Metal Unary: Add benchmarks and process kernels in a tile based fashion #2056
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Going off what @LaurentMazare found a week or so ago, this just changes the unary implementation to process kernels in tiles rather than a function call per index of the buffer. See below for benchmarks on my M3 for a variety of tile sizes, 1 being the effective current behaviour on main.
From the benchmarks below we see a falloff of improvement after a tile size of 4, but also a regression on f32 performance. The main benefits on this change appear to be on f16 and bf16 operations with f32 experiencing a minor efficiency gain at a tile size of 4.
(See latest benchmarks in the comments below)