1 /* { dg-do compile } */
2 /* { dg-options "-march=rv32gcv -mabi=ilp32 -fno-tree-vectorize" } */
3
4 #include "riscv_vector.h"
5
6 void f (int8_t * restrict in, int8_t * restrict out, int n)
7 {
8 if (n == 0) {
9 vint8mf8_t v = *(vint8mf8_t*)(in + 500);
10 *(vint8mf8_t*)(out + 500) = v;
11 } else if (n == 1) {
12 vint8mf8_t v = *(vint8mf8_t*)(in + 600);
13 *(vint8mf8_t*)(out + 600) = v;
14 }
15
16 for (int i = 0 ; i < n * n; i++)
17 out[i] = out[i] + out[i];
18
19 for (int i = 0 ; i < n * n * n; i++)
20 out[i] = out[i] * out[i];
21
22 for (int i = 0 ; i < n * n * n * n; i++)
23 out[i] = out[i] * out[i];
24
25 for (int i = 0 ; i < n * n * n * n; i++) {
26 vint8mf8_t v = *(vint8mf8_t*)(in + 900 + i);
27 *(vint8mf8_t*)(out + 900 + i) = v;
28 }
29 }
30
31 void f2 (int8_t * restrict in, int8_t * restrict out, int n)
32 {
33 if (n == 0) {
34 vint8mf4_t v = *(vint8mf4_t*)(in + 500);
35 *(vint8mf4_t*)(out + 500) = v;
36 } else if (n == 1) {
37 vint8mf4_t v = *(vint8mf4_t*)(in + 600);
38 *(vint8mf4_t*)(out + 600) = v;
39 }
40
41 for (int i = 0 ; i < n * n; i++)
42 out[i] = out[i] + out[i];
43
44 for (int i = 0 ; i < n * n * n; i++)
45 out[i] = out[i] * out[i];
46
47 for (int i = 0 ; i < n * n * n * n; i++)
48 out[i] = out[i] * out[i];
49
50 for (int i = 0 ; i < n * n * n * n; i++) {
51 vint8mf4_t v = *(vint8mf4_t*)(in + 900 + i);
52 *(vint8mf4_t*)(out + 900 + i) = v;
53 }
54 }
55
56 void f3 (int8_t * restrict in, int8_t * restrict out, int n)
57 {
58 if (n == 0) {
59 vint8mf2_t v = *(vint8mf2_t*)(in + 500);
60 *(vint8mf2_t*)(out + 500) = v;
61 } else if (n == 1) {
62 vint8mf2_t v = *(vint8mf2_t*)(in + 600);
63 *(vint8mf2_t*)(out + 600) = v;
64 }
65
66 for (int i = 0 ; i < n * n; i++)
67 out[i] = out[i] + out[i];
68
69 for (int i = 0 ; i < n * n * n; i++)
70 out[i] = out[i] * out[i];
71
72 for (int i = 0 ; i < n * n * n * n; i++)
73 out[i] = out[i] * out[i];
74
75 for (int i = 0 ; i < n * n * n * n; i++) {
76 vint8mf2_t v = *(vint8mf2_t*)(in + 900 + i);
77 *(vint8mf2_t*)(out + 900 + i) = v;
78 }
79 }
80
81 void f4 (int8_t * restrict in, int8_t * restrict out, int n)
82 {
83 if (n == 0) {
84 vint16mf4_t v = *(vint16mf4_t*)(in + 500);
85 *(vint16mf4_t*)(out + 500) = v;
86 } else if (n == 1) {
87 vint16mf4_t v = *(vint16mf4_t*)(in + 600);
88 *(vint16mf4_t*)(out + 600) = v;
89 }
90
91 for (int i = 0 ; i < n * n; i++)
92 out[i] = out[i] + out[i];
93
94 for (int i = 0 ; i < n * n * n; i++)
95 out[i] = out[i] * out[i];
96
97 for (int i = 0 ; i < n * n * n * n; i++)
98 out[i] = out[i] * out[i];
99
100 for (int i = 0 ; i < n * n * n * n; i++) {
101 vint16mf4_t v = *(vint16mf4_t*)(in + 900 + i);
102 *(vint16mf4_t*)(out + 900 + i) = v;
103 }
104 }
105
106 void f5 (int8_t * restrict in, int8_t * restrict out, int n)
107 {
108 if (n == 0) {
109 vint16mf2_t v = *(vint16mf2_t*)(in + 500);
110 *(vint16mf2_t*)(out + 500) = v;
111 } else if (n == 1) {
112 vint16mf2_t v = *(vint16mf2_t*)(in + 600);
113 *(vint16mf2_t*)(out + 600) = v;
114 }
115
116 for (int i = 0 ; i < n * n; i++)
117 out[i] = out[i] + out[i];
118
119 for (int i = 0 ; i < n * n * n; i++)
120 out[i] = out[i] * out[i];
121
122 for (int i = 0 ; i < n * n * n * n; i++)
123 out[i] = out[i] * out[i];
124
125 for (int i = 0 ; i < n * n * n * n; i++) {
126 vint16mf2_t v = *(vint16mf2_t*)(in + 900 + i);
127 *(vint16mf2_t*)(out + 900 + i) = v;
128 }
129 }
130
131 void f6 (int8_t * restrict in, int8_t * restrict out, int n)
132 {
133 if (n == 0) {
134 vint32mf2_t v = *(vint32mf2_t*)(in + 500);
135 *(vint32mf2_t*)(out + 500) = v;
136 } else if (n == 1) {
137 vint32mf2_t v = *(vint32mf2_t*)(in + 600);
138 *(vint32mf2_t*)(out + 600) = v;
139 }
140
141 for (int i = 0 ; i < n * n; i++)
142 out[i] = out[i] + out[i];
143
144 for (int i = 0 ; i < n * n * n; i++)
145 out[i] = out[i] * out[i];
146
147 for (int i = 0 ; i < n * n * n * n; i++)
148 out[i] = out[i] * out[i];
149
150 for (int i = 0 ; i < n * n * n * n; i++) {
151 vint32mf2_t v = *(vint32mf2_t*)(in + 900 + i);
152 *(vint32mf2_t*)(out + 900 + i) = v;
153 }
154 }
155
156 /* { dg-final { scan-assembler-times {vsetvli\s+[a-x0-9]+,\s*zero,\s*e8,\s*mf8,\s*t[au],\s*m[au]} 3 { target { no-opts "-O0" no-opts "-Os" no-opts "-Oz" no-opts "-funroll-loops" no-opts "-g" no-opts "-O1" } } } } */
157 /* { dg-final { scan-assembler-times {vsetvli\s+[a-x0-9]+,\s*zero,\s*e8,\s*mf4,\s*t[au],\s*m[au]} 3 { target { no-opts "-O0" no-opts "-Os" no-opts "-Oz" no-opts "-funroll-loops" no-opts "-g" no-opts "-O1" } } } } */
158 /* { dg-final { scan-assembler-times {vsetvli\s+[a-x0-9]+,\s*zero,\s*e8,\s*mf2,\s*t[au],\s*m[au]} 3 { target { no-opts "-O0" no-opts "-Os" no-opts "-Oz" no-opts "-funroll-loops" no-opts "-g" no-opts "-O1" } } } } */
159 /* { dg-final { scan-assembler-times {vsetvli\s+[a-x0-9]+,\s*zero,\s*e8,\s*mf8,\s*t[au],\s*m[au]\s+\.L[0-9]+\:} 1 { target { no-opts "-O0" no-opts "-Os" no-opts "-Oz" no-opts "-funroll-loops" no-opts "-g" no-opts "-O1" } } } } */
160 /* { dg-final { scan-assembler-times {vsetvli\s+[a-x0-9]+,\s*zero,\s*e8,\s*mf4,\s*t[au],\s*m[au]\s+\.L[0-9]+\:} 1 { target { no-opts "-O0" no-opts "-Os" no-opts "-Oz" no-opts "-funroll-loops" no-opts "-g" no-opts "-O1" } } } } */
161 /* { dg-final { scan-assembler-times {vsetvli\s+[a-x0-9]+,\s*zero,\s*e8,\s*mf2,\s*t[au],\s*m[au]\s+\.L[0-9]+\:} 1 { target { no-opts "-O0" no-opts "-Os" no-opts "-Oz" no-opts "-funroll-loops" no-opts "-g" no-opts "-O1" } } } } */
162 /* { dg-final { scan-assembler-times {vsetvli\s+[a-x0-9]+,\s*zero,\s*e16,\s*mf4,\s*t[au],\s*m[au]} 3 { target { no-opts "-O0" no-opts "-Os" no-opts "-Oz" no-opts "-funroll-loops" no-opts "-g" no-opts "-O1" } } } } */
163 /* { dg-final { scan-assembler-times {vsetvli\s+[a-x0-9]+,\s*zero,\s*e16,\s*mf2,\s*t[au],\s*m[au]} 3 { target { no-opts "-O0" no-opts "-Os" no-opts "-Oz" no-opts "-funroll-loops" no-opts "-g" no-opts "-O1" } } } } */
164 /* { dg-final { scan-assembler-times {vsetvli\s+[a-x0-9]+,\s*zero,\s*e16,\s*mf4,\s*t[au],\s*m[au]\s+\.L[0-9]+\:} 1 { target { no-opts "-O0" no-opts "-Os" no-opts "-Oz" no-opts "-funroll-loops" no-opts "-g" no-opts "-O1" } } } } */
165 /* { dg-final { scan-assembler-times {vsetvli\s+[a-x0-9]+,\s*zero,\s*e16,\s*mf2,\s*t[au],\s*m[au]\s+\.L[0-9]+\:} 1 { target { no-opts "-O0" no-opts "-Os" no-opts "-Oz" no-opts "-funroll-loops" no-opts "-g" no-opts "-O1" } } } } */
166 /* { dg-final { scan-assembler-times {vsetvli\s+[a-x0-9]+,\s*zero,\s*e32,\s*mf2,\s*t[au],\s*m[au]} 3 { target { no-opts "-O0" no-opts "-Os" no-opts "-Oz" no-opts "-funroll-loops" no-opts "-g" no-opts "-O1" } } } } */
167 /* { dg-final { scan-assembler-times {vsetvli\s+[a-x0-9]+,\s*zero,\s*e32,\s*mf2,\s*t[au],\s*m[au]\s+\.L[0-9]+\:} 1 { target { no-opts "-O0" no-opts "-Os" no-opts "-Oz" no-opts "-funroll-loops" no-opts "-g" no-opts "-O1" } } } } */