Loop Spreading

Loop Spreading

Loop spreading is an optimization of two non-nested sequential loops that have different iteration ranges. Typically, this refers to where the end ranges differ significantly. If the loop ranges only differ by an off-by-one issue, then only loop normalization is required.

Loop spreading modifies one of the loops, so that part of this loop fully overlaps with the other loop (i.e. ideally one loop “spreads out” further to match the other loop's end bounds). Hence, after loop spreading has occurred, this subloop can be fused with the other loop, and possibly parallelized. The remaining iterations that are not overlapping then have to be addressed in a followup partial loop (only for one of the loops).

Loop spreading mainly enables loop fusion as a followup optimization. For using loop fission on the two loops, it is not necessary to do loop spreading, since the two loops are already split apart, and each loop could already potentially be vectorized independently.

• Next: • Up: Table of Contents • Buy: Generative AI in C++: Coding Transformers and LLMs

The new AI programming book by Aussie AI co-founders: AI coding in C++ Transformer engine speedups LLM models Phone and desktop AI Code examples Research citations Get your copy from Amazon: Generative AI in C++