1; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI -check-prefix=FUNC %s
2; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI -check-prefix=FUNC %s
3
4; FIXME: This should be merged with uint_to_fp.ll, but s_uint_to_fp_v2i64 crashes on r600
5
6; FUNC-LABEL: {{^}}s_uint_to_fp_i64_to_f16:
7define amdgpu_kernel void @s_uint_to_fp_i64_to_f16(half addrspace(1)* %out, i64 %in) #0 {
8  %result = uitofp i64 %in to half
9  store half %result, half addrspace(1)* %out
10  ret void
11}
12
13; FUNC-LABEL: {{^}}v_uint_to_fp_i64_to_f16:
14; GCN: {{buffer|flat}}_load_dwordx2
15
16; GCN: v_ffbh_u32
17; GCN: v_ffbh_u32
18; GCN: v_cndmask
19; GCN: v_cndmask
20
21; GCN-DAG: v_cmp_eq_u64
22; GCN-DAG: v_cmp_gt_u64
23
24; GCN: v_add_{{[iu]}}32_e32 [[VR:v[0-9]+]]
25; GCN: v_cvt_f16_f32_e32 [[VR_F16:v[0-9]+]], [[VR]]
26; GCN: {{buffer|flat}}_store_short {{.*}}[[VR_F16]]
27define amdgpu_kernel void @v_uint_to_fp_i64_to_f16(half addrspace(1)* %out, i64 addrspace(1)* %in) #0 {
28  %tid = call i32 @llvm.amdgcn.workitem.id.x()
29  %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
30  %out.gep = getelementptr half, half addrspace(1)* %out, i32 %tid
31  %val = load i64, i64 addrspace(1)* %in.gep
32  %result = uitofp i64 %val to half
33  store half %result, half addrspace(1)* %out.gep
34  ret void
35}
36
37; FUNC-LABEL: {{^}}s_uint_to_fp_i64_to_f32:
38define amdgpu_kernel void @s_uint_to_fp_i64_to_f32(float addrspace(1)* %out, i64 %in) #0 {
39  %result = uitofp i64 %in to float
40  store float %result, float addrspace(1)* %out
41  ret void
42}
43
44; FUNC-LABEL: {{^}}v_uint_to_fp_i64_to_f32:
45; GCN: {{buffer|flat}}_load_dwordx2
46
47; GCN: v_ffbh_u32
48; GCN: v_ffbh_u32
49; GCN: v_cndmask
50; GCN: v_cndmask
51
52; GCN-DAG: v_cmp_eq_u64
53; GCN-DAG: v_cmp_gt_u64
54
55; GCN: v_add_{{[iu]}}32_e32 [[VR:v[0-9]+]]
56; GCN: {{buffer|flat}}_store_dword {{.*}}[[VR]]
57define amdgpu_kernel void @v_uint_to_fp_i64_to_f32(float addrspace(1)* %out, i64 addrspace(1)* %in) #0 {
58  %tid = call i32 @llvm.amdgcn.workitem.id.x()
59  %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
60  %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
61  %val = load i64, i64 addrspace(1)* %in.gep
62  %result = uitofp i64 %val to float
63  store float %result, float addrspace(1)* %out.gep
64  ret void
65}
66
67; FUNC-LABEL: {{^}}s_uint_to_fp_v2i64_to_v2f32:
68define amdgpu_kernel void @s_uint_to_fp_v2i64_to_v2f32(<2 x float> addrspace(1)* %out, <2 x i64> %in) #0{
69  %result = uitofp <2 x i64> %in to <2 x float>
70  store <2 x float> %result, <2 x float> addrspace(1)* %out
71  ret void
72}
73
74; FUNC-LABEL: {{^}}v_uint_to_fp_v4i64_to_v4f32:
75define amdgpu_kernel void @v_uint_to_fp_v4i64_to_v4f32(<4 x float> addrspace(1)* %out, <4 x i64> addrspace(1)* %in) #0 {
76  %tid = call i32 @llvm.amdgcn.workitem.id.x()
77  %in.gep = getelementptr <4 x i64>, <4 x i64> addrspace(1)* %in, i32 %tid
78  %out.gep = getelementptr <4 x float>, <4 x float> addrspace(1)* %out, i32 %tid
79  %value = load <4 x i64>, <4 x i64> addrspace(1)* %in.gep
80  %result = uitofp <4 x i64> %value to <4 x float>
81  store <4 x float> %result, <4 x float> addrspace(1)* %out.gep
82  ret void
83}
84
85; FUNC-LABEL: {{^}}s_uint_to_fp_v2i64_to_v2f16:
86define amdgpu_kernel void @s_uint_to_fp_v2i64_to_v2f16(<2 x half> addrspace(1)* %out, <2 x i64> %in) #0{
87  %result = uitofp <2 x i64> %in to <2 x half>
88  store <2 x half> %result, <2 x half> addrspace(1)* %out
89  ret void
90}
91
92; FUNC-LABEL: {{^}}v_uint_to_fp_v4i64_to_v4f16:
93define amdgpu_kernel void @v_uint_to_fp_v4i64_to_v4f16(<4 x half> addrspace(1)* %out, <4 x i64> addrspace(1)* %in) #0 {
94  %tid = call i32 @llvm.amdgcn.workitem.id.x()
95  %in.gep = getelementptr <4 x i64>, <4 x i64> addrspace(1)* %in, i32 %tid
96  %out.gep = getelementptr <4 x half>, <4 x half> addrspace(1)* %out, i32 %tid
97  %value = load <4 x i64>, <4 x i64> addrspace(1)* %in.gep
98  %result = uitofp <4 x i64> %value to <4 x half>
99  store <4 x half> %result, <4 x half> addrspace(1)* %out.gep
100  ret void
101}
102
103declare i32 @llvm.amdgcn.workitem.id.x() #1
104
105attributes #0 = { nounwind }
106attributes #1 = { nounwind readnone }
107