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 
21 /* Include CLBLAS header. It automatically includes needed OpenCL header,
22  * so we can drop out explicit inclusion of cl.h header.
23  */
24 #include <clBLAS.h>
25 
26 /* This example uses predefined matrices and their characteristics for
27  * simplicity purpose.
28  */
29 
30 static const clblasOrder order = clblasRowMajor;
31 static const clblasTranspose trans = clblasNoTrans;
32 
33 static const size_t M = 5;
34 static const size_t N = 5;
35 static const size_t KL = 1;
36 static const size_t KU = 2;
37 static const size_t lda = 4;    // lda = KL + KU + 1
38 
39 static const cl_float A[] = {
40     00, 12, 13, 14,
41     21, 22, 23, 24,
42     31, 32, 33, 34,
43     41, 42, 43, 00,
44     51, 62, 00, 00
45 };
46 
47 static const cl_float alpha = 10;
48 static const cl_float X[] = {
49     11,
50     21,
51     31,
52     41,
53     51,
54 };
55 static const int incx = 1;
56 
57 static const cl_float beta = 20;
58 static cl_float Y[] = {
59     11,
60     21,
61     31,
62     41,
63     51,
64 };
65 static const int incy = 1;
66 
67 
68 static void
printResult(const char * str)69 printResult(const char* str)
70 {
71     size_t i;
72 
73     printf("%s:\n", str);
74 
75     for (i = 0; i < M; i++) {
76         printf("%f\n", Y[ i * incy ]);
77     }
78 }
79 
80 int
main(void)81 main(void)
82 {
83     cl_int err;
84     cl_platform_id platform = 0;
85     cl_device_id device = 0;
86     cl_context_properties props[3] = { CL_CONTEXT_PLATFORM, 0, 0 };
87     cl_context ctx = 0;
88     cl_command_queue queue = 0;
89     cl_mem bufA, bufX, bufY;
90     cl_event event = NULL;
91     int ret = 0;
92 
93     /* Setup OpenCL environment. */
94     err = clGetPlatformIDs(1, &platform, NULL);
95     if (err != CL_SUCCESS) {
96         printf( "clGetPlatformIDs() failed with %d\n", err );
97         return 1;
98     }
99 
100     err = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, NULL);
101     if (err != CL_SUCCESS) {
102         printf( "clGetDeviceIDs() failed with %d\n", err );
103         return 1;
104     }
105 
106     props[1] = (cl_context_properties)platform;
107     ctx = clCreateContext(props, 1, &device, NULL, NULL, &err);
108     if (err != CL_SUCCESS) {
109         printf( "clCreateContext() failed with %d\n", err );
110         return 1;
111     }
112 
113     queue = clCreateCommandQueue(ctx, device, 0, &err);
114     if (err != CL_SUCCESS) {
115         printf( "clCreateCommandQueue() failed with %d\n", err );
116         clReleaseContext(ctx);
117         return 1;
118     }
119 
120     /* Setup clblas. */
121     err = clblasSetup();
122     if (err != CL_SUCCESS) {
123         printf("clblasSetup() failed with %d\n", err);
124         clReleaseCommandQueue(queue);
125         clReleaseContext(ctx);
126         return 1;
127     }
128 
129     /* Prepare OpenCL memory objects and place matrices inside them. */
130     bufA = clCreateBuffer(ctx, CL_MEM_READ_ONLY, M * lda * sizeof(cl_float),
131                           NULL, &err);
132     bufX = clCreateBuffer(ctx, CL_MEM_READ_ONLY, N * sizeof(cl_float),
133                           NULL, &err);
134     bufY = clCreateBuffer(ctx, CL_MEM_READ_WRITE, M * sizeof(cl_float),
135                           NULL, &err);
136 
137     err = clEnqueueWriteBuffer(queue, bufA, CL_TRUE, 0,
138                                 M * lda * sizeof(cl_float), A, 0, NULL, NULL);
139     err = clEnqueueWriteBuffer(queue, bufX, CL_TRUE, 0,
140                                 N * sizeof(cl_float), X, 0, NULL, NULL);
141     err = clEnqueueWriteBuffer(queue, bufY, CL_TRUE, 0,
142                                 M * sizeof(cl_float), Y, 0, NULL, NULL);
143 
144     /* Call clblas function. */
145     err = clblasSgbmv(order, trans, M, N, KL, KU, alpha, bufA, 0, lda, bufX, 0, incx,
146                             beta, bufY, 0, incy, 1, &queue, 0, NULL, &event);
147     if (err != CL_SUCCESS) {
148         printf("clblasSgbmv() failed with %d\n", err);
149         ret = 1;
150     }
151     else {
152         /* Wait for calculations to be finished. */
153         err = clWaitForEvents(1, &event);
154 
155         /* Fetch results of calculations from GPU memory. */
156         err = clEnqueueReadBuffer(queue, bufY, CL_TRUE, 0, M * sizeof(cl_float),
157                                       Y, 0, NULL, NULL);
158         /* At this point you will get the result of SGBMV placed in Y array. */
159         printResult("clblasSgbmv result");
160     }
161 
162     /* Release OpenCL events. */
163     clReleaseEvent(event);
164 
165     /* Release OpenCL memory objects. */
166     clReleaseMemObject(bufY);
167     clReleaseMemObject(bufX);
168     clReleaseMemObject(bufA);
169 
170     /* Finalize work with clblas. */
171     clblasTeardown();
172 
173     /* Release OpenCL working objects. */
174     clReleaseCommandQueue(queue);
175     clReleaseContext(ctx);
176 
177     return ret;
178 }
179