Skip to content

Batched sparse QR factorizations and solves with cusolver #1178

@bamos

Description

@bamos

These should significantly speed up qpth in some cases when sparse matrices can be used. Here is a cusolver reference on these functions. This issue depends on the other core sparse functions being added (I think at least PR #1147, please add other dependencies here so I'll know when I can start) so I won't immediately start trying to add these. Also please post a reference if a good cusolver wrapper usage example gets added that we can use as a reference for wrapping these QR functions.

Following the naming conventions of the batched LU factorizations and solves of btrifact and btrisolve, I propose bqrfact and bqrsolve for the names and for now we can just implement them for sparse tensors.

\cc @nikitaved @pearu @cpuhrsch @jianyuh @mruberry @walterddr @IvanYashchuk @xwang233 @lezcano @vincentqb @vishwakftw @ssnl @zkolter

Metadata

Metadata

Assignees

Labels

StalefeatureA request for a proper, new feature.module: linear algebraIssues related to specialized linear algebra operations in PyTorch; includes matrix multiply matmulmodule: sparseRelated to torch.sparsetriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate module

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions