76 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 52; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes=amdgpu-promote-alloca %s -o - | FileCheck %s3 4declare ptr @llvm.invariant.start.p5(i64, ptr addrspace(5) nocapture)5declare void @llvm.invariant.end.p5(ptr, i64, ptr addrspace(5) nocapture)6declare ptr addrspace(5) @llvm.launder.invariant.group.p5(ptr addrspace(5))7declare ptr addrspace(5) @llvm.strip.invariant.group.p5(ptr addrspace(5))8 9define amdgpu_kernel void @use_invariant_start_and_end() {10; CHECK-LABEL: define amdgpu_kernel void @use_invariant_start_and_end() {11; CHECK-NEXT: [[BB:.*:]]12; CHECK-NEXT: [[TMP0:%.*]] = call noalias nonnull dereferenceable(64) ptr addrspace(4) @llvm.amdgcn.dispatch.ptr()13; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr addrspace(4) [[TMP0]], i64 114; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr addrspace(4) [[TMP1]], align 4, !invariant.load [[META0:![0-9]+]]15; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr addrspace(4) [[TMP0]], i64 216; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr addrspace(4) [[TMP3]], align 4, !range [[RNG1:![0-9]+]], !invariant.load [[META0]]17; CHECK-NEXT: [[TMP5:%.*]] = lshr i32 [[TMP2]], 1618; CHECK-NEXT: [[TMP6:%.*]] = call range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.x()19; CHECK-NEXT: [[TMP7:%.*]] = call range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.y()20; CHECK-NEXT: [[TMP8:%.*]] = call range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.z()21; CHECK-NEXT: [[TMP9:%.*]] = mul nuw nsw i32 [[TMP5]], [[TMP4]]22; CHECK-NEXT: [[TMP10:%.*]] = mul i32 [[TMP9]], [[TMP6]]23; CHECK-NEXT: [[TMP11:%.*]] = mul nuw nsw i32 [[TMP7]], [[TMP4]]24; CHECK-NEXT: [[TMP12:%.*]] = add i32 [[TMP10]], [[TMP11]]25; CHECK-NEXT: [[TMP13:%.*]] = add i32 [[TMP12]], [[TMP8]]26; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1024 x i32], ptr addrspace(3) @use_invariant_start_and_end.alloca, i32 0, i32 [[TMP13]]27; CHECK-NEXT: [[INVARIANT1:%.*]] = call ptr @llvm.invariant.start.p3(i64 0, ptr addrspace(3) [[TMP14]])28; CHECK-NEXT: store <2 x i1> zeroinitializer, ptr [[INVARIANT1]], align 129; CHECK-NEXT: call void @llvm.invariant.end.p3(ptr [[INVARIANT1]], i64 0, ptr addrspace(3) [[TMP14]])30; CHECK-NEXT: ret void31;32bb:33 %alloca = alloca i32, align 4, addrspace(5)34 %invariant = call ptr @llvm.invariant.start.p5(i64 0, ptr addrspace(5) %alloca)35 store <2 x i1> zeroinitializer, ptr %invariant, align 136 call void @llvm.invariant.end.p5(ptr %invariant, i64 0, ptr addrspace(5) %alloca)37 ret void38}39 40define amdgpu_kernel void @use_invariant_group_and_strip() {41; CHECK-LABEL: define amdgpu_kernel void @use_invariant_group_and_strip() {42; CHECK-NEXT: [[BB:.*:]]43; CHECK-NEXT: [[TMP0:%.*]] = call noalias nonnull dereferenceable(64) ptr addrspace(4) @llvm.amdgcn.dispatch.ptr()44; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr addrspace(4) [[TMP0]], i64 145; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr addrspace(4) [[TMP1]], align 4, !invariant.load [[META0]]46; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr addrspace(4) [[TMP0]], i64 247; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr addrspace(4) [[TMP3]], align 4, !range [[RNG1]], !invariant.load [[META0]]48; CHECK-NEXT: [[TMP5:%.*]] = lshr i32 [[TMP2]], 1649; CHECK-NEXT: [[TMP6:%.*]] = call range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.x()50; CHECK-NEXT: [[TMP7:%.*]] = call range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.y()51; CHECK-NEXT: [[TMP8:%.*]] = call range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.z()52; CHECK-NEXT: [[TMP9:%.*]] = mul nuw nsw i32 [[TMP5]], [[TMP4]]53; CHECK-NEXT: [[TMP10:%.*]] = mul i32 [[TMP9]], [[TMP6]]54; CHECK-NEXT: [[TMP11:%.*]] = mul nuw nsw i32 [[TMP7]], [[TMP4]]55; CHECK-NEXT: [[TMP12:%.*]] = add i32 [[TMP10]], [[TMP11]]56; CHECK-NEXT: [[TMP13:%.*]] = add i32 [[TMP12]], [[TMP8]]57; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1024 x i32], ptr addrspace(3) @use_invariant_group_and_strip.alloca, i32 0, i32 [[TMP13]]58; CHECK-NEXT: [[INVARIANT2:%.*]] = call ptr addrspace(3) @llvm.launder.invariant.group.p3(ptr addrspace(3) [[TMP14]])59; CHECK-NEXT: store <2 x i1> zeroinitializer, ptr addrspace(3) [[INVARIANT2]], align 160; CHECK-NEXT: [[STRIP1:%.*]] = call ptr addrspace(3) @llvm.strip.invariant.group.p3(ptr addrspace(3) [[TMP14]])61; CHECK-NEXT: store <2 x i1> zeroinitializer, ptr addrspace(3) [[STRIP1]], align 162; CHECK-NEXT: ret void63;64bb:65 %alloca = alloca i32, align 4, addrspace(5)66 %invariant = call ptr addrspace(5) @llvm.launder.invariant.group.p5(ptr addrspace(5) %alloca)67 store <2 x i1> zeroinitializer, ptr addrspace(5) %invariant, align 168 %strip = call ptr addrspace(5) @llvm.strip.invariant.group.p5(ptr addrspace(5) %alloca)69 store <2 x i1> zeroinitializer, ptr addrspace(5) %strip, align 170 ret void71}72;.73; CHECK: [[META0]] = !{}74; CHECK: [[RNG1]] = !{i32 0, i32 1025}75;.76