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: v_and_b32
25; SI-NOT: v_and_b32
26; SI: s_endpgm
27define amdgpu_kernel void @fabs_f64(double addrspace(1)* %out, double %in) {
28  %fabs = call double @llvm.fabs.f64(double %in)
29  store double %fabs, double addrspace(1)* %out
30  ret void
31}
32
33; FUNC-LABEL: {{^}}fabs_v2f64:
34; SI: v_and_b32
35; SI: v_and_b32
36; SI: s_endpgm
37define amdgpu_kernel void @fabs_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %in) {
38  %fabs = call <2 x double> @llvm.fabs.v2f64(<2 x double> %in)
39  store <2 x double> %fabs, <2 x double> addrspace(1)* %out
40  ret void
41}
42
43; FUNC-LABEL: {{^}}fabs_v4f64:
44; SI: v_and_b32
45; SI: v_and_b32
46; SI: v_and_b32
47; SI: v_and_b32
48; SI: s_endpgm
49define amdgpu_kernel void @fabs_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %in) {
50  %fabs = call <4 x double> @llvm.fabs.v4f64(<4 x double> %in)
51  store <4 x double> %fabs, <4 x double> addrspace(1)* %out
52  ret void
53}
54
55; SI-LABEL: {{^}}fabs_fold_f64:
56; SI: s_load_dwordx2 [[ABS_VALUE:s\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, 0x13
57; SI-NOT: and
58; SI: v_mul_f64 {{v\[[0-9]+:[0-9]+\]}}, |[[ABS_VALUE]]|, {{v\[[0-9]+:[0-9]+\]}}
59; SI: s_endpgm
60define amdgpu_kernel void @fabs_fold_f64(double addrspace(1)* %out, [8 x i32], double %in0, [8 x i32], double %in1) {
61  %fabs = call double @llvm.fabs.f64(double %in0)
62  %fmul = fmul double %fabs, %in1
63  store double %fmul, double addrspace(1)* %out
64  ret void
65}
66
67; SI-LABEL: {{^}}fabs_fn_fold_f64:
68; SI: s_load_dwordx2 [[ABS_VALUE:s\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, 0x13
69; SI-NOT: and
70; SI: v_mul_f64 {{v\[[0-9]+:[0-9]+\]}}, |[[ABS_VALUE]]|, {{v\[[0-9]+:[0-9]+\]}}
71; SI: s_endpgm
72define amdgpu_kernel void @fabs_fn_fold_f64(double addrspace(1)* %out, [8 x i32], double %in0, [8 x i32], double %in1) {
73  %fabs = call double @fabs(double %in0)
74  %fmul = fmul double %fabs, %in1
75  store double %fmul, double addrspace(1)* %out
76  ret void
77}
78
79; FUNC-LABEL: {{^}}fabs_free_f64:
80; SI: v_and_b32
81; SI: s_endpgm
82define amdgpu_kernel void @fabs_free_f64(double addrspace(1)* %out, i64 %in) {
83  %bc= bitcast i64 %in to double
84  %fabs = call double @llvm.fabs.f64(double %bc)
85  store double %fabs, double addrspace(1)* %out
86  ret void
87}
88
89; FUNC-LABEL: {{^}}fabs_fn_free_f64:
90; SI: v_and_b32
91; SI: s_endpgm
92define amdgpu_kernel void @fabs_fn_free_f64(double addrspace(1)* %out, i64 %in) {
93  %bc= bitcast i64 %in to double
94  %fabs = call double @fabs(double %bc)
95  store double %fabs, double addrspace(1)* %out
96  ret void
97}
98