[CodeGen] Add a pass for testing finalizeBundle

[jayfoad]

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

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Jay Foad (jayfoad)

Changes

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

---

Full diff: https://github.com/llvm/llvm-project/pull/149813.diff

5 Files Affected:

• (modified) llvm/include/llvm/CodeGen/Passes.h (+3)
• (modified) llvm/include/llvm/InitializePasses.h (+1)
• (modified) llvm/lib/CodeGen/CodeGen.cpp (+1)
• (modified) llvm/lib/CodeGen/MachineInstrBundle.cpp (+23)
• (added) llvm/test/CodeGen/AMDGPU/finalize-bundle.mir (+18)

diff --git a/llvm/include/llvm/CodeGen/Passes.h b/llvm/include/llvm/CodeGen/Passes.h
index 714285eb5d5d4..cad8ea5a28dcc 100644
--- a/llvm/include/llvm/CodeGen/Passes.h
+++ b/llvm/include/llvm/CodeGen/Passes.h
@@ -623,6 +623,9 @@ LLVM_ABI ModulePass *createWindowsSecureHotPatchingPass();
 
 /// Lowers KCFI operand bundles for indirect calls.
 LLVM_ABI FunctionPass *createKCFIPass();
+
+/// A pass for testing finalizeBundle.
+LLVM_ABI extern char &FinalizeBundleTestID;
 } // namespace llvm
 
 #endif
diff --git a/llvm/include/llvm/InitializePasses.h b/llvm/include/llvm/InitializePasses.h
index 2e231cfba2443..ccb5711e55820 100644
--- a/llvm/include/llvm/InitializePasses.h
+++ b/llvm/include/llvm/InitializePasses.h
@@ -340,6 +340,7 @@ LLVM_ABI void initializeWindowsSecureHotPatchingPass(PassRegistry &);
 LLVM_ABI void initializeWinEHPreparePass(PassRegistry &);
 LLVM_ABI void initializeWriteBitcodePassPass(PassRegistry &);
 LLVM_ABI void initializeXRayInstrumentationLegacyPass(PassRegistry &);
+LLVM_ABI void initializeFinalizeBundleTestPass(PassRegistry &);
 
 } // end namespace llvm
 
diff --git a/llvm/lib/CodeGen/CodeGen.cpp b/llvm/lib/CodeGen/CodeGen.cpp
index c3b4077b27dd8..bd52edd66c547 100644
--- a/llvm/lib/CodeGen/CodeGen.cpp
+++ b/llvm/lib/CodeGen/CodeGen.cpp
@@ -146,4 +146,5 @@ void llvm::initializeCodeGen(PassRegistry &Registry) {
   initializeWasmEHPreparePass(Registry);
   initializeWinEHPreparePass(Registry);
   initializeXRayInstrumentationLegacyPass(Registry);
+  initializeFinalizeBundleTestPass(Registry);
 }
diff --git a/llvm/lib/CodeGen/MachineInstrBundle.cpp b/llvm/lib/CodeGen/MachineInstrBundle.cpp
index 34896c67144bc..741bfb08ef2fb 100644
--- a/llvm/lib/CodeGen/MachineInstrBundle.cpp
+++ b/llvm/lib/CodeGen/MachineInstrBundle.cpp
@@ -359,3 +359,26 @@ PhysRegInfo llvm::AnalyzePhysRegInBundle(const MachineInstr &MI, Register Reg,
 
   return PRI;
 }
+
+namespace {
+class FinalizeBundleTest : public MachineFunctionPass {
+public:
+  static char ID;
+
+  FinalizeBundleTest() : MachineFunctionPass(ID) {
+    initializeFinalizeBundleTestPass(*PassRegistry::getPassRegistry());
+  }
+
+  bool runOnMachineFunction(MachineFunction &MF) override {
+    // 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 true;
+  }
+};
+} // namespace
+
+char FinalizeBundleTest::ID = 0;
+INITIALIZE_PASS(FinalizeBundleTest, "finalizebundle-test",
+                "finalizeBundle test", false, false)
diff --git a/llvm/test/CodeGen/AMDGPU/finalize-bundle.mir b/llvm/test/CodeGen/AMDGPU/finalize-bundle.mir
new file mode 100644
index 0000000000000..be25e35765f67
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/finalize-bundle.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 -run-pass 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
+...

[arsenm]

Also do the new PM version?

[arsenm]

Or if this is just for tests, it could exclusively be new pm