1; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
2; RUN: llc -verify-machineinstrs -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
3
4; Test expansion of scalar selects on vectors.
5; Evergreen not enabled since it seems to be having problems with doubles.
6
7
8; FUNC-LABEL: {{^}}select_v4i8:
9; SI: v_cndmask_b32_e32
10; SI: v_cndmask_b32_e32
11; SI: v_cndmask_b32_e32
12; SI: v_cndmask_b32_e32
13define void @select_v4i8(<4 x i8> addrspace(1)* %out, <4 x i8> %a, <4 x i8> %b, i8 %c) nounwind {
14  %cmp = icmp eq i8 %c, 0
15  %select = select i1 %cmp, <4 x i8> %a, <4 x i8> %b
16  store <4 x i8> %select, <4 x i8> addrspace(1)* %out, align 4
17  ret void
18}
19
20; FUNC-LABEL: {{^}}select_v4i16:
21; SI: v_cndmask_b32_e32
22; SI: v_cndmask_b32_e32
23; SI: v_cndmask_b32_e32
24; SI: v_cndmask_b32_e32
25define void @select_v4i16(<4 x i16> addrspace(1)* %out, <4 x i16> %a, <4 x i16> %b, i32 %c) nounwind {
26  %cmp = icmp eq i32 %c, 0
27  %select = select i1 %cmp, <4 x i16> %a, <4 x i16> %b
28  store <4 x i16> %select, <4 x i16> addrspace(1)* %out, align 4
29  ret void
30}
31
32; FUNC-LABEL: {{^}}select_v2i32:
33; SI: v_cndmask_b32_e32
34; SI: v_cndmask_b32_e32
35; SI: buffer_store_dwordx2
36define void @select_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> %a, <2 x i32> %b, i32 %c) nounwind {
37  %cmp = icmp eq i32 %c, 0
38  %select = select i1 %cmp, <2 x i32> %a, <2 x i32> %b
39  store <2 x i32> %select, <2 x i32> addrspace(1)* %out, align 8
40  ret void
41}
42
43; FUNC-LABEL: {{^}}select_v4i32:
44; SI: v_cndmask_b32_e32
45; SI: v_cndmask_b32_e32
46; SI: v_cndmask_b32_e32
47; SI: v_cndmask_b32_e32
48; SI: buffer_store_dwordx4
49define void @select_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> %a, <4 x i32> %b, i32 %c) nounwind {
50  %cmp = icmp eq i32 %c, 0
51  %select = select i1 %cmp, <4 x i32> %a, <4 x i32> %b
52  store <4 x i32> %select, <4 x i32> addrspace(1)* %out, align 16
53  ret void
54}
55
56; FUNC-LABEL: {{^}}select_v8i32:
57; SI: v_cndmask_b32_e32
58; SI: v_cndmask_b32_e32
59; SI: v_cndmask_b32_e32
60; SI: v_cndmask_b32_e32
61; SI: v_cndmask_b32_e32
62; SI: v_cndmask_b32_e32
63; SI: v_cndmask_b32_e32
64; SI: v_cndmask_b32_e32
65define void @select_v8i32(<8 x i32> addrspace(1)* %out, <8 x i32> %a, <8 x i32> %b, i32 %c) nounwind {
66  %cmp = icmp eq i32 %c, 0
67  %select = select i1 %cmp, <8 x i32> %a, <8 x i32> %b
68  store <8 x i32> %select, <8 x i32> addrspace(1)* %out, align 16
69  ret void
70}
71
72; FUNC-LABEL: {{^}}select_v2f32:
73; SI: buffer_store_dwordx2
74define void @select_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b, i32 %c) nounwind {
75  %cmp = icmp eq i32 %c, 0
76  %select = select i1 %cmp, <2 x float> %a, <2 x float> %b
77  store <2 x float> %select, <2 x float> addrspace(1)* %out, align 16
78  ret void
79}
80
81; FUNC-LABEL: {{^}}select_v4f32:
82; SI: buffer_store_dwordx4
83define void @select_v4f32(<4 x float> addrspace(1)* %out, <4 x float> %a, <4 x float> %b, i32 %c) nounwind {
84  %cmp = icmp eq i32 %c, 0
85  %select = select i1 %cmp, <4 x float> %a, <4 x float> %b
86  store <4 x float> %select, <4 x float> addrspace(1)* %out, align 16
87  ret void
88}
89
90; FUNC-LABEL: {{^}}select_v8f32:
91; SI: v_cndmask_b32_e32
92; SI: v_cndmask_b32_e32
93; SI: v_cndmask_b32_e32
94; SI: v_cndmask_b32_e32
95; SI: v_cndmask_b32_e32
96; SI: v_cndmask_b32_e32
97; SI: v_cndmask_b32_e32
98; SI: v_cndmask_b32_e32
99define void @select_v8f32(<8 x float> addrspace(1)* %out, <8 x float> %a, <8 x float> %b, i32 %c) nounwind {
100  %cmp = icmp eq i32 %c, 0
101  %select = select i1 %cmp, <8 x float> %a, <8 x float> %b
102  store <8 x float> %select, <8 x float> addrspace(1)* %out, align 16
103  ret void
104}
105
106; FUNC-LABEL: {{^}}select_v2f64:
107; SI: v_cndmask_b32_e32
108; SI: v_cndmask_b32_e32
109; SI: v_cndmask_b32_e32
110; SI: v_cndmask_b32_e32
111define void @select_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %a, <2 x double> %b, i32 %c) nounwind {
112  %cmp = icmp eq i32 %c, 0
113  %select = select i1 %cmp, <2 x double> %a, <2 x double> %b
114  store <2 x double> %select, <2 x double> addrspace(1)* %out, align 16
115  ret void
116}
117
118; FUNC-LABEL: {{^}}select_v4f64:
119; SI: v_cndmask_b32_e32
120; SI: v_cndmask_b32_e32
121; SI: v_cndmask_b32_e32
122; SI: v_cndmask_b32_e32
123; SI: v_cndmask_b32_e32
124; SI: v_cndmask_b32_e32
125; SI: v_cndmask_b32_e32
126; SI: v_cndmask_b32_e32
127define void @select_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %a, <4 x double> %b, i32 %c) nounwind {
128  %cmp = icmp eq i32 %c, 0
129  %select = select i1 %cmp, <4 x double> %a, <4 x double> %b
130  store <4 x double> %select, <4 x double> addrspace(1)* %out, align 16
131  ret void
132}
133
134; FUNC-LABEL: {{^}}select_v8f64:
135; SI: v_cndmask_b32_e32
136; SI: v_cndmask_b32_e32
137; SI: v_cndmask_b32_e32
138; SI: v_cndmask_b32_e32
139; SI: v_cndmask_b32_e32
140; SI: v_cndmask_b32_e32
141; SI: v_cndmask_b32_e32
142; SI: v_cndmask_b32_e32
143; SI: v_cndmask_b32_e32
144; SI: v_cndmask_b32_e32
145; SI: v_cndmask_b32_e32
146; SI: v_cndmask_b32_e32
147; SI: v_cndmask_b32_e32
148; SI: v_cndmask_b32_e32
149; SI: v_cndmask_b32_e32
150; SI: v_cndmask_b32_e32
151define void @select_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %a, <8 x double> %b, i32 %c) nounwind {
152  %cmp = icmp eq i32 %c, 0
153  %select = select i1 %cmp, <8 x double> %a, <8 x double> %b
154  store <8 x double> %select, <8 x double> addrspace(1)* %out, align 16
155  ret void
156}
157