search

termi-official / Thunderbolt.jl

A modular shared-memory high-performance framework for multiscale cardiac multiphysics simulations.
MIT License
14 stars 1 forks source link

Parallel SpMV #9

Open termi-official opened 11 months ago

termi-official commented 11 months ago

Reintroduce https://github.com/BacAmorim/ThreadedSparseCSR.jl/blob/9a26079728f15e23221001569f6337849256d059/src/batch_matmul.jl and also introduce it for SparseCSC. This should be as easy as providing

  1. [x] a function like threaded_mul! (#39)
  2. [ ] benchmarks
  3. [ ] tests that threaded_mul! does give the same result as mul!
  4. [ ] a way to pass which multiplication should be used by the operator
termi-official commented 7 months ago

Using #39 with Krylov.cg! does scale extreme poorly. Needs investigation.