Evgeny Lazutkin <evgeny.lazut...@gmail.com> writes: > Dear all! > > I need your help! First of all, let me please introduce myself. My name > is Evgeny and currently I am scientific researcher at the university in > Germany. > > I am working with optimization of the large scale system. Based on the > algorithm, I have realized that I can use parallel programming with GPU. > The programming language is Python, but I found in Internet, that I can > use pyCUDA to solve my problem. > > The major time consumption in my program is to solve huge linear > equation system in the form *A*X = B*, where *X* and *B* are matrices. > The main idea is to divide the matrices (it is possible due to the > algorithm and structure) and to obtain the acceleration of the > calculation. Unfortunately, I cannot understand how to realize this > under pyCUDA.*Could you please provide the example: how to solve such > the system within pyCUDA?* > > I have mentioned, that there is "CUDA SciKit" [1], which provides > Python interfaces to a subset of the functions in the CUDA, CUDART, > CUBLAS, and CUFFT libraries distributed as part of NVIDIA's CUDA > Programming Toolkit, as well as interfaces to select functions > in the basic and premium versions of the CULA Toolkit. Is it a correct > way? Or did I make a mistake? Probably you have a better solution? But I > still need an example.
Here are two starting points: http://lebedov.github.io/scikits.cuda/generated/scikits.cuda.linalg.cho_factor.html (That's Cholesky, you want LU.) http://lebedov.github.io/scikits.cuda/generated/scikits.cuda.cula.culaDeviceDgetrf.html#scikits.cuda.cula.culaDeviceDgetrf (This performs LU.) HTH, Andreas
pgpoXEAq9SocS.pgp
Description: PGP signature
_______________________________________________ PyCUDA mailing list PyCUDA@tiker.net http://lists.tiker.net/listinfo/pycuda
