Question on BiCGStab Solver

Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)
Post Reply
dannnnn
Posts: 37
Joined: Tue Mar 03, 2015 3:41 pm

Question on BiCGStab Solver

Post by dannnnn » Tue Mar 31, 2015 4:02 pm

I run the BiCGStab solver with preconditioner ILU using NON SPD matrix bayer10.mtx with input:
--solver 5 --verbose 1 --format 0 --maxiter 1000 --mscale 0 --precond 2 bayer10.mtx

I got result:
MAGMA 1.6.1 compiled for CUDA capability >= 2.0
CUDA runtime 6050, driver 6050. OpenMP threads 8. MKL 11.2.2, MKL threads 4.
ndevices 1
device 0: GeForce GTX 980, 1342.0 MHz clock, 4096.0 MB memory, capability 5.2
Usage: C:\Users\Administrator\Documents\Visual Studio 2010\Projects\magma_test\x64\Debug\magmaTest.exe [options] [-h|--help] matrices

# Reading sparse matrix from file (bayer10.mtx): done

# matrix info: 13436-by-13436 with 71594 nonzeros

Operator A is not postive definite. (Ar,r) = -300187585.639264
#=============================================================#
# ILU-BiCGStab performance analysis every 1 iteration
# iter || residual-nrm2 || runtime
#=============================================================#
0 || -6.277439e+066 || -6277438562204192500000000000000000000000000000000000000000000000000.000000
#=============================================================#
# BiCGStab solver summary:
# initial residual: 6.478288e+003
# iterations: 0
# iterative residual: -9.255963e+061
# exact final residual: -9.255963e+061
# runtime: -92559631349317831000000000000000000000000000000000000000000000.0000 sec
# solver info: 0
#=============================================================#

Link to matrix bayer10.mtx : http://www.cise.ufl.edu/research/sparse ... yer10.html

I was told the BiCGStab solver in the library does not require matrix to be positive definite, but I got such error. I would like to know if there is anything I miss using this solver? Thank you!
Top
hartwig anzt
Posts: 90
Joined: Tue Sep 02, 2014 5:44 pm

Re: Question on BiCGStab Solver

Post by hartwig anzt » Tue Mar 31, 2015 4:08 pm

Sorry- this bug is resolved in the next release. Please just comment out the respective lines in the source file. Thanks!
Top
dannnnn
Posts: 37
Joined: Tue Mar 03, 2015 3:41 pm

Re: Question on BiCGStab Solver

Post by dannnnn » Tue Mar 31, 2015 4:30 pm

Thank you!
Top
hartwig anzt
Posts: 90
Joined: Tue Sep 02, 2014 5:44 pm

Re: Question on BiCGStab Solver

Post by hartwig anzt » Tue Mar 31, 2015 4:41 pm

I guess you have already figured out, but in case: the lines where the solver is exited due to non-SPD ness. Basically, the complete SPD check can be removed.
Top
Post Reply

Return to “User discussion”