Home My Page Projects PaStiX
Summary Activity Forums Lists Docs News Files

Forum: open-discussion

Monitor Forum | Start New Thread Start New Thread
RE: Bug in PETSc driver [ Reply ]
By: Serban Georgescu on 2014-08-20 15:23
[forum:148808]
Glad to be of service!
(I have one question but I will open another thread)

Serban

RE: Bug in PETSc driver [ Reply ]
By: Xavier Lacoste on 2014-08-20 10:52
[forum:148804]
Hello Serban,

You are completly right, i'll apply your fix.

Thanks,

XL.

Bug in PETSc driver [ Reply ]
By: Serban Georgescu on 2014-08-20 09:04
[forum:148803]
Hello XL,

I have found a small bug in the PETSc driver that resutled in symmetric matrices passed using " -petsc_s" to considered unsymmetric (and hence executing LU instead of LLt).

Here is the fix (against the latest version 5.2.2.12):
Inside /src/matrix_drivers/src/petscread.c, function PETScRead, the end should be:
(*Type)[0] = 'R';
(*Type)[1] = 'U';
(*Type)[2] = 'A';
(*Type)[3] = '\0';

Inside src/matrix_drivers/src/read_matrix.c, function read_matrix_common, line 224-225 should be:
if (driver_type == PETSCS) (*type)[1] = 'S';
if (driver_type == PETSCH) (*type)[1] = 'H';

After these modifications all works correctly.
Hopes this helps.

Regards,
Serban