cuspla - A CULA wrapper with a high-level interface for CUSP
cuspla is a CULA wrapper used in CUSP library that allows to detect where arrays or matrices are stored (in device or host memory) and the precision used (float or double). Doing so, it will call the appropriate unified function of cuspla that will call one of the associated routine of CULA. cuspla also ease the pain of specify all the parameter (leading dimension, numb of rows, ...) in CULA routines that are based in a LAPACK fashion because cuspla detect the properties of the matrix directly from the array2d object of CUSP.
CULA use a column major order of the matrix hence, when using matrix in CUSP you have to specify the cusp::column_major order for the matrix.
#include <cuspla.h>
cusp::array2d<cusp::double, cusp::device_memory, cusp::column_major> A, B, C; // Filling matrices A and B; // ....
// C = A*B cuspla::gemm(A, B, C);
// C = A'*B cuspla::gemm(A, B, C, true, false);
Copyright (C) 2008, 2009, 2010, 2011, 2012 Free Software
Foundation, Inc.
Permission is granted to make and distribute verbatim copies
of this document provided the copyright notice and this per‐
mission notice are preserved on all copies.
Permission is granted to copy and distribute modified versions
of this document under the conditions for verbatim copying,
provided that the entire resulting derived work is distributed
under the terms of a permission notice identical to this one.
Permission is granted to copy and distribute translations of
this document into another language, under the above condi‐
tions for modified versions, except that this permission
notice may be stated in a translation approved by the Free
Software Foundation.
Of course there is no bug in *cuspla*. But there may be unexpected behaviors.
Go to 'https://github.com/fsquillace/cuspla/issues' you can report directly
this unexpected behaviors.
cusp - Generic Parallel Algorithms for Sparse Matrix and Graph Computations on CUDA
Filippo Squillace sqoox85@gmail.com.