//
//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "subtarget"
#include "X86Subtarget.h"
#include "X86GenSubtarget.inc"
#include "llvm/Module.h"
#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
using namespace llvm;
AsmWriterFlavor("x86-asm-syntax", cl::init(X86Subtarget::Unset),
cl::desc("Choose style of code to emit from X86 backend:"),
cl::values(
- clEnumValN(X86Subtarget::ATT, "att", " Emit AT&T-style assembly"),
- clEnumValN(X86Subtarget::Intel, "intel", " Emit Intel-style assembly"),
+ clEnumValN(X86Subtarget::ATT, "att", "Emit AT&T-style assembly"),
+ clEnumValN(X86Subtarget::Intel, "intel", "Emit Intel-style assembly"),
clEnumValEnd));
bool isDirectCall) const
{
// FIXME: PIC
- if (TM.getRelocationModel() != Reloc::Static) {
+ if (TM.getRelocationModel() != Reloc::Static &&
+ TM.getCodeModel() != CodeModel::Large) {
if (isTargetDarwin()) {
- return (!isDirectCall &&
- (GV->hasWeakLinkage() || GV->hasLinkOnceLinkage() ||
- GV->hasCommonLinkage() ||
- (GV->isDeclaration() && !GV->hasNotBeenReadFromBitcode())));
+ 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()) {
- // Extra load is needed for all non-statics.
- return (!isDirectCall &&
- (GV->isDeclaration() || !GV->hasInternalLinkage()));
+ // 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());
}
}
-
return false;
}
-/// 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 passed as the second argument. Otherwise it
-/// returns null.
-const char *X86Subtarget::getBZeroEntry() const {
+/// 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;
+}
+/// 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
+/// passed as the second argument. Otherwise it returns null.
+const char *X86Subtarget::getBZeroEntry() const {
// Darwin 10 has a __bzero entry point for this purpose.
if (getDarwinVers() >= 10)
return "__bzero";
return 0;
}
+/// 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
+/// should be attempted.
+unsigned X86Subtarget::getSpecialAddressLatency() const {
+ // For x86 out-of-order targets, back-schedule address computations so
+ // that loads and stores aren't blocked.
+ // This value was chosen arbitrarily.
+ return 200;
+}
+
/// GetCpuIDAndInfo - Execute the specified cpuid and return the 4 values in the
/// specified arguments. If we can't run cpuid on the host, return true.
bool X86::GetCpuIDAndInfo(unsigned value, unsigned *rEAX, unsigned *rEBX,
return true;
}
+static void DetectFamilyModel(unsigned EAX, unsigned &Family, unsigned &Model) {
+ Family = (EAX >> 8) & 0xf; // Bits 8 - 11
+ Model = (EAX >> 4) & 0xf; // Bits 4 - 7
+ if (Family == 6 || Family == 0xf) {
+ if (Family == 0xf)
+ // Examine extended family ID if family ID is F.
+ Family += (EAX >> 20) & 0xff; // Bits 20 - 27
+ // Examine extended model ID if family ID is 6 or F.
+ Model += ((EAX >> 16) & 0xf) << 4; // Bits 16 - 19
+ }
+}
+
void X86Subtarget::AutoDetectSubtargetFeatures() {
unsigned EAX = 0, EBX = 0, ECX = 0, EDX = 0;
union {
if ((ECX >> 19) & 0x1) X86SSELevel = SSE41;
if ((ECX >> 20) & 0x1) X86SSELevel = SSE42;
- if (memcmp(text.c, "GenuineIntel", 12) == 0 ||
- memcmp(text.c, "AuthenticAMD", 12) == 0) {
+ bool IsIntel = memcmp(text.c, "GenuineIntel", 12) == 0;
+ bool IsAMD = !IsIntel && memcmp(text.c, "AuthenticAMD", 12) == 0;
+ if (IsIntel || IsAMD) {
+ // Determine if bit test memory instructions are slow.
+ unsigned Family = 0;
+ unsigned Model = 0;
+ DetectFamilyModel(EAX, Family, Model);
+ IsBTMemSlow = IsAMD || (Family == 6 && Model >= 13);
+
X86::GetCpuIDAndInfo(0x80000001, &EAX, &EBX, &ECX, &EDX);
HasX86_64 = (EDX >> 29) & 0x1;
}
unsigned EAX = 0, EBX = 0, ECX = 0, EDX = 0;
if (X86::GetCpuIDAndInfo(0x1, &EAX, &EBX, &ECX, &EDX))
return "generic";
- unsigned Family = (EAX >> 8) & 0xf; // Bits 8 - 11
- unsigned Model = (EAX >> 4) & 0xf; // Bits 4 - 7
+ unsigned Family = 0;
+ unsigned Model = 0;
+ DetectFamilyModel(EAX, Family, Model);
+
X86::GetCpuIDAndInfo(0x80000001, &EAX, &EBX, &ECX, &EDX);
bool Em64T = (EDX >> 29) & 0x1;
case 9:
case 13: return "pentium-m";
case 14: return "yonah";
- case 15: return "core2";
+ case 15:
+ case 22: // Celeron M 540
+ return "core2";
+ case 23: // 45nm: Penryn , Wolfdale, Yorkfield (XE)
+ return "penryn";
default: return "i686";
}
case 15: {
switch (Model) {
case 3:
case 4:
+ case 6: // same as 4, but 65nm
return (Em64T) ? "nocona" : "prescott";
+ case 26:
+ return "corei7";
+ case 28:
+ return "atom";
default:
return (Em64T) ? "x86-64" : "pentium4";
}
X86Subtarget::X86Subtarget(const Module &M, const std::string &FS, bool is64Bit)
: AsmFlavor(AsmWriterFlavor)
- , PICStyle(PICStyle::None)
+ , PICStyle(PICStyles::None)
, X86SSELevel(NoMMXSSE)
, X863DNowLevel(NoThreeDNow)
, HasX86_64(false)
+ , IsBTMemSlow(false)
, DarwinVers(0)
, IsLinux(false)
, stackAlignment(8)
// 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();
- }
-
- // 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) {
- HasX86_64 = true;
- if (X86SSELevel < SSE2)
+ // 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 features
+ // are enabled.
+ if (Is64Bit)
+ HasX86_64 = true;
+
+ DOUT << "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();