45 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 52; RUN: opt -mtriple amdgcn -passes=amdgpu-promote-alloca-to-vector -S < %s | FileCheck %s3 4; Skip promote-alloca in case of an index which is known to be out of bounds.5 6define amdgpu_kernel void @out_of_bounds() {7; CHECK-LABEL: define amdgpu_kernel void @out_of_bounds() {8; CHECK-NEXT: [[PTR:%.*]] = freeze <4 x float> poison9; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x float> [[PTR]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 2, i32 3>10; CHECK-NEXT: ret void11;12 %ptr = alloca [4 x float], align 4, addrspace(5)13 %elem_ptr = getelementptr [4 x float], ptr addrspace(5) %ptr, i32 0, i32 4214 call void @llvm.memcpy.p5.p5.i32(ptr addrspace(5) %ptr, ptr addrspace(5) %elem_ptr, i32 8, i1 false)15 ret void16}17 18define amdgpu_kernel void @memcpy_partially_out_of_bounds() {19; CHECK-LABEL: define amdgpu_kernel void @memcpy_partially_out_of_bounds() {20; CHECK-NEXT: [[PTR:%.*]] = freeze <3 x float> poison21; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <3 x float> [[PTR]], <3 x float> poison, <3 x i32> <i32 2, i32 poison, i32 2>22; CHECK-NEXT: ret void23;24 %ptr = alloca [3 x float], align 4, addrspace(5)25 %elem_ptr = getelementptr [3 x float], ptr addrspace(5) %ptr, i32 0, i32 226 call void @llvm.memcpy.p5.p5.i32(ptr addrspace(5) %ptr, ptr addrspace(5) %elem_ptr, i32 8, i1 false)27 ret void28}29 30define amdgpu_kernel void @in_bounds() {31; CHECK-LABEL: define amdgpu_kernel void @in_bounds() {32; CHECK-NEXT: [[PTR:%.*]] = freeze <4 x float> poison33; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x float> [[PTR]], <4 x float> poison, <4 x i32> <i32 2, i32 3, i32 2, i32 3>34; CHECK-NEXT: ret void35;36 %ptr = alloca [4 x float], align 4, addrspace(5)37 %elem_ptr = getelementptr [4 x float], ptr addrspace(5) %ptr, i32 0, i32 238 call void @llvm.memcpy.p5.p5.i32(ptr addrspace(5) %ptr, ptr addrspace(5) %elem_ptr, i32 8, i1 false)39 ret void40}41 42declare void @llvm.memcpy.p5.p5.i32(ptr addrspace(5) writeonly captures(none), ptr addrspace(5) readonly captures(none), i32, i1 immarg) #043 44attributes #0 = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }45