-
Notifications
You must be signed in to change notification settings - Fork 12.7k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Avoid unnecessary allocations in float_lit
and integer_lit
.
#55384
Avoid unnecessary allocations in float_lit
and integer_lit
.
#55384
Conversation
This commit avoids an allocation when parsing any float and integer literals that don't involved underscores. This reduces the number of allocations done for the `tuple-stress` benchmark by 10%, reducing its instruction count by just under 1%.
(rust_highfive has picked a reviewer for you, use r? to override) |
// Strip underscores without allocating a new String unless necessary. | ||
let s2; | ||
let s = if s.chars().any(|c| c == '_') { | ||
s2 = s.chars().filter(|&c| c != '_').collect::<String>(); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hm, since underscores are in ASCII, couldn't we into_bytes, retain, and then String::from_utf8, maybe unchecked?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That would work, but it's not clear to me that it would be any faster; it might even be slower because there are two allocations involved? (Because it converts to a vector, and then back to a new String?)
Besides, this is the cold path, so I'm satisfied with reusing the existing code, which until now had been considered good enough for the hot path :)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Just FYI, String::from_utf8 consumes the vector given to it, so no additional allocation would happen.
|
||
// Strip underscores without allocating a new String unless necessary. | ||
let s2; | ||
let s = if s.chars().any(|c| c == '_') { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It would be interesting to see if std::slice::memchr::memchr(b'_', s.as_bytes()).is_some()
is faster than s.chars().any()
.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I tried it. Cachegrind says it's marginally more instructions: 22,526,559,505 up from 22,516,683,388. So I think I'll stick with the original.
Although I haven't changed the code, IMO I've addressed the comments above, so this is ready for re-review. |
Thanks, @nnethercote! @bors r+ |
📌 Commit eb637d2 has been approved by |
…t_lit, r=michaelwoerister Avoid unnecessary allocations in `float_lit` and `integer_lit`. This commit avoids an allocation when parsing any float and integer literals that don't involved underscores. This reduces the number of allocations done for the `tuple-stress` benchmark by 10%, reducing its instruction count by just under 1%.
Rollup of 9 pull requests Successful merges: - #54965 (update tcp stream documentation) - #55269 (fix typos in various places) - #55384 (Avoid unnecessary allocations in `float_lit` and `integer_lit`.) - #55423 (back out bogus `Ok`-wrapping suggestion on `?` arm type mismatch) - #55426 (Make a bunch of trivial methods of NonNull be `#[inline]`) - #55438 (Avoid directly catching BaseException in bootstrap configure script) - #55439 (Remove unused sys import from generate-deriving-span-tests) - #55440 (Remove unreachable code in hasClass function in Rustdoc) - #55447 (Fix invalid path in generate-deriving-span-tests.py.) Failed merges: r? @ghost
This commit avoids an allocation when parsing any float and integer
literals that don't involved underscores.
This reduces the number of allocations done for the
tuple-stress
benchmark by 10%, reducing its instruction count by just under 1%.