search

anishida / lis

Lis (Library of Iterative Solvers for linear systems, pronounced [lis]) is a scalable parallel software library for solving linear equations and eigenvalue problems that arise in the numerical solution of partial differential equations using iterative methods.
http://www.ssisc.org/lis/
Other
98 stars 39 forks source link

Segmentation fault with quad precision #10

Closed yonggwi closed 8 years ago

yonggwi commented 8 years ago

I found that quad precision calculation stop when we choose "-initx_zeros 0" which mean user-defined initial vector.

for instance, we can reproduce it when execute a following example with some options.

./test5 200 2.0 -f quad -initx_zeros 0

Also I found that we can fix this problem by adding

xx = solver->x ;

to around 926 line (in function lis_solver_get_initial_residual() ) of src/solver/lis_solver.c . I think the problem was due to that a variable xx is not properly initialized. However I am not so sure whether this prescription is valid or not. I was wondering if you could give me some advice. Thanks in advance.

Yonggwi Cho

anishida commented 8 years ago

Hi,

Thank you for your notice. As you pointed out, xx must be predefined. Please check 1.6.29, in which the bug is fixed.

Sincerely, Akira Nishida

From: yonggwi [mailto:notifications@github.com] Sent: Friday, October 7, 2016 3:52 PM To: anishida/lis lis@noreply.github.com Subject: [anishida/lis] Segmentation fault with quad precision (#10)

I found that quad precision calculation stop when we choose "-initx_zeros 0" which mean user-defined initial vector. for instance, we can reproduce it when execute a following example with some options. ./test5 200 2 -f quad -initx_zeros 0 Also I found that we can fix this problem by adding xx = solver->x ; to around 926 line (in function lis_solver_get_initial_residual() ) of src/solver/lis_solver.c . I think the problem was due to that a variable xx is not properly initialized. However I am not so sure whether this prescription is valid or not. I was wondering if you could give me some advises. Thanks in advance. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

anishida commented 8 years ago

We have fixed your name in ChangeLog. Thank you again for your cooperation.

Sincerely, Akira Nishida

-----Original Message----- From: Akira NISHIDA [mailto:dolce@yo.rim.or.jp] Sent: Friday, October 7, 2016 6:06 PM To: 'anishida/lis' reply@reply.github.com; 'anishida/lis' lis@noreply.github.com Subject: RE: [anishida/lis] Segmentation fault with quad precision (#10)

Hi,

Thank you for your notice. As you pointed out, xx must be predefined. Please check 1.6.29, in which the bug is fixed.

Sincerely, Akira Nishida

From: yonggwi [mailto:notifications@github.com] Sent: Friday, October 7, 2016 3:52 PM To: anishida/lis lis@noreply.github.com Subject: [anishida/lis] Segmentation fault with quad precision (#10)

I found that quad precision calculation stop when we choose "-initx_zeros 0" which mean user-defined initial vector. for instance, we can reproduce it when execute a following example with some options. ./test5 200 2 -f quad -initx_zeros 0 Also I found that we can fix this problem by adding xx = solver->x ; to around 926 line (in function lis_solver_get_initial_residual() ) of src/solver/lis_solver.c . I think the problem was due to that a variable xx is not properly initialized. However I am not so sure whether this prescription is valid or not. I was wondering if you could give me some advises. Thanks in advance. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

yonggwi commented 8 years ago

Thanks a lot!

Sincerely, Yonggwi Cho