cs_matrix_spmv_hip.cpp File Reference

Sparse Matrix SpMV operations with HIP. More...

#include "base/cs_defs.h"
#include <stdarg.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <assert.h>
#include <math.h>
#include <mpi.h>
#include "bft/bft_error.h"
#include "bft/bft_printf.h"
#include "base/cs_base.h"
#include "base/cs_base_hip.h"
#include "base/cs_halo.h"
#include "base/cs_halo_perio.h"
#include "base/cs_log.h"
#include "base/cs_mem.h"
#include "base/cs_timer.h"
#include "alge/cs_matrix.h"
#include "alge/cs_matrix_priv.h"
#include "alge/cs_matrix_spmv.h"

Include dependency graph for cs_matrix_spmv_hip.cpp:

Functions
void	cs_matrix_spmv_hip_finalize (void)
	Finalize ROC matrix API. More...
void	cs_matrix_spmv_hip_set_stream (hipStream_t stream)
	Assign ROC stream for next ROC-based SpMV operations. More...
hipStream_t	cs_matrix_spmv_hip_get_stream (void)
	Return stream used for ROC-based SpMV operations. More...
void	cs_matrix_spmv_hip_native (const cs_matrix_t *matrix, bool exclude_diag, bool sync, cs_real_t d_x[], cs_real_t d_y[])
	Matrix.vector product y = A.x with MSR matrix, scalar ROC version. More...
void	cs_matrix_spmv_hip_csr (cs_matrix_t *matrix, bool exclude_diag, bool sync, cs_real_t d_x[], cs_real_t d_y[])
	Matrix.vector product y = A.x with CSR matrix, scalar ROC version. More...
void	cs_matrix_spmv_hip_msr (cs_matrix_t *matrix, bool exclude_diag, bool sync, cs_real_t d_x[], cs_real_t d_y[])
	Matrix.vector product y = A.x with MSR matrix, scalar ROC version. More...
void	cs_matrix_spmv_hip_msr_b (cs_matrix_t *matrix, bool exclude_diag, bool sync, cs_real_t d_x[], cs_real_t d_y[])
	Matrix.vector product y = A.x with MSR matrix, block diagonal ROC version. More...

Detailed Description

Sparse Matrix SpMV operations with HIP.

Function Documentation

cs_matrix_spmv_hip_csr()

void cs_matrix_spmv_hip_csr	(	cs_matrix_t *	matrix,
		bool	exclude_diag,
		bool	sync,
		cs_real_t	d_x[],
		cs_real_t	d_y[]
	)

Matrix.vector product y = A.x with CSR matrix, scalar ROC version.

Matrix.vector product y = A.x with CSR matrix, scalar HIP version.

Parameters

[in]	matrix	pointer to matrix structure
[in]	exclude_diag	exclude diagonal if true,
[in]	sync	synchronize ghost cells if true
[in]	d_x	multipliying vector values (on device)
[out]	d_y	resulting vector (on device)

cs_matrix_spmv_hip_finalize()

void cs_matrix_spmv_hip_finalize

(

void

)

Finalize ROC matrix API.

Finalize HIP matrix API.

This frees resources such as the rocSPARSE handle, if used.

cs_matrix_spmv_hip_get_stream()

hipStream_t cs_matrix_spmv_hip_get_stream

(

void

)

Return stream used for ROC-based SpMV operations.

This function is callable only from ROC code.

cs_matrix_spmv_hip_msr()

void cs_matrix_spmv_hip_msr	(	cs_matrix_t *	matrix,
		bool	exclude_diag,
		bool	sync,
		cs_real_t	d_x[],
		cs_real_t	d_y[]
	)

Matrix.vector product y = A.x with MSR matrix, scalar ROC version.

Matrix.vector product y = A.x with MSR matrix, scalar HIP version.

Parameters

[in]	matrix	pointer to matrix structure
[in]	exclude_diag	exclude diagonal if true,
[in]	sync	synchronize ghost cells if true
[in]	d_x	multipliying vector values (on device)
[out]	d_y	resulting vector (on device)

cs_matrix_spmv_hip_msr_b()

void cs_matrix_spmv_hip_msr_b	(	cs_matrix_t *	matrix,
		bool	exclude_diag,
		bool	sync,
		cs_real_t	d_x[],
		cs_real_t	d_y[]
	)

Matrix.vector product y = A.x with MSR matrix, block diagonal ROC version.

Matrix.vector product y = A.x with MSR matrix, block diagonal HIP version.

Parameters

[in]	matrix	pointer to matrix structure
[in]	exclude_diag	exclude diagonal if true,
[in]	sync	synchronize ghost cells if true
[in]	d_x	multipliying vector values (on device)
[out]	d_y	resulting vector (on device)

cs_matrix_spmv_hip_native()

void cs_matrix_spmv_hip_native	(	const cs_matrix_t *	matrix,
		bool	exclude_diag,
		bool	sync,
		cs_real_t	d_x[],
		cs_real_t	d_y[]
	)

Matrix.vector product y = A.x with MSR matrix, scalar ROC version.

Parameters

[in]	matrix	pointer to matrix structure
[in]	exclude_diag	exclude diagonal if true,
[in]	sync	synchronize ghost cells if true
[in]	d_x	multipliying vector values (on device)
[out]	d_y	resulting vector (on device)

cs_matrix_spmv_hip_set_stream()

void cs_matrix_spmv_hip_set_stream

(

hipStream_t

stream

)

Assign ROC stream for next ROC-based SpMV operations.

If a stream other than the default stream (0) is used, it will not be synchronized automatically after sparse matrix-vector products (so as to avoid the corresponding overhead), so the caller will need to manage stream syncronization manually.

This function is callable only from ROC code.