[AMDGPU][NewPM] Port SIFixVGPRCopies to NPM (#123592)

Extends NPM pipeline support till PostRegAlloc passes (greedy is in the
works)

Author: optimisan

llvm/lib/Target/AMDGPU/AMDGPU.h

@@ -177,7 +177,7 @@ extern char &SIShrinkInstructionsLegacyID;
 void initializeSIFixSGPRCopiesLegacyPass(PassRegistry &);
 extern char &SIFixSGPRCopiesLegacyID;
 
-void initializeSIFixVGPRCopiesPass(PassRegistry &);
+void initializeSIFixVGPRCopiesLegacyPass(PassRegistry &);
 extern char &SIFixVGPRCopiesID;
 
 void initializeSILowerWWMCopiesPass(PassRegistry &);

llvm/lib/Target/AMDGPU/AMDGPUPassRegistry.def

@@ -99,6 +99,7 @@ FUNCTION_PASS_WITH_PARAMS(
 MACHINE_FUNCTION_PASS("amdgpu-isel", AMDGPUISelDAGToDAGPass(*this))
 MACHINE_FUNCTION_PASS("si-fix-sgpr-copies", SIFixSGPRCopiesPass())
 MACHINE_FUNCTION_PASS("si-i1-copies", SILowerI1CopiesPass())
+MACHINE_FUNCTION_PASS("si-fix-vgpr-copies", SIFixVGPRCopiesPass())
 MACHINE_FUNCTION_PASS("si-fold-operands", SIFoldOperandsPass());
 MACHINE_FUNCTION_PASS("gcn-dpp-combine", GCNDPPCombinePass())
 MACHINE_FUNCTION_PASS("si-load-store-opt", SILoadStoreOptimizerPass())

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

@@ -36,6 +36,7 @@
 #include "R600.h"
 #include "R600TargetMachine.h"
 #include "SIFixSGPRCopies.h"
+#include "SIFixVGPRCopies.h"
 #include "SIFoldOperands.h"
 #include "SILoadStoreOptimizer.h"
 #include "SILowerControlFlow.h"
@@ -486,7 +487,7 @@ extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeAMDGPUTarget() {
   initializeAMDGPUMarkLastScratchLoadPass(*PR);
   initializeSILowerSGPRSpillsLegacyPass(*PR);
   initializeSIFixSGPRCopiesLegacyPass(*PR);
-  initializeSIFixVGPRCopiesPass(*PR);
+  initializeSIFixVGPRCopiesLegacyPass(*PR);
   initializeSIFoldOperandsLegacyPass(*PR);
   initializeSIPeepholeSDWALegacyPass(*PR);
   initializeSIShrinkInstructionsLegacyPass(*PR);
@@ -2107,7 +2108,7 @@ void AMDGPUCodeGenPassBuilder::addMachineSSAOptimization(
 }
 
 void AMDGPUCodeGenPassBuilder::addPostRegAlloc(AddMachinePass &addPass) const {
-  // addPass(SIFixVGPRCopiesID);
+  addPass(SIFixVGPRCopiesPass());
   if (TM.getOptLevel() > CodeGenOptLevel::None)
     addPass(SIOptimizeExecMaskingPass());
   Base::addPostRegAlloc(addPass);

llvm/lib/Target/AMDGPU/SIFixVGPRCopies.cpp

@@ -11,6 +11,7 @@
 ///
 //===----------------------------------------------------------------------===//
 
+#include "SIFixVGPRCopies.h"
 #include "AMDGPU.h"
 #include "GCNSubtarget.h"
 #include "MCTargetDesc/AMDGPUMCTargetDesc.h"
@@ -22,13 +23,12 @@ using namespace llvm;
 
 namespace {
 
-class SIFixVGPRCopies : public MachineFunctionPass {
+class SIFixVGPRCopiesLegacy : public MachineFunctionPass {
 public:
   static char ID;
 
-public:
-  SIFixVGPRCopies() : MachineFunctionPass(ID) {
-    initializeSIFixVGPRCopiesPass(*PassRegistry::getPassRegistry());
+  SIFixVGPRCopiesLegacy() : MachineFunctionPass(ID) {
+    initializeSIFixVGPRCopiesLegacyPass(*PassRegistry::getPassRegistry());
   }
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
@@ -41,15 +41,31 @@ class SIFixVGPRCopies : public MachineFunctionPass {
   StringRef getPassName() const override { return "SI Fix VGPR copies"; }
 };
 
+class SIFixVGPRCopies {
+public:
+  bool run(MachineFunction &MF);
+};
+
 } // End anonymous namespace.
 
-INITIALIZE_PASS(SIFixVGPRCopies, DEBUG_TYPE, "SI Fix VGPR copies", false, false)
+INITIALIZE_PASS(SIFixVGPRCopiesLegacy, DEBUG_TYPE, "SI Fix VGPR copies", false,
+                false)
 
-char SIFixVGPRCopies::ID = 0;
+char SIFixVGPRCopiesLegacy::ID = 0;
 
-char &llvm::SIFixVGPRCopiesID = SIFixVGPRCopies::ID;
+char &llvm::SIFixVGPRCopiesID = SIFixVGPRCopiesLegacy::ID;
+
+PreservedAnalyses SIFixVGPRCopiesPass::run(MachineFunction &MF,
+                                           MachineFunctionAnalysisManager &) {
+  SIFixVGPRCopies().run(MF);
+  return PreservedAnalyses::all();
+}
+
+bool SIFixVGPRCopiesLegacy::runOnMachineFunction(MachineFunction &MF) {
+  return SIFixVGPRCopies().run(MF);
+}
 
-bool SIFixVGPRCopies::runOnMachineFunction(MachineFunction &MF) {
+bool SIFixVGPRCopies::run(MachineFunction &MF) {
   const GCNSubtarget &ST = MF.getSubtarget<GCNSubtarget>();
   const SIRegisterInfo *TRI = ST.getRegisterInfo();
   const SIInstrInfo *TII = ST.getInstrInfo();

llvm/lib/Target/AMDGPU/SIFixVGPRCopies.h

@@ -0,0 +1,22 @@
+//===- SIFixVGPRCopies.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_SIFIXVGPRCOPIES_H
+#define LLVM_LIB_TARGET_AMDGPU_SIFIXVGPRCOPIES_H
+
+#include "llvm/CodeGen/MachinePassManager.h"
+
+namespace llvm {
+class SIFixVGPRCopiesPass : public PassInfoMixin<SIFixVGPRCopiesPass> {
+public:
+  PreservedAnalyses run(MachineFunction &MF,
+                        MachineFunctionAnalysisManager &MFAM);
+};
+} // namespace llvm
+
+#endif // LLVM_LIB_TARGET_AMDGPU_SIFIXVGPRCOPIES_H

llvm/test/CodeGen/AMDGPU/fix-vgpr-copies.mir

@@ -1,4 +1,6 @@
 # RUN: llc -mtriple=amdgcn -start-after=greedy -disable-copyprop -stop-after=si-optimize-exec-masking -o - %s | FileCheck %s
+# RUN: llc -mtriple=amdgcn -passes=si-fix-vgpr-copies,si-optimize-exec-masking -o - %s | FileCheck %s
+
 # Check that we first do all vector instructions and only then change exec
 # CHECK-DAG:  COPY $vgpr10_vgpr11
 # CHECK-DAG:  COPY $vgpr12_vgpr13