Standalone sparse-dense matrix multiplication benchmark

[pytorch]

Standalone sparse-dense matrix multiplication benchmark

[dskhudia]

@marsupialtail Such an API doesn't exist in FBGEMM at the moment.

[marsupialtail]

So is the easiest way of testing a SpMM fusing it with a quantized matmul at the moment?

[dskhudia]

@marsupialtail: Please take a look at the https://github.com/pytorch/FBGEMM/blob/master/bench/I8SpmdmBenchmark.cc

[marsupialtail]

So FBGEMM currently only supports int8 SpMM rn? Does it support fp32?

[dskhudia]

Currently it's int8 only.

[dskhudia]

fp32 and int8 dense-sparse exist at https://github.com/pytorch/FBGEMM/blob/master/bench/SparseDenseMMFP32Benchmark.cc and https://github.com/pytorch/FBGEMM/blob/master/bench/SparseDenseMMInt8Benchmark.cc