1 /* ************************************************************************
2 * Copyright 2013 Advanced Micro Devices, Inc.
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 * ************************************************************************/
16
17 #include <sys/types.h>
18 #include <stdio.h>
19 #include <string.h>
20 #include <math.h>
21
22 /* Include CLBLAS header. It automatically includes needed OpenCL header,
23 * so we can drop out explicit inclusion of cl.h header.
24 */
25 #include <clBLAS.h>
26
27 /* This example uses predefined matrices and their characteristics for
28 * simplicity purpose.
29 */
30 static const clblasOrder order = clblasRowMajor;
31
32 static const size_t N = 5;
33 static const cl_float alpha = 10;
34 static cl_float X[] = {
35 11,
36 21,
37 31,
38 41,
39 51,
40 };
41 static const int incx = 1;
42
43 static void
printResult(void)44 printResult(void)
45 {
46 size_t i;
47 printf("\nResult:\n");
48
49 for (i = 0; i < N; i++) {
50 printf("\t%f", X[i] );
51 printf("\n");
52 }
53 }
54
55 int
main(void)56 main(void)
57 {
58 cl_int err;
59 cl_platform_id platform = 0;
60 cl_device_id device = 0;
61 cl_context_properties props[3] = { CL_CONTEXT_PLATFORM, 0, 0 };
62 cl_context ctx = 0;
63 cl_command_queue queue = 0;
64 cl_mem bufX;
65 cl_event event = NULL;
66 int ret = 0;
67 int lenX = 1 + (N-1)*abs(incx);
68
69 /* Setup OpenCL environment. */
70 err = clGetPlatformIDs(1, &platform, NULL);
71 if (err != CL_SUCCESS) {
72 printf( "clGetPlatformIDs() failed with %d\n", err );
73 return 1;
74 }
75
76 err = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, NULL);
77 if (err != CL_SUCCESS) {
78 printf( "clGetDeviceIDs() failed with %d\n", err );
79 return 1;
80 }
81
82 props[1] = (cl_context_properties)platform;
83 ctx = clCreateContext(props, 1, &device, NULL, NULL, &err);
84 if (err != CL_SUCCESS) {
85 printf( "clCreateContext() failed with %d\n", err );
86 return 1;
87 }
88
89 queue = clCreateCommandQueue(ctx, device, 0, &err);
90 if (err != CL_SUCCESS) {
91 printf( "clCreateCommandQueue() failed with %d\n", err );
92 clReleaseContext(ctx);
93 return 1;
94 }
95
96 /* Setup clblas. */
97 err = clblasSetup();
98 if (err != CL_SUCCESS) {
99 printf("clblasSetup() failed with %d\n", err);
100 clReleaseCommandQueue(queue);
101 clReleaseContext(ctx);
102 return 1;
103 }
104
105 /* Prepare OpenCL memory objects and place vectors inside them. */
106 bufX = clCreateBuffer(ctx, CL_MEM_READ_WRITE, ( lenX * sizeof(cl_float)),
107 NULL, &err);
108
109 err = clEnqueueWriteBuffer(queue, bufX, CL_TRUE, 0,
110 ( lenX * sizeof(cl_float)), X, 0, NULL, NULL);
111
112 /* Call clblas function. */
113 err = clblasSscal( N, alpha, bufX, 0, incx, 1, &queue, 0, NULL, &event);
114
115 if (err != CL_SUCCESS) {
116 printf("clblasSscal() failed with %d\n", err);
117 ret = 1;
118 }
119 else {
120 /* Wait for calculations to be finished. */
121 err = clWaitForEvents(1, &event);
122
123 /* Fetch results of calculations from GPU memory. */
124 err = clEnqueueReadBuffer(queue, bufX, CL_TRUE, 0, (lenX * sizeof(cl_float)),
125 X, 0, NULL, NULL);
126 /* At this point you will get the result of SSCAL placed in X vector. */
127 printResult();
128 }
129
130 /* Release OpenCL events. */
131 clReleaseEvent(event);
132
133 /* Release OpenCL memory objects. */
134 clReleaseMemObject(bufX);
135
136 /* Finalize work with clblas. */
137 clblasTeardown();
138
139 /* Release OpenCL working objects. */
140 clReleaseCommandQueue(queue);
141 clReleaseContext(ctx);
142
143 return ret;
144 }
145