Mercurial > hg > Members > tobaru > cbc > CbC_llvm

comparison test/CodeGen/AMDGPU/srem.ll @ 121:803732b1fca8

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
LLVM 5.0
author kono
date Fri, 27 Oct 2017 17:07:41 +0900
parents afa8332a0e37
children
comparison
equal deleted inserted replaced
120:1172e4bd9c6f 121:803732b1fca8
1 ; RUN: llc -march=amdgcn -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 1 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
2 ; RUN: llc -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 2 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
3 ; RUN: llc -march=r600 -mcpu=redwood < %s 3 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=r600 -mcpu=redwood < %s
4 4
5 define void @srem_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 5 define amdgpu_kernel void @srem_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
6 %den_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1 6 %den_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1
7 %num = load i32, i32 addrspace(1) * %in 7 %num = load i32, i32 addrspace(1) * %in
8 %den = load i32, i32 addrspace(1) * %den_ptr 8 %den = load i32, i32 addrspace(1) * %den_ptr
9 %result = srem i32 %num, %den 9 %result = srem i32 %num, %den
10 store i32 %result, i32 addrspace(1)* %out 10 store i32 %result, i32 addrspace(1)* %out
11 ret void 11 ret void
12 } 12 }
13 13
14 define void @srem_i32_4(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 14 define amdgpu_kernel void @srem_i32_4(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
15 %num = load i32, i32 addrspace(1) * %in 15 %num = load i32, i32 addrspace(1) * %in
16 %result = srem i32 %num, 4 16 %result = srem i32 %num, 4
17 store i32 %result, i32 addrspace(1)* %out 17 store i32 %result, i32 addrspace(1)* %out
18 ret void 18 ret void
19 } 19 }
20 20
21 ; FUNC-LABEL: {{^}}srem_i32_7: 21 ; FUNC-LABEL: {{^}}srem_i32_7:
22 ; SI: v_mov_b32_e32 [[MAGIC:v[0-9]+]], 0x92492493 22 ; SI: v_mov_b32_e32 [[MAGIC:v[0-9]+]], 0x92492493
23 ; SI: v_mul_hi_i32 {{v[0-9]+}}, [[MAGIC]], 23 ; SI: v_mul_hi_i32 {{v[0-9]+}}, {{v[0-9]+}}, [[MAGIC]]
24 ; SI: v_mul_lo_i32 24 ; SI: v_mul_lo_i32
25 ; SI: v_sub_i32 25 ; SI: v_sub_i32
26 ; SI: s_endpgm 26 ; SI: s_endpgm
27 define void @srem_i32_7(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 27 define amdgpu_kernel void @srem_i32_7(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
28 %num = load i32, i32 addrspace(1) * %in 28 %num = load i32, i32 addrspace(1) * %in
29 %result = srem i32 %num, 7 29 %result = srem i32 %num, 7
30 store i32 %result, i32 addrspace(1)* %out 30 store i32 %result, i32 addrspace(1)* %out
31 ret void 31 ret void
32 } 32 }
33 33
34 define void @srem_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) { 34 define amdgpu_kernel void @srem_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
35 %den_ptr = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 1 35 %den_ptr = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 1
36 %num = load <2 x i32>, <2 x i32> addrspace(1) * %in 36 %num = load <2 x i32>, <2 x i32> addrspace(1) * %in
37 %den = load <2 x i32>, <2 x i32> addrspace(1) * %den_ptr 37 %den = load <2 x i32>, <2 x i32> addrspace(1) * %den_ptr
38 %result = srem <2 x i32> %num, %den 38 %result = srem <2 x i32> %num, %den
39 store <2 x i32> %result, <2 x i32> addrspace(1)* %out 39 store <2 x i32> %result, <2 x i32> addrspace(1)* %out
40 ret void 40 ret void
41 } 41 }
42 42
43 define void @srem_v2i32_4(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) { 43 define amdgpu_kernel void @srem_v2i32_4(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
44 %num = load <2 x i32>, <2 x i32> addrspace(1) * %in 44 %num = load <2 x i32>, <2 x i32> addrspace(1) * %in
45 %result = srem <2 x i32> %num, <i32 4, i32 4> 45 %result = srem <2 x i32> %num, <i32 4, i32 4>
46 store <2 x i32> %result, <2 x i32> addrspace(1)* %out 46 store <2 x i32> %result, <2 x i32> addrspace(1)* %out
47 ret void 47 ret void
48 } 48 }
49 49
50 define void @srem_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) { 50 define amdgpu_kernel void @srem_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
51 %den_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1 51 %den_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1
52 %num = load <4 x i32>, <4 x i32> addrspace(1) * %in 52 %num = load <4 x i32>, <4 x i32> addrspace(1) * %in
53 %den = load <4 x i32>, <4 x i32> addrspace(1) * %den_ptr 53 %den = load <4 x i32>, <4 x i32> addrspace(1) * %den_ptr
54 %result = srem <4 x i32> %num, %den 54 %result = srem <4 x i32> %num, %den
55 store <4 x i32> %result, <4 x i32> addrspace(1)* %out 55 store <4 x i32> %result, <4 x i32> addrspace(1)* %out
56 ret void 56 ret void
57 } 57 }
58 58
59 define void @srem_v4i32_4(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) { 59 define amdgpu_kernel void @srem_v4i32_4(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
60 %num = load <4 x i32>, <4 x i32> addrspace(1) * %in 60 %num = load <4 x i32>, <4 x i32> addrspace(1) * %in
61 %result = srem <4 x i32> %num, <i32 4, i32 4, i32 4, i32 4> 61 %result = srem <4 x i32> %num, <i32 4, i32 4, i32 4, i32 4>
62 store <4 x i32> %result, <4 x i32> addrspace(1)* %out 62 store <4 x i32> %result, <4 x i32> addrspace(1)* %out
63 ret void 63 ret void
64 } 64 }
65 65
66 define void @srem_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %in) { 66 define amdgpu_kernel void @srem_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %in) {
67 %den_ptr = getelementptr i64, i64 addrspace(1)* %in, i64 1 67 %den_ptr = getelementptr i64, i64 addrspace(1)* %in, i64 1
68 %num = load i64, i64 addrspace(1) * %in 68 %num = load i64, i64 addrspace(1) * %in
69 %den = load i64, i64 addrspace(1) * %den_ptr 69 %den = load i64, i64 addrspace(1) * %den_ptr
70 %result = srem i64 %num, %den 70 %result = srem i64 %num, %den
71 store i64 %result, i64 addrspace(1)* %out 71 store i64 %result, i64 addrspace(1)* %out
72 ret void 72 ret void
73 } 73 }
74 74
75 define void @srem_i64_4(i64 addrspace(1)* %out, i64 addrspace(1)* %in) { 75 define amdgpu_kernel void @srem_i64_4(i64 addrspace(1)* %out, i64 addrspace(1)* %in) {
76 %num = load i64, i64 addrspace(1) * %in 76 %num = load i64, i64 addrspace(1) * %in
77 %result = srem i64 %num, 4 77 %result = srem i64 %num, 4
78 store i64 %result, i64 addrspace(1)* %out 78 store i64 %result, i64 addrspace(1)* %out
79 ret void 79 ret void
80 } 80 }
81 81
82 define void @srem_v2i64(<2 x i64> addrspace(1)* %out, <2 x i64> addrspace(1)* %in) { 82 define amdgpu_kernel void @srem_v2i64(<2 x i64> addrspace(1)* %out, <2 x i64> addrspace(1)* %in) {
83 %den_ptr = getelementptr <2 x i64>, <2 x i64> addrspace(1)* %in, i64 1 83 %den_ptr = getelementptr <2 x i64>, <2 x i64> addrspace(1)* %in, i64 1
84 %num = load <2 x i64>, <2 x i64> addrspace(1) * %in 84 %num = load <2 x i64>, <2 x i64> addrspace(1) * %in
85 %den = load <2 x i64>, <2 x i64> addrspace(1) * %den_ptr 85 %den = load <2 x i64>, <2 x i64> addrspace(1) * %den_ptr
86 %result = srem <2 x i64> %num, %den 86 %result = srem <2 x i64> %num, %den
87 store <2 x i64> %result, <2 x i64> addrspace(1)* %out 87 store <2 x i64> %result, <2 x i64> addrspace(1)* %out
88 ret void 88 ret void
89 } 89 }
90 90
91 define void @srem_v2i64_4(<2 x i64> addrspace(1)* %out, <2 x i64> addrspace(1)* %in) { 91 define amdgpu_kernel void @srem_v2i64_4(<2 x i64> addrspace(1)* %out, <2 x i64> addrspace(1)* %in) {
92 %num = load <2 x i64>, <2 x i64> addrspace(1) * %in 92 %num = load <2 x i64>, <2 x i64> addrspace(1) * %in
93 %result = srem <2 x i64> %num, <i64 4, i64 4> 93 %result = srem <2 x i64> %num, <i64 4, i64 4>
94 store <2 x i64> %result, <2 x i64> addrspace(1)* %out 94 store <2 x i64> %result, <2 x i64> addrspace(1)* %out
95 ret void 95 ret void
96 } 96 }
97 97
98 define void @srem_v4i64(<4 x i64> addrspace(1)* %out, <4 x i64> addrspace(1)* %in) { 98 define amdgpu_kernel void @srem_v4i64(<4 x i64> addrspace(1)* %out, <4 x i64> addrspace(1)* %in) {
99 %den_ptr = getelementptr <4 x i64>, <4 x i64> addrspace(1)* %in, i64 1 99 %den_ptr = getelementptr <4 x i64>, <4 x i64> addrspace(1)* %in, i64 1
100 %num = load <4 x i64>, <4 x i64> addrspace(1) * %in 100 %num = load <4 x i64>, <4 x i64> addrspace(1) * %in
101 %den = load <4 x i64>, <4 x i64> addrspace(1) * %den_ptr 101 %den = load <4 x i64>, <4 x i64> addrspace(1) * %den_ptr
102 %result = srem <4 x i64> %num, %den 102 %result = srem <4 x i64> %num, %den
103 store <4 x i64> %result, <4 x i64> addrspace(1)* %out 103 store <4 x i64> %result, <4 x i64> addrspace(1)* %out
104 ret void 104 ret void
105 } 105 }
106 106
107 define void @srem_v4i64_4(<4 x i64> addrspace(1)* %out, <4 x i64> addrspace(1)* %in) { 107 define amdgpu_kernel void @srem_v4i64_4(<4 x i64> addrspace(1)* %out, <4 x i64> addrspace(1)* %in) {
108 %num = load <4 x i64>, <4 x i64> addrspace(1) * %in 108 %num = load <4 x i64>, <4 x i64> addrspace(1) * %in
109 %result = srem <4 x i64> %num, <i64 4, i64 4, i64 4, i64 4> 109 %result = srem <4 x i64> %num, <i64 4, i64 4, i64 4, i64 4>
110 store <4 x i64> %result, <4 x i64> addrspace(1)* %out 110 store <4 x i64> %result, <4 x i64> addrspace(1)* %out
111 ret void 111 ret void
112 } 112 }