Aarch64 simd #65
Draft
Aarch64 simd #65
Conversation
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
Remove unneeded `extern crate`s
Also, add some benchmarks against memchr
Add a test that looks for the first item in a long haystack
The memmap crate is unmaintained, instead, use the maintained memmap2 crate
Structs don't need the bounds, only the implementations
Mostly just adding #[must_use]
This speeds up the criteron benchmarks by almost 2x I believe this is needed because e.g. Bytes::find is inlined, and calls `find` generically, which will call PackedCompareControl methods. So the code calling the methods will be inlined into the calling crate, but the implemetations of the PackedCompareControl are not accessable to the code in the calling crate, so they will end up as actual function calls. However these functions are _super_ simple, and inlining them helps a LOT, so adding `#[inline]` to these functions, and making their implementation available to calling crates has a huge effect. This was only seen when moving to criterion because previously, nightly benchmarks were implemented in the library crate itself, and so these functions were already elegable for inlining. Criteron results were actually more accurate to what callers of the crate would actually see!
Per suggestion from @BurntSushi [here](tafia/quick-xml#664 (comment)) On my M1, tt appears to be slower but competitive with memchr up to memchr3, then start being the from 5-16
We may not want to be stuck with const-constructable implementations
Move the simd-only tests to the top level This allows testing even when sse4.2 isn't enabled: when it is available, it will still test the simd implementation, but will test the fallback otherwise.
This moves mentions of "simd" to be x86 specific. Also, do everything with #[cfg], rather than requiring custom cfgs populated in the build.rs
This includes pretty frequent instances
For aarch64, we can do quite a bit better than just calling the `find` function repeatedly: we build a bitset of 64 bits where we've already found if they match the set of bits we're looking for. We can then efficently iterate over those set bits. It may be possible to do something similar in the x86 simd implementation.
|
Looks like the unstable Pattern api changed |
|
Fwiw, I gave this a go in a hot loop in Comrak, and it works really well! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
Builds on #57
Closes #61
Currently still working, but opening as a draft for early looks