1; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
2; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=bonaire < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s
3; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s
4
5declare double @llvm.trunc.f64(double) nounwind readnone
6declare <2 x double> @llvm.trunc.v2f64(<2 x double>) nounwind readnone
7declare <3 x double> @llvm.trunc.v3f64(<3 x double>) nounwind readnone
8declare <4 x double> @llvm.trunc.v4f64(<4 x double>) nounwind readnone
9declare <8 x double> @llvm.trunc.v8f64(<8 x double>) nounwind readnone
10declare <16 x double> @llvm.trunc.v16f64(<16 x double>) nounwind readnone
11
12; FUNC-LABEL: {{^}}v_ftrunc_f64:
13; CI: v_trunc_f64
14; SI: v_bfe_u32 {{v[0-9]+}}, {{v[0-9]+}}, 20, 11
15; SI: s_endpgm
16define amdgpu_kernel void @v_ftrunc_f64(double addrspace(1)* %out, double addrspace(1)* %in) {
17  %x = load double, double addrspace(1)* %in, align 8
18  %y = call double @llvm.trunc.f64(double %x) nounwind readnone
19  store double %y, double addrspace(1)* %out, align 8
20  ret void
21}
22
23; FUNC-LABEL: {{^}}ftrunc_f64:
24; CI: v_trunc_f64_e32
25
26; SI: s_bfe_u32 [[SEXP:s[0-9]+]], {{s[0-9]+}}, 0xb0014
27; SI-DAG: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80000000
28; SI-DAG: s_add_i32 [[SEXP1:s[0-9]+]], [[SEXP]], 0xfffffc01
29; SI-DAG: s_lshr_b64 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], [[SEXP1]]
30; SI-DAG: s_not_b64
31; SI-DAG: s_and_b64
32; SI-DAG: cmp_gt_i32
33; SI-DAG: cndmask_b32
34; SI-DAG: cndmask_b32
35; SI-DAG: cmp_lt_i32
36; SI-DAG: cndmask_b32
37; SI-DAG: cndmask_b32
38; SI: s_endpgm
39define amdgpu_kernel void @ftrunc_f64(double addrspace(1)* %out, double %x) {
40  %y = call double @llvm.trunc.f64(double %x) nounwind readnone
41  store double %y, double addrspace(1)* %out
42  ret void
43}
44
45; FUNC-LABEL: {{^}}ftrunc_v2f64:
46; CI: v_trunc_f64_e32
47; CI: v_trunc_f64_e32
48define amdgpu_kernel void @ftrunc_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) {
49  %y = call <2 x double> @llvm.trunc.v2f64(<2 x double> %x) nounwind readnone
50  store <2 x double> %y, <2 x double> addrspace(1)* %out
51  ret void
52}
53
54; FIXME-FUNC-LABEL: {{^}}ftrunc_v3f64:
55; FIXME-CI: v_trunc_f64_e32
56; FIXME-CI: v_trunc_f64_e32
57; FIXME-CI: v_trunc_f64_e32
58; define amdgpu_kernel void @ftrunc_v3f64(<3 x double> addrspace(1)* %out, <3 x double> %x) {
59;   %y = call <3 x double> @llvm.trunc.v3f64(<3 x double> %x) nounwind readnone
60;   store <3 x double> %y, <3 x double> addrspace(1)* %out
61;   ret void
62; }
63
64; FUNC-LABEL: {{^}}ftrunc_v4f64:
65; CI: v_trunc_f64_e32
66; CI: v_trunc_f64_e32
67; CI: v_trunc_f64_e32
68; CI: v_trunc_f64_e32
69define amdgpu_kernel void @ftrunc_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %x) {
70  %y = call <4 x double> @llvm.trunc.v4f64(<4 x double> %x) nounwind readnone
71  store <4 x double> %y, <4 x double> addrspace(1)* %out
72  ret void
73}
74
75; FUNC-LABEL: {{^}}ftrunc_v8f64:
76; CI: v_trunc_f64_e32
77; CI: v_trunc_f64_e32
78; CI: v_trunc_f64_e32
79; CI: v_trunc_f64_e32
80; CI: v_trunc_f64_e32
81; CI: v_trunc_f64_e32
82; CI: v_trunc_f64_e32
83; CI: v_trunc_f64_e32
84define amdgpu_kernel void @ftrunc_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %x) {
85  %y = call <8 x double> @llvm.trunc.v8f64(<8 x double> %x) nounwind readnone
86  store <8 x double> %y, <8 x double> addrspace(1)* %out
87  ret void
88}
89
90; FUNC-LABEL: {{^}}ftrunc_v16f64:
91; CI: v_trunc_f64_e32
92; CI: v_trunc_f64_e32
93; CI: v_trunc_f64_e32
94; CI: v_trunc_f64_e32
95; CI: v_trunc_f64_e32
96; CI: v_trunc_f64_e32
97; CI: v_trunc_f64_e32
98; CI: v_trunc_f64_e32
99; CI: v_trunc_f64_e32
100; CI: v_trunc_f64_e32
101; CI: v_trunc_f64_e32
102; CI: v_trunc_f64_e32
103; CI: v_trunc_f64_e32
104; CI: v_trunc_f64_e32
105; CI: v_trunc_f64_e32
106; CI: v_trunc_f64_e32
107define amdgpu_kernel void @ftrunc_v16f64(<16 x double> addrspace(1)* %out, <16 x double> %x) {
108  %y = call <16 x double> @llvm.trunc.v16f64(<16 x double> %x) nounwind readnone
109  store <16 x double> %y, <16 x double> addrspace(1)* %out
110  ret void
111}
112