forked from hpc12/tools
-
Notifications
You must be signed in to change notification settings - Fork 1
/
memory.c
175 lines (140 loc) · 5.21 KB
/
memory.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
#include "timing.h"
#include "main-helper.h"
#include "cl-helper.h"
void testSumResult(
const ELEMENT_TYPE* A, const ELEMENT_TYPE* B, const ELEMENT_TYPE* C, size_t N)
{
size_t count = 0;
ELEMENT_TYPE value;
for (size_t i = 0; i < N; ++i) {
for (size_t j = 0; j < N; ++j) {
value = A[i * N + j] + B[i * N + j];
if (C[i * N + j] != value) {
printf(" C[%3u][%3u] is NOT OK!!! expected %f (was %f)\n",
(uint)i, (uint)j, value, C[i * N + j]);
count++;
}
}
}
if (count == 0)
printf("All values are OK :)\n");
}
int main(int argc, char **argv)
{
if (argc != 3)
{
fprintf(stderr, "USO: %s <dimension> <vueltas>\n", argv[0]);
abort();
}
const cl_long n = atol(argv[1]);
const int ntrips = atoi(argv[2]);
cl_context ctx;
cl_command_queue queue;
// create_context_on(CHOOSE_INTERACTIVELY, CHOOSE_INTERACTIVELY, 0, &ctx, &queue, 0);
// print_device_info_from_queue(queue);
print_platforms_devices();
create_context_on(NULL, NULL, 0, &ctx, &queue, 0);
// --------------------------------------------------------------------------
// load kernels
// --------------------------------------------------------------------------
char *knl_text = read_file("mat-sum.cl");
cl_kernel knl = kernel_from_string(ctx, knl_text, "sum", NULL);
free(knl_text);
// --------------------------------------------------------------------------
// allocate and initialize CPU memory
// --------------------------------------------------------------------------
cl_long sizeN = n*n;
ELEMENT_TYPE *a = (ELEMENT_TYPE *) malloc(sizeof(ELEMENT_TYPE) * sizeN);
if (!a) { perror("alloc x"); abort(); }
ELEMENT_TYPE *b = (ELEMENT_TYPE *) malloc(sizeof(ELEMENT_TYPE) * sizeN);
if (!b) { perror("alloc y"); abort(); }
ELEMENT_TYPE *c = (ELEMENT_TYPE *) malloc(sizeof(ELEMENT_TYPE) * sizeN);
if (!c) { perror("alloc z"); abort(); }
srand(2006);
randomInit(a, sizeN);
randomInit(b, sizeN);
zeroInit(c, sizeN);
// --------------------------------------------------------------------------
// allocate device memory
// --------------------------------------------------------------------------
cl_int status;
cl_mem buf_a = clCreateBuffer(ctx, CL_MEM_READ_WRITE,
sizeof(ELEMENT_TYPE) * sizeN, 0, &status);
CHECK_CL_ERROR(status, "clCreateBuffer");
cl_mem buf_b = clCreateBuffer(ctx, CL_MEM_READ_WRITE,
sizeof(ELEMENT_TYPE) * sizeN, 0, &status);
CHECK_CL_ERROR(status, "clCreateBuffer");
cl_mem buf_c = clCreateBuffer(ctx, CL_MEM_READ_WRITE,
sizeof(ELEMENT_TYPE) * sizeN, 0, &status);
CHECK_CL_ERROR(status, "clCreateBuffer");
// --------------------------------------------------------------------------
// transfer to device
// --------------------------------------------------------------------------
timestamp_type time0, time3;
get_timestamp(&time0);
CALL_CL_GUARDED(clEnqueueWriteBuffer, (
queue, buf_a, /*blocking*/ CL_TRUE, /*offset*/ 0,
sizeN * sizeof(ELEMENT_TYPE), a,
0, NULL, NULL));
CALL_CL_GUARDED(clEnqueueWriteBuffer, (
queue, buf_b, /*blocking*/ CL_TRUE, /*offset*/ 0,
sizeN * sizeof(ELEMENT_TYPE), b,
0, NULL, NULL));
// --------------------------------------------------------------------------
// run code on device
// --------------------------------------------------------------------------
CALL_CL_GUARDED(clFinish, (queue));
timestamp_type time1, time2;
get_timestamp(&time1);
// size_t globalWorkSize[] = {n,n};
size_t ldim[] = { 128 };
size_t gdim[] = { ((sizeN + ldim[0] - 1)/ldim[0])*ldim[0] };
printf("local=%ld\n", ldim[0]);
printf("group=%ld\n", gdim[0]);
SET_4_KERNEL_ARGS(knl, buf_a, buf_b, buf_c, sizeN);
for (int trip = 0; trip < ntrips; ++trip)
{
CALL_CL_GUARDED(clEnqueueNDRangeKernel,
(queue, knl,
/*dimensions*/ 1, NULL, gdim, ldim,
0, NULL, NULL));
}
CALL_CL_GUARDED(clFinish, (queue));
get_timestamp(&time2);
double elapsed = timestamp_diff_in_seconds(time0,time2);
printf("%12f s\n", elapsed);
printStatistics(time1, time2, ntrips, n);
// --------------------------------------------------------------------------
// transfer back & check
// --------------------------------------------------------------------------
if (! getenv("HIDE_CHECK_RESULTS"))
{
CALL_CL_GUARDED(clEnqueueReadBuffer, (
queue, buf_c, /*blocking*/ CL_TRUE, /*offset*/ 0,
sizeN * sizeof(ELEMENT_TYPE), c,
0, NULL, NULL));
if (n < 30)
{
printf("\nMatrix A\n");
printMatrix(a, n);
printf("\nMatrix B\n");
printMatrix(b, n);
printf("\nMatrix C = A · B\n");
printMatrix(c, n);
}
testSumResult(a,b,c,n);
}
// --------------------------------------------------------------------------
// clean up
// --------------------------------------------------------------------------
CALL_CL_GUARDED(clReleaseMemObject, (buf_a));
CALL_CL_GUARDED(clReleaseMemObject, (buf_b));
CALL_CL_GUARDED(clReleaseMemObject, (buf_c));
CALL_CL_GUARDED(clReleaseKernel, (knl));
CALL_CL_GUARDED(clReleaseCommandQueue, (queue));
CALL_CL_GUARDED(clReleaseContext, (ctx));
free(a);
free(b);
free(c);
return 0;
}