Added compiler support for executing element gather & scatter operations with multi-dimensional indices on SparseCore, achieving ∼60x speedup over TensorCore.

Implemented a compiler pass that offloads gather & scatter operations from TensorCore to SparseCore asynchronously and partitions the indices to run on multiple compute tiles in parallel.

Conducted performance analysis and optimization of NeRF (Neural Radiance Fields) Acceleration on TPUs.