expose sparse mv/mm algo selection

[Roger-luo]

this is a simple change to expose the algorithm parameter to mv/mm level

edit: I'm not sure if we should check if the selected algo match the input, e.g some algo only work on CSR with col/row-major

[Roger-luo]

I thought the current tests cover this already tho? Or we should test on other algorithms?

[maleadt]

I thought the current tests cover this already tho?

How? Didn't you introduce the algo keyword here?

[Roger-luo]

hmm I'm a bit confused why the test is failing on CUDA 11.1, I've filter the tests to not run below CUDA 11.2, but somehow it still get executed on those versions, what would be right way of doing this?

[maleadt]

CUDA.version() returns the driver version. You need to check against CUSPARSE.version() (which requires you to use the library version, so look up what the appropriate version is corresponding with CUDA toolkit 11.2 -- it's in the release notes, or you can check yourself by setting JULIA_CUDA_VERSION=11.2).

[maleadt]

Please rebase on master to not require Windows to pass.

[codecov]

Codecov Report

Merging #1201 (3bad02d) into master (07e8bed) will decrease coverage by 0.74%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #1201      +/-   ##
==========================================
- Coverage   77.36%   76.62%   -0.75%     
==========================================
  Files         120      120              
  Lines        9274     9273       -1     
==========================================
- Hits         7175     7105      -70     
- Misses       2099     2168      +69     

Impacted Files	Coverage Δ
lib/cusparse/generic.jl	93.06% <ø> (ø)
lib/cutensor/error.jl	27.27% <0.00%> (-64.40%)	⬇️
lib/cudnn/CUDNN.jl	37.50% <0.00%> (-35.94%)	⬇️
lib/cublas/CUBLAS.jl	50.00% <0.00%> (-25.44%)	⬇️
src/utilities.jl	68.91% <0.00%> (-4.06%)	⬇️
lib/cudadrv/CUDAdrv.jl	51.66% <0.00%> (-3.34%)	⬇️
lib/cudadrv/module/linker.jl	68.75% <0.00%> (-3.13%)	⬇️
lib/cudadrv/memory.jl	78.59% <0.00%> (-1.01%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 07e8bed...3bad02d. Read the comment docs.

[maleadt]

LGTM, thanks!