[llvm] [AMDGPU][NewPM] Port AMDGPUOpenCLEnqueuedBlockLowering to NPM (PR #122434)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-amdgpu

Author: Akshat Oke (optimisan)

<details>
<summary>Changes</summary>

Since this is an IR pass, it could be tested as a standalone pass with `opt` - but it cannot be since the generic PassBuilder does not know about it in opt.

I tested if it is working in the new pipeline with llc by passing `-stop-after=amdgpu-lower-enqueued-block`.

---
Full diff: https://github.com/llvm/llvm-project/pull/122434.diff


5 Files Affected:

- (modified) llvm/lib/Target/AMDGPU/AMDGPU.h (+3-3) 
- (modified) llvm/lib/Target/AMDGPU/AMDGPUOpenCLEnqueuedBlockLowering.cpp (+28-9) 
- (added) llvm/lib/Target/AMDGPU/AMDGPUOpenCLEnqueuedBlockLowering.h (+23) 
- (modified) llvm/lib/Target/AMDGPU/AMDGPUPassRegistry.def (+1) 
- (modified) llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp (+4-3) 


``````````diff
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.h b/llvm/lib/Target/AMDGPU/AMDGPU.h
index b9769a1baf4d17..f7a7e48295ae5b 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.h
@@ -440,9 +440,9 @@ void initializeAMDGPUExternalAAWrapperPass(PassRegistry&);
 
 void initializeAMDGPUArgumentUsageInfoPass(PassRegistry &);
 
-ModulePass *createAMDGPUOpenCLEnqueuedBlockLoweringPass();
-void initializeAMDGPUOpenCLEnqueuedBlockLoweringPass(PassRegistry &);
-extern char &AMDGPUOpenCLEnqueuedBlockLoweringID;
+ModulePass *createAMDGPUOpenCLEnqueuedBlockLoweringLegacyPass();
+void initializeAMDGPUOpenCLEnqueuedBlockLoweringLegacyPass(PassRegistry &);
+extern char &AMDGPUOpenCLEnqueuedBlockLoweringLegacyID;
 
 void initializeGCNNSAReassignPass(PassRegistry &);
 extern char &GCNNSAReassignID;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUOpenCLEnqueuedBlockLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUOpenCLEnqueuedBlockLowering.cpp
index 4f5ca08b46c139..fbd15ad176e3bd 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUOpenCLEnqueuedBlockLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUOpenCLEnqueuedBlockLowering.cpp
@@ -31,6 +31,7 @@
 //
 //===----------------------------------------------------------------------===//
 
+#include "AMDGPUOpenCLEnqueuedBlockLowering.h"
 #include "AMDGPU.h"
 #include "llvm/ADT/DenseSet.h"
 #include "llvm/ADT/SmallString.h"
@@ -48,11 +49,16 @@ using namespace llvm;
 namespace {
 
 /// Lower enqueued blocks.
-class AMDGPUOpenCLEnqueuedBlockLowering : public ModulePass {
+class AMDGPUOpenCLEnqueuedBlockLowering {
+public:
+  bool run(Module &M);
+};
+
+class AMDGPUOpenCLEnqueuedBlockLoweringLegacy : public ModulePass {
 public:
   static char ID;
 
-  explicit AMDGPUOpenCLEnqueuedBlockLowering() : ModulePass(ID) {}
+  explicit AMDGPUOpenCLEnqueuedBlockLoweringLegacy() : ModulePass(ID) {}
 
 private:
   bool runOnModule(Module &M) override;
@@ -60,19 +66,32 @@ class AMDGPUOpenCLEnqueuedBlockLowering : public ModulePass {
 
 } // end anonymous namespace
 
-char AMDGPUOpenCLEnqueuedBlockLowering::ID = 0;
+char AMDGPUOpenCLEnqueuedBlockLoweringLegacy::ID = 0;
 
-char &llvm::AMDGPUOpenCLEnqueuedBlockLoweringID =
-    AMDGPUOpenCLEnqueuedBlockLowering::ID;
+char &llvm::AMDGPUOpenCLEnqueuedBlockLoweringLegacyID =
+    AMDGPUOpenCLEnqueuedBlockLoweringLegacy::ID;
 
-INITIALIZE_PASS(AMDGPUOpenCLEnqueuedBlockLowering, DEBUG_TYPE,
+INITIALIZE_PASS(AMDGPUOpenCLEnqueuedBlockLoweringLegacy, DEBUG_TYPE,
                 "Lower OpenCL enqueued blocks", false, false)
 
-ModulePass* llvm::createAMDGPUOpenCLEnqueuedBlockLoweringPass() {
-  return new AMDGPUOpenCLEnqueuedBlockLowering();
+ModulePass *llvm::createAMDGPUOpenCLEnqueuedBlockLoweringLegacyPass() {
+  return new AMDGPUOpenCLEnqueuedBlockLoweringLegacy();
+}
+
+bool AMDGPUOpenCLEnqueuedBlockLoweringLegacy::runOnModule(Module &M) {
+  AMDGPUOpenCLEnqueuedBlockLowering Impl;
+  return Impl.run(M);
+}
+
+PreservedAnalyses
+AMDGPUOpenCLEnqueuedBlockLoweringPass::run(Module &M, ModuleAnalysisManager &) {
+  AMDGPUOpenCLEnqueuedBlockLowering Impl;
+  if (Impl.run(M))
+    return PreservedAnalyses::none();
+  return PreservedAnalyses::all();
 }
 
-bool AMDGPUOpenCLEnqueuedBlockLowering::runOnModule(Module &M) {
+bool AMDGPUOpenCLEnqueuedBlockLowering::run(Module &M) {
   DenseSet<Function *> Callers;
   auto &C = M.getContext();
   bool Changed = false;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUOpenCLEnqueuedBlockLowering.h b/llvm/lib/Target/AMDGPU/AMDGPUOpenCLEnqueuedBlockLowering.h
new file mode 100644
index 00000000000000..16ed7c18d8523e
--- /dev/null
+++ b/llvm/lib/Target/AMDGPU/AMDGPUOpenCLEnqueuedBlockLowering.h
@@ -0,0 +1,23 @@
+//===- AMDGPUOpenCLEnqueuedBlockLowering.h -----------------------*- C++-*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_LIB_TARGET_AMDGPU_OPENCLENQUEUEDBLOCKLOWERING_H
+#define LLVM_LIB_TARGET_AMDGPU_OPENCLENQUEUEDBLOCKLOWERING_H
+
+#include "llvm/IR/PassManager.h"
+
+namespace llvm {
+class AMDGPUOpenCLEnqueuedBlockLoweringPass
+    : public PassInfoMixin<AMDGPUOpenCLEnqueuedBlockLoweringPass> {
+public:
+  AMDGPUOpenCLEnqueuedBlockLoweringPass() = default;
+  PreservedAnalyses run(Module &M, ModuleAnalysisManager &MAM);
+};
+} // namespace llvm
+
+#endif // LLVM_LIB_TARGET_AMDGPU_OPENCLENQUEUEDBLOCKLOWERING_H
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUPassRegistry.def b/llvm/lib/Target/AMDGPU/AMDGPUPassRegistry.def
index 182e825a59a41b..1e04f30fa3e7dc 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUPassRegistry.def
+++ b/llvm/lib/Target/AMDGPU/AMDGPUPassRegistry.def
@@ -22,6 +22,7 @@ MODULE_PASS("amdgpu-lower-buffer-fat-pointers",
 MODULE_PASS("amdgpu-lower-ctor-dtor", AMDGPUCtorDtorLoweringPass())
 MODULE_PASS("amdgpu-sw-lower-lds", AMDGPUSwLowerLDSPass(*this))
 MODULE_PASS("amdgpu-lower-module-lds", AMDGPULowerModuleLDSPass(*this))
+MODULE_PASS("amdgpu-lower-enqueued-block", AMDGPUOpenCLEnqueuedBlockLoweringPass())
 MODULE_PASS("amdgpu-perf-hint",
             AMDGPUPerfHintAnalysisPass(
               *static_cast<const GCNTargetMachine *>(this)))
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 7256eec89008a5..2cba3860cfce09 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -22,6 +22,7 @@
 #include "AMDGPUIGroupLP.h"
 #include "AMDGPUISelDAGToDAG.h"
 #include "AMDGPUMacroFusion.h"
+#include "AMDGPUOpenCLEnqueuedBlockLowering.h"
 #include "AMDGPUPerfHintAnalysis.h"
 #include "AMDGPUSplitModule.h"
 #include "AMDGPUTargetObjectFile.h"
@@ -499,7 +500,7 @@ extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeAMDGPUTarget() {
   initializeAMDGPULowerKernelArgumentsPass(*PR);
   initializeAMDGPUPromoteKernelArgumentsPass(*PR);
   initializeAMDGPULowerKernelAttributesPass(*PR);
-  initializeAMDGPUOpenCLEnqueuedBlockLoweringPass(*PR);
+  initializeAMDGPUOpenCLEnqueuedBlockLoweringLegacyPass(*PR);
   initializeAMDGPUPostLegalizerCombinerPass(*PR);
   initializeAMDGPUPreLegalizerCombinerPass(*PR);
   initializeAMDGPURegBankCombinerPass(*PR);
@@ -1172,7 +1173,7 @@ void AMDGPUPassConfig::addIRPasses() {
     addPass(createR600OpenCLImageTypeLoweringPass());
 
   // Replace OpenCL enqueued block function pointers with global variables.
-  addPass(createAMDGPUOpenCLEnqueuedBlockLoweringPass());
+  addPass(createAMDGPUOpenCLEnqueuedBlockLoweringLegacyPass());
 
   // Lower LDS accesses to global memory pass if address sanitizer is enabled.
   if (EnableSwLowerLDS)
@@ -1939,7 +1940,7 @@ void AMDGPUCodeGenPassBuilder::addIRPasses(AddIRPass &addPass) const {
   addPass(AMDGPUAlwaysInlinePass());
   addPass(AlwaysInlinerPass());
 
-  // TODO: Missing OpenCLEnqueuedBlockLowering
+  addPass(AMDGPUOpenCLEnqueuedBlockLoweringPass());
 
   // Runs before PromoteAlloca so the latter can account for function uses
   if (EnableLowerModuleLDS)

``````````

</details>


https://github.com/llvm/llvm-project/pull/122434