[CodeGen] Add a pass for testing finalizeBundle (#149813)

This allows for unit testing of finalizeBundle with standard MIR tests
using update_mir_test_checks.py.

Author: jayfoad

llvm/include/llvm/CodeGen/MachineInstrBundle.h

@@ -15,6 +15,7 @@
 #define LLVM_CODEGEN_MACHINEINSTRBUNDLE_H
 
 #include "llvm/CodeGen/MachineBasicBlock.h"
+#include "llvm/CodeGen/MachinePassManager.h"
 #include "llvm/Support/Compiler.h"
 
 namespace llvm {
@@ -294,6 +295,12 @@ LLVM_ABI PhysRegInfo AnalyzePhysRegInBundle(const MachineInstr &MI,
                                             Register Reg,
                                             const TargetRegisterInfo *TRI);
 
+class FinalizeBundleTestPass : public PassInfoMixin<FinalizeBundleTestPass> {
+public:
+  PreservedAnalyses run(MachineFunction &MF,
+                        MachineFunctionAnalysisManager &MFAM);
+};
+
 } // End llvm namespace
 
 #endif

llvm/include/llvm/Passes/MachinePassRegistry.def

@@ -113,6 +113,7 @@ MACHINE_FUNCTION_PASS("early-machinelicm", EarlyMachineLICMPass())
 MACHINE_FUNCTION_PASS("early-tailduplication", EarlyTailDuplicatePass())
 MACHINE_FUNCTION_PASS("fentry-insert", FEntryInserterPass())
 MACHINE_FUNCTION_PASS("finalize-isel", FinalizeISelPass())
+MACHINE_FUNCTION_PASS("finalizebundle-test", FinalizeBundleTestPass())
 MACHINE_FUNCTION_PASS("fixup-statepoint-caller-saved", FixupStatepointCallerSavedPass())
 MACHINE_FUNCTION_PASS("init-undef", InitUndefPass())
 MACHINE_FUNCTION_PASS("localstackalloc", LocalStackSlotAllocationPass())

llvm/lib/CodeGen/MachineInstrBundle.cpp

@@ -359,3 +359,13 @@ PhysRegInfo llvm::AnalyzePhysRegInBundle(const MachineInstr &MI, Register Reg,
 
   return PRI;
 }
+
+PreservedAnalyses
+llvm::FinalizeBundleTestPass::run(MachineFunction &MF,
+                                  MachineFunctionAnalysisManager &) {
+  // For testing purposes, bundle the entire contents of each basic block
+  // except for terminators.
+  for (MachineBasicBlock &MBB : MF)
+    finalizeBundle(MBB, MBB.instr_begin(), MBB.getFirstInstrTerminator());
+  return PreservedAnalyses::none();
+}

llvm/lib/Passes/PassBuilder.cpp

@@ -124,6 +124,7 @@
 #include "llvm/CodeGen/MachineCopyPropagation.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/MachineFunctionAnalysis.h"
+#include "llvm/CodeGen/MachineInstrBundle.h"
 #include "llvm/CodeGen/MachineLICM.h"
 #include "llvm/CodeGen/MachineLateInstrsCleanup.h"
 #include "llvm/CodeGen/MachinePassManager.h"

llvm/test/CodeGen/AMDGPU/finalizebundle.mir

@@ -0,0 +1,18 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -passes=finalizebundle-test %s -o - | FileCheck %s
+
+---
+name: test_overlap
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+    ; CHECK-LABEL: name: test_overlap
+    ; CHECK: liveins: $vgpr0_vgpr1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: BUNDLE implicit-def $vgpr2_vgpr3, implicit-def $vgpr2, implicit-def $vgpr2_lo16, implicit-def $vgpr2_hi16, implicit-def $vgpr3, implicit-def $vgpr3_lo16, implicit-def $vgpr3_hi16, implicit-def $vgpr3_vgpr4, implicit-def $vgpr4, implicit-def $vgpr4_lo16, implicit-def $vgpr4_hi16, implicit $vgpr0_vgpr1, implicit $exec, implicit $vgpr1_vgpr2 {
+    ; CHECK-NEXT:   $vgpr2_vgpr3 = V_LSHLREV_B64_pseudo_e32 1, $vgpr0_vgpr1, implicit $exec
+    ; CHECK-NEXT:   $vgpr3_vgpr4 = V_LSHLREV_B64_pseudo_e32 1, $vgpr1_vgpr2, implicit $exec
+    ; CHECK-NEXT: }
+    $vgpr2_vgpr3 = V_LSHLREV_B64_pseudo_e32 1, $vgpr0_vgpr1, implicit $exec
+    $vgpr3_vgpr4 = V_LSHLREV_B64_pseudo_e32 1, $vgpr1_vgpr2, implicit $exec
+...