#include "AMDGPUTargetMachine.h"
#include "AMDGPU.h"
+#include "AMDGPUTargetTransformInfo.h"
#include "R600ISelLowering.h"
#include "R600InstrInfo.h"
+#include "R600MachineScheduler.h"
#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/IR/LegacyPassManager.h"
#include "llvm/Support/TargetRegistry.h"
#include "llvm/Support/raw_os_ostream.h"
#include "llvm/Transforms/IPO.h"
extern "C" void LLVMInitializeR600Target() {
// Register the target
- RegisterTargetMachine<AMDGPUTargetMachine> X(TheAMDGPUTarget);
+ RegisterTargetMachine<R600TargetMachine> X(TheAMDGPUTarget);
+ RegisterTargetMachine<GCNTargetMachine> Y(TheGCNTarget);
}
-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);
+static ScheduleDAGInstrs *createR600MachineScheduler(MachineSchedContext *C) {
+ return new ScheduleDAGMILive(C, make_unique<R600SchedStrategy>());
+}
+
+static MachineSchedRegistry
+SchedCustomRegistry("r600", "Run R600's custom scheduler",
+ createR600MachineScheduler);
+
+static std::string computeDataLayout(StringRef TT) {
+ Triple Triple(TT);
+ std::string Ret = "e-p:32:32";
+
+ if (Triple.getArch() == Triple::amdgcn) {
+ // 32-bit private, local, and region pointers. 64-bit global and constant.
+ Ret += "-p1:64:64-p2:64:64-p3:32:32-p4:64:64-p5:32:32-p24:64:64";
}
+
+ Ret += "-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256"
+ "-v512:512-v1024:1024-v2048:2048-n32:64";
+
+ return Ret;
+}
+
+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),
+ DL(computeDataLayout(TT)),
+ TLOF(new TargetLoweringObjectFileELF()),
+ Subtarget(TT, CPU, FS, *this), IntrinsicInfo() {
+ setRequiresStructuredCFG(true);
+ initAsmInfo();
}
AMDGPUTargetMachine::~AMDGPUTargetMachine() {
+ delete TLOF;
}
+//===----------------------------------------------------------------------===//
+// R600 Target Machine (R600 -> Cayman)
+//===----------------------------------------------------------------------===//
+
+R600TargetMachine::R600TargetMachine(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) { }
+
+
+//===----------------------------------------------------------------------===//
+// 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) { }
+
+//===----------------------------------------------------------------------===//
+// AMDGPU Pass Setup
+//===----------------------------------------------------------------------===//
+
namespace {
class AMDGPUPassConfig : public TargetPassConfig {
public:
- AMDGPUPassConfig(AMDGPUTargetMachine *TM, PassManagerBase &PM)
+ AMDGPUPassConfig(TargetMachine *TM, PassManagerBase &PM)
: 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;
+ virtual bool addPreISel() override;
+ virtual bool addInstSelector() override;
+};
+
+class R600PassConfig : public AMDGPUPassConfig {
+public:
+ R600PassConfig(TargetMachine *TM, PassManagerBase &PM)
+ : AMDGPUPassConfig(TM, PM) { }
+
+ bool addPreISel() override;
+ void addPreRegAlloc() override;
+ void addPreSched2() override;
+ void addPreEmitPass() override;
+};
+
+class GCNPassConfig : public AMDGPUPassConfig {
+public:
+ GCNPassConfig(TargetMachine *TM, PassManagerBase &PM)
+ : AMDGPUPassConfig(TM, PM) { }
+ bool addPreISel() override;
+ bool addInstSelector() override;
+ void addPreRegAlloc() override;
+ void addPostRegAlloc() override;
+ void addPreSched2() override;
+ void addPreEmitPass() override;
};
+
} // End of anonymous namespace
-TargetPassConfig *AMDGPUTargetMachine::createPassConfig(PassManagerBase &PM) {
- return new AMDGPUPassConfig(this, PM);
+TargetIRAnalysis AMDGPUTargetMachine::getTargetIRAnalysis() {
+ return TargetIRAnalysis(
+ [this](Function &F) { return TargetTransformInfo(AMDGPUTTIImpl(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(createSIAnnotateControlFlowPass());
- }
+ addPass(createFlattenCFGPass());
+ if (ST.IsIRStructurizerEnabled())
+ addPass(createStructurizeCFGPass());
return false;
}
bool AMDGPUPassConfig::addInstSelector() {
- addPass(createAMDGPUPeepholeOpt(*TM));
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));
- }
return false;
}
-bool AMDGPUPassConfig::addPreRegAlloc() {
- const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
+//===----------------------------------------------------------------------===//
+// R600 Pass Setup
+//===----------------------------------------------------------------------===//
- if (ST.device()->getGeneration() > AMDGPUDeviceInfo::HD6XXX) {
- addPass(createSIAssignInterpRegsPass(*TM));
- }
- addPass(createAMDGPUConvertToISAPass(*TM));
+bool R600PassConfig::addPreISel() {
+ AMDGPUPassConfig::addPreISel();
+ addPass(createR600TextureIntrinsicsReplacer());
return false;
}
-bool AMDGPUPassConfig::addPostRegAlloc() {
+void R600PassConfig::addPreRegAlloc() {
+ addPass(createR600VectorRegMerger(*TM));
+}
+
+void R600PassConfig::addPreSched2() {
const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
+ addPass(createR600EmitClauseMarkers(), false);
+ if (ST.isIfCvtEnabled())
+ addPass(&IfConverterID, false);
+ addPass(createR600ClauseMergePass(*TM), false);
+}
- if (ST.device()->getGeneration() > AMDGPUDeviceInfo::HD6XXX) {
- addPass(createSIInsertWaits(*TM));
- }
- return false;
+void R600PassConfig::addPreEmitPass() {
+ addPass(createAMDGPUCFGStructurizerPass(), false);
+ addPass(createR600ExpandSpecialInstrsPass(*TM), false);
+ addPass(&FinalizeMachineBundlesID, false);
+ addPass(createR600Packetizer(*TM), false);
+ addPass(createR600ControlFlowFinalizer(*TM), false);
+}
+
+TargetPassConfig *R600TargetMachine::createPassConfig(PassManagerBase &PM) {
+ return new R600PassConfig(this, PM);
}
-bool AMDGPUPassConfig::addPreSched2() {
+//===----------------------------------------------------------------------===//
+// GCN Pass Setup
+//===----------------------------------------------------------------------===//
+
+bool GCNPassConfig::addPreISel() {
+ AMDGPUPassConfig::addPreISel();
+ addPass(createSinkingPass());
+ addPass(createSITypeRewriter());
+ addPass(createSIAnnotateControlFlowPass());
+ return false;
+}
- addPass(&IfConverterID);
+bool GCNPassConfig::addInstSelector() {
+ AMDGPUPassConfig::addInstSelector();
+ addPass(createSILowerI1CopiesPass());
+ addPass(createSIFixSGPRCopiesPass(*TM));
+ addPass(createSIFoldOperandsPass());
return false;
}
-bool AMDGPUPassConfig::addPreEmitPass() {
+void GCNPassConfig::addPreRegAlloc() {
const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
- if (ST.device()->getGeneration() <= AMDGPUDeviceInfo::HD6XXX) {
- addPass(createAMDGPUCFGPreparationPass(*TM));
- addPass(createAMDGPUCFGStructurizerPass(*TM));
- addPass(createR600ExpandSpecialInstrsPass(*TM));
- addPass(&FinalizeMachineBundlesID);
- addPass(createR600LowerConstCopy(*TM));
- } else {
- addPass(createSILowerLiteralConstantsPass(*TM));
- addPass(createSILowerControlFlowPass(*TM));
+ 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);
+}
- return false;
+void GCNPassConfig::addPostRegAlloc() {
+ addPass(createSIPrepareScratchRegs(), false);
+ addPass(createSIShrinkInstructionsPass(), false);
+}
+
+void GCNPassConfig::addPreSched2() {
+ addPass(createSIInsertWaits(*TM), false);
+}
+
+void GCNPassConfig::addPreEmitPass() {
+ addPass(createSILowerControlFlowPass(*TM), false);
}
+TargetPassConfig *GCNTargetMachine::createPassConfig(PassManagerBase &PM) {
+ return new GCNPassConfig(this, PM);
+}
projects / oota-llvm.git / blobdiff