1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -march=amdgcn -mcpu=tahiti -amdgpu-dce-in-ra=0 -o - %s | FileCheck %s
3 ; Don't crash when the use of an undefined value is only detected by the
4 ; register coalescer because it is hidden with subregister insert/extract.
5 target triple="amdgcn--"
7 define amdgpu_kernel void @foobar(float %a0, float %a1, float addrspace(1)* %out) nounwind {
9 ; CHECK: ; %bb.0: ; %entry
10 ; CHECK-NEXT: s_load_dwordx2 s[4:5], s[0:1], 0x9
11 ; CHECK-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xb
12 ; CHECK-NEXT: v_mbcnt_lo_u32_b32_e64 v0, -1, 0
13 ; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
14 ; CHECK-NEXT: s_mov_b32 s2, -1
15 ; CHECK-NEXT: s_waitcnt lgkmcnt(0)
17 ; FIXME: The change related to the fact that
18 ; DetectDeadLanes pass hit "Copy across incompatible class" SGPR -> VGPR in analysis
19 ; and hence it cannot derive the fact that the vector element is unused.
20 ; Such a copies appear because the float4 vectors and their elements in the test are uniform
21 ; but the PHI node in "ife" block is divergent because of the CF dependency (divergent branch in bb0)
23 ; CHECK-NEXT: v_mov_b32_e32 v0, s4
24 ; CHECK-NEXT: v_mov_b32_e32 v1, s5
25 ; CHECK-NEXT: v_mov_b32_e32 v2, s6
26 ; CHECK-NEXT: v_mov_b32_e32 v3, s7
28 ; CHECK-NEXT: s_and_saveexec_b64 s[6:7], vcc
29 ; CHECK-NEXT: ; mask branch BB0_2
30 ; CHECK-NEXT: BB0_1: ; %ift
31 ; CHECK-NEXT: s_mov_b32 s4, s5
32 ; CHECK-NEXT: v_mov_b32_e32 v0, s4
33 ; CHECK-NEXT: v_mov_b32_e32 v1, s5
34 ; CHECK-NEXT: v_mov_b32_e32 v2, s6
35 ; CHECK-NEXT: v_mov_b32_e32 v3, s7
36 ; CHECK-NEXT: BB0_2: ; %ife
37 ; CHECK-NEXT: s_or_b64 exec, exec, s[6:7]
38 ; CHECK-NEXT: s_mov_b32 s3, 0xf000
39 ; CHECK-NEXT: buffer_store_dword v1, off, s[0:3], 0
40 ; CHECK-NEXT: s_endpgm
42 %v0 = insertelement <4 x float> undef, float %a0, i32 0
43 %tid = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) #0
44 %cnd = icmp eq i32 %tid, 0
45 br i1 %cnd, label %ift, label %ife
48 %v1 = insertelement <4 x float> undef, float %a1, i32 0
52 %val = phi <4 x float> [ %v1, %ift ], [ %v0, %entry ]
53 %v2 = extractelement <4 x float> %val, i32 1
54 store float %v2, float addrspace(1)* %out, align 4
58 declare i32 @llvm.amdgcn.mbcnt.lo(i32, i32) #0
60 attributes #0 = { nounwind readnone }