Rupanshu Soi

Rupanshu Soi

Hi, reader! I am a fourth-year CS PhD student at Stanford. My advisor is Alex Aiken. I also work on the Programming Systems and Applications Research Group at NVIDIA, managed by Michael Garland.

I am broadly interested in the theory and practice of program optimization and compilation.

Our paper on Twill, the first optimal approach to software pipelining and warp specialization on modern GPUs, will appear at OSDI 2026. If you are the conference, come say hi! If you are interested in learning more about Twill, the recording of a talk I gave at the Triton community meeting is available.

I also study equality saturation (EqSat). My research includes the first extensive evaluation of EqSat against another program optimization technique (under submission).

I have worked on various aspects of the Legion parallel programming model and its domain-specific language Regent.

I received a B.E. in CS from BITS Pilani in 2022.

Other Writing

  1. VLIWs, and How to Schedule Them
  2. Compilers Should Have a Superoptimize Flag
  3. The Unspoken Expectation