• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX10 %s
2; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX9 %s
3
4; GCN-LABEL: {{^}}test_add_lit:
5; GFX10: v_add_co_u32_e64 v{{[0-9]+}}, vcc_lo, 0x80992bff, v{{[0-9]+}}
6; GFX10: v_add_co_ci_u32_e32 v{{[0-9]+}}, vcc_lo, 0xe7, v{{[0-9]+}}, vcc_lo
7; GFX9:  v_mov_b32_e32 [[C2:v[0-9]+]], 0xe7
8; GFX9:  v_add_co_u32_e32 v{{[0-9]+}}, vcc, 0x80992bff, v{{[0-9]+}}
9; GFX9:  v_addc_co_u32_e32 v{{[0-9]+}}, vcc, v{{[0-9]+}}, [[C2]], vcc
10define amdgpu_kernel void @test_add_lit(i64 addrspace(1)* %p) {
11  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
12  %ptr = getelementptr inbounds i64, i64 addrspace(1)* %p, i32 %id
13  %load = load i64, i64 addrspace(1)* %ptr, align 8
14  %add = add nsw i64 %load, 994294967295
15  store i64 %add, i64 addrspace(1)* %ptr, align 8
16  ret void
17}
18
19; GCN-LABEL: {{^}}test_cndmask_lit:
20; GFX10: v_cndmask_b32_e32 v{{[0-9]+}}, 0x3039, v{{[0-9]+}}, vcc_lo
21; GFX9:  v_mov_b32_e32 [[C:v[0-9]+]], 0x3039
22; GFX9:  v_cndmask_b32_e32 v{{[0-9]+}}, [[C]], v{{[0-9]+}}, vcc
23define amdgpu_kernel void @test_cndmask_lit(i32 addrspace(1)* %p) {
24  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
25  %n = add nuw nsw i32 %id, 1
26  %p1 = getelementptr inbounds i32, i32 addrspace(1)* %p, i32 %id
27  %v1 = load i32, i32 addrspace(1)* %p1, align 4
28  %p2 = getelementptr inbounds i32, i32 addrspace(1)* %p, i32 %n
29  %v2 = load i32, i32 addrspace(1)* %p2, align 4
30  %cmp = icmp sgt i32 %v1, 0
31  %sel = select i1 %cmp, i32 12345, i32 %v2
32  store i32 %sel, i32 addrspace(1)* %p1, align 4
33  ret void
34}
35
36; GCN-LABEL: {{^}}test_bfe_2lit_s:
37; GFX10: v_mov_b32_e32 [[C1:v[0-9]+]], 0xddd5
38; GFX10: v_bfe_u32 v{{[0-9]+}}, 0x3039, s{{[0-9]+}}, [[C1]]
39; GFX9-DAG: v_mov_b32_e32 [[C2:v[0-9]+]], 0xddd5
40; GFX9-DAG: s_movk_i32 [[C1:s[0-9]+]], 0x3039
41; GFX9:     v_bfe_u32 v{{[0-9]+}}, [[C1]], v{{[0-9]+}}, [[C2]]
42define amdgpu_kernel void @test_bfe_2lit_s(i32 addrspace(1)* %p, i32 %src) {
43  %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 12345, i32 %src, i32 56789)
44  store i32 %bfe, i32 addrspace(1)* %p, align 4
45  ret void
46}
47
48; GCN-LABEL: {{^}}test_bfe_2lit_v:
49; GFX10: s_movk_i32 [[C1:s[0-9]+]], 0x3039
50; GFX10: v_bfe_u32 v{{[0-9]+}}, [[C1]], v{{[0-9]+}}, 0xddd5
51; GFX9-DAG: v_mov_b32_e32 [[C2:v[0-9]+]], 0xddd5
52; GFX9-DAG: s_movk_i32 [[C1:s[0-9]+]], 0x3039
53; GFX9:     v_bfe_u32 v{{[0-9]+}}, [[C1]], v{{[0-9]+}}, [[C2]]
54define amdgpu_kernel void @test_bfe_2lit_v(i32 addrspace(1)* %p) {
55  %id = tail call i32 @llvm.amdgcn.workitem.id.x()
56  %ptr = getelementptr inbounds i32, i32 addrspace(1)* %p, i32 %id
57  %load = load i32, i32 addrspace(1)* %ptr, align 4
58  %bfe = call i32 @llvm.amdgcn.ubfe.i32(i32 12345, i32 %load, i32 56789)
59  store i32 %bfe, i32 addrspace(1)* %ptr, align 4
60  ret void
61}
62
63declare i32 @llvm.amdgcn.workitem.id.x()
64declare i32 @llvm.amdgcn.ubfe.i32(i32, i32, i32)
65