1; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
2
3declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
4
5declare double @fabs(double) readnone
6declare double @llvm.fabs.f64(double) readnone
7declare <2 x double> @llvm.fabs.v2f64(<2 x double>) readnone
8declare <4 x double> @llvm.fabs.v4f64(<4 x double>) readnone
9
10; FUNC-LABEL: {{^}}v_fabs_f64:
11; SI: v_and_b32
12; SI: s_endpgm
13define amdgpu_kernel void @v_fabs_f64(double addrspace(1)* %out, double addrspace(1)* %in) {
14  %tid = call i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
15  %tidext = sext i32 %tid to i64
16  %gep = getelementptr double, double addrspace(1)* %in, i64 %tidext
17  %val = load double, double addrspace(1)* %gep, align 8
18  %fabs = call double @llvm.fabs.f64(double %val)
19  store double %fabs, double addrspace(1)* %out
20  ret void
21}
22
23; FUNC-LABEL: {{^}}fabs_f64:
24; SI: s_bitset0_b32
25; SI: s_endpgm
26define amdgpu_kernel void @fabs_f64(double addrspace(1)* %out, double %in) {
27  %fabs = call double @llvm.fabs.f64(double %in)
28  store double %fabs, double addrspace(1)* %out
29  ret void
30}
31
32; FUNC-LABEL: {{^}}fabs_v2f64:
33; SI: s_bitset0_b32
34; SI: s_bitset0_b32
35; SI: s_endpgm
36define amdgpu_kernel void @fabs_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %in) {
37  %fabs = call <2 x double> @llvm.fabs.v2f64(<2 x double> %in)
38  store <2 x double> %fabs, <2 x double> addrspace(1)* %out
39  ret void
40}
41
42; FUNC-LABEL: {{^}}fabs_v4f64:
43; SI: s_bitset0_b32
44; SI: s_bitset0_b32
45; SI: s_bitset0_b32
46; SI: s_bitset0_b32
47; SI: s_endpgm
48define amdgpu_kernel void @fabs_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %in) {
49  %fabs = call <4 x double> @llvm.fabs.v4f64(<4 x double> %in)
50  store <4 x double> %fabs, <4 x double> addrspace(1)* %out
51  ret void
52}
53
54; SI-LABEL: {{^}}fabs_fold_f64:
55; SI: s_load_dwordx2 [[ABS_VALUE:s\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, 0x13
56; SI-NOT: and
57; SI: v_mul_f64 {{v\[[0-9]+:[0-9]+\]}}, |[[ABS_VALUE]]|, {{v\[[0-9]+:[0-9]+\]}}
58; SI: s_endpgm
59define amdgpu_kernel void @fabs_fold_f64(double addrspace(1)* %out, [8 x i32], double %in0, [8 x i32], double %in1) {
60  %fabs = call double @llvm.fabs.f64(double %in0)
61  %fmul = fmul double %fabs, %in1
62  store double %fmul, double addrspace(1)* %out
63  ret void
64}
65
66; SI-LABEL: {{^}}fabs_fn_fold_f64:
67; SI: s_load_dwordx2 [[ABS_VALUE:s\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, 0x13
68; SI-NOT: and
69; SI: v_mul_f64 {{v\[[0-9]+:[0-9]+\]}}, |[[ABS_VALUE]]|, {{v\[[0-9]+:[0-9]+\]}}
70; SI: s_endpgm
71define amdgpu_kernel void @fabs_fn_fold_f64(double addrspace(1)* %out, [8 x i32], double %in0, [8 x i32], double %in1) {
72  %fabs = call double @fabs(double %in0)
73  %fmul = fmul double %fabs, %in1
74  store double %fmul, double addrspace(1)* %out
75  ret void
76}
77
78; FUNC-LABEL: {{^}}fabs_free_f64:
79; SI: s_bitset0_b32
80; SI: s_endpgm
81define amdgpu_kernel void @fabs_free_f64(double addrspace(1)* %out, i64 %in) {
82  %bc= bitcast i64 %in to double
83  %fabs = call double @llvm.fabs.f64(double %bc)
84  store double %fabs, double addrspace(1)* %out
85  ret void
86}
87
88; FUNC-LABEL: {{^}}fabs_fn_free_f64:
89; SI: s_bitset0_b32
90; SI: s_endpgm
91define amdgpu_kernel void @fabs_fn_free_f64(double addrspace(1)* %out, i64 %in) {
92  %bc= bitcast i64 %in to double
93  %fabs = call double @fabs(double %bc)
94  store double %fabs, double addrspace(1)* %out
95  ret void
96}
97