Part of requirement for the course E0358: Advanced Techniques in Compilation and Programming for Parallel Architectures

Summary

• Loop optimization: Implemented a loop interchange pass within the Affine dialect in MLIR, driven by an analytical cost model that optimises for locality (spatial, temporal - both self and group). The pass handles perfectly nested loop nests and detects conditionals to bail out of optimization (to preserve correctness)

• Accurate Reductions: Designed and implemented infrastructure and compiler passes for end-to-end support in MLIR for atomic reduction operations on a CPU (ex: sum, mul, etc.). The passes automatically detect sequential operations that follow the RMW template and replace the instructions in place.

The applied patches and commits can be found here.