AMDGPU/SI: Add SI Machine Scheduler

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

diff --git a/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 16fff50398ea5e6199fac01fe8a202a8a134fcc0..b1be6197a6c6a0a01a8b55d876f045a8af5a7fc4 100644 (file)
--- a/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -14,7 +14,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "AMDGPUTargetMachine.h"
-#include "AMDGPUHSATargetObjectFile.h"
+#include "AMDGPUTargetObjectFile.h"
 #include "AMDGPU.h"
 #include "AMDGPUTargetTransformInfo.h"
 #include "R600ISelLowering.h"
@@ -44,16 +44,21 @@ extern "C" void LLVMInitializeAMDGPUTarget() {
   RegisterTargetMachine<GCNTargetMachine> Y(TheGCNTarget);
 
   PassRegistry *PR = PassRegistry::getPassRegistry();
+  initializeSILowerI1CopiesPass(*PR);
+  initializeSIFixSGPRCopiesPass(*PR);
+  initializeSIFoldOperandsPass(*PR);
   initializeSIFixSGPRLiveRangesPass(*PR);
   initializeSIFixControlFlowLiveIntervalsPass(*PR);
   initializeSILoadStoreOptimizerPass(*PR);
+  initializeAMDGPUAnnotateKernelFeaturesPass(*PR);
+  initializeAMDGPUAnnotateUniformValuesPass(*PR);
 }
 
 static std::unique_ptr<TargetLoweringObjectFile> createTLOF(const Triple &TT) {
   if (TT.getOS() == Triple::AMDHSA)
     return make_unique<AMDGPUHSATargetObjectFile>();
 
-  return make_unique<TargetLoweringObjectFileELF>();
+  return make_unique<AMDGPUTargetObjectFile>();
 }
 
 static ScheduleDAGInstrs *createR600MachineScheduler(MachineSchedContext *C) {
@@ -61,8 +66,12 @@ static ScheduleDAGInstrs *createR600MachineScheduler(MachineSchedContext *C) {
 }
 
 static MachineSchedRegistry
-SchedCustomRegistry("r600", "Run R600's custom scheduler",
-                    createR600MachineScheduler);
+R600SchedRegistry("r600", "Run R600's custom scheduler",
+                   createR600MachineScheduler);
+
+static MachineSchedRegistry
+SISchedRegistry("si", "Run SI's custom scheduler",
+                createSIMachineScheduler);
 
 static std::string computeDataLayout(const Triple &TT) {
   std::string Ret = "e-p:32:32";
@@ -192,8 +201,10 @@ void AMDGPUPassConfig::addIRPasses() {
   // functions, then we will generate code for the first function
   // without ever running any passes on the second.
   addPass(createBarrierNoopPass());
+
   // Handle uses of OpenCL image2d_t, image3d_t and sampler_t arguments.
   addPass(createAMDGPUOpenCLImageTypeLoweringPass());
+
   TargetPassConfig::addIRPasses();
 }
 
@@ -265,16 +276,23 @@ TargetPassConfig *R600TargetMachine::createPassConfig(PassManagerBase &PM) {
 
 bool GCNPassConfig::addPreISel() {
   AMDGPUPassConfig::addPreISel();
+
+  // FIXME: We need to run a pass to propagate the attributes when calls are
+  // supported.
+  addPass(&AMDGPUAnnotateKernelFeaturesID);
+
   addPass(createSinkingPass());
   addPass(createSITypeRewriter());
   addPass(createSIAnnotateControlFlowPass());
+  addPass(createAMDGPUAnnotateUniformValues());
+
   return false;
 }
 
 bool GCNPassConfig::addInstSelector() {
   AMDGPUPassConfig::addInstSelector();
   addPass(createSILowerI1CopiesPass());
-  addPass(createSIFixSGPRCopiesPass(*TM));
+  addPass(&SIFixSGPRCopiesID);
   addPass(createSIFoldOperandsPass());
   return false;
 }
@@ -309,12 +327,13 @@ void GCNPassConfig::addFastRegAlloc(FunctionPass *RegAllocPass) {
 void GCNPassConfig::addOptimizedRegAlloc(FunctionPass *RegAllocPass) {
   // We want to run this after LiveVariables is computed to avoid computing them
   // twice.
-  insertPass(&LiveVariablesID, &SIFixSGPRLiveRangesID);
+  // FIXME: We shouldn't disable the verifier here. r249087 introduced a failure
+  // that needs to be fixed.
+  insertPass(&LiveVariablesID, &SIFixSGPRLiveRangesID, /*VerifyAfter=*/false);
   TargetPassConfig::addOptimizedRegAlloc(RegAllocPass);
 }
 
 void GCNPassConfig::addPostRegAlloc() {
-  addPass(createSIPrepareScratchRegs(), false);
   addPass(createSIShrinkInstructionsPass(), false);
 }