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 static const clblasOrder order = clblasRowMajor;
30 static const size_t N = 5;
31 static const cl_float alpha = 10;
32 
33 static const clblasUplo uplo = clblasUpper;
34 static const cl_float AP[] = {
35     11, 12, 13, 14, 15,
36         22, 23, 24, 25,
37             33, 34, 35,
38                 44, 45,
39                     55
40 };
41 static const size_t lda = 5;    /* i.e. lda = N */
42 
43 static const cl_float X[] = {
44     11,
45     21,
46     31,
47     41,
48     51
49 };
50 static const int incx = 1;
51 
52 static const cl_float beta = 20;
53 static cl_float Y[] = {
54     11,
55     21,
56     31,
57     41,
58     51
59 };
60 static const int incy = 1;
61 
62 static void
printResult(void)63 printResult(void)
64 {
65     size_t i;
66 
67     printf("Result:\n\n");
68     for (i = 0; i < N; i++) {
69         printf("%.3f\n", Y[i * incy]);
70     }
71 }
72 
73 int
main(void)74 main(void)
75 {
76     cl_int err;
77     cl_platform_id platform = 0;
78     cl_device_id device = 0;
79     cl_context_properties props[3] = { CL_CONTEXT_PLATFORM, 0, 0 };
80     cl_context ctx = 0;
81     cl_command_queue queue = 0;
82     cl_mem bufAP, bufX, bufY;
83     cl_event event = NULL;
84     int ret = 0, numElementsAP;
85 
86     /* Setup OpenCL environment. */
87     err = clGetPlatformIDs(1, &platform, NULL);
88     if (err != CL_SUCCESS) {
89         printf( "clGetPlatformIDs() failed with %d\n", err );
90         return 1;
91     }
92 
93     err = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, NULL);
94     if (err != CL_SUCCESS) {
95         printf( "clGetDeviceIDs() failed with %d\n", err );
96         return 1;
97     }
98 
99     props[1] = (cl_context_properties)platform;
100     ctx = clCreateContext(props, 1, &device, NULL, NULL, &err);
101     if (err != CL_SUCCESS) {
102         printf( "clCreateContext() failed with %d\n", err );
103         return 1;
104     }
105 
106     queue = clCreateCommandQueue(ctx, device, 0, &err);
107     if (err != CL_SUCCESS) {
108         printf( "clCreateCommandQueue() failed with %d\n", err );
109         clReleaseContext(ctx);
110         return 1;
111     }
112 
113     /* Setup clblas. */
114     err = clblasSetup();
115     if (err != CL_SUCCESS) {
116         printf("clblasSetup() failed with %d\n", err);
117         clReleaseCommandQueue(queue);
118         clReleaseContext(ctx);
119         return 1;
120     }
121 
122     numElementsAP = (N * (N+1)) / 2;	// To get number of elements in a packed matrix
123 
124     /* Prepare OpenCL memory objects and place matrices inside them. */
125     bufAP = clCreateBuffer(ctx, CL_MEM_READ_ONLY, numElementsAP * sizeof(cl_float),
126                           NULL, &err);
127     bufX = clCreateBuffer(ctx, CL_MEM_READ_ONLY, N * sizeof(cl_float),
128                           NULL, &err);
129     bufY = clCreateBuffer(ctx, CL_MEM_READ_WRITE, N * sizeof(cl_float),
130                           NULL, &err);
131 
132     err = clEnqueueWriteBuffer(queue, bufAP, CL_TRUE, 0,
133                         numElementsAP * sizeof(cl_float), AP, 0, NULL, NULL);
134     err = clEnqueueWriteBuffer(queue, bufX, CL_TRUE, 0,
135                         N * sizeof(cl_float), X, 0, NULL, NULL);
136     err = clEnqueueWriteBuffer(queue, bufY, CL_TRUE, 0,
137                         N * sizeof(cl_float), Y, 0, NULL, NULL);
138 
139     /* Call clblas function. */
140     err = clblasSspmv(order, uplo, N, alpha, bufAP, 0, bufX, 0, incx,
141                                 beta, bufY, 0, incy, 1, &queue, 0, NULL, &event);
142     if (err != CL_SUCCESS) {
143         printf("clblasSspmv() failed with %d\n", err);
144         ret = 1;
145     }
146     else {
147         /* Wait for calculations to be finished. */
148         err = clWaitForEvents(1, &event);
149 
150         /* Fetch results of calculations from GPU memory. */
151         err = clEnqueueReadBuffer(queue, bufY, CL_TRUE, 0, N * sizeof(cl_float),
152                                             Y, 0, NULL, NULL);
153         /* At this point you will get the result of SSPMV placed in Y array. */
154         printResult();
155     }
156 
157     /* Release OpenCL events. */
158     clReleaseEvent(event);
159 
160     /* Release OpenCL memory objects. */
161     clReleaseMemObject(bufY);
162     clReleaseMemObject(bufX);
163     clReleaseMemObject(bufAP);
164 
165     /* Finalize work with clblas. */
166     clblasTeardown();
167 
168     /* Release OpenCL working objects. */
169     clReleaseCommandQueue(queue);
170     clReleaseContext(ctx);
171 
172     return ret;
173 }
174