[llvm][OpenMPOpt] Remove no-op ptr-to-ptr bitcast (NFC) (#73869)

JOE1994 · web-flow · commit c57ef2c69846 · 2023-11-29T20:47:37.000-05:00
* Remove a call to CreatePointerBitCastOrAddrSpaceCast which merely adds
a no-op ptr-to-ptr bitcast.

* Most of the diff is from removing checks for no-op ptr-to-ptr bitcasts
in relevant LIT tests
diff --git a/llvm/lib/Transforms/IPO/OpenMPOpt.cpp b/llvm/lib/Transforms/IPO/OpenMPOpt.cpp
@@ -4527,9 +4527,6 @@ struct AAKernelInfoFunction : AAKernelInfo {
     FunctionType *ParallelRegionFnTy = FunctionType::get(
         Type::getVoidTy(Ctx), {Type::getInt16Ty(Ctx), Type::getInt32Ty(Ctx)},
         false);
-    Value *WorkFnCast = BitCastInst::CreatePointerBitCastOrAddrSpaceCast(
-        WorkFn, ParallelRegionFnTy->getPointerTo(), "worker.work_fn.addr_cast",
-        StateMachineBeginBB);
 
     Instruction *IsDone =
         ICmpInst::Create(ICmpInst::ICmp, llvm::CmpInst::ICMP_EQ, WorkFn,
@@ -4576,7 +4573,7 @@ struct AAKernelInfoFunction : AAKernelInfo {
       Value *IsPR;
       if (I + 1 < E || !ReachedUnknownParallelRegions.empty()) {
         Instruction *CmpI = ICmpInst::Create(
-            ICmpInst::ICmp, llvm::CmpInst::ICMP_EQ, WorkFnCast, ParallelRegion,
+            ICmpInst::ICmp, llvm::CmpInst::ICMP_EQ, WorkFn, ParallelRegion,
             "worker.check_parallel_region", StateMachineIfCascadeCurrentBB);
         CmpI->setDebugLoc(DLoc);
         IsPR = CmpI;
@@ -4596,7 +4593,7 @@ struct AAKernelInfoFunction : AAKernelInfo {
     if (!ReachedUnknownParallelRegions.empty()) {
       StateMachineIfCascadeCurrentBB->setName(
           "worker_state_machine.parallel_region.fallback.execute");
-      CallInst::Create(ParallelRegionFnTy, WorkFnCast, {ZeroArg, GTid}, "",
+      CallInst::Create(ParallelRegionFnTy, WorkFn, {ZeroArg, GTid}, "",
                        StateMachineIfCascadeCurrentBB)
           ->setDebugLoc(DLoc);
     }
diff --git a/llvm/test/Transforms/OpenMP/custom_state_machines.ll b/llvm/test/Transforms/OpenMP/custom_state_machines.ll
diff --git a/llvm/test/Transforms/OpenMP/gpu_state_machine_function_ptr_replacement.ll b/llvm/test/Transforms/OpenMP/gpu_state_machine_function_ptr_replacement.ll
@@ -26,8 +26,8 @@
 ; CHECK-DAG: @__omp_outlined__1_wrapper.ID = private constant i8 undef
 ; CHECK-DAG: @__omp_outlined__2_wrapper.ID = private constant i8 undef
 
-; CHECK-DAG:   icmp eq ptr %worker.work_fn.addr_cast, @__omp_outlined__1_wrapper.ID
-; CHECK-DAG:   icmp eq ptr %worker.work_fn.addr_cast, @__omp_outlined__2_wrapper.ID
+; CHECK-DAG:   icmp eq ptr %worker.work_fn, @__omp_outlined__1_wrapper.ID
+; CHECK-DAG:   icmp eq ptr %worker.work_fn, @__omp_outlined__2_wrapper.ID
 
 
 ; CHECK-DAG:   call void @__kmpc_parallel_51(ptr @1, i32 %{{.*}}, i32 1, i32 -1, i32 -1, ptr @__omp_outlined__1, ptr @__omp_outlined__1_wrapper.ID, ptr %{{.*}}, i64 0)
diff --git a/llvm/test/Transforms/OpenMP/spmdization.ll b/llvm/test/Transforms/OpenMP/spmdization.ll
diff --git a/llvm/test/Transforms/OpenMP/spmdization_guarding.ll b/llvm/test/Transforms/OpenMP/spmdization_guarding.ll
@@ -203,7 +203,6 @@ define weak void @__omp_offloading_2a_fbfa7a_sequential_loop_l6(ptr %dyn, ptr %x
 ; CHECK-DISABLED-NEXT:    call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
 ; CHECK-DISABLED-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(ptr [[WORKER_WORK_FN_ADDR]])
 ; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN:%.*]] = load ptr, ptr [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-DISABLED-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast ptr [[WORKER_WORK_FN]] to ptr
 ; CHECK-DISABLED-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq ptr [[WORKER_WORK_FN]], null
 ; CHECK-DISABLED-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
 ; CHECK-DISABLED:       worker_state_machine.finished:
diff --git a/llvm/test/Transforms/OpenMP/spmdization_indirect.ll b/llvm/test/Transforms/OpenMP/spmdization_indirect.ll
@@ -390,15 +390,14 @@ define weak void @spmd_and_non_spmd_callee(i1 %c) #0 {
 ; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast ptr addrspace(5) [[WORKER_WORK_FN_ADDR]] to ptr
 ; AMDGPU-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(ptr [[WORKER_WORK_FN_ADDR_GENERIC]])
 ; AMDGPU-NEXT:    [[WORKER_WORK_FN:%.*]] = load ptr, ptr [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
-; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast ptr [[WORKER_WORK_FN]] to ptr
 ; AMDGPU-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq ptr [[WORKER_WORK_FN]], null
 ; AMDGPU-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
 ; AMDGPU:       worker_state_machine.finished:
 ; AMDGPU-NEXT:    ret void
 ; AMDGPU:       worker_state_machine.is_active.check:
 ; AMDGPU-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
 ; AMDGPU:       worker_state_machine.parallel_region.fallback.execute:
-; AMDGPU-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    call void [[WORKER_WORK_FN]](i16 0, i32 [[TMP0]])
 ; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
 ; AMDGPU:       worker_state_machine.parallel_region.end:
 ; AMDGPU-NEXT:    call void @__kmpc_kernel_end_parallel()
@@ -451,15 +450,14 @@ define weak void @spmd_and_non_spmd_callee(i1 %c) #0 {
 ; NVPTX-NEXT:    call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
 ; NVPTX-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(ptr [[WORKER_WORK_FN_ADDR]])
 ; NVPTX-NEXT:    [[WORKER_WORK_FN:%.*]] = load ptr, ptr [[WORKER_WORK_FN_ADDR]], align 8
-; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast ptr [[WORKER_WORK_FN]] to ptr
 ; NVPTX-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq ptr [[WORKER_WORK_FN]], null
 ; NVPTX-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
 ; NVPTX:       worker_state_machine.finished:
 ; NVPTX-NEXT:    ret void
 ; NVPTX:       worker_state_machine.is_active.check:
 ; NVPTX-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
 ; NVPTX:       worker_state_machine.parallel_region.fallback.execute:
-; NVPTX-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    call void [[WORKER_WORK_FN]](i16 0, i32 [[TMP0]])
 ; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
 ; NVPTX:       worker_state_machine.parallel_region.end:
 ; NVPTX-NEXT:    call void @__kmpc_kernel_end_parallel()
@@ -736,15 +734,14 @@ define weak void @spmd_and_non_spmd_callees_metadata(ptr %fp) #0 {
 ; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_GENERIC:%.*]] = addrspacecast ptr addrspace(5) [[WORKER_WORK_FN_ADDR]] to ptr
 ; AMDGPU-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(ptr [[WORKER_WORK_FN_ADDR_GENERIC]])
 ; AMDGPU-NEXT:    [[WORKER_WORK_FN:%.*]] = load ptr, ptr [[WORKER_WORK_FN_ADDR_GENERIC]], align 8
-; AMDGPU-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast ptr [[WORKER_WORK_FN]] to ptr
 ; AMDGPU-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq ptr [[WORKER_WORK_FN]], null
 ; AMDGPU-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
 ; AMDGPU:       worker_state_machine.finished:
 ; AMDGPU-NEXT:    ret void
 ; AMDGPU:       worker_state_machine.is_active.check:
 ; AMDGPU-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
 ; AMDGPU:       worker_state_machine.parallel_region.fallback.execute:
-; AMDGPU-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; AMDGPU-NEXT:    call void [[WORKER_WORK_FN]](i16 0, i32 [[TMP0]])
 ; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
 ; AMDGPU:       worker_state_machine.parallel_region.end:
 ; AMDGPU-NEXT:    call void @__kmpc_kernel_end_parallel()
@@ -796,15 +793,14 @@ define weak void @spmd_and_non_spmd_callees_metadata(ptr %fp) #0 {
 ; NVPTX-NEXT:    call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
 ; NVPTX-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(ptr [[WORKER_WORK_FN_ADDR]])
 ; NVPTX-NEXT:    [[WORKER_WORK_FN:%.*]] = load ptr, ptr [[WORKER_WORK_FN_ADDR]], align 8
-; NVPTX-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast ptr [[WORKER_WORK_FN]] to ptr
 ; NVPTX-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq ptr [[WORKER_WORK_FN]], null
 ; NVPTX-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
 ; NVPTX:       worker_state_machine.finished:
 ; NVPTX-NEXT:    ret void
 ; NVPTX:       worker_state_machine.is_active.check:
 ; NVPTX-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
 ; NVPTX:       worker_state_machine.parallel_region.fallback.execute:
-; NVPTX-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; NVPTX-NEXT:    call void [[WORKER_WORK_FN]](i16 0, i32 [[TMP0]])
 ; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
 ; NVPTX:       worker_state_machine.parallel_region.end:
 ; NVPTX-NEXT:    call void @__kmpc_kernel_end_parallel()
diff --git a/llvm/test/Transforms/OpenMP/spmdization_no_guarding_two_reaching_kernels.ll b/llvm/test/Transforms/OpenMP/spmdization_no_guarding_two_reaching_kernels.ll
@@ -88,7 +88,6 @@ define weak void @__omp_offloading_2b_10393b5_spmd_l12(ptr %dyn) #0 {
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(ptr [[WORKER_WORK_FN_ADDR]])
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    [[WORKER_WORK_FN:%.*]] = load ptr, ptr [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-DISABLE-SPMDIZATION-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast ptr [[WORKER_WORK_FN]] to ptr
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq ptr [[WORKER_WORK_FN]], null
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
 ; CHECK-DISABLE-SPMDIZATION:       worker_state_machine.finished:
@@ -166,15 +165,14 @@ define weak void @__omp_offloading_2b_10393b5_generic_l20(ptr %dyn) #0 {
 ; CHECK-NEXT:    call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
 ; CHECK-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(ptr [[WORKER_WORK_FN_ADDR]])
 ; CHECK-NEXT:    [[WORKER_WORK_FN:%.*]] = load ptr, ptr [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast ptr [[WORKER_WORK_FN]] to ptr
 ; CHECK-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq ptr [[WORKER_WORK_FN]], null
 ; CHECK-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
 ; CHECK:       worker_state_machine.finished:
 ; CHECK-NEXT:    ret void
 ; CHECK:       worker_state_machine.is_active.check:
 ; CHECK-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
 ; CHECK:       worker_state_machine.parallel_region.fallback.execute:
-; CHECK-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; CHECK-NEXT:    call void [[WORKER_WORK_FN]](i16 0, i32 [[TMP0]])
 ; CHECK-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
 ; CHECK:       worker_state_machine.parallel_region.end:
 ; CHECK-NEXT:    call void @__kmpc_kernel_end_parallel()
@@ -209,15 +207,14 @@ define weak void @__omp_offloading_2b_10393b5_generic_l20(ptr %dyn) #0 {
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    call void @__kmpc_barrier_simple_generic(ptr @[[GLOB1]], i32 [[TMP0]])
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    [[WORKER_IS_ACTIVE:%.*]] = call i1 @__kmpc_kernel_parallel(ptr [[WORKER_WORK_FN_ADDR]])
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    [[WORKER_WORK_FN:%.*]] = load ptr, ptr [[WORKER_WORK_FN_ADDR]], align 8
-; CHECK-DISABLE-SPMDIZATION-NEXT:    [[WORKER_WORK_FN_ADDR_CAST:%.*]] = bitcast ptr [[WORKER_WORK_FN]] to ptr
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    [[WORKER_IS_DONE:%.*]] = icmp eq ptr [[WORKER_WORK_FN]], null
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    br i1 [[WORKER_IS_DONE]], label [[WORKER_STATE_MACHINE_FINISHED]], label [[WORKER_STATE_MACHINE_IS_ACTIVE_CHECK:%.*]]
 ; CHECK-DISABLE-SPMDIZATION:       worker_state_machine.finished:
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    ret void
 ; CHECK-DISABLE-SPMDIZATION:       worker_state_machine.is_active.check:
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
 ; CHECK-DISABLE-SPMDIZATION:       worker_state_machine.parallel_region.fallback.execute:
-; CHECK-DISABLE-SPMDIZATION-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; CHECK-DISABLE-SPMDIZATION-NEXT:    call void [[WORKER_WORK_FN]](i16 0, i32 [[TMP0]])
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
 ; CHECK-DISABLE-SPMDIZATION:       worker_state_machine.parallel_region.end:
 ; CHECK-DISABLE-SPMDIZATION-NEXT:    call void @__kmpc_kernel_end_parallel()
