AMDGPU: Add pass to detect used kernel features

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

diff --git a/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index ab40d170a4cc91fc583ff98e85bb258348f3899b..4e31c7ab4d4c46c91fd17221638cc791046cecaf 100644 (file)
--- a/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -50,6 +50,7 @@ extern "C" void LLVMInitializeAMDGPUTarget() {
   initializeSIFixSGPRLiveRangesPass(*PR);
   initializeSIFixControlFlowLiveIntervalsPass(*PR);
   initializeSILoadStoreOptimizerPass(*PR);
   initializeSIFixSGPRLiveRangesPass(*PR);
   initializeSIFixControlFlowLiveIntervalsPass(*PR);
   initializeSILoadStoreOptimizerPass(*PR);

+  initializeAMDGPUAnnotateKernelFeaturesPass(*PR);

 }
 
 static std::unique_ptr<TargetLoweringObjectFile> createTLOF(const Triple &TT) {
 }
 
 static std::unique_ptr<TargetLoweringObjectFile> createTLOF(const Triple &TT) {


@@ -195,8 +196,10 @@ void AMDGPUPassConfig::addIRPasses() {
   // functions, then we will generate code for the first function
   // without ever running any passes on the second.
   addPass(createBarrierNoopPass());
   // 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());
   // Handle uses of OpenCL image2d_t, image3d_t and sampler_t arguments.
   addPass(createAMDGPUOpenCLImageTypeLoweringPass());

+

   TargetPassConfig::addIRPasses();
 }
 
   TargetPassConfig::addIRPasses();
 }
 


@@ -268,6 +271,11 @@ TargetPassConfig *R600TargetMachine::createPassConfig(PassManagerBase &PM) {
 
 bool GCNPassConfig::addPreISel() {
   AMDGPUPassConfig::addPreISel();
 
 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(createSinkingPass());
   addPass(createSITypeRewriter());
   addPass(createSIAnnotateControlFlowPass());