opencl.h

Go to the documentation of this file.

/* ------------------------------------------------------------
 This is the file "opencl.h" of the H2Lib package.
 All rights reserved, Sven Christophersen 2015
 ------------------------------------------------------------ */

#ifndef OPENCL_H_
#define OPENCL_H_

#ifdef USE_OPENCL

/* C STD LIBRARY */
#include <stdio.h>
#include <assert.h>
#include <CL/cl.h>
#include <omp.h>
/* CORE 0 */
#include "basic.h"
/* CORE 1 */
/* CORE 2 */
/* CORE 3 */
/* SIMPLE */
/* PARTICLES */
/* BEM */

struct _ocl {
  cl_uint num_platforms;
  cl_platform_id *platforms;
  cl_platform_id *devplatforms;
  cl_uint num_devices;
  cl_device_id *devices;
  cl_uint num_contexts;
  cl_context *contexts;
  cl_uint queues_per_device;
  cl_command_queue *queues;
  size_t max_package_size;
};

extern struct _ocl ocl_system;

#define CL_CHECK(res) \
    {if (res != CL_SUCCESS) {fprintf(stderr,"Error \"%s\" (%d) in file %s on line %d\n", \
        get_error_string(res), res, __FILE__,__LINE__); abort();}}

HEADER_PREFIX void get_opencl_devices(cl_device_id **devices, cl_uint *ndevices);

HEADER_PREFIX void set_opencl_devices(cl_device_id *devices, cl_uint ndevices,
    cl_uint queues_per_device);

HEADER_PREFIX const char *get_error_string(cl_int error);

HEADER_PREFIX void setup_kernels(const char *src_str, const uint num_kernels,
    const char **kernel_names, cl_kernel **kernels);

typedef enum {
  CPU_ONLY, 
  GPU_ONLY, 
  CPU_FIRST,
  GPU_FIRST 
}task_affinity;

typedef struct _task task;

typedef task *ptask;

struct _task {
  void *data;
  ptask next;
};

typedef struct _taskgroup taskgroup;

typedef taskgroup *ptaskgroup;

struct _taskgroup {
  ptask tasks;
  size_t size;
  task_affinity affinity;
  ptaskgroup next;
  void *data;

  void (*merge_tasks)(ptaskgroup t, void **data);
  void (*cleanup_merge)(void *data);
  void (*distribute_results)(ptaskgroup t, void *data);
  void (*close_taskgroup)(ptaskgroup tg);

  void (*callback_cpu)(void *data);
  void (*callback_gpu)(void *data);
  size_t (*getsize_task)(void *data);
  void (*split_task)(void *data, void ***splits, uint *n);
  void (*cleanup_task)(void *data);
};

HEADER_PREFIX ptask new_ocltask(void *data, ptask next);

HEADER_PREFIX void del_ocltask(ptaskgroup tg, ptask t);

HEADER_PREFIX ptaskgroup new_ocltaskgroup(task_affinity affinity, ptaskgroup next,
    void (*merge_tasks)(ptaskgroup tg, void **data),
    void (*cleanup_merge)(void *data),
    void (*distribute_results)(ptaskgroup tg, void *data),
    void (*close_taskgroup)(ptaskgroup tg), void (*callback_cpu)(void *data),
    void (*callback_gpu)(void *data), size_t (*getsize_task)(void *data),
    void (*split_task)(void *data, void ***split, uint *n),
    void (*cleanup_task)(void *data), void *data);

HEADER_PREFIX void del_taskgroup(ptaskgroup tg);

HEADER_PREFIX void add_task_taskgroup(ptaskgroup *tg, void *data);

HEADER_PREFIX void add_taskgroup_to_scheduler(ptaskgroup tg);

HEADER_PREFIX void start_scheduler(uint cpu_threads, uint gpu_threads);

HEADER_PREFIX void stop_scheduler();

#define SCHEDULE_OPENCL(cpu_threads, gpu_threads, func, ...) \
    _Pragma("omp parallel num_threads(2)") \
    { \
  start_scheduler(cpu_threads, gpu_threads); \
  if (omp_get_thread_num() == 0) { \
    func(__VA_ARGS__); \
  } \
  stop_scheduler(); \
    }

#endif

#endif /* OPENCL_H_ */