cs_base_cuda.h File Reference

#include "cs_defs.h"
#include <stdio.h>
#include "cs_log.h"

Include dependency graph for cs_base_cuda.h:

Go to the source code of this file.

Macros
#define	CS_CUDA_CHECK(a)
#define	CS_CUDA_CHECK_CALL(a, file_name, line_num)
#define	CS_CUDA_WARP_SIZE   32

Functions
static unsigned int	cs_cuda_grid_size (cs_lnum_t n, unsigned int block_size)
	Compute grid size for given array and block sizes. More...

Variables
int	cs_glob_cuda_device_id
int	cs_glob_cuda_max_threads_per_block
int	cs_glob_cuda_max_block_size
int	cs_glob_cuda_max_blocks
int	cs_glob_cuda_n_mp

Macro Definition Documentation

CS_CUDA_CHECK

#define CS_CUDA_CHECK

(

a

)

Value:

{ \
    cudaError_t ret_code = a; \
    if (cudaSuccess != ret_code) { \
      bft_error(__FILE__, __LINE__, 0, "[CUDA error] %d: %s\n  running: %s", \
                ret_code, ::cudaGetErrorString(ret_code), #a); \
    } \
  }

CS_CUDA_CHECK_CALL

#define CS_CUDA_CHECK_CALL	(		a,
			file_name,
			line_num
	)

Value:

{ \
    cudaError_t ret_code = a; \
    if (cudaSuccess != ret_code) { \
      bft_error(file_name, line_num, 0, "[CUDA error] %d: %s\n  running: %s", \
                ret_code, ::cudaGetErrorString(ret_code), #a); \
    } \
  }

CS_CUDA_WARP_SIZE

#define CS_CUDA_WARP_SIZE   32

Function Documentation

cs_cuda_grid_size()

static unsigned int cs_cuda_grid_size ( cs_lnum_t  n, unsigned int  block_size  )

inlinestatic

Compute grid size for given array and block sizes.

This assumes each thread on a given block handles a single array element. For kernels in which each thread handles multiple elements, a grid size divided by that multiple is sufficient.

Parameters

[in]	n	size of arrays
[in]	block_size	block size for kernels

Returns

grid size for kernels

Variable Documentation

cs_glob_cuda_device_id

int cs_glob_cuda_device_id

cs_glob_cuda_max_block_size

int cs_glob_cuda_max_block_size

cs_glob_cuda_max_blocks

int cs_glob_cuda_max_blocks

cs_glob_cuda_max_threads_per_block

int cs_glob_cuda_max_threads_per_block

cs_glob_cuda_n_mp

int cs_glob_cuda_n_mp