• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; XFAIL: *
2; RUN: llc -march=amdgcn -mcpu=SI < %s
3
4; GCN-LABEL: {{^}}global_truncstore_f64_to_f16:
5; GCN: s_endpgm
6define void @global_truncstore_f64_to_f16(half addrspace(1)* %out, double addrspace(1)* %in) #0 {
7  %val = load double, double addrspace(1)* %in
8  %cvt = fptrunc double %val to half
9  store half %cvt, half addrspace(1)* %out
10  ret void
11}
12
13; GCN-LABEL: {{^}}global_truncstore_v2f64_to_v2f16:
14; GCN: s_endpgm
15define void @global_truncstore_v2f64_to_v2f16(<2 x half> addrspace(1)* %out, <2 x double> addrspace(1)* %in) #0 {
16  %val = load <2 x double>, <2 x double> addrspace(1)* %in
17  %cvt = fptrunc <2 x double> %val to <2 x half>
18  store <2 x half> %cvt, <2 x half> addrspace(1)* %out
19  ret void
20}
21
22; GCN-LABEL: {{^}}global_truncstore_v3f64_to_v3f16:
23; GCN: s_endpgm
24define void @global_truncstore_v3f64_to_v3f16(<3 x half> addrspace(1)* %out, <3 x double> addrspace(1)* %in) #0 {
25  %val = load <3 x double>, <3 x double> addrspace(1)* %in
26  %cvt = fptrunc <3 x double> %val to <3 x half>
27  store <3 x half> %cvt, <3 x half> addrspace(1)* %out
28  ret void
29}
30
31; GCN-LABEL: {{^}}global_truncstore_v4f64_to_v4f16:
32; GCN: s_endpgm
33define void @global_truncstore_v4f64_to_v4f16(<4 x half> addrspace(1)* %out, <4 x double> addrspace(1)* %in) #0 {
34  %val = load <4 x double>, <4 x double> addrspace(1)* %in
35  %cvt = fptrunc <4 x double> %val to <4 x half>
36  store <4 x half> %cvt, <4 x half> addrspace(1)* %out
37  ret void
38}
39
40; GCN-LABEL: {{^}}global_truncstore_v8f64_to_v8f16:
41; GCN: s_endpgm
42define void @global_truncstore_v8f64_to_v8f16(<8 x half> addrspace(1)* %out, <8 x double> addrspace(1)* %in) #0 {
43  %val = load <8 x double>, <8 x double> addrspace(1)* %in
44  %cvt = fptrunc <8 x double> %val to <8 x half>
45  store <8 x half> %cvt, <8 x half> addrspace(1)* %out
46  ret void
47}
48
49; GCN-LABEL: {{^}}global_truncstore_v16f64_to_v16f16:
50; GCN: s_endpgm
51define void @global_truncstore_v16f64_to_v16f16(<16 x half> addrspace(1)* %out, <16 x double> addrspace(1)* %in) #0 {
52  %val = load <16 x double>, <16 x double> addrspace(1)* %in
53  %cvt = fptrunc <16 x double> %val to <16 x half>
54  store <16 x half> %cvt, <16 x half> addrspace(1)* %out
55  ret void
56}
57