GitHub

Feature Request: Expose saddmm (Sampled Dense-Dense Matrix Multiplication) for sparse tensors

I would like to request exposing the at::saddmm (or sampled_addmm) function from the C++ libtorch backend to the R torch interface. This function performs a Sampled Dense-Dense Matrix Multiplication.
IEEE

DenSparSA: A Balanced Systolic Array Approach for Dense and Sparse Matrix Multiplication

Abstract: Numerous studies have proposed hardware architectures to accelerate sparse matrix multiplication, but these approaches often incur substantial area and power overhead, significantly ...
IEEE

Optimizing High-Level Synthesis Sparse Matrix-Vector Kernel for Alveo FPGAs

Abstract: Sparse matrix-vector multiplication (SpMV) is a critical kernel in scientific computing and high-performance applications, yet it remains challenging to optimize due to irregular memory ...