28 lines · plain
1; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes=amdgpu-promote-alloca -disable-promote-alloca-to-vector < %s | FileCheck %s2 3; This kernel starts with the amdgpu-no-workitem-id-* attributes, but4; need to be removed when these intrinsic uses are introduced.5 6; CHECK-LABEL: define amdgpu_kernel void @promote_to_lds(ptr addrspace(1) %out, i32 %in) #0 {7; CHECK: call noalias nonnull dereferenceable(64) ptr addrspace(4) @llvm.amdgcn.dispatch.ptr()8; CHECK: call range(i32 0, 256) i32 @llvm.amdgcn.workitem.id.x()9; CHECK: call range(i32 0, 256) i32 @llvm.amdgcn.workitem.id.y()10; CHECK: call range(i32 0, 256) i32 @llvm.amdgcn.workitem.id.z()11define amdgpu_kernel void @promote_to_lds(ptr addrspace(1) %out, i32 %in) #0 {12entry:13 %tmp = alloca [2 x i32], addrspace(5)14 %tmp2 = getelementptr inbounds [2 x i32], ptr addrspace(5) %tmp, i32 0, i32 115 store i32 0, ptr addrspace(5) %tmp16 store i32 1, ptr addrspace(5) %tmp217 %tmp3 = getelementptr inbounds [2 x i32], ptr addrspace(5) %tmp, i32 0, i32 %in18 %tmp4 = load i32, ptr addrspace(5) %tmp319 %tmp5 = load volatile i32, ptr addrspace(1) poison20 %tmp6 = add i32 %tmp4, %tmp521 store i32 %tmp6, ptr addrspace(1) %out22 ret void23}24 25attributes #0 = { "amdgpu-flat-work-group-size"="256,256" "amdgpu-waves-per-eu"="1,5" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-no-dispatch-ptr" }26 27; CHECK: attributes #0 = { "amdgpu-flat-work-group-size"="256,256" "amdgpu-waves-per-eu"="1,5" }28