blob: 67ab2607acd5fd6aa7f28d4f65368bbce38c421b [file] [log] [blame]
Matt Arsenault3c5e4232017-05-10 21:29:33 +00001; 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
Matt Arsenault59767ce2016-03-25 00:14:11 +00004
Matt Arsenault3c5e4232017-05-10 21:29:33 +00005; GCN-LABEL: 'insertelement_v2i32'
6; GCN: estimated cost of 0 for {{.*}} insertelement <2 x i32>
Matt Arsenault3dbeefa2017-03-21 21:39:51 +00007define amdgpu_kernel void @insertelement_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %vaddr) {
Matt Arsenault59767ce2016-03-25 00:14:11 +00008 %vec = load <2 x i32>, <2 x i32> addrspace(1)* %vaddr
Matt Arsenault3c5e4232017-05-10 21:29:33 +00009 %insert = insertelement <2 x i32> %vec, i32 123, i32 1
Matt Arsenault59767ce2016-03-25 00:14:11 +000010 store <2 x i32> %insert, <2 x i32> addrspace(1)* %out
11 ret void
12}
13
Matt Arsenault3c5e4232017-05-10 21:29:33 +000014; GCN-LABEL: 'insertelement_v2i64'
15; GCN: estimated cost of 0 for {{.*}} insertelement <2 x i64>
Matt Arsenault3dbeefa2017-03-21 21:39:51 +000016define amdgpu_kernel void @insertelement_v2i64(<2 x i64> addrspace(1)* %out, <2 x i64> addrspace(1)* %vaddr) {
Matt Arsenault59767ce2016-03-25 00:14:11 +000017 %vec = load <2 x i64>, <2 x i64> addrspace(1)* %vaddr
Matt Arsenault3c5e4232017-05-10 21:29:33 +000018 %insert = insertelement <2 x i64> %vec, i64 123, i64 1
Matt Arsenault59767ce2016-03-25 00:14:11 +000019 store <2 x i64> %insert, <2 x i64> addrspace(1)* %out
20 ret void
21}
22
Matt Arsenault3c5e4232017-05-10 21:29:33 +000023; 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) {
Matt Arsenault59767ce2016-03-25 00:14:11 +000028 %vec = load <2 x i16>, <2 x i16> addrspace(1)* %vaddr
Matt Arsenault3c5e4232017-05-10 21:29:33 +000029 %insert = insertelement <2 x i16> %vec, i16 123, i16 0
Matt Arsenault59767ce2016-03-25 00:14:11 +000030 store <2 x i16> %insert, <2 x i16> addrspace(1)* %out
31 ret void
32}
33
Matt Arsenault3c5e4232017-05-10 21:29:33 +000034; 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) {
Matt Arsenault59767ce2016-03-25 00:14:11 +000046 %vec = load <2 x i8>, <2 x i8> addrspace(1)* %vaddr
Matt Arsenault3c5e4232017-05-10 21:29:33 +000047 %insert = insertelement <2 x i8> %vec, i8 123, i8 1
Matt Arsenault59767ce2016-03-25 00:14:11 +000048 store <2 x i8> %insert, <2 x i8> addrspace(1)* %out
49 ret void
50}