1 ; RUN: llc -march=amdgcn -mcpu=tahiti -o - %s | FileCheck %s 2 ; Don't crash when the use of an undefined value is only detected by the 3 ; register coalescer because it is hidden with subregister insert/extract. 4 target triple="amdgcn--" 5 6 ; CHECK-LABEL: foobar: 7 ; CHECK: s_load_dwordx2 s[2:3], s[0:1], 0x9 8 ; CHECK-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0xb 9 ; CHECK-NEXT: v_mbcnt_lo_u32_b32_e64 10 ; CHECK-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0 11 ; CHECK-NEXT: s_waitcnt lgkmcnt(0) 12 ; CHECK-NEXT: v_mov_b32_e32 v1, s3 13 ; CHECK-NEXT: s_and_saveexec_b64 s[2:3], vcc 14 15 ; CHECK: BB0_1: 16 ; CHECK-NEXT: ; kill: def $vgpr0_vgpr1 killed $sgpr2_sgpr3 killed $exec 17 ; CHECK-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3 18 19 ; CHECK: BB0_2: 20 ; CHECK: s_or_b64 exec, exec, s[2:3] 21 ; CHECK-NEXT: s_mov_b32 s3, 0xf000 22 ; CHECK-NEXT: s_mov_b32 s2, -1 23 ; CHECK-NEXT: buffer_store_dword v1, off, s[0:3], 0 24 ; CHECK-NEXT: s_endpgm 25 define amdgpu_kernel void @foobar(float %a0, float %a1, float addrspace(1)* %out) nounwind { 26 entry: 27 %v0 = insertelement <4 x float> undef, float %a0, i32 0 28 %tid = call i32 @llvm.amdgcn.mbcnt.lo(i32 -1, i32 0) #0 29 %cnd = icmp eq i32 %tid, 0 30 br i1 %cnd, label %ift, label %ife 31 32 ift: 33 %v1 = insertelement <4 x float> undef, float %a1, i32 0 34 br label %ife 35 36 ife: 37 %val = phi <4 x float> [ %v1, %ift ], [ %v0, %entry ] 38 %v2 = extractelement <4 x float> %val, i32 1 39 store float %v2, float addrspace(1)* %out, align 4 40 ret void 41 } 42 43 declare i32 @llvm.amdgcn.mbcnt.lo(i32, i32) #0 44 45 attributes #0 = { nounwind readnone } 46
