• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=R600-CHECK
2; RUN: llc < %s -march=r600 -mcpu=SI -verify-machineinstrs | FileCheck %s --check-prefix=SI-CHECK
3
4; DAGCombiner will transform:
5; (fabs (f32 bitcast (i32 a))) => (f32 bitcast (and (i32 a), 0x7FFFFFFF))
6; unless isFabsFree returns true
7
8; R600-CHECK-LABEL: @fabs_free
9; R600-CHECK-NOT: AND
10; R600-CHECK: |PV.{{[XYZW]}}|
11; SI-CHECK-LABEL: @fabs_free
12; SI-CHECK: V_AND_B32
13
14define void @fabs_free(float addrspace(1)* %out, i32 %in) {
15entry:
16  %0 = bitcast i32 %in to float
17  %1 = call float @fabs(float %0)
18  store float %1, float addrspace(1)* %out
19  ret void
20}
21
22; R600-CHECK-LABEL: @fabs_v2
23; R600-CHECK: |{{(PV|T[0-9])\.[XYZW]}}|
24; R600-CHECK: |{{(PV|T[0-9])\.[XYZW]}}|
25; SI-CHECK-LABEL: @fabs_v2
26; SI-CHECK: V_AND_B32
27; SI-CHECK: V_AND_B32
28define void @fabs_v2(<2 x float> addrspace(1)* %out, <2 x float> %in) {
29entry:
30  %0 = call <2 x float> @llvm.fabs.v2f32(<2 x float> %in)
31  store <2 x float> %0, <2 x float> addrspace(1)* %out
32  ret void
33}
34
35; R600-CHECK-LABEL: @fabs_v4
36; R600-CHECK: |{{(PV|T[0-9])\.[XYZW]}}|
37; R600-CHECK: |{{(PV|T[0-9])\.[XYZW]}}|
38; R600-CHECK: |{{(PV|T[0-9])\.[XYZW]}}|
39; R600-CHECK: |{{(PV|T[0-9])\.[XYZW]}}|
40; SI-CHECK-LABEL: @fabs_v4
41; SI-CHECK: V_AND_B32
42; SI-CHECK: V_AND_B32
43; SI-CHECK: V_AND_B32
44; SI-CHECK: V_AND_B32
45define void @fabs_v4(<4 x float> addrspace(1)* %out, <4 x float> %in) {
46entry:
47  %0 = call <4 x float> @llvm.fabs.v4f32(<4 x float> %in)
48  store <4 x float> %0, <4 x float> addrspace(1)* %out
49  ret void
50}
51
52; SI-CHECK-LABEL: @fabs_fold
53; SI-CHECK-NOT: V_AND_B32_e32
54; SI-CHECK: V_MUL_F32_e64 v{{[0-9]+}}, s{{[0-9]+}}, |v{{[0-9]+}}|
55define void @fabs_fold(float addrspace(1)* %out, float %in0, float %in1) {
56entry:
57  %0 = call float @fabs(float %in0)
58  %1 = fmul float %0, %in1
59  store float %1, float addrspace(1)* %out
60  ret void
61}
62
63declare float @fabs(float ) readnone
64declare <2 x float> @llvm.fabs.v2f32(<2 x float> ) readnone
65declare <4 x float> @llvm.fabs.v4f32(<4 x float> ) readnone
66