• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 sint_to_fp.ll, but s_sint_to_fp_v2i64 crashes on r600
5
6; FUNC-LABEL: {{^}}s_sint_to_fp_i64_to_f16:
7define amdgpu_kernel void @s_sint_to_fp_i64_to_f16(half addrspace(1)* %out, i64 %in) #0 {
8  %result = sitofp i64 %in to half
9  store half %result, half addrspace(1)* %out
10  ret void
11}
12
13; FUNC-LABEL: {{^}}v_sint_to_fp_i64_to_f16:
14; GCN: {{buffer|flat}}_load_dwordx2
15
16; GCN: v_ashrrev_i32_e32 v{{[0-9]+}}, 31, v{{[0-9]+}}
17; GCN: v_xor_b32
18
19; GCN: v_ffbh_u32
20; GCN: v_ffbh_u32
21; GCN: v_cndmask
22; GCN: v_cndmask
23
24; GCN-DAG: v_cmp_eq_u64
25; GCN-DAG: v_cmp_gt_u64
26
27; GCN: v_cndmask_b32_e64 [[SIGN_SEL:v[0-9]+]], v{{[0-9]+}}, -v{{[0-9]+}}
28; GCN: v_cvt_f16_f32_e32 [[SIGN_SEL_F16:v[0-9]+]], [[SIGN_SEL]]
29; GCN: {{buffer|flat}}_store_short {{.*}}[[SIGN_SEL_F16]]
30define amdgpu_kernel void @v_sint_to_fp_i64_to_f16(half addrspace(1)* %out, i64 addrspace(1)* %in) #0 {
31  %tid = call i32 @llvm.amdgcn.workitem.id.x()
32  %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
33  %out.gep = getelementptr half, half addrspace(1)* %out, i32 %tid
34  %val = load i64, i64 addrspace(1)* %in.gep
35  %result = sitofp i64 %val to half
36  store half %result, half addrspace(1)* %out.gep
37  ret void
38}
39
40; FUNC-LABEL: {{^}}s_sint_to_fp_i64_to_f32:
41define amdgpu_kernel void @s_sint_to_fp_i64_to_f32(float addrspace(1)* %out, i64 %in) #0 {
42  %result = sitofp i64 %in to float
43  store float %result, float addrspace(1)* %out
44  ret void
45}
46
47; FUNC-LABEL: {{^}}v_sint_to_fp_i64_to_f32:
48; GCN: {{buffer|flat}}_load_dwordx2
49
50; GCN: v_ashrrev_i32_e32 v{{[0-9]+}}, 31, v{{[0-9]+}}
51; GCN: v_xor_b32
52
53; GCN: v_ffbh_u32
54; GCN: v_ffbh_u32
55; GCN: v_cndmask
56; GCN: v_cndmask
57
58; GCN-DAG: v_cmp_eq_u64
59; GCN-DAG: v_cmp_gt_u64
60
61; GCN: v_cndmask_b32_e64 [[SIGN_SEL:v[0-9]+]], v{{[0-9]+}}, -v{{[0-9]+}}
62; GCN: {{buffer|flat}}_store_dword {{.*}}[[SIGN_SEL]]
63define amdgpu_kernel void @v_sint_to_fp_i64_to_f32(float addrspace(1)* %out, i64 addrspace(1)* %in) #0 {
64  %tid = call i32 @llvm.amdgcn.workitem.id.x()
65  %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
66  %out.gep = getelementptr float, float addrspace(1)* %out, i32 %tid
67  %val = load i64, i64 addrspace(1)* %in.gep
68  %result = sitofp i64 %val to float
69  store float %result, float addrspace(1)* %out.gep
70  ret void
71}
72
73; FUNC-LABEL: {{^}}s_sint_to_fp_v2i64_to_v2f32:
74; GCN-NOT: v_and_b32_e32 v{{[0-9]+}}, -1,
75define amdgpu_kernel void @s_sint_to_fp_v2i64_to_v2f32(<2 x float> addrspace(1)* %out, <2 x i64> %in) #0{
76  %result = sitofp <2 x i64> %in to <2 x float>
77  store <2 x float> %result, <2 x float> addrspace(1)* %out
78  ret void
79}
80
81; FUNC-LABEL: {{^}}v_sint_to_fp_v4i64_to_v4f32:
82define amdgpu_kernel void @v_sint_to_fp_v4i64_to_v4f32(<4 x float> addrspace(1)* %out, <4 x i64> addrspace(1)* %in) #0 {
83  %tid = call i32 @llvm.amdgcn.workitem.id.x()
84  %in.gep = getelementptr <4 x i64>, <4 x i64> addrspace(1)* %in, i32 %tid
85  %out.gep = getelementptr <4 x float>, <4 x float> addrspace(1)* %out, i32 %tid
86  %value = load <4 x i64>, <4 x i64> addrspace(1)* %in.gep
87  %result = sitofp <4 x i64> %value to <4 x float>
88  store <4 x float> %result, <4 x float> addrspace(1)* %out.gep
89  ret void
90}
91
92; FUNC-LABEL: {{^}}s_sint_to_fp_v2i64_to_v2f16:
93; GCN-NOT: v_and_b32_e32 v{{[0-9]+}}, -1,
94define amdgpu_kernel void @s_sint_to_fp_v2i64_to_v2f16(<2 x half> addrspace(1)* %out, <2 x i64> %in) #0{
95  %result = sitofp <2 x i64> %in to <2 x half>
96  store <2 x half> %result, <2 x half> addrspace(1)* %out
97  ret void
98}
99
100; FUNC-LABEL: {{^}}v_sint_to_fp_v4i64_to_v4f16:
101define amdgpu_kernel void @v_sint_to_fp_v4i64_to_v4f16(<4 x half> addrspace(1)* %out, <4 x i64> addrspace(1)* %in) #0 {
102  %tid = call i32 @llvm.amdgcn.workitem.id.x()
103  %in.gep = getelementptr <4 x i64>, <4 x i64> addrspace(1)* %in, i32 %tid
104  %out.gep = getelementptr <4 x half>, <4 x half> addrspace(1)* %out, i32 %tid
105  %value = load <4 x i64>, <4 x i64> addrspace(1)* %in.gep
106  %result = sitofp <4 x i64> %value to <4 x half>
107  store <4 x half> %result, <4 x half> addrspace(1)* %out.gep
108  ret void
109}
110
111declare i32 @llvm.amdgcn.workitem.id.x() #1
112
113attributes #0 = { nounwind }
114attributes #1 = { nounwind readnone }
115