Option for deterministic sparse matrix vector multiplication #1607
Labels
cuda libraries
Stuff about CUDA library wrappers.
enhancement
New feature or request
good first issue
Good for newcomers
I would like to request deterministic sparse matrix vector multiplications. Essentially, the following code should always return the same result:
A similar issue is #938.
What I found out so far
We need to use
CUSPARSE_SPMV_COO_ALG2
orCUSPARSE_SPMV_CSR_ALG2
. This will probably be related to changing the argumentalgo
ofmv!()
:CUDA.jl/lib/cusparse/generic.jl
Lines 112 to 113 in 33a7187
Unfortunately, the algorithm information seems to get lost in the wrapper function:
CUDA.jl/lib/cusparse/interfaces.jl
Lines 6 to 9 in 99f962a
The wrapper seems to be used for implementing the operators:
CUDA.jl/lib/cusparse/interfaces.jl
Lines 38 to 60 in 99f962a
The remaining question for the maintainers is now: Where do we pass in the algorithm?
The text was updated successfully, but these errors were encountered: