Is there any plan to make NPUs support BLAS/LAPACK for accelerated matrix multiplications?

[kmshort]

Hi AMD Team,

I’m interested in knowing if there are any plans to support BLAS/LAPACK operations, particularly matrix multiplications, on AMD NPUs. With the increasing presence of NPUs in modern systems, integrating BLAS/LAPACK support would open up opportunities for accelerating a wide range of scientific computations, especially in environments like R statistics and Python that heavily rely on these libraries, particularly for Single Cell RNA sequencing computations.

Is there any roadmap or ongoing work to enable BLAS/LAPACK acceleration on NPUs?
Would such support come through existing libraries (e.g., rocBLAS) or require a new framework?
Any insights would be greatly appreciated.