1 ; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 2 3 declare i32 @llvm.ctlz.i32(i32, i1) nounwind readnone 4 declare <2 x i32> @llvm.ctlz.v2i32(<2 x i32>, i1) nounwind readnone 5 declare <4 x i32> @llvm.ctlz.v4i32(<4 x i32>, i1) nounwind readnone 6 7 ; FUNC-LABEL: @s_ctlz_zero_undef_i32: 8 ; SI: S_LOAD_DWORD [[VAL:s[0-9]+]], 9 ; SI: S_FLBIT_I32_B32 [[SRESULT:s[0-9]+]], [[VAL]] 10 ; SI: V_MOV_B32_e32 [[VRESULT:v[0-9]+]], [[SRESULT]] 11 ; SI: BUFFER_STORE_DWORD [[VRESULT]], 12 ; SI: S_ENDPGM 13 define void @s_ctlz_zero_undef_i32(i32 addrspace(1)* noalias %out, i32 %val) nounwind { 14 %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone 15 store i32 %ctlz, i32 addrspace(1)* %out, align 4 16 ret void 17 } 18 19 ; FUNC-LABEL: @v_ctlz_zero_undef_i32: 20 ; SI: BUFFER_LOAD_DWORD [[VAL:v[0-9]+]], 21 ; SI: V_FFBH_U32_e32 [[RESULT:v[0-9]+]], [[VAL]] 22 ; SI: BUFFER_STORE_DWORD [[RESULT]], 23 ; SI: S_ENDPGM 24 define void @v_ctlz_zero_undef_i32(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind { 25 %val = load i32 addrspace(1)* %valptr, align 4 26 %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 true) nounwind readnone 27 store i32 %ctlz, i32 addrspace(1)* %out, align 4 28 ret void 29 } 30 31 ; FUNC-LABEL: @v_ctlz_zero_undef_v2i32: 32 ; SI: BUFFER_LOAD_DWORDX2 33 ; SI: V_FFBH_U32_e32 34 ; SI: V_FFBH_U32_e32 35 ; SI: BUFFER_STORE_DWORDX2 36 ; SI: S_ENDPGM 37 define void @v_ctlz_zero_undef_v2i32(<2 x i32> addrspace(1)* noalias %out, <2 x i32> addrspace(1)* noalias %valptr) nounwind { 38 %val = load <2 x i32> addrspace(1)* %valptr, align 8 39 %ctlz = call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %val, i1 true) nounwind readnone 40 store <2 x i32> %ctlz, <2 x i32> addrspace(1)* %out, align 8 41 ret void 42 } 43 44 ; FUNC-LABEL: @v_ctlz_zero_undef_v4i32: 45 ; SI: BUFFER_LOAD_DWORDX4 46 ; SI: V_FFBH_U32_e32 47 ; SI: V_FFBH_U32_e32 48 ; SI: V_FFBH_U32_e32 49 ; SI: V_FFBH_U32_e32 50 ; SI: BUFFER_STORE_DWORDX4 51 ; SI: S_ENDPGM 52 define void @v_ctlz_zero_undef_v4i32(<4 x i32> addrspace(1)* noalias %out, <4 x i32> addrspace(1)* noalias %valptr) nounwind { 53 %val = load <4 x i32> addrspace(1)* %valptr, align 16 54 %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %val, i1 true) nounwind readnone 55 store <4 x i32> %ctlz, <4 x i32> addrspace(1)* %out, align 16 56 ret void 57 } 58
