• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GFX9 -check-prefix=GCN %s
2
3; GCN-LABEL: {{^}}inline_asm_input_v2i16:
4; GCN: s_mov_b32 s{{[0-9]+}}, s{{[0-9]+}}
5define amdgpu_kernel void @inline_asm_input_v2i16(i32 addrspace(1)* %out, <2 x i16> %in) #0 {
6entry:
7  %val = call i32 asm "s_mov_b32 $0, $1", "=r,r"(<2 x i16> %in) #0
8  store i32 %val, i32 addrspace(1)* %out
9  ret void
10}
11
12; GCN-LABEL: {{^}}inline_asm_input_v2f16:
13; GCN: s_mov_b32 s0, s{{[0-9]+}}
14define amdgpu_kernel void @inline_asm_input_v2f16(i32 addrspace(1)* %out, <2 x half> %in) #0 {
15entry:
16  %val = call i32 asm "s_mov_b32 $0, $1", "=r,r"(<2 x half> %in) #0
17  store i32 %val, i32 addrspace(1)* %out
18  ret void
19}
20
21; GCN-LABEL: {{^}}inline_asm_output_v2i16:
22; GCN: s_mov_b32 s{{[0-9]+}}, s{{[0-9]+}}
23define amdgpu_kernel void @inline_asm_output_v2i16(<2 x i16> addrspace(1)* %out, i32 %in) #0 {
24entry:
25  %val = call <2 x i16> asm "s_mov_b32 $0, $1", "=r,r"(i32 %in) #0
26  store <2 x i16> %val, <2 x i16> addrspace(1)* %out
27  ret void
28}
29
30; GCN-LABEL: {{^}}inline_asm_output_v2f16:
31; GCN: v_mov_b32 v{{[0-9]+}}, s{{[0-9]+}}
32define amdgpu_kernel void @inline_asm_output_v2f16(<2 x half> addrspace(1)* %out, i32 %in) #0 {
33entry:
34  %val = call <2 x half> asm "v_mov_b32 $0, $1", "=v,r"(i32 %in) #0
35  store <2 x half> %val, <2 x half> addrspace(1)* %out
36  ret void
37}
38
39; GCN-LABEL: {{^}}inline_asm_packed_v2i16:
40; GCN: v_pk_add_u16 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}
41define amdgpu_kernel void @inline_asm_packed_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> %in0, <2 x i16> %in1) #0 {
42entry:
43  %val = call <2 x i16> asm "v_pk_add_u16 $0, $1, $2", "=v,r,v"(<2 x i16> %in0, <2 x i16> %in1) #0
44  store <2 x i16> %val, <2 x i16> addrspace(1)* %out
45  ret void
46}
47
48; GCN-LABEL: {{^}}inline_asm_packed_v2f16:
49; GCN: v_pk_add_f16 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}
50define amdgpu_kernel void @inline_asm_packed_v2f16(<2 x half> addrspace(1)* %out, <2 x half> %in0, <2 x half> %in1) #0 {
51entry:
52  %val = call <2 x half> asm "v_pk_add_f16 $0, $1, $2", "=v,r,v"(<2 x half> %in0, <2 x half> %in1) #0
53  store <2 x half> %val, <2 x half> addrspace(1)* %out
54  ret void
55}
56
57attributes #0 = { nounwind }
58