• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa %s | FileCheck -check-prefixes=GCN,CI %s
2; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=fiji %s | FileCheck -check-prefixes=GCN,VI %s
3; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx900 %s | FileCheck -check-prefixes=GCN,GFX9 %s
4
5; GCN-LABEL: 'insertelement_v2i32'
6; GCN: estimated cost of 0 for {{.*}} insertelement <2 x i32>
7define amdgpu_kernel void @insertelement_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %vaddr) {
8  %vec = load <2 x i32>, <2 x i32> addrspace(1)* %vaddr
9  %insert = insertelement <2 x i32> %vec, i32 123, i32 1
10  store <2 x i32> %insert, <2 x i32> addrspace(1)* %out
11  ret void
12}
13
14; GCN-LABEL: 'insertelement_v2i64'
15; GCN: estimated cost of 0 for {{.*}} insertelement <2 x i64>
16define amdgpu_kernel void @insertelement_v2i64(<2 x i64> addrspace(1)* %out, <2 x i64> addrspace(1)* %vaddr) {
17  %vec = load <2 x i64>, <2 x i64> addrspace(1)* %vaddr
18  %insert = insertelement <2 x i64> %vec, i64 123, i64 1
19  store <2 x i64> %insert, <2 x i64> addrspace(1)* %out
20  ret void
21}
22
23; GCN-LABEL: 'insertelement_0_v2i16'
24; CI: estimated cost of 1 for {{.*}} insertelement <2 x i16>
25; VI: estimated cost of 0 for {{.*}} insertelement <2 x i16>
26; GFX9: estimated cost of 0 for {{.*}} insertelement <2 x i16>
27define amdgpu_kernel void @insertelement_0_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %vaddr) {
28  %vec = load <2 x i16>, <2 x i16> addrspace(1)* %vaddr
29  %insert = insertelement <2 x i16> %vec, i16 123, i16 0
30  store <2 x i16> %insert, <2 x i16> addrspace(1)* %out
31  ret void
32}
33
34; GCN-LABEL: 'insertelement_1_v2i16'
35; GCN: estimated cost of 1 for {{.*}} insertelement <2 x i16>
36define amdgpu_kernel void @insertelement_1_v2i16(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %vaddr) {
37  %vec = load <2 x i16>, <2 x i16> addrspace(1)* %vaddr
38  %insert = insertelement <2 x i16> %vec, i16 123, i16 1
39  store <2 x i16> %insert, <2 x i16> addrspace(1)* %out
40  ret void
41}
42
43; GCN-LABEL: 'insertelement_1_v2i8'
44; GCN: estimated cost of 1 for {{.*}} insertelement <2 x i8>
45define amdgpu_kernel void @insertelement_1_v2i8(<2 x i8> addrspace(1)* %out, <2 x i8> addrspace(1)* %vaddr) {
46  %vec = load <2 x i8>, <2 x i8> addrspace(1)* %vaddr
47  %insert = insertelement <2 x i8> %vec, i8 123, i8 1
48  store <2 x i8> %insert, <2 x i8> addrspace(1)* %out
49  ret void
50}
51