benchmark more blas/lapack functions

OpenMathLib / BLAS-Benchmarks

MIT License

0 stars 0 forks source link

benchmark more blas/lapack functions #11

Open ev-br opened 5 months ago

ev-br commented 5 months ago

Currently we benchmark:

BLAS level 1

ddot
?nrm2
?axpy

BLAS level 3

?gemm
?syrk

LAPACK:

?gesv (linalg.solve)
?gesdd (linalg.svd)
?syev (linalg.eigh)

@martin-frbg suggested it'd be useful to add

[x] ?gemv
[x] ?gbmv

for the banded matrix-vector multiply, what is more interesting to scale: matrix size or the number of bands?

[ ] cdotc -- complex-valued BLAS level 1

martin-frbg commented 5 months ago

good question - maybe matrix size for now to keep it simple, although number of bands would also be interesting. GBMV is a curious case where the implementation is unchanged from the original GotoBLAS and may not have been evaluated for (multithreaded) performance at all.

ev-br commented 5 months ago

OK, gbmv benchmarks now run for a standard range of sizes from 100 to 1000 and kl=1, 2, 3 (kl=ku, so it's 3, 5, and 7 bands). Web display is at http://www.openmathlib.org/BLAS-Benchmarks/#benchmarks.gbmv.time_gbmv and raw data is in the usual places from GH actions, https://github.com/OpenMathLib/BLAS-Benchmarks/actions/runs/9693385959/job/26748720100 for the graviton run etc.