Add 64-bit indices support to csrmm2 #58899
Labels
module: sparse
Related to torch.sparse
open source
triaged
This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Projects
Motivation
NVIDIA documentation mentions that with
cusparseSpMM
cuSparse APICUSPARSE_INDEX_64I
can be used withCUSPARSE_SPMM_CSR_ALG2/CUSPARSE_SPMM_CSR_ALG3
options, currently it uses CUSPARSE_CSRMM_ALG1.See: https://docs.nvidia.com/cuda/cusparse/index.html
Support for 64-bit indices will improve performance if
csrmm2
could work also with int64_t csr indices.pytorch/aten/src/ATen/native/sparse/cuda/SparseCUDABlas.cu
Line 200 in e4be80c
The task boils down to generalizing
template<typename T> void _csrmm2
... totemplate<typename T, typename I> void _csrmm2
whereI
is index type.cc @aocsa @nikitaved @pearu @mruberry @IvanYashchuk
The text was updated successfully, but these errors were encountered: