() == Triple::x86 &&
355                   TargetTriple.getEnvironment() == Triple::CODE16),
356       TSInfo(), InstrInfo(initializeSubtargetDependencies(CPU, FS)),
357       TLInfo(TM, *this), FrameLowering(*this, getStackAlignment()) {
358   // Determine the PICStyle based on the target selected.
359   if (TM.getRelocationModel() == Reloc::Static) {
360     // Unless we're in PIC or DynamicNoPIC mode, set the PIC style to None.
361     setPICStyle(PICStyles::None);
362   } else if (is64Bit()) {
363     // PIC in 64 bit mode is always rip-rel.
364     setPICStyle(PICStyles::RIPRel);
365   } else if (isTargetCOFF()) {
366     setPICStyle(PICStyles::None);
367   } else if (isTargetDarwin()) {
368     if (TM.getRelocationModel() == Reloc::PIC_)
369       setPICStyle(PICStyles::StubPIC);
370     else {
371       assert(TM.getRelocationModel() == Reloc::DynamicNoPIC);
372       setPICStyle(PICStyles::StubDynamicNoPIC);
373     }
374   } else if (isTargetELF()) {
375     setPICStyle(PICStyles::GOT);
376   }
377 }
378 
379 bool X86Subtarget::enableEarlyIfConversion() const {
380   return hasCMov() && X86EarlyIfConv;
381 }
382 
383