-
Notifications
You must be signed in to change notification settings - Fork 12.8k
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
Optimize decimal formatting of 128-bit integers #81484
Conversation
Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @dtolnay (or someone else) soon. If any changes to this PR are deemed necessary, please add them as extra commits. This ensures that the reviewer can see what has changed since they last reviewed the code. Due to the way GitHub handles out-of-date commits, this should also make it reasonably obvious what issues have or haven't been addressed. Large or tricky changes may require several passes of review and changes. Please see the contribution instructions for more information. |
r? @nagisa |
((q << 1) | carry as u128, r as u64) | ||
const FACTOR: u128 = 156927543384667019095894735580191660403; | ||
|
||
let quot = if n < 1 << 83 { |
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.
Is there an actual value in this condition? I think on majority of the targets we support u128_mulhi
will be faster than a 64-bit division anyway.
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.
Of course if you're using laptop/desktop PC u128_mulhi
works very fast and thus this branch may be needless. However integer multiplication is still expensive operation on some modern processors. For example, on the Intel Knights Landing micro architecture (which is widely used for supercomputers and high-performance workstations), MUL
instruction needs >7 cycles for generating result. Also it has only 1 specific port for multiplication.
In addition, As you can see in assembler output, this conditional branch generates only 2 instructions, and second call to this function always fallback to fast path because u128::MAX / 10^19 < 2^83
. That means, if fast path exists, then these 2 calls results in 44 instructions in total, 58 otherwise.
Broadly LGTM, pending answer to the q above. |
@bors r+ |
📌 Commit ada714d has been approved by |
… r=nagisa Optimize decimal formatting of 128-bit integers ## Description This PR optimizes the `udivmod_1e19` function, which is used for formatting 128-bit integers, based on the algorithm provided in \[1\]. This optimization improves performance of formatting 128-bit integers, especially on 64-bit architectures. It also slightly reduces the output binary size. ## Assembler comparison https://godbolt.org/z/YrG5zY ## Performance #### previous results ``` test fmt::write_u128_max ... bench: 552 ns/iter (+/- 4) test fmt::write_u128_min ... bench: 125 ns/iter (+/- 2) ``` #### new results ``` test fmt::write_u128_max ... bench: 205 ns/iter (+/- 13) test fmt::write_u128_min ... bench: 129 ns/iter (+/- 5) ``` ## Reference \[1\] T. Granlund and P. Montgomery, “Division by Invariant Integers Using Multiplication” in Proc. of the SIGPLAN94 Conference on Programming Language Design and Implementation, 1994, pp. 61–72
…as-schievink Rollup of 18 pull requests Successful merges: - rust-lang#78044 (Implement io::Seek for io::Empty) - rust-lang#79285 (Stabilize Arc::{increment,decrement}_strong_count) - rust-lang#80053 (stabilise `cargo test -- --include-ignored`) - rust-lang#80279 (Implement missing `AsMut<str>` for `str`) - rust-lang#80470 (Stabilize by-value `[T; N]` iterator `core::array::IntoIter`) - rust-lang#80945 (Add Box::downcast() for dyn Any + Send + Sync) - rust-lang#81048 (Stabilize `core::slice::fill_with`) - rust-lang#81198 (Remove requirement that forces symmetric and transitive PartialEq impls to exist) - rust-lang#81422 (Account for existing `_` field pattern when suggesting `..`) - rust-lang#81472 (Clone entire `TokenCursor` when collecting tokens) - rust-lang#81484 (Optimize decimal formatting of 128-bit integers) - rust-lang#81491 (Balance sidebar `Deref` cycle check with main content) - rust-lang#81509 (Add a regression test for ICE of bad_placeholder_type) - rust-lang#81547 (Edit rustc_typeck top-level docs) - rust-lang#81550 (Replace predecessor with range in collections documentation) - rust-lang#81558 (Fix ascii art text wrapping in mobile) - rust-lang#81562 (Clarify that InPlaceIterable guarantees extend to all advancing iterator methods.) - rust-lang#81563 (Improve docblock readability on small screen) Failed merges: r? `@ghost` `@rustbot` modify labels: rollup
Description
This PR optimizes the
udivmod_1e19
function, which is used for formatting 128-bit integers, based on the algorithm provided in [1]. This optimization improves performance of formatting 128-bit integers, especially on 64-bit architectures. It also slightly reduces the output binary size.Assembler comparison
https://godbolt.org/z/YrG5zY
Performance
previous results
new results
Reference
[1] T. Granlund and P. Montgomery, “Division by Invariant Integers Using Multiplication” in Proc. of the SIGPLAN94 Conference on Programming Language Design and Implementation, 1994, pp. 61–72