Research project, 2026
The Implicit Control Plane
Scheduling warp-specialised persistent kernels as concurrent systems. Manuscript in preparation. Advisor: Dr. Sheel Sindhu Manohar, Dept. of Computer Science and Engineering, Shiv Nadar Institution of Eminence.
What I did
- Formalised warp-specialised persistent GPU kernels (Hopper/Blackwell MLA, FlashAttention-3, persistent GEMM) as timed marked graphs.
- Derived a closed-form steady-state initiation interval via maximum cycle mean, predicting measured per-tile latency to 1.2% mean error with no per-configuration fitting.
- Identified the per-SM TMA copy engine as an unarbitrated shared resource causing 21% inter-CTA lifetime dispersion, and designed PhaseSkew, a zero-cost deterministic scheduling fix.
Get in touch
Email animeshmishra0567@gmail.com if you work in any of these areas and want to talk shop:
- NLP and ML evaluation
- Scientific AI and retrieval-augmented generation
- Research roles ahead of a PhD