[AMDGPU] Pass to propagate ABI attributes from kernels to the functions

The pass works in two modes:

Mode 1: Just set attributes starting from kernels. This can work at
the very beginning of opt and llc pipeline, but cannot clone functions
because it must be a function pass.

Mode 2: Actually clone functions for new attributes. This can only work
after all function passes in the opt pipeline because it has to be a
module pass.

Differential Revision: https://reviews.llvm.org/D63208

llvm-svn: 363586

1 files changed, 11 insertions, 4 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 41876eb7125..41a075756ee 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -217,6 +217,8 @@ extern "C" void LLVMInitializeAMDGPUTarget() {
   initializeAMDGPUOpenCLEnqueuedBlockLoweringPass(*PR);
   initializeAMDGPUPromoteAllocaPass(*PR);
   initializeAMDGPUCodeGenPreparePass(*PR);
+  initializeAMDGPUPropagateAttributesEarlyPass(*PR);
+  initializeAMDGPUPropagateAttributesLatePass(*PR);
   initializeAMDGPURewriteOutArgumentsPass(*PR);
   initializeAMDGPUUnifyMetadataPass(*PR);
   initializeSIAnnotateControlFlowPass(*PR);
@@ -402,13 +404,14 @@ void AMDGPUTargetMachine::adjustPassManager(PassManagerBuilder &Builder) {
 
   Builder.addExtension(
     PassManagerBuilder::EP_ModuleOptimizerEarly,
-    [Internalize, EarlyInline, AMDGPUAA](const PassManagerBuilder &,
-                                         legacy::PassManagerBase &PM) {
+    [Internalize, EarlyInline, AMDGPUAA, this](const PassManagerBuilder &,
+                                               legacy::PassManagerBase &PM) {
       if (AMDGPUAA) {
         PM.add(createAMDGPUAAWrapperPass());
         PM.add(createAMDGPUExternalAAWrapperPass());
       }
       PM.add(createAMDGPUUnifyMetadataPass());
+      PM.add(createAMDGPUPropagateAttributesLatePass(this));
       if (Internalize) {
         PM.add(createInternalizePass(mustPreserveGV));
         PM.add(createGlobalDCEPass());
@@ -420,12 +423,13 @@ void AMDGPUTargetMachine::adjustPassManager(PassManagerBuilder &Builder) {
   const auto &Opt = Options;
   Builder.addExtension(
     PassManagerBuilder::EP_EarlyAsPossible,
-    [AMDGPUAA, LibCallSimplify, &Opt](const PassManagerBuilder &,
-                                      legacy::PassManagerBase &PM) {
+    [AMDGPUAA, LibCallSimplify, &Opt, this](const PassManagerBuilder &,
+                                            legacy::PassManagerBase &PM) {
       if (AMDGPUAA) {
         PM.add(createAMDGPUAAWrapperPass());
         PM.add(createAMDGPUExternalAAWrapperPass());
       }
+      PM.add(llvm::createAMDGPUPropagateAttributesEarlyPass(this));
       PM.add(llvm::createAMDGPUUseNativeCallsPass());
       if (LibCallSimplify)
         PM.add(llvm::createAMDGPUSimplifyLibCallsPass(Opt));
@@ -654,6 +658,9 @@ void AMDGPUPassConfig::addIRPasses() {
   disablePass(&FuncletLayoutID);
   disablePass(&PatchableFunctionID);
 
+  // A call to propagate attributes pass in the backend in case opt was not run.
+  addPass(createAMDGPUPropagateAttributesEarlyPass(&TM));
+
   addPass(createAtomicExpandPass());
 
   // This must occur before inlining, as the inliner will not look through