[AMDGPU][Attributor] Add ThinOrFullLTOPhase as an argument (#123994)

Author: shiltian

llvm/include/llvm/Pass.h

@@ -86,6 +86,23 @@ enum class ThinOrFullLTOPhase {
   FullLTOPostLink
 };
 
+#ifndef NDEBUG
+constexpr const char *to_string(ThinOrFullLTOPhase Phase) {
+  switch (Phase) {
+  case ThinOrFullLTOPhase::None:
+    return "None";
+  case ThinOrFullLTOPhase::ThinLTOPreLink:
+    return "ThinLTOPreLink";
+  case ThinOrFullLTOPhase::ThinLTOPostLink:
+    return "ThinLTOPostLink";
+  case ThinOrFullLTOPhase::FullLTOPreLink:
+    return "FullLTOPreLink";
+  case ThinOrFullLTOPhase::FullLTOPostLink:
+    return "FullLTOPostLink";
+  }
+}
+#endif
+
 //===----------------------------------------------------------------------===//
 /// Pass interface - Implemented by all 'passes'.  Subclass this if you are an
 /// interprocedural optimization or you do not fit into any of the more

llvm/lib/Target/AMDGPU/AMDGPU.h

@@ -336,9 +336,12 @@ class AMDGPUAttributorPass : public PassInfoMixin<AMDGPUAttributorPass> {
 
   AMDGPUAttributorOptions Options;
 
+  const ThinOrFullLTOPhase LTOPhase;
+
 public:
-  AMDGPUAttributorPass(TargetMachine &TM, AMDGPUAttributorOptions Options = {})
-      : TM(TM), Options(Options) {};
+  AMDGPUAttributorPass(TargetMachine &TM, AMDGPUAttributorOptions Options,
+                       ThinOrFullLTOPhase LTOPhase = ThinOrFullLTOPhase::None)
+      : TM(TM), Options(Options), LTOPhase(LTOPhase) {};
   PreservedAnalyses run(Module &M, ModuleAnalysisManager &AM);
 };
 

llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp

@@ -1343,7 +1343,8 @@ static void addPreloadKernArgHint(Function &F, TargetMachine &TM) {
 }
 
 static bool runImpl(Module &M, AnalysisGetter &AG, TargetMachine &TM,
-                    AMDGPUAttributorOptions Options) {
+                    AMDGPUAttributorOptions Options,
+                    ThinOrFullLTOPhase LTOPhase) {
   SetVector<Function *> Functions;
   for (Function &F : M) {
     if (!F.isIntrinsic())
@@ -1378,9 +1379,13 @@ static bool runImpl(Module &M, AnalysisGetter &AG, TargetMachine &TM,
 
   Attributor A(Functions, InfoCache, AC);
 
-  LLVM_DEBUG(dbgs() << "[AMDGPUAttributor] Module " << M.getName() << " is "
-                    << (AC.IsClosedWorldModule ? "" : "not ")
-                    << "assumed to be a closed world.\n");
+  LLVM_DEBUG({
+    StringRef LTOPhaseStr = to_string(LTOPhase);
+    dbgs() << "[AMDGPUAttributor] Running at phase " << LTOPhaseStr << '\n'
+           << "[AMDGPUAttributor] Module " << M.getName() << " is "
+           << (AC.IsClosedWorldModule ? "" : "not ")
+           << "assumed to be a closed world.\n";
+  });
 
   for (auto *F : Functions) {
     A.getOrCreateAAFor<AAAMDAttributes>(IRPosition::function(*F));
@@ -1433,7 +1438,8 @@ class AMDGPUAttributorLegacy : public ModulePass {
 
   bool runOnModule(Module &M) override {
     AnalysisGetter AG(this);
-    return runImpl(M, AG, *TM, /*Options=*/{});
+    return runImpl(M, AG, *TM, /*Options=*/{},
+                   /*LTOPhase=*/ThinOrFullLTOPhase::None);
   }
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
@@ -1454,8 +1460,8 @@ PreservedAnalyses llvm::AMDGPUAttributorPass::run(Module &M,
   AnalysisGetter AG(FAM);
 
   // TODO: Probably preserves CFG
-  return runImpl(M, AG, TM, Options) ? PreservedAnalyses::none()
-                                     : PreservedAnalyses::all();
+  return runImpl(M, AG, TM, Options, LTOPhase) ? PreservedAnalyses::none()
+                                               : PreservedAnalyses::all();
 }
 
 char AMDGPUAttributorLegacy::ID = 0;

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

@@ -884,8 +884,10 @@ void AMDGPUTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
                                             OptimizationLevel Level,
                                             ThinOrFullLTOPhase Phase) {
     if (Level != OptimizationLevel::O0) {
-      if (!isLTOPreLink(Phase))
-        MPM.addPass(AMDGPUAttributorPass(*this));
+      if (!isLTOPreLink(Phase)) {
+        AMDGPUAttributorOptions Opts;
+        MPM.addPass(AMDGPUAttributorPass(*this, Opts, Phase));
+      }
     }
   });
 
@@ -914,7 +916,8 @@ void AMDGPUTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
             AMDGPUAttributorOptions Opt;
             if (HasClosedWorldAssumption)
               Opt.IsClosedWorld = true;
-            PM.addPass(AMDGPUAttributorPass(*this, Opt));
+            PM.addPass(AMDGPUAttributorPass(
+                *this, Opt, ThinOrFullLTOPhase::FullLTOPostLink));
           }
         }
         if (!NoKernelInfoEndLTO) {

llvm/test/LTO/AMDGPU/closed-world-assumption.ll

@@ -1,10 +1,12 @@
-; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -O3 -debug-only=amdgpu-attributor -o - %s 2>&1 | FileCheck %s --check-prefix=NO-CW
-; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes="lto<O3>" -debug-only=amdgpu-attributor -o - %s 2>&1 | FileCheck %s --check-prefix=NO-CW
-; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes="lto<O3>" -debug-only=amdgpu-attributor -amdgpu-link-time-closed-world=1 -o - %s 2>&1 | FileCheck %s --check-prefix=CW
+; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -O3 -debug-only=amdgpu-attributor -o - %s 2>&1 | FileCheck %s --check-prefixes=NO-CW,NO-LTO
+; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes="lto<O3>" -debug-only=amdgpu-attributor -o - %s 2>&1 | FileCheck %s --check-prefixes=NO-CW,LTO
+; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes="lto<O3>" -debug-only=amdgpu-attributor -amdgpu-link-time-closed-world=1 -o - %s 2>&1 | FileCheck %s --check-prefixes=CW,LTO
 
 ; REQUIRES: amdgpu-registered-target
 ; REQUIRES: asserts
 
+; NO-LTO: Running at phase None
+; LTO: Running at phase FullLTOPostLink
 ; NO-CW: Module {{.*}} is not assumed to be a closed world.
 ; CW: Module {{.*}} is assumed to be a closed world.
 define hidden noundef i32 @_Z3foov() {