#include "SIISelLowering.h"
#include "SIInstrInfo.h"
#include "llvm/Analysis/Passes.h"
-#include "llvm/Analysis/Verifier.h"
#include "llvm/CodeGen/MachineFunctionAnalysis.h"
+#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/Passes.h"
+#include "llvm/IR/Verifier.h"
#include "llvm/MC/MCAsmInfo.h"
#include "llvm/PassManager.h"
#include "llvm/Support/TargetRegistry.h"
extern "C" void LLVMInitializeR600Target() {
// Register the target
RegisterTargetMachine<AMDGPUTargetMachine> X(TheAMDGPUTarget);
+ RegisterTargetMachine<GCNTargetMachine> Y(TheGCNTarget);
}
static ScheduleDAGInstrs *createR600MachineScheduler(MachineSchedContext *C) {
- return new ScheduleDAGMI(C, new R600SchedStrategy());
+ return new ScheduleDAGMILive(C, make_unique<R600SchedStrategy>());
}
static MachineSchedRegistry
createR600MachineScheduler);
AMDGPUTargetMachine::AMDGPUTargetMachine(const Target &T, StringRef TT,
- StringRef CPU, StringRef FS,
- TargetOptions Options,
- Reloc::Model RM, CodeModel::Model CM,
- CodeGenOpt::Level OptLevel
-)
-:
- LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OptLevel),
- Subtarget(TT, CPU, FS),
- Layout(Subtarget.getDataLayout()),
- FrameLowering(TargetFrameLowering::StackGrowsUp,
- Subtarget.device()->getStackAlignment(), 0),
- IntrinsicInfo(this),
- InstrItins(&Subtarget.getInstrItineraryData()) {
- // TLInfo uses InstrInfo so it must be initialized after.
- if (Subtarget.device()->getGeneration() <= AMDGPUDeviceInfo::HD6XXX) {
- InstrInfo = new R600InstrInfo(*this);
- TLInfo = new R600TargetLowering(*this);
- } else {
- InstrInfo = new SIInstrInfo(*this);
- TLInfo = new SITargetLowering(*this);
- }
+ StringRef CPU, StringRef FS,
+ TargetOptions Options, Reloc::Model RM,
+ CodeModel::Model CM,
+ CodeGenOpt::Level OptLevel)
+ : LLVMTargetMachine(T, TT, CPU, FS, Options, RM, CM, OptLevel),
+ TLOF(new TargetLoweringObjectFileELF()),
+ Subtarget(TT, CPU, FS, *this), IntrinsicInfo() {
+ setRequiresStructuredCFG(true);
+ initAsmInfo();
}
AMDGPUTargetMachine::~AMDGPUTargetMachine() {
+ delete TLOF;
}
namespace {
class AMDGPUPassConfig : public TargetPassConfig {
public:
AMDGPUPassConfig(AMDGPUTargetMachine *TM, PassManagerBase &PM)
- : TargetPassConfig(TM, PM) {
- const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
- if (ST.device()->getGeneration() <= AMDGPUDeviceInfo::HD6XXX) {
- enablePass(&MachineSchedulerID);
- MachineSchedRegistry::setDefault(createR600MachineScheduler);
- }
- }
+ : TargetPassConfig(TM, PM) {}
AMDGPUTargetMachine &getAMDGPUTargetMachine() const {
return getTM<AMDGPUTargetMachine>();
}
- virtual bool addPreISel();
- virtual bool addInstSelector();
- virtual bool addPreRegAlloc();
- virtual bool addPostRegAlloc();
- virtual bool addPreSched2();
- virtual bool addPreEmitPass();
+ ScheduleDAGInstrs *
+ createMachineScheduler(MachineSchedContext *C) const override {
+ const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
+ if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS)
+ return createR600MachineScheduler(C);
+ return nullptr;
+ }
+
+ void addIRPasses() override;
+ void addCodeGenPrepare() override;
+ bool addPreISel() override;
+ bool addInstSelector() override;
+ void addPreRegAlloc() override;
+ void addPostRegAlloc() override;
+ void addPreSched2() override;
+ void addPreEmitPass() override;
};
} // End of anonymous namespace
return new AMDGPUPassConfig(this, PM);
}
+//===----------------------------------------------------------------------===//
+// AMDGPU Analysis Pass Setup
+//===----------------------------------------------------------------------===//
+
+void AMDGPUTargetMachine::addAnalysisPasses(PassManagerBase &PM) {
+ // Add first the target-independent BasicTTI pass, then our AMDGPU pass. This
+ // allows the AMDGPU pass to delegate to the target independent layer when
+ // appropriate.
+ PM.add(createBasicTargetTransformInfoPass(this));
+ PM.add(createAMDGPUTargetTransformInfoPass(this));
+}
+
+void AMDGPUPassConfig::addIRPasses() {
+ // Function calls are not supported, so make sure we inline everything.
+ addPass(createAMDGPUAlwaysInlinePass());
+ addPass(createAlwaysInlinerPass());
+ // We need to add the barrier noop pass, otherwise adding the function
+ // inlining pass will cause all of the PassConfigs passes to be run
+ // one function at a time, which means if we have a nodule with two
+ // functions, then we will generate code for the first function
+ // without ever running any passes on the second.
+ addPass(createBarrierNoopPass());
+ TargetPassConfig::addIRPasses();
+}
+
+void AMDGPUPassConfig::addCodeGenPrepare() {
+ const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
+ if (ST.isPromoteAllocaEnabled()) {
+ addPass(createAMDGPUPromoteAlloca(ST));
+ addPass(createSROAPass());
+ }
+
+ TargetPassConfig::addCodeGenPrepare();
+}
+
bool
AMDGPUPassConfig::addPreISel() {
const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
- if (ST.device()->getGeneration() > AMDGPUDeviceInfo::HD6XXX) {
- addPass(createAMDGPUStructurizeCFGPass());
+ addPass(createFlattenCFGPass());
+ if (ST.IsIRStructurizerEnabled())
+ addPass(createStructurizeCFGPass());
+ if (ST.getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS) {
+ addPass(createSinkingPass());
+ addPass(createSITypeRewriter());
addPass(createSIAnnotateControlFlowPass());
+ } else {
+ addPass(createR600TextureIntrinsicsReplacer());
}
return false;
}
bool AMDGPUPassConfig::addInstSelector() {
+ const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
+
addPass(createAMDGPUISelDag(getAMDGPUTargetMachine()));
- const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
- if (ST.device()->getGeneration() <= AMDGPUDeviceInfo::HD6XXX) {
- // This callbacks this pass uses are not implemented yet on SI.
- addPass(createAMDGPUIndirectAddressingPass(*TM));
+ if (ST.getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS) {
+ addPass(createSILowerI1CopiesPass());
+ addPass(createSIFixSGPRCopiesPass(*TM));
+ addPass(createSIFoldOperandsPass());
}
+
return false;
}
-bool AMDGPUPassConfig::addPreRegAlloc() {
- addPass(createAMDGPUConvertToISAPass(*TM));
- return false;
+void AMDGPUPassConfig::addPreRegAlloc() {
+ const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
+
+ if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS) {
+ addPass(createR600VectorRegMerger(*TM));
+ } else {
+ if (getOptLevel() > CodeGenOpt::None && ST.loadStoreOptEnabled()) {
+ // Don't do this with no optimizations since it throws away debug info by
+ // merging nonadjacent loads.
+
+ // This should be run after scheduling, but before register allocation. It
+ // also need extra copies to the address operand to be eliminated.
+ initializeSILoadStoreOptimizerPass(*PassRegistry::getPassRegistry());
+ insertPass(&MachineSchedulerID, &SILoadStoreOptimizerID);
+ }
+
+ addPass(createSIShrinkInstructionsPass(), false);
+ addPass(createSIFixSGPRLiveRangesPass(), false);
+ }
}
-bool AMDGPUPassConfig::addPostRegAlloc() {
+void AMDGPUPassConfig::addPostRegAlloc() {
const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
- if (ST.device()->getGeneration() > AMDGPUDeviceInfo::HD6XXX) {
- addPass(createSIInsertWaits(*TM));
+ if (ST.getGeneration() > AMDGPUSubtarget::NORTHERN_ISLANDS) {
+ addPass(createSIPrepareScratchRegs(), false);
+ addPass(createSIShrinkInstructionsPass(), false);
}
- return false;
}
-bool AMDGPUPassConfig::addPreSched2() {
+void AMDGPUPassConfig::addPreSched2() {
+ const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
- addPass(&IfConverterID);
- return false;
+ if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS)
+ addPass(createR600EmitClauseMarkers(), false);
+ if (ST.isIfCvtEnabled())
+ addPass(&IfConverterID, false);
+ if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS)
+ addPass(createR600ClauseMergePass(*TM), false);
+ if (ST.getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS) {
+ addPass(createSIInsertWaits(*TM), false);
+ }
}
-bool AMDGPUPassConfig::addPreEmitPass() {
+void AMDGPUPassConfig::addPreEmitPass() {
const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
- if (ST.device()->getGeneration() <= AMDGPUDeviceInfo::HD6XXX) {
- addPass(createAMDGPUCFGPreparationPass(*TM));
- addPass(createAMDGPUCFGStructurizerPass(*TM));
- addPass(createR600EmitClauseMarkers(*TM));
- addPass(createR600ExpandSpecialInstrsPass(*TM));
- addPass(&FinalizeMachineBundlesID);
- addPass(createR600Packetizer(*TM));
- addPass(createR600ControlFlowFinalizer(*TM));
+ if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS) {
+ addPass(createAMDGPUCFGStructurizerPass(), false);
+ addPass(createR600ExpandSpecialInstrsPass(*TM), false);
+ addPass(&FinalizeMachineBundlesID, false);
+ addPass(createR600Packetizer(*TM), false);
+ addPass(createR600ControlFlowFinalizer(*TM), false);
} else {
- addPass(createSILowerControlFlowPass(*TM));
+ addPass(createSILowerControlFlowPass(*TM), false);
}
-
- return false;
}
+
+//===----------------------------------------------------------------------===//
+// GCN Target Machine (SI+)
+//===----------------------------------------------------------------------===//
+
+GCNTargetMachine::GCNTargetMachine(const Target &T, StringRef TT, StringRef FS,
+ StringRef CPU, TargetOptions Options, Reloc::Model RM,
+ CodeModel::Model CM, CodeGenOpt::Level OL) :
+ AMDGPUTargetMachine(T, TT, FS, CPU, Options, RM, CM, OL) { }
projects / oota-llvm.git / blobdiff