150
|
1 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
|
|
2 ; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
|
|
3
|
|
4 ; FUNC-LABEL: {{^}}fpext_f32_to_f64:
|
|
5 ; SI: v_cvt_f64_f32_e32 {{v\[[0-9]+:[0-9]+\]}}, s{{[0-9]+}}
|
252
|
6 define amdgpu_kernel void @fpext_f32_to_f64(ptr addrspace(1) %out, float %in) {
|
150
|
7 %result = fpext float %in to double
|
252
|
8 store double %result, ptr addrspace(1) %out
|
150
|
9 ret void
|
|
10 }
|
|
11
|
|
12 ; FUNC-LABEL: {{^}}fpext_v2f32_to_v2f64:
|
|
13 ; SI: v_cvt_f64_f32_e32
|
|
14 ; SI: v_cvt_f64_f32_e32
|
252
|
15 define amdgpu_kernel void @fpext_v2f32_to_v2f64(ptr addrspace(1) %out, <2 x float> %in) {
|
150
|
16 %result = fpext <2 x float> %in to <2 x double>
|
252
|
17 store <2 x double> %result, ptr addrspace(1) %out
|
150
|
18 ret void
|
|
19 }
|
|
20
|
|
21 ; FUNC-LABEL: {{^}}fpext_v3f32_to_v3f64:
|
|
22 ; SI: v_cvt_f64_f32_e32
|
|
23 ; SI: v_cvt_f64_f32_e32
|
|
24 ; SI: v_cvt_f64_f32_e32
|
252
|
25 define amdgpu_kernel void @fpext_v3f32_to_v3f64(ptr addrspace(1) %out, <3 x float> %in) {
|
150
|
26 %result = fpext <3 x float> %in to <3 x double>
|
252
|
27 store <3 x double> %result, ptr addrspace(1) %out
|
150
|
28 ret void
|
|
29 }
|
|
30
|
|
31 ; FUNC-LABEL: {{^}}fpext_v4f32_to_v4f64:
|
|
32 ; SI: v_cvt_f64_f32_e32
|
|
33 ; SI: v_cvt_f64_f32_e32
|
|
34 ; SI: v_cvt_f64_f32_e32
|
|
35 ; SI: v_cvt_f64_f32_e32
|
252
|
36 define amdgpu_kernel void @fpext_v4f32_to_v4f64(ptr addrspace(1) %out, <4 x float> %in) {
|
150
|
37 %result = fpext <4 x float> %in to <4 x double>
|
252
|
38 store <4 x double> %result, ptr addrspace(1) %out
|
150
|
39 ret void
|
|
40 }
|
|
41
|
|
42 ; FUNC-LABEL: {{^}}fpext_v8f32_to_v8f64:
|
|
43 ; SI: v_cvt_f64_f32_e32
|
|
44 ; SI: v_cvt_f64_f32_e32
|
|
45 ; SI: v_cvt_f64_f32_e32
|
|
46 ; SI: v_cvt_f64_f32_e32
|
|
47 ; SI: v_cvt_f64_f32_e32
|
|
48 ; SI: v_cvt_f64_f32_e32
|
|
49 ; SI: v_cvt_f64_f32_e32
|
|
50 ; SI: v_cvt_f64_f32_e32
|
252
|
51 define amdgpu_kernel void @fpext_v8f32_to_v8f64(ptr addrspace(1) %out, <8 x float> %in) {
|
150
|
52 %result = fpext <8 x float> %in to <8 x double>
|
252
|
53 store <8 x double> %result, ptr addrspace(1) %out
|
150
|
54 ret void
|
|
55 }
|