Rewriteis_ascii using slice::as_chunks
#144837
Open
+82
−222
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
rustbot
added
S-waiting-on-review
This comment has been minimized.
This comment has been minimized.
Kmeakin
force-pushed
the
km/optimize-is-ascii
branch
from
c43111d to
7cbbbc6
Compare
|
The simpler source code and shorter assembly seem to boil down to two changes:
The first one seems quite reasonable in many cases. It probably causes a huge performance regression for targets that don't have efficient unaligned loads, but to be fair, those are becoming less common and less important over time. The second change may be quite problematic for some common input sizes, though. Try benchmarking before vs. after on an input that's |
|
There are some benchmarks in When I originally made that PR, new uses of However, the usize-aligned path is probably still needed for targets without SIMD like |
Kmeakin
force-pushed
the
km/optimize-is-ascii
branch
from
7cbbbc6 to
ebb9522
Compare
rustbot
added
S-waiting-on-author
This comment has been minimized.
This comment has been minimized.
Kmeakin
force-pushed
the
km/optimize-is-ascii
branch
from
ebb9522 to
e4222eb
Compare
Benchmark resultsAArch64 (Apple M3): x86 (AMD Ryzen 9 9950X): |
This comment has been minimized.
This comment has been minimized.
Kmeakin
force-pushed
the
km/optimize-is-ascii
branch
from
e4222eb to
6ac4350
Compare
This comment has been minimized.
This comment has been minimized.
Kmeakin
force-pushed
the
km/optimize-is-ascii
branch
from
6ac4350 to
5edf425
Compare
rustbot
added
S-waiting-on-review
tgross35
reviewed
Aug 7, 2025
There was a problem hiding this comment.
This is indeed much more straightforward and I think the results speak reasonably well for themselves.
The previous implementation had a lot useful of info, could you keep / add some more docs here? E.g.:
- How dispatching to simd vs. swar was chosen
- What we expect
is_ascii_simdto turn into, or what is required for it to become efficient simd - How
UNROLL_FACTORandCHUNK_SIZEwere picked
It would also be good to rebase and rerun the benchmarks/codegen demos, we just got a LLVM upgrade so the exact optimizations may be slightly different.
Comment on lines
-330
to
-338
| /// ASCII test *without* the chunk-at-a-time optimizations. | ||
| /// | ||
| /// This is carefully structured to produce nice small code -- it's smaller in | ||
| /// `-O` than what the "obvious" ways produces under `-C opt-level=s`. If you | ||
| /// touch it, be sure to run (and update if needed) the assembly test. | ||
| #[unstable(feature = "str_internals", issue = "none")] | ||
| #[doc(hidden)] | ||
| #[inline] | ||
| pub const fn is_ascii_simple(mut bytes: &[u8]) -> bool { |
There was a problem hiding this comment.
Could you check the code size for this compared to the other implementations? I wonder if it may be worth using for cfg(feature = "optimize_for_size").
Also, any idea which test it is talking about?
Comment on lines
+361
to
+364
| #[inline(always)] | ||
| fn is_ascii_scalar(bytes: &[u8]) -> bool { | ||
| bytes.iter().all(u8::is_ascii) | ||
| } |
There was a problem hiding this comment.
is_ascii_const (formerly is_ascii_simple) seems to optimize pretty well based on the comments, does this get better codegen?
| // sufficient alignment for `usize`, because it's a weird edge case. | ||
| if len < USIZE_SIZE || len < align_offset || USIZE_SIZE < align_of::<usize>() { | ||
| return is_ascii_simple(s); | ||
| if cfg!(all(target_arch = "x86_64", target_feature = "sse2")) { |
There was a problem hiding this comment.
i686 also has sse2, would there be any advantage here?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Generalize the x86-64+sse2 version of
is_asciito be architecture-neutral, and rewrite it usingslice::as_chunks. The new version is both shorter (in terms of Rust source code) and smaller (in terms of produced assembly).Compare the assembly generated before and after:
https://godbolt.org/z/MWKdnaYoK