Re: [Viennacl-support] Possible bug in inplace_solve

[Karl R. <ru...@iu...>]

Hi!

Sorry for the late approval of your support email, the junk filter 
caught it...

Hmm, I'll have a look at the inplace_solve issue and let you know by Monday.

Best regards,
Karli


On 06/30/2010 11:51 PM, mel...@ms... wrote:
> Hi, I've just recently discovered your project and I was toying around
> LU factorization and forward-back substitution, and in my programs, the
> inplace_solve function seems to break down when I invert a matrix larger
> than about 225x225.
>
> I'm using linux and an NVIDIA tesla C1060.
>
> Basically, when I run the viennacl inplace solve and the ublas inplace
> solve they produce results differing by about one index (if my matrix is
> (200-400)^2 ) and more indices if the matrix is larger. I tested the
> lu_factorize function, and that produces similar results to ublas (up to
> a relatively high level of significance), so I'm fairly certain the
> problem I'm having lies with the inplace_solve function.
>
> Here is the code I'm running:
>
> --------------------------------------------------------------------------------------------------
>
> #include <boost/numeric/ublas/triangular.hpp>
> #include <boost/numeric/ublas/matrix_sparse.hpp>
> #include <boost/numeric/ublas/vector.hpp>
> #include <boost/numeric/ublas/matrix.hpp>
> #include <boost/numeric/ublas/io.hpp>
> #include <boost/numeric/ublas/lu.hpp>
> #include <fstream>
>
> #include "viennacl/vector.hpp"
> #include "viennacl/matrix.hpp"
> #include "viennacl/linalg/direct_solve.hpp"
> #include "benchmark-utils.hpp"
>
> using namespace boost::numeric;
>
> int main()
> {
> int N = 500;
> ublas::matrix<double> A(N,N);
> ublas::vector<double> b(N);
> ublas::vector<double> b2(N);
> viennacl::matrix<double> gpu_A(N,N);
> viennacl::vector<double> gpu_b(N);
>
> for (int i=0;i<N;i++)
> {
> for (int j=i;j<N;j++)
> {
> A(i,j)=1;
> }
> b(i)=N-i;
> }
> copy(A,gpu_A);
> copy(b,gpu_b);
>
> ublas::inplace_solve(A,b,ublas::upper_tag());
> viennacl::linalg::inplace_solve(gpu_A,gpu_b,viennacl::linalg::upper_tag());
> ublas::inplace_solve(A2,b3,ublas::upper_tag());
> copy(gpu_b,b2);
>
>
> double err = 0;
>
>
> for (int i=0;i<N;i++)
> {
> if (abs(b2(i)-b(i))>err)
> {
> err=abs(b2(i)-b(i));
> std::cout<<"At index "<<i<<std::endl;
> std::cout<<"CPU= "<<b(i)<<"  and"<<std::endl;
> std::cout<<"GPU= "<<b2(i)<<std::endl;
> }
> }
>
> std::cout<<err<<std::endl;
> }
>
> --------------------------------------------------------------------------------------------------
>
> Every time I run this there are one or two errors at different indices.
> Am I doing something wrong? Do you have any suggestions?
>
> Thanks in advance,
> Andrew Melfi
>
>
>
>
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Sprint
> What will you do first with EVO, the first 4G phone?
> Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first
>
>
>
> _______________________________________________
> Viennacl-support mailing list
> Vie...@li...
> https://lists.sourceforge.net/lists/listinfo/viennacl-support