1; RUN: llc -amdgpu-scalarize-global-loads=false --verify-machineinstrs -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=SI,GFX9 -check-prefix=FUNC %s
2; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
3; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
4; RUN: llc -amdgpu-scalarize-global-loads=false -march=r600 -mcpu=redwood -verify-machineinstrs < %s
5
6; FUNC-LABEL: {{^}}srem_i16_7:
7; GFX9: s_movk_i32 {{s[0-9]+}}, 0x4925
8; GFX9: v_mul_lo_u32
9define amdgpu_kernel void @srem_i16_7(i16 addrspace(1)* %out, i16 addrspace(1)* %in) {
10  %num = load i16, i16 addrspace(1) * %in
11  %result = srem i16 %num, 7
12  store i16 %result, i16 addrspace(1)* %out
13  ret void
14}
15
16define amdgpu_kernel void @srem_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
17  %den_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1
18  %num = load i32, i32 addrspace(1) * %in
19  %den = load i32, i32 addrspace(1) * %den_ptr
20  %result = srem i32 %num, %den
21  store i32 %result, i32 addrspace(1)* %out
22  ret void
23}
24
25define amdgpu_kernel void @srem_i32_4(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
26  %num = load i32, i32 addrspace(1) * %in
27  %result = srem i32 %num, 4
28  store i32 %result, i32 addrspace(1)* %out
29  ret void
30}
31
32; FIXME: uniform i16 srem should not use VALU instructions
33; FUNC-LABEL: {{^}}srem_i32_7:
34; SI: s_mov_b32 [[MAGIC:s[0-9]+]], 0x92492493
35; SI: v_mul_hi_i32 {{v[0-9]+}}, {{v[0-9]+}}, [[MAGIC]]
36; SI: v_mul_lo_u32
37; SI: v_sub_{{[iu]}}32
38; SI: s_endpgm
39define amdgpu_kernel void @srem_i32_7(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
40  %num = load i32, i32 addrspace(1) * %in
41  %result = srem i32 %num, 7
42  store i32 %result, i32 addrspace(1)* %out
43  ret void
44}
45
46define amdgpu_kernel void @srem_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
47  %den_ptr = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 1
48  %num = load <2 x i32>, <2 x i32> addrspace(1) * %in
49  %den = load <2 x i32>, <2 x i32> addrspace(1) * %den_ptr
50  %result = srem <2 x i32> %num, %den
51  store <2 x i32> %result, <2 x i32> addrspace(1)* %out
52  ret void
53}
54
55define amdgpu_kernel void @srem_v2i32_4(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
56  %num = load <2 x i32>, <2 x i32> addrspace(1) * %in
57  %result = srem <2 x i32> %num, <i32 4, i32 4>
58  store <2 x i32> %result, <2 x i32> addrspace(1)* %out
59  ret void
60}
61
62define amdgpu_kernel void @srem_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
63  %den_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1
64  %num = load <4 x i32>, <4 x i32> addrspace(1) * %in
65  %den = load <4 x i32>, <4 x i32> addrspace(1) * %den_ptr
66  %result = srem <4 x i32> %num, %den
67  store <4 x i32> %result, <4 x i32> addrspace(1)* %out
68  ret void
69}
70
71define amdgpu_kernel void @srem_v4i32_4(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
72  %num = load <4 x i32>, <4 x i32> addrspace(1) * %in
73  %result = srem <4 x i32> %num, <i32 4, i32 4, i32 4, i32 4>
74  store <4 x i32> %result, <4 x i32> addrspace(1)* %out
75  ret void
76}
77
78define amdgpu_kernel void @srem_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %in) {
79  %den_ptr = getelementptr i64, i64 addrspace(1)* %in, i64 1
80  %num = load i64, i64 addrspace(1) * %in
81  %den = load i64, i64 addrspace(1) * %den_ptr
82  %result = srem i64 %num, %den
83  store i64 %result, i64 addrspace(1)* %out
84  ret void
85}
86
87define amdgpu_kernel void @srem_i64_4(i64 addrspace(1)* %out, i64 addrspace(1)* %in) {
88  %num = load i64, i64 addrspace(1) * %in
89  %result = srem i64 %num, 4
90  store i64 %result, i64 addrspace(1)* %out
91  ret void
92}
93
94define amdgpu_kernel void @srem_v2i64(<2 x i64> addrspace(1)* %out, <2 x i64> addrspace(1)* %in) {
95  %den_ptr = getelementptr <2 x i64>, <2 x i64> addrspace(1)* %in, i64 1
96  %num = load <2 x i64>, <2 x i64> addrspace(1) * %in
97  %den = load <2 x i64>, <2 x i64> addrspace(1) * %den_ptr
98  %result = srem <2 x i64> %num, %den
99  store <2 x i64> %result, <2 x i64> addrspace(1)* %out
100  ret void
101}
102
103define amdgpu_kernel void @srem_v2i64_4(<2 x i64> addrspace(1)* %out, <2 x i64> addrspace(1)* %in) {
104  %num = load <2 x i64>, <2 x i64> addrspace(1) * %in
105  %result = srem <2 x i64> %num, <i64 4, i64 4>
106  store <2 x i64> %result, <2 x i64> addrspace(1)* %out
107  ret void
108}
109
110define amdgpu_kernel void @srem_v4i64(<4 x i64> addrspace(1)* %out, <4 x i64> addrspace(1)* %in) {
111  %den_ptr = getelementptr <4 x i64>, <4 x i64> addrspace(1)* %in, i64 1
112  %num = load <4 x i64>, <4 x i64> addrspace(1) * %in
113  %den = load <4 x i64>, <4 x i64> addrspace(1) * %den_ptr
114  %result = srem <4 x i64> %num, %den
115  store <4 x i64> %result, <4 x i64> addrspace(1)* %out
116  ret void
117}
118
119define amdgpu_kernel void @srem_v4i64_4(<4 x i64> addrspace(1)* %out, <4 x i64> addrspace(1)* %in) {
120  %num = load <4 x i64>, <4 x i64> addrspace(1) * %in
121  %result = srem <4 x i64> %num, <i64 4, i64 4, i64 4, i64 4>
122  store <4 x i64> %result, <4 x i64> addrspace(1)* %out
123  ret void
124}
125