#define DEBUG_TYPE "subtarget"
#include "X86Subtarget.h"
+#include "X86InstrInfo.h"
#include "X86GenSubtarget.inc"
-#include "llvm/Module.h"
+#include "llvm/GlobalValue.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
using namespace llvm;
+#if defined(_MSC_VER)
+#include <intrin.h>
+#endif
+
static cl::opt<X86Subtarget::AsmWriterFlavorTy>
AsmWriterFlavor("x86-asm-syntax", cl::init(X86Subtarget::Unset),
cl::desc("Choose style of code to emit from X86 backend:"),
clEnumValN(X86Subtarget::Intel, "intel", "Emit Intel-style assembly"),
clEnumValEnd));
+/// ClassifyGlobalReference - Classify a global variable reference for the
+/// current subtarget according to how we should reference it in a non-pcrel
+/// context.
+unsigned char X86Subtarget::
+ClassifyGlobalReference(const GlobalValue *GV, const TargetMachine &TM) const {
+ // DLLImport only exists on windows, it is implemented as a load from a
+ // DLLIMPORT stub.
+ if (GV->hasDLLImportLinkage())
+ return X86II::MO_DLLIMPORT;
+
+ // GV with ghost linkage (in JIT lazy compilation mode) do not require an
+ // extra load from stub.
+ bool isDecl = GV->isDeclaration() && !GV->hasNotBeenReadFromBitcode();
-/// True if accessing the GV requires an extra load. For Windows, dllimported
-/// symbols are indirect, loading the value at address GV rather then the
-/// value of GV itself. This means that the GlobalAddress must be in the base
-/// or index register of the address, not the GV offset field.
-bool X86Subtarget::GVRequiresExtraLoad(const GlobalValue* GV,
- const TargetMachine& TM,
- bool isDirectCall) const
-{
- // FIXME: PIC
- if (TM.getRelocationModel() != Reloc::Static &&
- TM.getCodeModel() != CodeModel::Large) {
+ // X86-64 in PIC mode.
+ if (isPICStyleRIPRel()) {
+ // Large model never uses stubs.
+ if (TM.getCodeModel() == CodeModel::Large)
+ return X86II::MO_NO_FLAG;
+
if (isTargetDarwin()) {
- if (isDirectCall)
- return false;
- bool isDecl = GV->isDeclaration() && !GV->hasNotBeenReadFromBitcode();
- if (GV->hasHiddenVisibility() &&
- (Is64Bit || (!isDecl && !GV->hasCommonLinkage())))
- // If symbol visibility is hidden, the extra load is not needed if
- // target is x86-64 or the symbol is definitely defined in the current
- // translation unit.
- return false;
- return !isDirectCall && (isDecl || GV->mayBeOverridden());
- } else if (isTargetELF()) {
+ // If symbol visibility is hidden, the extra load is not needed if
+ // target is x86-64 or the symbol is definitely defined in the current
+ // translation unit.
+ if (GV->hasDefaultVisibility() &&
+ (isDecl || GV->isWeakForLinker()))
+ return X86II::MO_GOTPCREL;
+ } else {
+ assert(isTargetELF() && "Unknown rip-relative target");
+
// Extra load is needed for all externally visible.
- if (isDirectCall)
- return false;
- if (GV->hasLocalLinkage() || GV->hasHiddenVisibility())
- return false;
- return true;
- } else if (isTargetCygMing() || isTargetWindows()) {
- return (GV->hasDLLImportLinkage());
+ if (!GV->hasLocalLinkage() && GV->hasDefaultVisibility())
+ return X86II::MO_GOTPCREL;
}
+
+ return X86II::MO_NO_FLAG;
}
- return false;
-}
+
+ if (isPICStyleGOT()) { // 32-bit ELF targets.
+ // Extra load is needed for all externally visible.
+ if (GV->hasLocalLinkage() || GV->hasHiddenVisibility())
+ return X86II::MO_GOTOFF;
+ return X86II::MO_GOT;
+ }
+
+ if (isPICStyleStubPIC()) { // Darwin/32 in PIC mode.
+ // Determine whether we have a stub reference and/or whether the reference
+ // is relative to the PIC base or not.
+
+ // If this is a strong reference to a definition, it is definitely not
+ // through a stub.
+ if (!isDecl && !GV->isWeakForLinker())
+ return X86II::MO_PIC_BASE_OFFSET;
-/// True if accessing the GV requires a register. This is a superset of the
-/// cases where GVRequiresExtraLoad is true. Some variations of PIC require
-/// a register, but not an extra load.
-bool X86Subtarget::GVRequiresRegister(const GlobalValue *GV,
- const TargetMachine& TM,
- bool isDirectCall) const
-{
- if (GVRequiresExtraLoad(GV, TM, isDirectCall))
- return true;
- // Code below here need only consider cases where GVRequiresExtraLoad
- // returns false.
- if (TM.getRelocationModel() == Reloc::PIC_)
- return !isDirectCall &&
- (GV->hasLocalLinkage() || GV->hasExternalLinkage());
- return false;
+ // Unless we have a symbol with hidden visibility, we have to go through a
+ // normal $non_lazy_ptr stub because this symbol might be resolved late.
+ if (!GV->hasHiddenVisibility()) // Non-hidden $non_lazy_ptr reference.
+ return X86II::MO_DARWIN_NONLAZY_PIC_BASE;
+
+ // If symbol visibility is hidden, we have a stub for common symbol
+ // references and external declarations.
+ if (isDecl || GV->hasCommonLinkage()) {
+ // Hidden $non_lazy_ptr reference.
+ return X86II::MO_DARWIN_HIDDEN_NONLAZY_PIC_BASE;
+ }
+
+ // Otherwise, no stub.
+ return X86II::MO_PIC_BASE_OFFSET;
+ }
+
+ if (isPICStyleStubNoDynamic()) { // Darwin/32 in -mdynamic-no-pic mode.
+ // Determine whether we have a stub reference.
+
+ // If this is a strong reference to a definition, it is definitely not
+ // through a stub.
+ if (!isDecl && !GV->isWeakForLinker())
+ return X86II::MO_NO_FLAG;
+
+ // Unless we have a symbol with hidden visibility, we have to go through a
+ // normal $non_lazy_ptr stub because this symbol might be resolved late.
+ if (!GV->hasHiddenVisibility()) // Non-hidden $non_lazy_ptr reference.
+ return X86II::MO_DARWIN_NONLAZY;
+
+ // If symbol visibility is hidden, we have a stub for common symbol
+ // references and external declarations.
+ if (isDecl || GV->hasCommonLinkage()) {
+ // Hidden $non_lazy_ptr reference.
+ return X86II::MO_DARWIN_HIDDEN_NONLAZY;
+ }
+
+ // Otherwise, no stub.
+ return X86II::MO_NO_FLAG;
+ }
+
+ // Direct static reference to global.
+ return X86II::MO_NO_FLAG;
}
+
/// getBZeroEntry - This function returns the name of a function which has an
/// interface like the non-standard bzero function, if such a function exists on
/// the current subtarget and it is considered prefereable over memset with zero
return 0;
}
+/// IsLegalToCallImmediateAddr - Return true if the subtarget allows calls
+/// to immediate address.
+bool X86Subtarget::IsLegalToCallImmediateAddr(const TargetMachine &TM) const {
+ if (Is64Bit)
+ return false;
+ return isTargetELF() || TM.getRelocationModel() == Reloc::Static;
+}
+
/// getSpecialAddressLatency - For targets where it is beneficial to
/// backschedule instructions that compute addresses, return a value
/// indicating the number of scheduling cycles of backscheduling that
/// specified arguments. If we can't run cpuid on the host, return true.
bool X86::GetCpuIDAndInfo(unsigned value, unsigned *rEAX, unsigned *rEBX,
unsigned *rECX, unsigned *rEDX) {
-#if defined(__x86_64__)
- // gcc doesn't know cpuid would clobber ebx/rbx. Preseve it manually.
- asm ("movq\t%%rbx, %%rsi\n\t"
- "cpuid\n\t"
- "xchgq\t%%rbx, %%rsi\n\t"
- : "=a" (*rEAX),
- "=S" (*rEBX),
- "=c" (*rECX),
- "=d" (*rEDX)
- : "a" (value));
- return false;
+#if defined(__x86_64__) || defined(_M_AMD64)
+ #if defined(__GNUC__)
+ // gcc doesn't know cpuid would clobber ebx/rbx. Preseve it manually.
+ asm ("movq\t%%rbx, %%rsi\n\t"
+ "cpuid\n\t"
+ "xchgq\t%%rbx, %%rsi\n\t"
+ : "=a" (*rEAX),
+ "=S" (*rEBX),
+ "=c" (*rECX),
+ "=d" (*rEDX)
+ : "a" (value));
+ return false;
+ #elif defined(_MSC_VER)
+ int registers[4];
+ __cpuid(registers, value);
+ *rEAX = registers[0];
+ *rEBX = registers[1];
+ *rECX = registers[2];
+ *rEDX = registers[3];
+ return false;
+ #endif
#elif defined(i386) || defined(__i386__) || defined(__x86__) || defined(_M_IX86)
-#if defined(__GNUC__)
- asm ("movl\t%%ebx, %%esi\n\t"
- "cpuid\n\t"
- "xchgl\t%%ebx, %%esi\n\t"
- : "=a" (*rEAX),
- "=S" (*rEBX),
- "=c" (*rECX),
- "=d" (*rEDX)
- : "a" (value));
- return false;
-#elif defined(_MSC_VER)
- __asm {
- mov eax,value
- cpuid
- mov esi,rEAX
- mov dword ptr [esi],eax
- mov esi,rEBX
- mov dword ptr [esi],ebx
- mov esi,rECX
- mov dword ptr [esi],ecx
- mov esi,rEDX
- mov dword ptr [esi],edx
- }
- return false;
-#endif
+ #if defined(__GNUC__)
+ asm ("movl\t%%ebx, %%esi\n\t"
+ "cpuid\n\t"
+ "xchgl\t%%ebx, %%esi\n\t"
+ : "=a" (*rEAX),
+ "=S" (*rEBX),
+ "=c" (*rECX),
+ "=d" (*rEDX)
+ : "a" (value));
+ return false;
+ #elif defined(_MSC_VER)
+ __asm {
+ mov eax,value
+ cpuid
+ mov esi,rEAX
+ mov dword ptr [esi],eax
+ mov esi,rEBX
+ mov dword ptr [esi],ebx
+ mov esi,rECX
+ mov dword ptr [esi],ecx
+ mov esi,rEDX
+ mov dword ptr [esi],edx
+ }
+ return false;
+ #endif
#endif
return true;
}
bool IsIntel = memcmp(text.c, "GenuineIntel", 12) == 0;
bool IsAMD = !IsIntel && memcmp(text.c, "AuthenticAMD", 12) == 0;
+
+ HasFMA3 = IsIntel && ((ECX >> 12) & 0x1);
+ HasAVX = ((ECX >> 28) & 0x1);
+
if (IsIntel || IsAMD) {
// Determine if bit test memory instructions are slow.
unsigned Family = 0;
X86::GetCpuIDAndInfo(0x80000001, &EAX, &EBX, &ECX, &EDX);
HasX86_64 = (EDX >> 29) & 0x1;
+ HasSSE4A = IsAMD && ((ECX >> 6) & 0x1);
+ HasFMA4 = IsAMD && ((ECX >> 16) & 0x1);
}
}
X86::GetCpuIDAndInfo(0x80000001, &EAX, &EBX, &ECX, &EDX);
bool Em64T = (EDX >> 29) & 0x1;
+ bool HasSSE3 = (ECX & 0x1);
union {
unsigned u[3];
default: return "athlon";
}
case 15:
- switch (Model) {
- case 1: return "opteron";
- case 5: return "athlon-fx"; // also opteron
- default: return "athlon64";
+ if (HasSSE3) {
+ return "k8-sse3";
+ } else {
+ switch (Model) {
+ case 1: return "opteron";
+ case 5: return "athlon-fx"; // also opteron
+ default: return "athlon64";
+ }
}
+ case 16:
+ return "amdfam10";
default:
return "generic";
}
}
}
-X86Subtarget::X86Subtarget(const Module &M, const std::string &FS, bool is64Bit)
+X86Subtarget::X86Subtarget(const std::string &TT, const std::string &FS,
+ bool is64Bit)
: AsmFlavor(AsmWriterFlavor)
, PICStyle(PICStyles::None)
, X86SSELevel(NoMMXSSE)
, X863DNowLevel(NoThreeDNow)
, HasX86_64(false)
+ , HasSSE4A(false)
+ , HasAVX(false)
+ , HasFMA3(false)
+ , HasFMA4(false)
, IsBTMemSlow(false)
, DarwinVers(0)
, IsLinux(false)
, MaxInlineSizeThreshold(128)
, Is64Bit(is64Bit)
, TargetType(isELF) { // Default to ELF unless otherwise specified.
+
+ // default to hard float ABI
+ if (FloatABIType == FloatABI::Default)
+ FloatABIType = FloatABI::Hard;
// Determine default and user specified characteristics
if (!FS.empty()) {
// If feature string is not empty, parse features string.
std::string CPU = GetCurrentX86CPU();
ParseSubtargetFeatures(FS, CPU);
+ // All X86-64 CPUs also have SSE2, however user might request no SSE via
+ // -mattr, so don't force SSELevel here.
} else {
// Otherwise, use CPUID to auto-detect feature set.
AutoDetectSubtargetFeatures();
+ // Make sure SSE2 is enabled; it is available on all X86-64 CPUs.
+ if (Is64Bit && X86SSELevel < SSE2)
+ X86SSELevel = SSE2;
}
-
- // If requesting codegen for X86-64, make sure that 64-bit and SSE2 features
- // are enabled. These are available on all x86-64 CPUs.
- if (Is64Bit) {
+
+ // If requesting codegen for X86-64, make sure that 64-bit features
+ // are enabled.
+ if (Is64Bit)
HasX86_64 = true;
- }
- DOUT << "Subtarget features: SSELevel " << X86SSELevel
- << ", 3DNowLevel " << X863DNowLevel
- << ", 64bit " << HasX86_64 << "\n";
+
+ DEBUG(errs() << "Subtarget features: SSELevel " << X86SSELevel
+ << ", 3DNowLevel " << X863DNowLevel
+ << ", 64bit " << HasX86_64 << "\n");
+ assert((!Is64Bit || HasX86_64) &&
+ "64-bit code requested on a subtarget that doesn't support it!");
// Set the boolean corresponding to the current target triple, or the default
// if one cannot be determined, to true.
- const std::string& TT = M.getTargetTriple();
if (TT.length() > 5) {
size_t Pos;
if ((Pos = TT.find("-darwin")) != std::string::npos) {
} else if (TT.find("windows") != std::string::npos) {
TargetType = isWindows;
}
+ else if (TT.find("-cl") != std::string::npos) {
+ TargetType = isDarwin;
+ DarwinVers = 9;
+ }
} else if (TT.empty()) {
#if defined(__CYGWIN__)
TargetType = isCygwin;