diff options
author | CNugteren <web@cedricnugteren.nl> | 2015-08-13 13:47:15 +0200 |
---|---|---|
committer | CNugteren <web@cedricnugteren.nl> | 2015-08-13 13:47:15 +0200 |
commit | a6c104ef207f4ba6ef41161905c8d8dd29e97ebf (patch) | |
tree | f946d7c8166836efcbeb1db92fa998f6a7a5d36d /samples | |
parent | 8617195ac581058a936bee59d738ebc22d2c651e (diff) |
Added SGEMM example using the C API
Diffstat (limited to 'samples')
-rw-r--r-- | samples/sgemm.c | 108 |
1 files changed, 108 insertions, 0 deletions
diff --git a/samples/sgemm.c b/samples/sgemm.c new file mode 100644 index 00000000..f43fb147 --- /dev/null +++ b/samples/sgemm.c @@ -0,0 +1,108 @@ + +// ================================================================================================= +// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This +// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max- +// width of 100 characters per line. +// +// Author(s): +// Cedric Nugteren <www.cedricnugteren.nl> +// +// This file demonstrates the use of the SGEMM routine. It is pure C99 and demonstrates the use of +// the C API to the CLBlast library. +// +// Note that this example is meant for illustration purposes only. CLBlast provides other programs +// for performance benchmarking ('client_xxxxx') and for correctness testing ('test_xxxxx'). +// +// ================================================================================================= + +#include <stdio.h> +#include <string.h> + +// Includes the CLBlast library (C interface) +#include <clblast_c.h> + +// ================================================================================================= + +// Example use of the single-precision routine SGEMM +int main(void) { + + // OpenCL platform/device settings + const size_t platform_id = 0; + const size_t device_id = 0; + + // Example SGEMM arguments + const size_t m = 128; + const size_t n = 64; + const size_t k = 512; + const float alpha = 0.7f; + const float beta = 1.0f; + const size_t a_ld = k; + const size_t b_ld = n; + const size_t c_ld = n; + + // Initializes the OpenCL platform + cl_uint num_platforms; + clGetPlatformIDs(0, NULL, &num_platforms); + cl_platform_id* platforms = (cl_platform_id*)malloc(num_platforms*sizeof(cl_platform_id)); + clGetPlatformIDs(num_platforms, platforms, NULL); + cl_platform_id platform = platforms[platform_id]; + + // Initializes the OpenCL device (note: example for GPU devices only) + cl_uint num_devices; + clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 0, NULL, &num_devices); + cl_device_id* devices = (cl_device_id*)malloc(num_devices*sizeof(cl_device_id)); + clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, num_devices, devices, NULL); + cl_device_id device = devices[device_id]; + + // Creates the OpenCL context, queue, and an event + cl_context context = clCreateContext(NULL, 1, &device, NULL, NULL, NULL); + cl_command_queue queue = clCreateCommandQueue(context, device, 0, NULL); + cl_event event = NULL; + + // Populate host matrices with some example data + float* host_a = (float*)malloc(sizeof(float)*m*k); + float* host_b = (float*)malloc(sizeof(float)*n*k); + float* host_c = (float*)malloc(sizeof(float)*m*n); + for (size_t i=0; i<m*k; ++i) { host_a[i] = 12.193f; } + for (size_t i=0; i<n*k; ++i) { host_b[i] = -8.199f; } + for (size_t i=0; i<m*n; ++i) { host_c[i] = 0.0f; } + + // Copy the matrices to the device + cl_mem device_a = clCreateBuffer(context, CL_MEM_READ_WRITE, m*k*sizeof(float), NULL, NULL); + cl_mem device_b = clCreateBuffer(context, CL_MEM_READ_WRITE, n*k*sizeof(float), NULL, NULL); + cl_mem device_c = clCreateBuffer(context, CL_MEM_READ_WRITE, m*n*sizeof(float), NULL, NULL); + clEnqueueWriteBuffer(queue, device_a, CL_TRUE, 0, m*k*sizeof(float), host_a, 0, NULL, NULL); + clEnqueueWriteBuffer(queue, device_b, CL_TRUE, 0, n*k*sizeof(float), host_b, 0, NULL, NULL); + clEnqueueWriteBuffer(queue, device_c, CL_TRUE, 0, m*n*sizeof(float), host_c, 0, NULL, NULL); + + // Call the SGEMM routine. + StatusCode status = CLBlastSgemm(kRowMajor, kNo, kNo, + m, n, k, + alpha, + device_a, 0, a_ld, + device_b, 0, b_ld, + beta, + device_c, 0, c_ld, + &queue, &event); + + // Wait for completion + clWaitForEvents(1, &event); + + // Example completed. See "clblast_c.h" for status codes (0 -> success). + printf("Completed with status %d\n", status); + + // Clean-up + free(platforms); + free(devices); + free(host_a); + free(host_b); + free(host_c); + clReleaseMemObject(device_a); + clReleaseMemObject(device_b); + clReleaseMemObject(device_c); + clReleaseCommandQueue(queue); + clReleaseContext(context); + return 0; +} + +// ================================================================================================= |