(double)GMRES Segfaults on Tesla C2050

Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)
Post Reply
JosefWeinbub
Posts: 4
Joined: Tue Aug 05, 2014 10:22 am

(double)GMRES Segfaults on Tesla C2050

Post by JosefWeinbub » Tue Aug 05, 2014 10:43 am

Hi Magma-Team,

recently I played around with the iterative solvers shipped with MAGMA 1.5.0-beta3.
In particular, I discovered that the (double)GMRES implementation fails to solve (i.e. Segmentation fault) a coupel of matrices from the Florida Sparse Matrix collection,
among those are:
http://www.cise.ufl.edu/research/sparse ... 20k_A.html
http://www.cise.ufl.edu/research/sparse ... RM07R.html

I am using default solver parameters, a zero-vector initial guess, and a one-vector RHS.
I took the liberty of attaching a minimal example which I used to test MAGMA's GMRES and which throws the segmentation faults on my platform for the above mentioned matrices.

My platform:
NVCC 6.0 (using arch=sm_20 flag)
Fedora 64-bit
Tesla C2050

I am happy to provide additional details if needed or to perform further test runs.

Kind regards, Josef Weinbub
Attachments
magma_gmres.cu.txt
Minimal test application
(686 Bytes) Downloaded 134 times

JosefWeinbub
Posts: 4
Joined: Tue Aug 05, 2014 10:22 am

Re: (double)GMRES Segfaults on Tesla C2050

Post by JosefWeinbub » Thu Sep 04, 2014 7:38 am

Short Update: The listed issues are resolved with the final 1.5.0 release.

Post Reply