[X86][SSE] Vector integer/float conversion memory folding

[oota-llvm.git] / lib / Target / R600 / AMDGPUTargetMachine.cpp

diff --git a/lib/Target/R600/AMDGPUTargetMachine.cpp b/lib/Target/R600/AMDGPUTargetMachine.cpp
index 7b069e7760471aa631f3efa43b75e50d0c8469c5..1b4fe832f20b80d1867a02160a8398233db0b366 100644 (file)
--- a/lib/Target/R600/AMDGPUTargetMachine.cpp
+++ b/lib/Target/R600/AMDGPUTargetMachine.cpp
@@ -17,13 +17,14 @@
 #include "AMDGPU.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/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"
@@ -39,28 +40,23 @@ extern "C" void LLVMInitializeR600Target() {
   RegisterTargetMachine<AMDGPUTargetMachine> X(TheAMDGPUTarget);
 }
 
+static ScheduleDAGInstrs *createR600MachineScheduler(MachineSchedContext *C) {
+  return new ScheduleDAGMILive(C, make_unique<R600SchedStrategy>());
+}
+
+static MachineSchedRegistry
+SchedCustomRegistry("r600", "Run R600's custom scheduler",
+                    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),
+      Subtarget(TT, CPU, FS, *this), IntrinsicInfo() {
+  setRequiresStructuredCFG(true);
+  initAsmInfo();
 }
 
 AMDGPUTargetMachine::~AMDGPUTargetMachine() {
@@ -76,12 +72,21 @@ public:
     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 addCodeGenPrepare() override;
+  bool addPreISel() override;
+  bool addInstSelector() override;
+  bool addPreRegAlloc() override;
+  bool addPostRegAlloc() override;
+  bool addPreSched2() override;
+  bool addPreEmitPass() override;
 };
 } // End of anonymous namespace
 
@@ -89,60 +94,110 @@ TargetPassConfig *AMDGPUTargetMachine::createPassConfig(PassManagerBase &PM) {
   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::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() {
-  addPass(createAMDGPUPeepholeOpt(*TM));
   addPass(createAMDGPUISelDag(getAMDGPUTargetMachine()));
+  addPass(createSILowerI1CopiesPass());
   return false;
 }
 
 bool AMDGPUPassConfig::addPreRegAlloc() {
   const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
 
-  if (ST.device()->getGeneration() > AMDGPUDeviceInfo::HD6XXX) {
-    addPass(createSIAssignInterpRegsPass(*TM));
+  if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS) {
+    addPass(createR600VectorRegMerger(*TM));
+  } else {
+    addPass(createSIFixSGPRCopiesPass(*TM));
+    // SIFixSGPRCopies can generate a lot of duplicate instructions,
+    // so we need to run MachineCSE afterwards.
+    addPass(&MachineCSEID);
+
+    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());
+    addPass(createSIFixSGPRLiveRangesPass());
   }
-  addPass(createAMDGPUConvertToISAPass(*TM));
   return false;
 }
 
 bool AMDGPUPassConfig::addPostRegAlloc() {
   const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
 
-  if (ST.device()->getGeneration() > AMDGPUDeviceInfo::HD6XXX) {
+  addPass(createSIShrinkInstructionsPass());
+  if (ST.getGeneration() > AMDGPUSubtarget::NORTHERN_ISLANDS) {
     addPass(createSIInsertWaits(*TM));
   }
   return false;
 }
 
 bool AMDGPUPassConfig::addPreSched2() {
+  const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
 
-  addPass(&IfConverterID);
+  if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS)
+    addPass(createR600EmitClauseMarkers());
+  if (ST.isIfCvtEnabled())
+    addPass(&IfConverterID);
+  if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS)
+    addPass(createR600ClauseMergePass(*TM));
   return false;
 }
 
 bool AMDGPUPassConfig::addPreEmitPass() {
   const AMDGPUSubtarget &ST = TM->getSubtarget<AMDGPUSubtarget>();
-  if (ST.device()->getGeneration() <= AMDGPUDeviceInfo::HD6XXX) {
-    addPass(createAMDGPUCFGPreparationPass(*TM));
-    addPass(createAMDGPUCFGStructurizerPass(*TM));
+  if (ST.getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS) {
+    addPass(createAMDGPUCFGStructurizerPass());
     addPass(createR600ExpandSpecialInstrsPass(*TM));
-    addPass(createR600LowerConstCopy(*TM));
     addPass(&FinalizeMachineBundlesID);
+    addPass(createR600Packetizer(*TM));
+    addPass(createR600ControlFlowFinalizer(*TM));
   } else {
-    addPass(createSILowerLiteralConstantsPass(*TM));
     addPass(createSILowerControlFlowPass(*TM));
   }
 
   return false;
 }
-