search

NVIDIA / Fuser

A Fusion Code Generator for NVIDIA GPUs (commonly known as "nvFuser")
Other
271 stars 53 forks source link

Support 2D inner reduction scheduler with autotuning #3456

Open rdspring1 opened 1 day ago

rdspring1 commented 1 day ago

Summary:

This PR creates an SOL autotuning script for the 2d inner reduction scheduler. It trains a random forest to predict the best performing configuration for the reduction scheduler.

Inner Reduction Fusions:

  1. Sum --- y = sum(x, dim=-1)
  2. Add Sum --- z = sum(x1 + x2 + x3 + x4, dim=-1)
  3. Tanh Sum --- y = sum(tanh(x), dim=-1)
  4. Exp Sum --- z = sum(exp(x), dim=-1)
rdspring1 commented 18 hours ago

!test