perf(reverse): use a dedicated kernel for the dims reversal - #118
Open
shreyas-omkar wants to merge 4 commits into
Open
perf(reverse): use a dedicated kernel for the dims reversal#118shreyas-omkar wants to merge 4 commits into
shreyas-omkar wants to merge 4 commits into
Conversation
Add a `dims` keyword to `reverse!`/`reverse`, reaching parity with `Base.reverse` and the vendor reverse kernels. `dims=:` (the default) keeps the fast flat path - each thread swaps one mirrored pair - while `dims=d` (an integer or iterable) reverses only along those dimensions via a general ND kernel written on `foreachindex`, so it runs on every backend (CUDA/AMDGPU/oneAPI/Metal/POCL) and the CPU-threaded path from one implementation. Invalid dims throw `ArgumentError`. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Exercise single-dim, multi-dim ((1,2)/(2,3)/(1,3)/:), size-1 degenerate dims and 3-D arrays across in-place, out-of-place and allocating forms, plus ArgumentError on out-of-range dims. Verified on CPU-threaded, AMDGPU (ROCm) and POCL backends. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Benchmarks for the dedicated-kernel
dimsreversal.reverse(A; dims), Float32, min of 30 runs, ms.NVIDIA RTX 5080 (CUDA 6.3.1)
AMD RX 9060 XT (AMDGPU 2.8.0)
On CUDA the dedicated kernel runs the index math inline, which the
foreachindexclosure did not on NVPTX, giving a 3x to 6x speedup and reaching native parity. On AMD the closure was already inlined, so the numbers are unchanged (no regression).