[Feature Request] Extend Sparse GEMM Support for SM80 and SM89 Architectures

[LeiWang1999]

The current implementation of T.gemm_sp primarily targets SM90. But we can dig further to extend support to additional architectures like SM80 and SM89. We're looking for volunteers to help push this initiative forward.

[botbw]

I would like to continue working on gemm_sp, feel free to ping me if anyone is interested in collaborating!

Some related references we can look at:

#526

https://github.com/NVIDIA/cutlass/blob/main/examples/15_ampere_sparse_tensorop_gemm/ampere_sparse_tensorop_gemm.cu

https://github.com/pytorch/pytorch/blob/a8d6afb511a69687bbb2b7e88a3cf67917e1697e/aten/src/ATen/native/sparse/cuda/SparseSemiStructuredOps.cu#L47

[LeiWang1999]

closed as now support via #691 , thanks @botbw