#include "xmp_internal.h"
#include "tca-api.h"
#include "xmp.h"
#include "cuda_runtime.h"

Include dependency graph for xmp_reflect_tca.c:

Macros
#define	_XMP_TCA_DMAC 0

#define	_XMP_TCA_USE_PACK_SIZE 8

#define	_XMP_TCA_DEBUG(...)

Functions
void	_XMP_gpu_pack_vector_async (char restrict dst, char restrict src, int count, int blocklength, long stride, size_t typesize, void *async_id)

void	_XMP_gpu_unpack_vector_async (char restrict dst, char restrict src, int count, int blocklength, long stride, size_t typesize, void *async_id)

void	_XMP_reflect_init_tca (void acc_addr, _XMP_array_t adesc)

void	_XMP_reflect_do_tca (_XMP_array_t *adesc)

Macro Definition Documentation

◆ _XMP_TCA_DEBUG

#define _XMP_TCA_DEBUG ( ... )

◆ _XMP_TCA_DMAC

#define _XMP_TCA_DMAC 0

◆ _XMP_TCA_USE_PACK_SIZE

#define _XMP_TCA_USE_PACK_SIZE 8

Function Documentation

◆ _XMP_gpu_pack_vector_async()

void _XMP_gpu_pack_vector_async	(	char *restrict	dst,
		char *restrict	src,
		int	count,
		int	blocklength,
		long	stride,
		size_t	typesize,
		void *	async_id
	)

◆ _XMP_gpu_unpack_vector_async()

void _XMP_gpu_unpack_vector_async	(	char *restrict	dst,
		char *restrict	src,
		int	count,
		int	blocklength,
		long	stride,
		size_t	typesize,
		void *	async_id
	)

◆ _XMP_reflect_do_tca()

void _XMP_reflect_do_tca ( _XMP_array_t * adesc )

 {
   _XMP_tca_pack_vector(adesc);
   TCA_CHECK(tcaStartDMADesc(_XMP_TCA_DMAC));
   _XMP_refect_wait_tca(adesc);
   _XMP_tca_unpack_vector(adesc);
   MPI_Barrier(MPI_COMM_WORLD);
 }

Here is the caller graph for this function:

◆ _XMP_reflect_init_tca()

void _XMP_reflect_init_tca	(	void *	acc_addr,
		_XMP_array_t *	adesc
	)

 {
   _XMP_create_TCA_handle(acc_addr, adesc);
   _XMP_create_TCA_reflect_desc(adesc, acc_addr);
 }