150
|
1 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
|
|
2 ; RUN: llc -march=amdgcn -mcpu=fiji -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
|
|
3
|
|
4 declare i64 @llvm.amdgcn.qsad.pk.u16.u8(i64, i32, i64) #0
|
|
5
|
|
6 ; GCN-LABEL: {{^}}v_qsad_pk_u16_u8:
|
|
7 ; GCN: v_qsad_pk_u16_u8 v[0:1], v[4:5], s{{[0-9]+}}, v[{{[0-9]+:[0-9]+}}]
|
|
8 ; GCN-DAG: v_mov_b32_e32 v5, v1
|
|
9 ; GCN-DAG: v_mov_b32_e32 v4, v0
|
|
10 define amdgpu_kernel void @v_qsad_pk_u16_u8(i64 addrspace(1)* %out, i64 %src) {
|
|
11 %tmp = call i64 asm "v_lsrlrev_b64 $0, $1, 1", "={v[4:5]},v"(i64 %src) #0
|
|
12 %tmp1 = call i64 @llvm.amdgcn.qsad.pk.u16.u8(i64 %tmp, i32 100, i64 100) #0
|
|
13 %tmp2 = call i64 asm ";; force constraint", "=v,{v[4:5]}"(i64 %tmp1) #0
|
|
14 store i64 %tmp2, i64 addrspace(1)* %out, align 4
|
|
15 ret void
|
|
16 }
|
|
17
|
|
18 ; GCN-LABEL: {{^}}v_qsad_pk_u16_u8_non_immediate:
|
|
19 ; GCN: v_qsad_pk_u16_u8 v[0:1], v[2:3], v4, v[6:7]
|
|
20 ; GCN-DAG: v_mov_b32_e32 v3, v1
|
|
21 ; GCN-DAG: v_mov_b32_e32 v2, v0
|
|
22 define amdgpu_kernel void @v_qsad_pk_u16_u8_non_immediate(i64 addrspace(1)* %out, i64 %src, i32 %a, i64 %b) {
|
|
23 %tmp = call i64 asm "v_lsrlrev_b64 $0, $1, 1", "={v[2:3]},v"(i64 %src) #0
|
|
24 %tmp1 = call i32 asm "v_mov_b32 $0, $1", "={v4},v"(i32 %a) #0
|
|
25 %tmp2 = call i64 asm "v_lshlrev_b64 $0, $1, 1", "={v[6:7]},v"(i64 %b) #0
|
|
26 %tmp3 = call i64 @llvm.amdgcn.qsad.pk.u16.u8(i64 %tmp, i32 %tmp1, i64 %tmp2) #0
|
|
27 %tmp4 = call i64 asm ";; force constraint", "=v,{v[2:3]}"(i64 %tmp3) #0
|
|
28 store i64 %tmp4, i64 addrspace(1)* %out, align 4
|
|
29 ret void
|
|
30 }
|
|
31
|
|
32 attributes #0 = { nounwind readnone }
|