PETSc vs Scipy linear system discrepancy

BlueBrain / AstroVascPy

Vasculature blood flow computation and impact of astrocytic endfeet on vessels

https://astrovascpy.rtfd.io

Apache License 2.0

8 stars 2 forks source link

PETSc vs Scipy linear system discrepancy #20

Closed nico-canta closed 9 months ago

nico-canta commented 10 months ago

Fix for the issue #18

bbpbuildbot commented 10 months ago

Logfiles from GitLab pipeline #181495 (:white_check_mark:) have been uploaded here!

Status and direct links:

bbpbuildbot commented 10 months ago

Logfiles from GitLab pipeline #181752 (:white_check_mark:) have been uploaded here!

Status and direct links:

bbpbuildbot commented 10 months ago

Logfiles from GitLab pipeline #182249 (:white_check_mark:) have been uploaded here!

Status and direct links:

bbpbuildbot commented 10 months ago

Logfiles from GitLab pipeline #183310 (:white_check_mark:) have been uploaded here!

Status and direct links:

nico-canta commented 9 months ago

I added a short comment on issue #18 with a summary of the tests I did.

The new important improvements are here: https://github.com/BlueBrain/AstroVascPy/blob/3b5d7af9bc92592db7b804724ff4e9531909bac1/astrovascpy/bloodflow.py#L710

The more we inform PETSc about the structure of our matrix, the more we get performance improvements. I think there is still room for improvement because the Laplacian is a very "special" matrix. This will be useful when running with the mouse dataset.