|
1 |
| -; RUN: llc -mtriple=amdgcn < %s |
| 1 | +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 |
| 2 | +; RUN: llc -mtriple=amdgcn < %s | FileCheck %s |
2 | 3 |
|
3 | 4 | ; Tests for a bug in SelectionDAG::UpdateNodeOperands exposed by VectorLegalizer
|
4 | 5 | ; where divergence information is not updated.
|
5 | 6 |
|
6 | 7 | declare i32 @llvm.amdgcn.workitem.id.x()
|
7 | 8 |
|
8 | 9 | define amdgpu_kernel void @spam(ptr addrspace(1) noalias %arg) {
|
| 10 | +; CHECK-LABEL: spam: |
| 11 | +; CHECK: ; %bb.0: |
| 12 | +; CHECK-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x9 |
| 13 | +; CHECK-NEXT: v_lshlrev_b32_e32 v4, 3, v0 |
| 14 | +; CHECK-NEXT: v_mov_b32_e32 v5, 0 |
| 15 | +; CHECK-NEXT: s_mov_b32 s3, 0xf000 |
| 16 | +; CHECK-NEXT: s_mov_b32 s2, 0 |
| 17 | +; CHECK-NEXT: v_mov_b32_e32 v1, 0x7ff80000 |
| 18 | +; CHECK-NEXT: v_mov_b32_e32 v0, v5 |
| 19 | +; CHECK-NEXT: s_waitcnt lgkmcnt(0) |
| 20 | +; CHECK-NEXT: buffer_store_dwordx4 v[0:3], v[4:5], s[0:3], 0 addr64 offset:16 |
| 21 | +; CHECK-NEXT: s_waitcnt expcnt(0) |
| 22 | +; CHECK-NEXT: v_mov_b32_e32 v2, v5 |
| 23 | +; CHECK-NEXT: v_mov_b32_e32 v3, v5 |
| 24 | +; CHECK-NEXT: buffer_store_dwordx4 v[0:3], v[4:5], s[0:3], 0 addr64 offset:48 |
| 25 | +; CHECK-NEXT: s_endpgm |
9 | 26 | %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
|
10 | 27 | %tmp1 = zext i32 %tmp to i64
|
11 | 28 | %tmp2 = getelementptr inbounds double, ptr addrspace(1) %arg, i64 %tmp1
|
12 | 29 | %tmp3 = load double, ptr addrspace(1) %tmp2, align 8
|
13 |
| - %tmp4 = fadd double undef, 0.000000e+00 |
| 30 | + %tmp4 = fadd double 0x7FF8000000000000, 0.000000e+00 |
14 | 31 | %tmp5 = insertelement <2 x double> poison, double %tmp4, i64 0
|
15 | 32 | %tmp6 = insertelement <2 x double> %tmp5, double %tmp3, i64 1
|
16 | 33 | %tmp7 = insertelement <2 x double> %tmp6, double 0.000000e+00, i64 1
|
17 |
| - %tmp8 = fadd <2 x double> zeroinitializer, undef |
| 34 | + %tmp8 = fadd <2 x double> zeroinitializer, splat (double 0x7FF8000000000000) |
18 | 35 | %tmp9 = fadd <2 x double> %tmp7, zeroinitializer
|
19 | 36 | %tmp10 = extractelement <2 x double> %tmp8, i64 0
|
20 | 37 | %tmp11 = getelementptr inbounds double, ptr addrspace(1) %tmp2, i64 2
|
21 | 38 | store double %tmp10, ptr addrspace(1) %tmp11, align 8
|
22 | 39 | %tmp12 = getelementptr inbounds double, ptr addrspace(1) %tmp2, i64 3
|
23 |
| - store double undef, ptr addrspace(1) %tmp12, align 8 |
| 40 | + store double poison, ptr addrspace(1) %tmp12, align 8 |
24 | 41 | %tmp13 = extractelement <2 x double> %tmp9, i64 0
|
25 | 42 | %tmp14 = getelementptr inbounds double, ptr addrspace(1) %tmp2, i64 6
|
26 | 43 | store double %tmp13, ptr addrspace(1) %tmp14, align 8
|
|
0 commit comments