CLBlast_trsv_event

sgemv.c

// compile and run with 
// gcc -I../include -O2 -Wall sgemv.c -o sgemv -lOpenCL -L . -lclblast; ./sgemv
// in the build directory of CLBlast

#include <stdlib.h>
#include <stdio.h>
#include <string.h>

#define CL_USE_DEPRECATED_OPENCL_1_2_APIS // to disable deprecation warnings
#include <clblast_c.h>


int main(void) {

  // OpenCL platform/device settings
  const size_t platform_id = 0;
  const size_t device_id = 0;

  // Example arguments
  const size_t m = 128;
  const size_t n = 289;
  const float alpha = 0.7;
  const float beta = 0.0;
  const size_t a_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
  cl_uint num_devices;
  clGetDeviceIDs(platform, CL_DEVICE_TYPE_ALL, 0, NULL, &num_devices);
  cl_device_id* devices = (cl_device_id*)malloc(num_devices*sizeof(cl_device_id));
  clGetDeviceIDs(platform, CL_DEVICE_TYPE_ALL, 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 data structures with some example data
  float* host_a = (float*)malloc(sizeof(float)*m*n);
  float* host_x = (float*)malloc(sizeof(float)*n);
  float* host_y = (float*)malloc(sizeof(float)*m);
  for (size_t i=0; i<m*n; ++i) { host_a[i] = 12.193; }
  for (size_t i=0; i<n; ++i) { host_x[i] = -8.199; }
  for (size_t i=0; i<m; ++i) { host_y[i] = 0.0; }

  // Copy the data-structures to the device
  cl_mem device_a = clCreateBuffer(context, CL_MEM_READ_WRITE, m*n*sizeof(float), NULL, NULL);
  cl_mem device_x = clCreateBuffer(context, CL_MEM_READ_WRITE, n*sizeof(float), NULL, NULL);
  cl_mem device_y = clCreateBuffer(context, CL_MEM_READ_WRITE, m*sizeof(float), NULL, NULL);
  clEnqueueWriteBuffer(queue, device_a, CL_TRUE, 0, m*n*sizeof(float), host_a, 0, NULL, NULL);
  clEnqueueWriteBuffer(queue, device_x, CL_TRUE, 0, n*sizeof(float), host_x, 0, NULL, NULL);
  clEnqueueWriteBuffer(queue, device_y, CL_TRUE, 0, m*sizeof(float), host_y, 0, NULL, NULL);

  // Call the SGEMV routine.
  CLBlastStatusCode status = CLBlastSgemv(CLBlastLayoutRowMajor, CLBlastTransposeNo,
                                          m, n,
                                          alpha,
                                          device_a, 0, a_ld,
                                          device_x, 0, 1,
                                          beta,
                                          device_y, 0, 1,
                                          &queue, &event);

  // Wait for completion
  if (status == CLBlastSuccess) {
    printf("Event: %ld\n", (long)event);
    cl_int wait_status = clWaitForEvents(1, &event);
    printf("Status: %d\n", wait_status);
    cl_int release_status = clReleaseEvent(event);
    printf("Status: %d\n", release_status);
  }

  // Example completed. See "clblast_c.h" for status codes (0 -> success).
  printf("Completed SGEMV with status %d\n", status);

  // Clean-up
  free(platforms);
  free(devices);
  free(host_a);
  free(host_x);
  free(host_y);
  clReleaseMemObject(device_a);
  clReleaseMemObject(device_x);
  clReleaseMemObject(device_y);
  clReleaseCommandQueue(queue);
  clReleaseContext(context);
  return 0;
}

strsv.c

// compile and run with 
// gcc -I../include -O2 -Wall strsv.c -o strsv -lOpenCL -L . -lclblast; ./strsv
// in the build directory of CLBlast

#include <stdlib.h>
#include <stdio.h>
#include <string.h>

#define CL_USE_DEPRECATED_OPENCL_1_2_APIS // to disable deprecation warnings
#include <clblast_c.h>


int main(void) {

  // OpenCL platform/device settings
  const size_t platform_id = 0;
  const size_t device_id = 0;

  // Example arguments
  const size_t n = 50;
  const size_t a_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
  cl_uint num_devices;
  clGetDeviceIDs(platform, CL_DEVICE_TYPE_ALL, 0, NULL, &num_devices);
  cl_device_id* devices = (cl_device_id*)malloc(num_devices*sizeof(cl_device_id));
  clGetDeviceIDs(platform, CL_DEVICE_TYPE_ALL, 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 data structures with some example data
  float* host_a = (float*)malloc(sizeof(float)*n*n);
  float* host_x = (float*)malloc(sizeof(float)*n);
  for (size_t i=0; i<n*n; ++i) { host_a[i] = i+1; }
  for (size_t i=0; i<n; ++i) { host_x[i] = 1.0; }

  // Copy the data-structures to the device
  cl_mem device_a = clCreateBuffer(context, CL_MEM_READ_WRITE, n*n*sizeof(float), NULL, NULL);
  cl_mem device_x = clCreateBuffer(context, CL_MEM_READ_WRITE, n*sizeof(float), NULL, NULL);
  clEnqueueWriteBuffer(queue, device_a, CL_TRUE, 0, n*n*sizeof(float), host_a, 0, NULL, NULL);
  clEnqueueWriteBuffer(queue, device_x, CL_TRUE, 0, n*sizeof(float), host_x, 0, NULL, NULL);

  // Call the STRSV routine.
  CLBlastStatusCode status = CLBlastStrsv(CLBlastLayoutRowMajor, CLBlastTriangleUpper, 
                                          CLBlastTransposeNo, CLBlastDiagonalNonUnit,
                                          n,
                                          device_a, 0, a_ld,
                                          device_x, 0, 1,
                                          &queue, &event);

  // Wait for completion
  if (status == CLBlastSuccess) {
    printf("Event: %ld\n", (long)event);
    int wait_status = clWaitForEvents(1, &event);
    printf("Status: %d\n", wait_status);
    cl_int release_status = clReleaseEvent(event);
    printf("Status: %d\n", release_status);
  }

  // Example completed. See "clblast_c.h" for status codes (0 -> success).
  printf("Completed STRSV with status %d\n", status);

  // Clean-up
  free(platforms);
  free(devices);
  free(host_a);
  free(host_x);
  clReleaseMemObject(device_a);
  clReleaseMemObject(device_x);
  clReleaseCommandQueue(queue);
  clReleaseContext(context);
  return 0;
}