1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -march=amdgcn -verify-machineinstrs -o - %s | FileCheck -check-prefix=GCN %s
4 declare i32 @llvm.cttz.i32(i32, i1) nounwind readnone
5 declare i32 @llvm.amdgcn.sffbh.i32(i32) nounwind readnone speculatable
6 define amdgpu_kernel void @select_constant_cttz(i32 addrspace(1)* noalias %out, i32 addrspace(1)* nocapture readonly %arrayidx) nounwind {
7 ; GCN-LABEL: select_constant_cttz:
9 ; GCN-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0xb
10 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
11 ; GCN-NEXT: s_load_dword s2, s[2:3], 0x0
12 ; GCN-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9
13 ; GCN-NEXT: s_mov_b32 s7, 0xf000
14 ; GCN-NEXT: s_waitcnt lgkmcnt(0)
15 ; GCN-NEXT: s_lshr_b32 s0, 1, s2
16 ; GCN-NEXT: s_ff1_i32_b32 s0, s0
17 ; GCN-NEXT: s_mov_b32 s6, -1
18 ; GCN-NEXT: v_mov_b32_e32 v0, s0
19 ; GCN-NEXT: v_cmp_ne_u32_e64 s[2:3], s2, 0
20 ; GCN-NEXT: v_cndmask_b32_e64 v0, v0, -1, s[2:3]
21 ; GCN-NEXT: v_ffbh_i32_e32 v1, v0
22 ; GCN-NEXT: v_cmp_eq_u32_e64 s[0:1], 0, v0
23 ; GCN-NEXT: v_sub_i32_e32 v0, vcc, 31, v1
24 ; GCN-NEXT: s_or_b64 s[0:1], s[2:3], s[0:1]
25 ; GCN-NEXT: v_cndmask_b32_e64 v0, v0, -1, s[0:1]
26 ; GCN-NEXT: buffer_store_dword v0, off, s[4:7], 0
28 %v = load i32, i32 addrspace(1)* %arrayidx, align 4
30 %cmp = icmp ne i32 %v, 0
31 %cttz = call i32 @llvm.cttz.i32(i32 %sr, i1 true), !range !0
32 %sel = select i1 %cmp, i32 -1, i32 %cttz
33 %ffbh = call i32 @llvm.amdgcn.sffbh.i32(i32 %sel)
34 %sub = sub i32 31, %ffbh
35 %cmp2 = icmp eq i32 %sel, 0
36 %or = or i1 %cmp, %cmp2
37 %sel2 = select i1 %or, i32 -1, i32 %sub
38 store i32 %sel2, i32 addrspace(1)* %out