//===----------------------------------------------------------------------===//
#include "AMDGPUSubtarget.h"
+#include "R600ISelLowering.h"
+#include "R600InstrInfo.h"
+#include "R600MachineScheduler.h"
+#include "SIISelLowering.h"
+#include "SIInstrInfo.h"
+#include "SIMachineFunctionInfo.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/CodeGen/MachineScheduler.h"
using namespace llvm;
+#define DEBUG_TYPE "amdgpu-subtarget"
+
#define GET_SUBTARGETINFO_ENUM
#define GET_SUBTARGETINFO_TARGET_DESC
#define GET_SUBTARGETINFO_CTOR
#include "AMDGPUGenSubtargetInfo.inc"
-AMDGPUSubtarget::AMDGPUSubtarget(StringRef TT, StringRef CPU, StringRef FS) :
- AMDGPUGenSubtargetInfo(TT, CPU, FS), DumpCode(false) {
- InstrItins = getInstrItineraryForCPU(CPU);
-
- // Default card
- StringRef GPU = CPU;
- Is64bit = false;
- DefaultSize[0] = 64;
- DefaultSize[1] = 1;
- DefaultSize[2] = 1;
- HasVertexCache = false;
- TexVTXClauseSize = 0;
- Gen = AMDGPUSubtarget::R600;
- FP64 = false;
- CaymanISA = false;
- EnableIRStructurizer = true;
- EnableIfCvt = true;
- WavefrontSize = 0;
- CFALUBug = false;
- ParseSubtargetFeatures(GPU, FS);
- DevName = GPU;
-}
+AMDGPUSubtarget &
+AMDGPUSubtarget::initializeSubtargetDependencies(StringRef TT, StringRef GPU, StringRef FS) {
+ // Determine default and user-specified characteristics
+ // On SI+, we want FP64 denormals to be on by default. FP32 denormals can be
+ // enabled, but some instructions do not respect them and they run at the
+ // double precision rate, so don't enable by default.
+ //
+ // We want to be able to turn these off, but making this a subtarget feature
+ // for SI has the unhelpful behavior that it unsets everything else if you
+ // disable it.
-bool
-AMDGPUSubtarget::is64bit() const {
- return Is64bit;
-}
-bool
-AMDGPUSubtarget::hasVertexCache() const {
- return HasVertexCache;
-}
-short
-AMDGPUSubtarget::getTexVTXClauseSize() const {
- return TexVTXClauseSize;
-}
-enum AMDGPUSubtarget::Generation
-AMDGPUSubtarget::getGeneration() const {
- return Gen;
-}
-bool
-AMDGPUSubtarget::hasHWFP64() const {
- return FP64;
-}
-bool
-AMDGPUSubtarget::hasCaymanISA() const {
- return CaymanISA;
-}
-bool
-AMDGPUSubtarget::IsIRStructurizerEnabled() const {
- return EnableIRStructurizer;
-}
-bool
-AMDGPUSubtarget::isIfCvtEnabled() const {
- return EnableIfCvt;
+ SmallString<256> FullFS("+promote-alloca,+fp64-denormals,");
+ FullFS += FS;
+
+ if (GPU == "" && Triple(TT).getArch() == Triple::amdgcn)
+ GPU = "SI";
+
+ ParseSubtargetFeatures(GPU, FullFS);
+
+ // FIXME: I don't think think Evergreen has any useful support for
+ // denormals, but should be checked. Should we issue a warning somewhere
+ // if someone tries to enable these?
+ if (getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS) {
+ FP32Denormals = false;
+ FP64Denormals = false;
+ }
+ return *this;
}
-unsigned
-AMDGPUSubtarget::getWavefrontSize() const {
- return WavefrontSize;
+
+AMDGPUSubtarget::AMDGPUSubtarget(StringRef TT, StringRef GPU, StringRef FS,
+ TargetMachine &TM)
+ : AMDGPUGenSubtargetInfo(TT, GPU, FS), DevName(GPU), Is64bit(false),
+ DumpCode(false), R600ALUInst(false), HasVertexCache(false),
+ TexVTXClauseSize(0), Gen(AMDGPUSubtarget::R600), FP64(false),
+ FP64Denormals(false), FP32Denormals(false),
+ FastFMAF32(false), CaymanISA(false),
+ FlatAddressSpace(false), EnableIRStructurizer(true),
+ EnablePromoteAlloca(false), EnableIfCvt(true),
+ EnableLoadStoreOpt(false), WavefrontSize(0), CFALUBug(false), LocalMemorySize(0),
+ EnableVGPRSpilling(false),
+ FrameLowering(TargetFrameLowering::StackGrowsUp,
+ 64 * 16, // Maximum stack alignment (long16)
+ 0),
+ InstrItins(getInstrItineraryForCPU(GPU)),
+ TargetTriple(TT) {
+
+ initializeSubtargetDependencies(TT, GPU, FS);
+
+ if (getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS) {
+ InstrInfo.reset(new R600InstrInfo(*this));
+ TLInfo.reset(new R600TargetLowering(TM, *this));
+ } else {
+ InstrInfo.reset(new SIInstrInfo(*this));
+ TLInfo.reset(new SITargetLowering(TM, *this));
+ }
}
-unsigned
-AMDGPUSubtarget::getStackEntrySize() const {
+
+unsigned AMDGPUSubtarget::getStackEntrySize() const {
assert(getGeneration() <= NORTHERN_ISLANDS);
switch(getWavefrontSize()) {
case 16:
return 8;
case 32:
- if (hasCaymanISA())
- return 4;
- else
- return 8;
+ return hasCaymanISA() ? 4 : 8;
case 64:
return 4;
default:
llvm_unreachable("Illegal wavefront size.");
}
}
-bool
-AMDGPUSubtarget::hasCFAluBug() const {
- assert(getGeneration() <= NORTHERN_ISLANDS);
- return CFALUBug;
-}
-bool
-AMDGPUSubtarget::isTargetELF() const {
- return false;
-}
-size_t
-AMDGPUSubtarget::getDefaultSize(uint32_t dim) const {
- if (dim > 2) {
- return 1;
- } else {
- return DefaultSize[dim];
+
+unsigned AMDGPUSubtarget::getAmdKernelCodeChipID() const {
+ switch(getGeneration()) {
+ default: llvm_unreachable("ChipID unknown");
+ case SEA_ISLANDS: return 12;
}
}
-std::string
-AMDGPUSubtarget::getDeviceName() const {
- return DevName;
+bool AMDGPUSubtarget::isVGPRSpillingEnabled(
+ const SIMachineFunctionInfo *MFI) const {
+ return MFI->getShaderType() == ShaderType::COMPUTE || EnableVGPRSpilling;
+}
+
+void AMDGPUSubtarget::overrideSchedPolicy(MachineSchedPolicy &Policy,
+ MachineInstr *begin,
+ MachineInstr *end,
+ unsigned NumRegionInstrs) const {
+ if (getGeneration() >= SOUTHERN_ISLANDS) {
+
+ // Track register pressure so the scheduler can try to decrease
+ // pressure once register usage is above the threshold defined by
+ // SIRegisterInfo::getRegPressureSetLimit()
+ Policy.ShouldTrackPressure = true;
+
+ // Enabling both top down and bottom up scheduling seems to give us less
+ // register spills than just using one of these approaches on its own.
+ Policy.OnlyTopDown = false;
+ Policy.OnlyBottomUp = false;
+ }
}