Matrix assembly speed-up

Registered by Anders Logg on 2009-10-22

Instead of computing and inserting the element tensor on each cell. We may compute the element tensor for a batch of elements in one single operation. This may potentially speedup both the element tensor computation and insertion.

In particular for tensor representation, the element tensors for a batch of elements may be computed using a single Level 3 BLAS call (dgemm).

This may require the addition of a new UFC function called tabulate_tensors.

Blueprint information

Status:
Not started
Approver:
None
Priority:
Low
Drafter:
None
Direction:
Needs approval
Assignee:
None
Definition:
Discussion
Series goal:
None
Implementation:
Unknown
Milestone target:
None

Related branches

Sprints

Whiteboard

GNW: A batch insertion function for matrices seems to have appeared in PETSc 3.2:

http://www.mcs.anl.gov/petsc/petsc-as/snapshots/petsc-current/docs/manualpages/Mat/MatSetValuesBatch.html

Fredrik V: I am in the progress of integrating MatSetValuesBatch with the PETScMatrix wrapper. That function only yields speedup on GPUs as far as I have understood from the PETSc team .

(?)

Work Items

This blueprint contains Public information 
Everyone can see this information.

Subscribers

No subscribers.