xref: /llvm-project/llvm/test/CodeGen/AMDGPU/select-constant-cttz.ll (revision 6548b6354d1d990e1c98736f5e7c3de876bedc8e)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=amdgcn -verify-machineinstrs -o - %s | FileCheck -check-prefix=GCN %s
3
4declare i32 @llvm.cttz.i32(i32, i1) nounwind readnone
5declare i32 @llvm.amdgcn.sffbh.i32(i32) nounwind readnone speculatable
6define amdgpu_kernel void @select_constant_cttz(ptr addrspace(1) noalias %out, ptr addrspace(1) nocapture readonly %arrayidx) nounwind {
7; GCN-LABEL: select_constant_cttz:
8; GCN:       ; %bb.0:
9; GCN-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x9
10; GCN-NEXT:    s_waitcnt lgkmcnt(0)
11; GCN-NEXT:    s_load_dword s2, s[2:3], 0x0
12; GCN-NEXT:    s_mov_b32 s3, 0xf000
13; GCN-NEXT:    s_waitcnt lgkmcnt(0)
14; GCN-NEXT:    s_lshr_b32 s4, 1, s2
15; GCN-NEXT:    s_cmp_lg_u32 s2, 0
16; GCN-NEXT:    s_ff1_i32_b32 s2, s4
17; GCN-NEXT:    s_cselect_b64 s[4:5], -1, 0
18; GCN-NEXT:    s_and_b64 s[6:7], s[4:5], exec
19; GCN-NEXT:    s_cselect_b32 s2, -1, s2
20; GCN-NEXT:    s_flbit_i32 s6, s2
21; GCN-NEXT:    s_sub_i32 s8, 31, s6
22; GCN-NEXT:    s_cmp_eq_u32 s2, 0
23; GCN-NEXT:    s_cselect_b64 s[6:7], -1, 0
24; GCN-NEXT:    s_or_b64 s[4:5], s[4:5], s[6:7]
25; GCN-NEXT:    s_and_b64 s[4:5], s[4:5], exec
26; GCN-NEXT:    s_cselect_b32 s4, -1, s8
27; GCN-NEXT:    s_mov_b32 s2, -1
28; GCN-NEXT:    v_mov_b32_e32 v0, s4
29; GCN-NEXT:    buffer_store_dword v0, off, s[0:3], 0
30; GCN-NEXT:    s_endpgm
31  %v    = load i32, ptr addrspace(1) %arrayidx, align 4
32  %sr   = lshr i32 1, %v
33  %cmp  = icmp ne i32 %v, 0
34  %cttz = call i32 @llvm.cttz.i32(i32 %sr, i1 true), !range !0
35  %sel  = select i1 %cmp, i32 -1, i32 %cttz
36  %ffbh = call i32 @llvm.amdgcn.sffbh.i32(i32 %sel)
37  %sub  = sub i32 31, %ffbh
38  %cmp2 = icmp eq i32 %sel, 0
39  %or   = or i1 %cmp, %cmp2
40  %sel2 = select i1 %or, i32 -1, i32 %sub
41  store i32 %sel2, ptr addrspace(1) %out
42  ret void
43}
44
45!0 = !{i32 0, i32 33}
46