search

pytorch / glow

Compiler for Neural Network hardware accelerators
Apache License 2.0
3.19k stars 685 forks source link

Implementing Matrix Factorizations (SVD, QR, LU) #4881

Open anujgupt-github opened 3 years ago

anujgupt-github commented 3 years ago

I need to implement SVD, QR, LU decompositions.

These are not part of standard C++ libraries, but are part of other libraries (like Eigen3). For implementing for Interpreter backend, what is your recommendation:

  1. Use/include libraries like Eigen3 and import the in-built functions? Or,
  2. Implement using brute-force method?

@jfix71, WDYT?

jfix71 commented 3 years ago

For the Interpreter, we usually like to have simple reference implementations. Our goal is not to have super performant kernels -- other backends that are not references go for performance, such as CPU, OpenCL, NNPI, etc. So I'd suggest 2.

anujgupt-github commented 3 years ago

Thanks for your recommendations, @jfix71