Functions
void	magmablas_csetmatrix_transpose (magma_int_t m, magma_int_t n, magma_int_t nb, const magmaFloatComplex *hA, magma_int_t lda, magmaFloatComplex_ptr dAT, magma_int_t ldda, magmaFloatComplex_ptr dwork, magma_int_t lddw, magma_queue_t queues[2])
	Copy and transpose matrix hA on CPU host to dAT on GPU device. More...

void	magmablas_csetmatrix_transpose_mgpu (magma_int_t ngpu, magma_int_t m, magma_int_t n, magma_int_t nb, const magmaFloatComplex *hA, magma_int_t lda, magmaFloatComplex_ptr dAT[], magma_int_t ldda, magmaFloatComplex_ptr dwork[], magma_int_t lddw, magma_queue_t queues[][2])
	Copy and transpose matrix hA on CPU host to dAT, which is distributed row block cyclic over multiple GPUs. More...

void	magmablas_dsetmatrix_transpose (magma_int_t m, magma_int_t n, magma_int_t nb, const double *hA, magma_int_t lda, magmaDouble_ptr dAT, magma_int_t ldda, magmaDouble_ptr dwork, magma_int_t lddw, magma_queue_t queues[2])
	Copy and transpose matrix hA on CPU host to dAT on GPU device. More...

void	magmablas_dsetmatrix_transpose_mgpu (magma_int_t ngpu, magma_int_t m, magma_int_t n, magma_int_t nb, const double *hA, magma_int_t lda, magmaDouble_ptr dAT[], magma_int_t ldda, magmaDouble_ptr dwork[], magma_int_t lddw, magma_queue_t queues[][2])
	Copy and transpose matrix hA on CPU host to dAT, which is distributed row block cyclic over multiple GPUs. More...

void	magmablas_ssetmatrix_transpose (magma_int_t m, magma_int_t n, magma_int_t nb, const float *hA, magma_int_t lda, magmaFloat_ptr dAT, magma_int_t ldda, magmaFloat_ptr dwork, magma_int_t lddw, magma_queue_t queues[2])
	Copy and transpose matrix hA on CPU host to dAT on GPU device. More...

void	magmablas_ssetmatrix_transpose_mgpu (magma_int_t ngpu, magma_int_t m, magma_int_t n, magma_int_t nb, const float *hA, magma_int_t lda, magmaFloat_ptr dAT[], magma_int_t ldda, magmaFloat_ptr dwork[], magma_int_t lddw, magma_queue_t queues[][2])
	Copy and transpose matrix hA on CPU host to dAT, which is distributed row block cyclic over multiple GPUs. More...

void	magmablas_zsetmatrix_transpose (magma_int_t m, magma_int_t n, magma_int_t nb, const magmaDoubleComplex *hA, magma_int_t lda, magmaDoubleComplex_ptr dAT, magma_int_t ldda, magmaDoubleComplex_ptr dwork, magma_int_t lddw, magma_queue_t queues[2])
	Copy and transpose matrix hA on CPU host to dAT on GPU device. More...

void	magmablas_zsetmatrix_transpose_mgpu (magma_int_t ngpu, magma_int_t m, magma_int_t n, magma_int_t nb, const magmaDoubleComplex *hA, magma_int_t lda, magmaDoubleComplex_ptr dAT[], magma_int_t ldda, magmaDoubleComplex_ptr dwork[], magma_int_t lddw, magma_queue_t queues[][2])
	Copy and transpose matrix hA on CPU host to dAT, which is distributed row block cyclic over multiple GPUs. More...

Detailed Description

Function Documentation

void magmablas_csetmatrix_transpose	(	magma_int_t	m,
		magma_int_t	n,
		magma_int_t	nb,
		const magmaFloatComplex *	hA,
		magma_int_t	lda,
		magmaFloatComplex_ptr	dAT,
		magma_int_t	ldda,
		magmaFloatComplex_ptr	dwork,
		magma_int_t	lddw,
		magma_queue_t	queues[2]
	)

Copy and transpose matrix hA on CPU host to dAT on GPU device.

Parameters

[in]	m	Number of rows of input matrix hA. m >= 0.
[in]	n	Number of columns of input matrix hA. n >= 0.
[in]	nb	Block size. nb >= 0.
[in]	hA	The m-by-n matrix A on the CPU, of dimension (lda,n).
[in]	lda	Leading dimension of matrix hA. lda >= m.
[out]	dAT	The n-by-m matrix A^T on the GPU, of dimension (ldda,m).
[in]	ldda	Leading dimension of matrix dAT. ldda >= n.
[out]	dwork	Workspace on the GPU, of dimension (2lddwnb).
[in]	lddw	Leading dimension of dwork. lddw >= m.
[in]	queues	Array of two queues, to pipeline operation.

void magmablas_csetmatrix_transpose_mgpu	(	magma_int_t	ngpu,
		magma_int_t	m,
		magma_int_t	n,
		magma_int_t	nb,
		const magmaFloatComplex *	hA,
		magma_int_t	lda,
		magmaFloatComplex_ptr	dAT[],
		magma_int_t	ldda,
		magmaFloatComplex_ptr	dwork[],
		magma_int_t	lddw,
		magma_queue_t	queues[][2]
	)

Copy and transpose matrix hA on CPU host to dAT, which is distributed row block cyclic over multiple GPUs.

Parameters

[in]	ngpu	Number of GPUs over which dAT is distributed.
[in]	m	Number of rows of input matrix hA. m >= 0.
[in]	n	Number of columns of input matrix hA. n >= 0.
[in]	nb	Block size. nb >= 0.
[out]	hA	The m-by-n matrix A on the CPU, of dimension (lda,n).
[in]	lda	Leading dimension of matrix hA. lda >= m.
[in]	dAT	Array of ngpu pointers, one per GPU, that store the disributed n-by-m matrix A^T on the GPUs, each of dimension (ldda,m).
[in]	ldda	Leading dimension of each matrix dAT on each GPU. ngpu*ldda >= n.
[out]	dwork	Array of ngpu pointers, one per GPU, that store the workspaces on each GPU, each of dimension (2lddwnb).
[in]	lddw	Leading dimension of dwork. lddw >= m.
[in]	queues	2D array of dimension (ngpu,2), with two queues per GPU.

void magmablas_dsetmatrix_transpose	(	magma_int_t	m,
		magma_int_t	n,
		magma_int_t	nb,
		const double *	hA,
		magma_int_t	lda,
		magmaDouble_ptr	dAT,
		magma_int_t	ldda,
		magmaDouble_ptr	dwork,
		magma_int_t	lddw,
		magma_queue_t	queues[2]
	)

Copy and transpose matrix hA on CPU host to dAT on GPU device.

Parameters

[in]	m	Number of rows of input matrix hA. m >= 0.
[in]	n	Number of columns of input matrix hA. n >= 0.
[in]	nb	Block size. nb >= 0.
[in]	hA	The m-by-n matrix A on the CPU, of dimension (lda,n).
[in]	lda	Leading dimension of matrix hA. lda >= m.
[out]	dAT	The n-by-m matrix A^T on the GPU, of dimension (ldda,m).
[in]	ldda	Leading dimension of matrix dAT. ldda >= n.
[out]	dwork	Workspace on the GPU, of dimension (2lddwnb).
[in]	lddw	Leading dimension of dwork. lddw >= m.
[in]	queues	Array of two queues, to pipeline operation.

void magmablas_dsetmatrix_transpose_mgpu	(	magma_int_t	ngpu,
		magma_int_t	m,
		magma_int_t	n,
		magma_int_t	nb,
		const double *	hA,
		magma_int_t	lda,
		magmaDouble_ptr	dAT[],
		magma_int_t	ldda,
		magmaDouble_ptr	dwork[],
		magma_int_t	lddw,
		magma_queue_t	queues[][2]
	)

Copy and transpose matrix hA on CPU host to dAT, which is distributed row block cyclic over multiple GPUs.

Parameters

[in]	ngpu	Number of GPUs over which dAT is distributed.
[in]	m	Number of rows of input matrix hA. m >= 0.
[in]	n	Number of columns of input matrix hA. n >= 0.
[in]	nb	Block size. nb >= 0.
[out]	hA	The m-by-n matrix A on the CPU, of dimension (lda,n).
[in]	lda	Leading dimension of matrix hA. lda >= m.
[in]	dAT	Array of ngpu pointers, one per GPU, that store the disributed n-by-m matrix A^T on the GPUs, each of dimension (ldda,m).
[in]	ldda	Leading dimension of each matrix dAT on each GPU. ngpu*ldda >= n.
[out]	dwork	Array of ngpu pointers, one per GPU, that store the workspaces on each GPU, each of dimension (2lddwnb).
[in]	lddw	Leading dimension of dwork. lddw >= m.
[in]	queues	2D array of dimension (ngpu,2), with two queues per GPU.

void magmablas_ssetmatrix_transpose	(	magma_int_t	m,
		magma_int_t	n,
		magma_int_t	nb,
		const float *	hA,
		magma_int_t	lda,
		magmaFloat_ptr	dAT,
		magma_int_t	ldda,
		magmaFloat_ptr	dwork,
		magma_int_t	lddw,
		magma_queue_t	queues[2]
	)

Copy and transpose matrix hA on CPU host to dAT on GPU device.

Parameters

[in]	m	Number of rows of input matrix hA. m >= 0.
[in]	n	Number of columns of input matrix hA. n >= 0.
[in]	nb	Block size. nb >= 0.
[in]	hA	The m-by-n matrix A on the CPU, of dimension (lda,n).
[in]	lda	Leading dimension of matrix hA. lda >= m.
[out]	dAT	The n-by-m matrix A^T on the GPU, of dimension (ldda,m).
[in]	ldda	Leading dimension of matrix dAT. ldda >= n.
[out]	dwork	Workspace on the GPU, of dimension (2lddwnb).
[in]	lddw	Leading dimension of dwork. lddw >= m.
[in]	queues	Array of two queues, to pipeline operation.

void magmablas_ssetmatrix_transpose_mgpu	(	magma_int_t	ngpu,
		magma_int_t	m,
		magma_int_t	n,
		magma_int_t	nb,
		const float *	hA,
		magma_int_t	lda,
		magmaFloat_ptr	dAT[],
		magma_int_t	ldda,
		magmaFloat_ptr	dwork[],
		magma_int_t	lddw,
		magma_queue_t	queues[][2]
	)

Copy and transpose matrix hA on CPU host to dAT, which is distributed row block cyclic over multiple GPUs.

Parameters

[in]	ngpu	Number of GPUs over which dAT is distributed.
[in]	m	Number of rows of input matrix hA. m >= 0.
[in]	n	Number of columns of input matrix hA. n >= 0.
[in]	nb	Block size. nb >= 0.
[out]	hA	The m-by-n matrix A on the CPU, of dimension (lda,n).
[in]	lda	Leading dimension of matrix hA. lda >= m.
[in]	dAT	Array of ngpu pointers, one per GPU, that store the disributed n-by-m matrix A^T on the GPUs, each of dimension (ldda,m).
[in]	ldda	Leading dimension of each matrix dAT on each GPU. ngpu*ldda >= n.
[out]	dwork	Array of ngpu pointers, one per GPU, that store the workspaces on each GPU, each of dimension (2lddwnb).
[in]	lddw	Leading dimension of dwork. lddw >= m.
[in]	queues	2D array of dimension (ngpu,2), with two queues per GPU.

void magmablas_zsetmatrix_transpose	(	magma_int_t	m,
		magma_int_t	n,
		magma_int_t	nb,
		const magmaDoubleComplex *	hA,
		magma_int_t	lda,
		magmaDoubleComplex_ptr	dAT,
		magma_int_t	ldda,
		magmaDoubleComplex_ptr	dwork,
		magma_int_t	lddw,
		magma_queue_t	queues[2]
	)

Copy and transpose matrix hA on CPU host to dAT on GPU device.

Parameters

[in]	m	Number of rows of input matrix hA. m >= 0.
[in]	n	Number of columns of input matrix hA. n >= 0.
[in]	nb	Block size. nb >= 0.
[in]	hA	The m-by-n matrix A on the CPU, of dimension (lda,n).
[in]	lda	Leading dimension of matrix hA. lda >= m.
[out]	dAT	The n-by-m matrix A^T on the GPU, of dimension (ldda,m).
[in]	ldda	Leading dimension of matrix dAT. ldda >= n.
[out]	dwork	Workspace on the GPU, of dimension (2lddwnb).
[in]	lddw	Leading dimension of dwork. lddw >= m.
[in]	queues	Array of two queues, to pipeline operation.

void magmablas_zsetmatrix_transpose_mgpu	(	magma_int_t	ngpu,
		magma_int_t	m,
		magma_int_t	n,
		magma_int_t	nb,
		const magmaDoubleComplex *	hA,
		magma_int_t	lda,
		magmaDoubleComplex_ptr	dAT[],
		magma_int_t	ldda,
		magmaDoubleComplex_ptr	dwork[],
		magma_int_t	lddw,
		magma_queue_t	queues[][2]
	)

Copy and transpose matrix hA on CPU host to dAT, which is distributed row block cyclic over multiple GPUs.

Parameters

[in]	ngpu	Number of GPUs over which dAT is distributed.
[in]	m	Number of rows of input matrix hA. m >= 0.
[in]	n	Number of columns of input matrix hA. n >= 0.
[in]	nb	Block size. nb >= 0.
[out]	hA	The m-by-n matrix A on the CPU, of dimension (lda,n).
[in]	lda	Leading dimension of matrix hA. lda >= m.
[in]	dAT	Array of ngpu pointers, one per GPU, that store the disributed n-by-m matrix A^T on the GPUs, each of dimension (ldda,m).
[in]	ldda	Leading dimension of each matrix dAT on each GPU. ngpu*ldda >= n.
[out]	dwork	Array of ngpu pointers, one per GPU, that store the workspaces on each GPU, each of dimension (2lddwnb).
[in]	lddw	Leading dimension of dwork. lddw >= m.
[in]	queues	2D array of dimension (ngpu,2), with two queues per GPU.

Functions

Detailed Description

Function Documentation