[llvm] [AMDGPU] Enable AMDGPUAttributorPass in full LTO (PR #102673)

[Shilei Tian via llvm-commits]

https://github.com/shiltian updated https://github.com/llvm/llvm-project/pull/102673

>From 85aab7855c0bb0f4316a0259a541fcd900ef53fe Mon Sep 17 00:00:00 2001
From: Shilei Tian <i at tianshilei.me>
Date: Mon, 12 Aug 2024 13:38:23 -0400
Subject: [PATCH] [AMDGPU] Enable AMDGPUAttributorPass in full LTO

---
 llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp    |  7 +++++++
 llvm/test/CodeGen/AMDGPU/print-pipeline-passes.ll | 12 ++++++++++++
 2 files changed, 19 insertions(+)
 create mode 100644 llvm/test/CodeGen/AMDGPU/print-pipeline-passes.ll

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 0523fee5bcf9f4..d82be9a7e9041a 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -383,6 +383,11 @@ static cl::opt<bool> EnableHipStdPar(
   cl::desc("Enable HIP Standard Parallelism Offload support"), cl::init(false),
   cl::Hidden);
 
+static cl::opt<bool>
+    EnableAMDGPUAttributor("amdgpu-attributor-enable",
+                           cl::desc("Enable AMDGPUAttributorPass"),
+                           cl::init(true), cl::Hidden);
+
 extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeAMDGPUTarget() {
   // Register the target
   RegisterTargetMachine<R600TargetMachine> X(getTheR600Target());
@@ -762,6 +767,8 @@ void AMDGPUTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
         // module is partitioned for codegen.
         if (EnableLowerModuleLDS)
           PM.addPass(AMDGPULowerModuleLDSPass(*this));
+        if (EnableAMDGPUAttributor && Level != OptimizationLevel::O0)
+          PM.addPass(AMDGPUAttributorPass(*this));
       });
 
   PB.registerRegClassFilterParsingCallback(
diff --git a/llvm/test/CodeGen/AMDGPU/print-pipeline-passes.ll b/llvm/test/CodeGen/AMDGPU/print-pipeline-passes.ll
new file mode 100644
index 00000000000000..b9eda0c1cd3bb6
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/print-pipeline-passes.ll
@@ -0,0 +1,12 @@
+; RUN: opt -mtriple=amdgcn--amdhsa -S -passes="lto<O0>" -print-pipeline-passes %s -o - | FileCheck --check-prefix=O0 %s
+; RUN: opt -mtriple=amdgcn--amdhsa -S -passes="lto<O1>" -print-pipeline-passes %s -o - | FileCheck %s
+; RUN: opt -mtriple=amdgcn--amdhsa -S -passes="lto<O2>" -print-pipeline-passes %s -o - | FileCheck %s
+; RUN: opt -mtriple=amdgcn--amdhsa -S -passes="lto<O3>" -print-pipeline-passes %s -o - | FileCheck %s
+
+; CHECK: amdgpu-attributor
+; O0-NOT: amdgpu-attributor
+
+define amdgpu_kernel void @kernel() {
+entry:
+  ret void
+}