diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/fp_to_uint.f64.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/fp_to_uint.f64.ll | 16 |
1 files changed, 8 insertions, 8 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/fp_to_uint.f64.ll b/llvm/test/CodeGen/AMDGPU/fp_to_uint.f64.ll index d5bc416434d..4f597eb3f32 100644 --- a/llvm/test/CodeGen/AMDGPU/fp_to_uint.f64.ll +++ b/llvm/test/CodeGen/AMDGPU/fp_to_uint.f64.ll @@ -6,7 +6,7 @@ declare double @llvm.fabs.f64(double) #1 ; SI-LABEL: {{^}}fp_to_uint_i32_f64: ; SI: v_cvt_u32_f64_e32 -define void @fp_to_uint_i32_f64(i32 addrspace(1)* %out, double %in) { +define amdgpu_kernel void @fp_to_uint_i32_f64(i32 addrspace(1)* %out, double %in) { %cast = fptoui double %in to i32 store i32 %cast, i32 addrspace(1)* %out, align 4 ret void @@ -15,7 +15,7 @@ define void @fp_to_uint_i32_f64(i32 addrspace(1)* %out, double %in) { ; SI-LABEL: @fp_to_uint_v2i32_v2f64 ; SI: v_cvt_u32_f64_e32 ; SI: v_cvt_u32_f64_e32 -define void @fp_to_uint_v2i32_v2f64(<2 x i32> addrspace(1)* %out, <2 x double> %in) { +define amdgpu_kernel void @fp_to_uint_v2i32_v2f64(<2 x i32> addrspace(1)* %out, <2 x double> %in) { %cast = fptoui <2 x double> %in to <2 x i32> store <2 x i32> %cast, <2 x i32> addrspace(1)* %out, align 8 ret void @@ -26,7 +26,7 @@ define void @fp_to_uint_v2i32_v2f64(<2 x i32> addrspace(1)* %out, <2 x double> % ; SI: v_cvt_u32_f64_e32 ; SI: v_cvt_u32_f64_e32 ; SI: v_cvt_u32_f64_e32 -define void @fp_to_uint_v4i32_v4f64(<4 x i32> addrspace(1)* %out, <4 x double> %in) { +define amdgpu_kernel void @fp_to_uint_v4i32_v4f64(<4 x i32> addrspace(1)* %out, <4 x double> %in) { %cast = fptoui <4 x double> %in to <4 x i32> store <4 x i32> %cast, <4 x i32> addrspace(1)* %out, align 8 ret void @@ -47,7 +47,7 @@ define void @fp_to_uint_v4i32_v4f64(<4 x i32> addrspace(1)* %out, <4 x double> % ; CI-DAG: v_cvt_u32_f64_e32 v[[LO:[0-9]+]], [[FMA]] ; CI-DAG: v_cvt_u32_f64_e32 v[[HI:[0-9]+]], [[FLOOR]] ; CI: buffer_store_dwordx2 v{{\[}}[[LO]]:[[HI]]{{\]}} -define void @fp_to_uint_i64_f64(i64 addrspace(1)* %out, double addrspace(1)* %in) { +define amdgpu_kernel void @fp_to_uint_i64_f64(i64 addrspace(1)* %out, double addrspace(1)* %in) { %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone %gep = getelementptr double, double addrspace(1)* %in, i32 %tid %val = load double, double addrspace(1)* %gep, align 8 @@ -57,14 +57,14 @@ define void @fp_to_uint_i64_f64(i64 addrspace(1)* %out, double addrspace(1)* %in } ; SI-LABEL: @fp_to_uint_v2i64_v2f64 -define void @fp_to_uint_v2i64_v2f64(<2 x i64> addrspace(1)* %out, <2 x double> %in) { +define amdgpu_kernel void @fp_to_uint_v2i64_v2f64(<2 x i64> addrspace(1)* %out, <2 x double> %in) { %cast = fptoui <2 x double> %in to <2 x i64> store <2 x i64> %cast, <2 x i64> addrspace(1)* %out, align 16 ret void } ; SI-LABEL: @fp_to_uint_v4i64_v4f64 -define void @fp_to_uint_v4i64_v4f64(<4 x i64> addrspace(1)* %out, <4 x double> %in) { +define amdgpu_kernel void @fp_to_uint_v4i64_v4f64(<4 x i64> addrspace(1)* %out, <4 x double> %in) { %cast = fptoui <4 x double> %in to <4 x i64> store <4 x i64> %cast, <4 x i64> addrspace(1)* %out, align 32 ret void @@ -72,7 +72,7 @@ define void @fp_to_uint_v4i64_v4f64(<4 x i64> addrspace(1)* %out, <4 x double> % ; FUNC-LABEL: {{^}}fp_to_uint_f64_to_i1: ; SI: v_cmp_eq_f64_e64 s{{\[[0-9]+:[0-9]+\]}}, 1.0, s{{\[[0-9]+:[0-9]+\]}} -define void @fp_to_uint_f64_to_i1(i1 addrspace(1)* %out, double %in) #0 { +define amdgpu_kernel void @fp_to_uint_f64_to_i1(i1 addrspace(1)* %out, double %in) #0 { %conv = fptoui double %in to i1 store i1 %conv, i1 addrspace(1)* %out ret void @@ -80,7 +80,7 @@ define void @fp_to_uint_f64_to_i1(i1 addrspace(1)* %out, double %in) #0 { ; FUNC-LABEL: {{^}}fp_to_uint_fabs_f64_to_i1: ; SI: v_cmp_eq_f64_e64 s{{\[[0-9]+:[0-9]+\]}}, 1.0, |s{{\[[0-9]+:[0-9]+\]}}| -define void @fp_to_uint_fabs_f64_to_i1(i1 addrspace(1)* %out, double %in) #0 { +define amdgpu_kernel void @fp_to_uint_fabs_f64_to_i1(i1 addrspace(1)* %out, double %in) #0 { %in.fabs = call double @llvm.fabs.f64(double %in) %conv = fptoui double %in.fabs to i1 store i1 %conv, i1 addrspace(1)* %out |
