1 ; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s 2 ; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s 3 4 declare float @llvm.amdgcn.cubeid(float, float, float) #0 5 declare float @llvm.amdgcn.cubesc(float, float, float) #0 6 declare float @llvm.amdgcn.cubetc(float, float, float) #0 7 declare float @llvm.amdgcn.cubema(float, float, float) #0 8 9 declare <4 x float> @llvm.AMDGPU.cube(<4 x float>) #0 10 11 12 ; GCN-LABEL: {{^}}cube: 13 ; GCN-DAG: v_cubeid_f32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} 14 ; GCN-DAG: v_cubesc_f32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} 15 ; GCN-DAG: v_cubetc_f32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} 16 ; GCN-DAG: v_cubema_f32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} 17 ; GCN: buffer_store_dwordx4 18 define void @cube(<4 x float> addrspace(1)* %out, float %a, float %b, float %c) #1 { 19 %cubeid = call float @llvm.amdgcn.cubeid(float %a, float %b, float %c) 20 %cubesc = call float @llvm.amdgcn.cubesc(float %a, float %b, float %c) 21 %cubetc = call float @llvm.amdgcn.cubetc(float %a, float %b, float %c) 22 %cubema = call float @llvm.amdgcn.cubema(float %a, float %b, float %c) 23 24 %vec0 = insertelement <4 x float> undef, float %cubeid, i32 0 25 %vec1 = insertelement <4 x float> %vec0, float %cubesc, i32 1 26 %vec2 = insertelement <4 x float> %vec1, float %cubetc, i32 2 27 %vec3 = insertelement <4 x float> %vec2, float %cubema, i32 3 28 store <4 x float> %vec3, <4 x float> addrspace(1)* %out 29 ret void 30 } 31 32 ; GCN-LABEL: {{^}}legacy_cube: 33 ; GCN-DAG: v_cubeid_f32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} 34 ; GCN-DAG: v_cubesc_f32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} 35 ; GCN-DAG: v_cubetc_f32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} 36 ; GCN-DAG: v_cubema_f32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}} 37 ; GCN: buffer_store_dwordx4 38 define void @legacy_cube(<4 x float> addrspace(1)* %out, <4 x float> %abcx) #1 { 39 %cube = call <4 x float> @llvm.AMDGPU.cube(<4 x float> %abcx) 40 store <4 x float> %cube, <4 x float> addrspace(1)* %out 41 ret void 42 } 43 44 attributes #0 = { nounwind readnone } 45 attributes #1 = { nounwind } 46 47
