Intellectual Property India Publishes Patent Application for A High-Throughput Gpu Warp-Shuffle Compaction Kernel And Dynamic In-Register Attention Masking System

MUMBAI, India, Nov. 6 -- Intellectual Property India has published a patent application (202641128763 A) filed by Tirja Consulting LLP on October 28, 2026, for A High-Throughput Gpu Warp-Shuffle Compaction Kernel And Dynamic In-Register Attention Masking System.

Inventor includes Shreos Roychowdhury.

The application for the patent was published on November 06, 2026, under issue no. 45/2026.

Abstract: A high-throughput GPU architecture for massive parallel signature search and constrained neural generation on streaming multiprocessors is disclosed. The kernel performs 128-bit vectorized reads, hardware Bloom filter intersection, and intra-warp ballot synchronization electing, conditioned on a non-empty qualification mask, a single thread ...