MAGMA  2.3.0
Matrix Algebra for GPU and Multicore Architectures
 All Classes Files Functions Friends Groups Pages
getmatrix_transpose: GPU => CPU

Functions

void magmablas_cgetmatrix_transpose (magma_int_t m, magma_int_t n, magma_int_t nb, magmaFloatComplex_const_ptr dAT, magma_int_t ldda, magmaFloatComplex *hA, magma_int_t lda, magmaFloatComplex_ptr dwork, magma_int_t lddw, magma_queue_t queues[2])
 Copy and transpose matrix dAT on GPU device to hA on CPU host. More...
 
void magmablas_cgetmatrix_transpose_mgpu (magma_int_t ngpu, magma_int_t m, magma_int_t n, magma_int_t nb, magmaFloatComplex_const_ptr const dAT[], magma_int_t ldda, magmaFloatComplex *hA, magma_int_t lda, magmaFloatComplex_ptr dwork[], magma_int_t lddw, magma_queue_t queues[][2])
 Copy and transpose matrix dAT, which is distributed row block cyclic over multiple GPUs, to hA on CPU host. More...
 
void magmablas_dgetmatrix_transpose (magma_int_t m, magma_int_t n, magma_int_t nb, magmaDouble_const_ptr dAT, magma_int_t ldda, double *hA, magma_int_t lda, magmaDouble_ptr dwork, magma_int_t lddw, magma_queue_t queues[2])
 Copy and transpose matrix dAT on GPU device to hA on CPU host. More...
 
void magmablas_dgetmatrix_transpose_mgpu (magma_int_t ngpu, magma_int_t m, magma_int_t n, magma_int_t nb, magmaDouble_const_ptr const dAT[], magma_int_t ldda, double *hA, magma_int_t lda, magmaDouble_ptr dwork[], magma_int_t lddw, magma_queue_t queues[][2])
 Copy and transpose matrix dAT, which is distributed row block cyclic over multiple GPUs, to hA on CPU host. More...
 
void magmablas_sgetmatrix_transpose (magma_int_t m, magma_int_t n, magma_int_t nb, magmaFloat_const_ptr dAT, magma_int_t ldda, float *hA, magma_int_t lda, magmaFloat_ptr dwork, magma_int_t lddw, magma_queue_t queues[2])
 Copy and transpose matrix dAT on GPU device to hA on CPU host. More...
 
void magmablas_sgetmatrix_transpose_mgpu (magma_int_t ngpu, magma_int_t m, magma_int_t n, magma_int_t nb, magmaFloat_const_ptr const dAT[], magma_int_t ldda, float *hA, magma_int_t lda, magmaFloat_ptr dwork[], magma_int_t lddw, magma_queue_t queues[][2])
 Copy and transpose matrix dAT, which is distributed row block cyclic over multiple GPUs, to hA on CPU host. More...
 
void magmablas_zgetmatrix_transpose (magma_int_t m, magma_int_t n, magma_int_t nb, magmaDoubleComplex_const_ptr dAT, magma_int_t ldda, magmaDoubleComplex *hA, magma_int_t lda, magmaDoubleComplex_ptr dwork, magma_int_t lddw, magma_queue_t queues[2])
 Copy and transpose matrix dAT on GPU device to hA on CPU host. More...
 
void magmablas_zgetmatrix_transpose_mgpu (magma_int_t ngpu, magma_int_t m, magma_int_t n, magma_int_t nb, magmaDoubleComplex_const_ptr const dAT[], magma_int_t ldda, magmaDoubleComplex *hA, magma_int_t lda, magmaDoubleComplex_ptr dwork[], magma_int_t lddw, magma_queue_t queues[][2])
 Copy and transpose matrix dAT, which is distributed row block cyclic over multiple GPUs, to hA on CPU host. More...
 

Detailed Description

Function Documentation

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

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

Parameters
[in]mNumber of rows of output matrix hA. m >= 0.
[in]nNumber of columns of output matrix hA. n >= 0.
[in]nbBlock size. nb >= 0.
[in]dATThe n-by-m matrix A^T on the GPU, of dimension (ldda,m).
[in]lddaLeading dimension of matrix dAT. ldda >= n.
[out]hAThe m-by-n matrix A on the CPU, of dimension (lda,n).
[in]ldaLeading dimension of matrix hA. lda >= m.
[out]dworkWorkspace on the GPU, of dimension (2*lddw*nb).
[in]lddwLeading dimension of dwork. lddw >= m.
[in]queuesArray of two queues, to pipeline operation.
void magmablas_cgetmatrix_transpose_mgpu ( magma_int_t  ngpu,
magma_int_t  m,
magma_int_t  n,
magma_int_t  nb,
magmaFloatComplex_const_ptr const  dAT[],
magma_int_t  ldda,
magmaFloatComplex *  hA,
magma_int_t  lda,
magmaFloatComplex_ptr  dwork[],
magma_int_t  lddw,
magma_queue_t  queues[][2] 
)

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

Parameters
[in]ngpuNumber of GPUs over which dAT is distributed.
[in]mNumber of rows of output matrix hA. m >= 0.
[in]nNumber of columns of output matrix hA. n >= 0.
[in]nbBlock size. nb >= 0.
[in]dATArray 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]lddaLeading dimension of each matrix dAT on each GPU. ngpu*ldda >= n.
[out]hAThe m-by-n matrix A on the CPU, of dimension (lda,n).
[in]ldaLeading dimension of matrix hA. lda >= m.
[out]dworkArray of ngpu pointers, one per GPU, that store the workspaces on each GPU, each of dimension (2*lddw*nb).
[in]lddwLeading dimension of dwork. lddw >= m.
[in]queues2D array of dimension (ngpu,2), with two queues per GPU.
void magmablas_dgetmatrix_transpose ( magma_int_t  m,
magma_int_t  n,
magma_int_t  nb,
magmaDouble_const_ptr  dAT,
magma_int_t  ldda,
double *  hA,
magma_int_t  lda,
magmaDouble_ptr  dwork,
magma_int_t  lddw,
magma_queue_t  queues[2] 
)

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

Parameters
[in]mNumber of rows of output matrix hA. m >= 0.
[in]nNumber of columns of output matrix hA. n >= 0.
[in]nbBlock size. nb >= 0.
[in]dATThe n-by-m matrix A^T on the GPU, of dimension (ldda,m).
[in]lddaLeading dimension of matrix dAT. ldda >= n.
[out]hAThe m-by-n matrix A on the CPU, of dimension (lda,n).
[in]ldaLeading dimension of matrix hA. lda >= m.
[out]dworkWorkspace on the GPU, of dimension (2*lddw*nb).
[in]lddwLeading dimension of dwork. lddw >= m.
[in]queuesArray of two queues, to pipeline operation.
void magmablas_dgetmatrix_transpose_mgpu ( magma_int_t  ngpu,
magma_int_t  m,
magma_int_t  n,
magma_int_t  nb,
magmaDouble_const_ptr const  dAT[],
magma_int_t  ldda,
double *  hA,
magma_int_t  lda,
magmaDouble_ptr  dwork[],
magma_int_t  lddw,
magma_queue_t  queues[][2] 
)

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

Parameters
[in]ngpuNumber of GPUs over which dAT is distributed.
[in]mNumber of rows of output matrix hA. m >= 0.
[in]nNumber of columns of output matrix hA. n >= 0.
[in]nbBlock size. nb >= 0.
[in]dATArray 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]lddaLeading dimension of each matrix dAT on each GPU. ngpu*ldda >= n.
[out]hAThe m-by-n matrix A on the CPU, of dimension (lda,n).
[in]ldaLeading dimension of matrix hA. lda >= m.
[out]dworkArray of ngpu pointers, one per GPU, that store the workspaces on each GPU, each of dimension (2*lddw*nb).
[in]lddwLeading dimension of dwork. lddw >= m.
[in]queues2D array of dimension (ngpu,2), with two queues per GPU.
void magmablas_sgetmatrix_transpose ( magma_int_t  m,
magma_int_t  n,
magma_int_t  nb,
magmaFloat_const_ptr  dAT,
magma_int_t  ldda,
float *  hA,
magma_int_t  lda,
magmaFloat_ptr  dwork,
magma_int_t  lddw,
magma_queue_t  queues[2] 
)

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

Parameters
[in]mNumber of rows of output matrix hA. m >= 0.
[in]nNumber of columns of output matrix hA. n >= 0.
[in]nbBlock size. nb >= 0.
[in]dATThe n-by-m matrix A^T on the GPU, of dimension (ldda,m).
[in]lddaLeading dimension of matrix dAT. ldda >= n.
[out]hAThe m-by-n matrix A on the CPU, of dimension (lda,n).
[in]ldaLeading dimension of matrix hA. lda >= m.
[out]dworkWorkspace on the GPU, of dimension (2*lddw*nb).
[in]lddwLeading dimension of dwork. lddw >= m.
[in]queuesArray of two queues, to pipeline operation.
void magmablas_sgetmatrix_transpose_mgpu ( magma_int_t  ngpu,
magma_int_t  m,
magma_int_t  n,
magma_int_t  nb,
magmaFloat_const_ptr const  dAT[],
magma_int_t  ldda,
float *  hA,
magma_int_t  lda,
magmaFloat_ptr  dwork[],
magma_int_t  lddw,
magma_queue_t  queues[][2] 
)

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

Parameters
[in]ngpuNumber of GPUs over which dAT is distributed.
[in]mNumber of rows of output matrix hA. m >= 0.
[in]nNumber of columns of output matrix hA. n >= 0.
[in]nbBlock size. nb >= 0.
[in]dATArray 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]lddaLeading dimension of each matrix dAT on each GPU. ngpu*ldda >= n.
[out]hAThe m-by-n matrix A on the CPU, of dimension (lda,n).
[in]ldaLeading dimension of matrix hA. lda >= m.
[out]dworkArray of ngpu pointers, one per GPU, that store the workspaces on each GPU, each of dimension (2*lddw*nb).
[in]lddwLeading dimension of dwork. lddw >= m.
[in]queues2D array of dimension (ngpu,2), with two queues per GPU.
void magmablas_zgetmatrix_transpose ( magma_int_t  m,
magma_int_t  n,
magma_int_t  nb,
magmaDoubleComplex_const_ptr  dAT,
magma_int_t  ldda,
magmaDoubleComplex *  hA,
magma_int_t  lda,
magmaDoubleComplex_ptr  dwork,
magma_int_t  lddw,
magma_queue_t  queues[2] 
)

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

Parameters
[in]mNumber of rows of output matrix hA. m >= 0.
[in]nNumber of columns of output matrix hA. n >= 0.
[in]nbBlock size. nb >= 0.
[in]dATThe n-by-m matrix A^T on the GPU, of dimension (ldda,m).
[in]lddaLeading dimension of matrix dAT. ldda >= n.
[out]hAThe m-by-n matrix A on the CPU, of dimension (lda,n).
[in]ldaLeading dimension of matrix hA. lda >= m.
[out]dworkWorkspace on the GPU, of dimension (2*lddw*nb).
[in]lddwLeading dimension of dwork. lddw >= m.
[in]queuesArray of two queues, to pipeline operation.
void magmablas_zgetmatrix_transpose_mgpu ( magma_int_t  ngpu,
magma_int_t  m,
magma_int_t  n,
magma_int_t  nb,
magmaDoubleComplex_const_ptr const  dAT[],
magma_int_t  ldda,
magmaDoubleComplex *  hA,
magma_int_t  lda,
magmaDoubleComplex_ptr  dwork[],
magma_int_t  lddw,
magma_queue_t  queues[][2] 
)

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

Parameters
[in]ngpuNumber of GPUs over which dAT is distributed.
[in]mNumber of rows of output matrix hA. m >= 0.
[in]nNumber of columns of output matrix hA. n >= 0.
[in]nbBlock size. nb >= 0.
[in]dATArray 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]lddaLeading dimension of each matrix dAT on each GPU. ngpu*ldda >= n.
[out]hAThe m-by-n matrix A on the CPU, of dimension (lda,n).
[in]ldaLeading dimension of matrix hA. lda >= m.
[out]dworkArray of ngpu pointers, one per GPU, that store the workspaces on each GPU, each of dimension (2*lddw*nb).
[in]lddwLeading dimension of dwork. lddw >= m.
[in]queues2D array of dimension (ngpu,2), with two queues per GPU.