#include "ARMSubtarget.h"
#include "ARMGenSubtarget.inc"
+#include "ARMBaseRegisterInfo.h"
#include "llvm/GlobalValue.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/CommandLine.h"
static cl::opt<bool>
ReserveR9("arm-reserve-r9", cl::Hidden,
cl::desc("Reserve R9, making it unavailable as GPR"));
+
static cl::opt<bool>
-UseNEONFP("arm-use-neon-fp",
- cl::desc("Use NEON for single-precision FP"),
- cl::init(false), cl::Hidden);
+DarwinUseMOVT("arm-darwin-use-movt", cl::init(true), cl::Hidden);
static cl::opt<bool>
-UseMOVT("arm-use-movt",
- cl::init(true), cl::Hidden);
+StrictAlign("arm-strict-align", cl::Hidden,
+ cl::desc("Disallow all unaligned memory accesses"));
ARMSubtarget::ARMSubtarget(const std::string &TT, const std::string &FS,
bool isT)
- : ARMArchVersion(V4T)
+ : ARMArchVersion(V4)
+ , ARMProcFamily(Others)
, ARMFPUType(None)
- , UseNEONForSinglePrecisionFP(UseNEONFP)
+ , UseNEONForSinglePrecisionFP(false)
+ , SlowFPVMLx(false)
+ , HasVMLxForwarding(false)
+ , SlowFPBrcc(false)
, IsThumb(isT)
, ThumbMode(Thumb1)
+ , NoARM(false)
, PostRAScheduler(false)
, IsR9Reserved(ReserveR9)
- , UseMovt(UseMOVT)
+ , UseMovt(false)
+ , HasFP16(false)
+ , HasD16(false)
+ , HasHardwareDivide(false)
+ , HasT2ExtractPack(false)
+ , HasDataBarrier(false)
+ , Pref32BitThumb(false)
+ , AvoidCPSRPartialUpdate(false)
+ , HasMPExtension(false)
+ , FPOnlySP(false)
+ , AllowsUnalignedMem(false)
, stackAlignment(4)
, CPUString("generic")
- , TargetType(isELF) // Default to ELF unless otherwise specified.
+ , TargetTriple(TT)
, TargetABI(ARM_ABI_APCS) {
- // default to soft float ABI
+ // Default to soft float ABI
if (FloatABIType == FloatABI::Default)
FloatABIType = FloatABI::Soft;
// Determine default and user specified characteristics
- // Parse features string.
- CPUString = ParseSubtargetFeatures(FS, CPUString);
+ // When no arch is specified either by CPU or by attributes, make the default
+ // ARMv4T.
+ const char *ARMArchFeature = "";
+ if (CPUString == "generic" && (FS.empty() || FS == "generic")) {
+ ARMArchVersion = V4T;
+ ARMArchFeature = ",+v4t";
+ }
// Set the boolean corresponding to the current target triple, or the default
// if one cannot be determined, to true.
}
if (Idx) {
unsigned SubVer = TT[Idx];
- if (SubVer > '4' && SubVer <= '9') {
- if (SubVer >= '7') {
- ARMArchVersion = V7A;
- } else if (SubVer == '6') {
- ARMArchVersion = V6;
- if (Len >= Idx+3 && TT[Idx+1] == 't' && TT[Idx+2] == '2')
- ARMArchVersion = V6T2;
- } else if (SubVer == '5') {
- ARMArchVersion = V5T;
- if (Len >= Idx+3 && TT[Idx+1] == 't' && TT[Idx+2] == 'e')
- ARMArchVersion = V5TE;
+ if (SubVer >= '7' && SubVer <= '9') {
+ ARMArchVersion = V7A;
+ ARMArchFeature = ",+v7a";
+ if (Len >= Idx+2 && TT[Idx+1] == 'm') {
+ ARMArchVersion = V7M;
+ ARMArchFeature = ",+v7m";
+ }
+ } else if (SubVer == '6') {
+ ARMArchVersion = V6;
+ ARMArchFeature = ",+v6";
+ if (Len >= Idx+3 && TT[Idx+1] == 't' && TT[Idx+2] == '2') {
+ ARMArchVersion = V6T2;
+ ARMArchFeature = ",+v6t2";
+ }
+ } else if (SubVer == '5') {
+ ARMArchVersion = V5T;
+ ARMArchFeature = ",+v5t";
+ if (Len >= Idx+3 && TT[Idx+1] == 't' && TT[Idx+2] == 'e') {
+ ARMArchVersion = V5TE;
+ ARMArchFeature = ",+v5te";
+ }
+ } else if (SubVer == '4') {
+ if (Len >= Idx+2 && TT[Idx+1] == 't') {
+ ARMArchVersion = V4T;
+ ARMArchFeature = ",+v4t";
+ } else {
+ ARMArchVersion = V4;
+ ARMArchFeature = "";
}
- if (ARMArchVersion >= V6T2)
- ThumbMode = Thumb2;
}
}
- // Thumb2 implies at least V6T2.
- if (ARMArchVersion < V6T2 && ThumbMode >= Thumb2)
- ARMArchVersion = V6T2;
-
- if (Len >= 10) {
- if (TT.find("-darwin") != std::string::npos)
- // arm-darwin
- TargetType = isDarwin;
- }
-
if (TT.find("eabi") != std::string::npos)
TargetABI = ARM_ABI_AAPCS;
+ // Parse features string. If the first entry in FS (the CPU) is missing,
+ // insert the architecture feature derived from the target triple. This is
+ // important for setting features that are implied based on the architecture
+ // version.
+ std::string FSWithArch;
+ if (FS.empty())
+ FSWithArch = std::string(ARMArchFeature);
+ else if (FS.find(',') == 0)
+ FSWithArch = std::string(ARMArchFeature) + FS;
+ else
+ FSWithArch = FS;
+ CPUString = ParseSubtargetFeatures(FSWithArch, CPUString);
+
+ // After parsing Itineraries, set ItinData.IssueWidth.
+ computeIssueWidth();
+
+ // Thumb2 implies at least V6T2.
+ if (ARMArchVersion >= V6T2)
+ ThumbMode = Thumb2;
+ else if (ThumbMode >= Thumb2)
+ ARMArchVersion = V6T2;
+
if (isAAPCS_ABI())
stackAlignment = 8;
- if (isTargetDarwin())
+ if (!isTargetDarwin())
+ UseMovt = hasV6T2Ops();
+ else {
IsR9Reserved = ReserveR9 | (ARMArchVersion < V6);
+ UseMovt = DarwinUseMOVT && hasV6T2Ops();
+ }
if (!isThumb() || hasThumb2())
PostRAScheduler = true;
- // Set CPU specific features.
- if (CPUString == "cortex-a8") {
- // On Cortex-a8, it's faster to perform some single-precision FP
- // operations with NEON instructions.
- if (UseNEONFP.getPosition() == 0)
- UseNEONForSinglePrecisionFP = true;
- }
+ // v6+ may or may not support unaligned mem access depending on the system
+ // configuration.
+ if (!StrictAlign && hasV6Ops() && isTargetDarwin())
+ AllowsUnalignedMem = true;
}
/// GVIsIndirectSymbol - true if the GV will be accessed via an indirect symbol.
bool
-ARMSubtarget::GVIsIndirectSymbol(GlobalValue *GV, Reloc::Model RelocM) const {
+ARMSubtarget::GVIsIndirectSymbol(const GlobalValue *GV,
+ Reloc::Model RelocM) const {
if (RelocM == Reloc::Static)
return false;
// Materializable GVs (in JIT lazy compilation mode) do not require an extra
// load from stub.
- bool isDecl = GV->isDeclaration() && !GV->isMaterializable();
+ bool isDecl = GV->hasAvailableExternallyLinkage();
+ if (GV->isDeclaration() && !GV->isMaterializable())
+ isDecl = true;
if (!isTargetDarwin()) {
// Extra load is needed for all externally visible.
// through a stub.
if (!isDecl && !GV->isWeakForLinker())
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 false;
}
+unsigned ARMSubtarget::getMispredictionPenalty() const {
+ // If we have a reasonable estimate of the pipeline depth, then we can
+ // estimate the penalty of a misprediction based on that.
+ if (isCortexA8())
+ return 13;
+ else if (isCortexA9())
+ return 8;
+
+ // Otherwise, just return a sensible default.
+ return 10;
+}
+
+void ARMSubtarget::computeIssueWidth() {
+ unsigned allStage1Units = 0;
+ for (const InstrItinerary *itin = InstrItins.Itineraries;
+ itin->FirstStage != ~0U; ++itin) {
+ const InstrStage *IS = InstrItins.Stages + itin->FirstStage;
+ allStage1Units |= IS->getUnits();
+ }
+ InstrItins.IssueWidth = 0;
+ while (allStage1Units) {
+ ++InstrItins.IssueWidth;
+ // clear the lowest bit
+ allStage1Units ^= allStage1Units & ~(allStage1Units - 1);
+ }
+ assert(InstrItins.IssueWidth <= 2 && "itinerary bug, too many stage 1 units");
+}
+
bool ARMSubtarget::enablePostRAScheduler(
CodeGenOpt::Level OptLevel,
TargetSubtarget::AntiDepBreakMode& Mode,