perf: remove DisjointMut guards in DSP hot loops, use raw pointers in fn signatures#1493
Open
silentnoisehun wants to merge 5 commits into
Open
perf: remove DisjointMut guards in DSP hot loops, use raw pointers in fn signatures#1493silentnoisehun wants to merge 5 commits into
silentnoisehun wants to merge 5 commits into
Conversation
…_rust - no DisjointMut guards in loops
Replace filter_8tap (Rav1dPictureDataComponentOffset overhead) with filter_8tap_raw in put_8tap_scaled_rust, matching prep_8tap_scaled_rust. Extract sptr/dptr/strides before loops; convert output pass to raw ptr. filter_8tap and filter_bilin are now dead code (no callers remain). Result: C gap 6.4% → ~4.6% on 1920x1080 900-frame bench. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
… in DSP fn signatures
On Windows x64 ABI, structs larger than 8 bytes are passed via hidden
pointer (caller allocates on stack, passes address). The trailing
`FFISafeRav1dPictureDataComponentOffset` parameter (16 bytes:
`WithOffset<*const Rav1dPictureDataComponent>`) therefore costs a stack
allocation + hidden pointer indirection on every DSP call.
Replace with `*const Rav1dPictureDataComponentOffset` (8 bytes, fits in
register) in all affected wrap_fn_ptr signatures:
- src/cdef.rs
- src/filmgrain.rs
- src/ipred.rs
- src/itx.rs
- src/loopfilter.rs
- src/looprestoration.rs
- src/mc.rs
In Fn::call: pass `&offset_value` (pointer to stack local).
In c_erased: reconstruct via `unsafe { *ref }`.
On arm/aarch64 neon_erased: ignored (neon uses raw ptrs directly).
Correctness verified: decoded YUV output matches C dav1d reference
(identical MD5 on test_long.ivf).
Co-Authored-By: Claude Sonnet 4.6 <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
1 participant
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.
Summary
This PR removes
DisjointMutguards from the innermost DSP decoding loops and replacesFFISafeRav1dPictureDataComponentOffsetwith raw pointers in DSP function signatures, eliminating per-iteration overhead in hot paths.Performance
Measured on 1920x1080, 900-frame AV1 decode benchmark:
~28% reduction in the performance gap vs C dav1d.
Correctness verified: decoded YUV output matches C dav1d reference (identical MD5 on test_long.ivf).
Changes
FFISafeRav1dPictureDataComponentOffset(16-byte struct, Windows x64 passes via hidden pointer) with*const Rav1dPictureDataComponentOffset(8 bytes, register-sized) in all DSP fn signatures:cdef.rs,filmgrain.rs,ipred.rs,itx.rs,loopfilter.rs,looprestoration.rs,mc.rsDisjointMutguards input_rust,prep_rust,filter_8tap,put_8tap_rustprep_8tap_rustput_bilin_rust,prep_bilin_rust,filter_bilinput_8tap_scaled_rustSafety
The raw pointer accesses are bounded by the same invariants as the existing C dav1d implementation. No new unsafety is introduced beyond what was already present in the DSP layer.