Inside LTO improvements. #627
There are no files selected for viewing
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
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,91 @@ | ||
| --- | ||
| layout: post | ||
| title: "Disk space and LTO improvements" | ||
| author: Eric Huss | ||
| description: "Disk space and LTO improvements" | ||
| team: the Cargo team <https://www.rust-lang.org/governance/teams/dev-tools#cargo> | ||
| --- | ||
|
|
||
| Thanks to the work of [Nicholas Nethercote] and [Alex Crichton], there have been some recent improvements that reduce the size of compiled libraries, and improves the compile-time performance, particularly when using LTO. This post dives into some of the details of what changed, and an estimation of the benefits. | ||
|
|
||
| These changes have been added incrementally over the past three months, with the latest changes landing just a few days ago on the nightly channel. The bulk of the improvements will be found in the 1.46 stable release. It would be great for any projects that use LTO to test it out on the nightly channel and report any issues that arise. | ||
|
|
||
| [Nicholas Nethercote]: https://github.com/nnethercote | ||
| [Alex Crichton]: https://github.com/alexcrichton/ | ||
|
|
||
| ## Background | ||
|
|
||
| When compiling a library, `rustc` saves the output in an `rlib` file which is an [archive file]. This has historically contained the following: | ||
|
|
||
| * Object code, which is the result of code generation. This is used during regular linking. | ||
| * [LLVM bitcode], which is a binary representation of LLVM's intermediate representation. This can be used for [Link Time Optimization] (LTO). | ||
| * Rust-specific metadata, which covers [a wide range of data][metadata] about the crate. | ||
|
|
||
| LTO is an optimization technique that can perform whole-program analysis. It analyzes all of the bitcode from every library at once, and performs optimizations and code generation. `rustc` supports several forms of LTO: | ||
|
|
||
| * Fat LTO. This performs "full" LTO, which can take a long time to complete and may require a significant amount of memory. | ||
| * [Thin LTO]. This is a lightweight version of "fat" LTO that can achieve similar performance improvements while taking much less time to complete. | ||
|
There was a problem hiding this comment. "much less time" isn't necessarily right but rather "it does stuff in parallel while fat LTO is single threaded". I'd suspect that thin LTO does more work in terms of CPU hours but no one building Rust has only one core nowadays! There was a problem hiding this comment. @alexcrichton Because of the phrasing of this sentence, is there ever a time where fat LTO is better than thin LTO? There was a problem hiding this comment. Heh I wondered the same thing when ThinLTO came along! Fat LTO, however, produces smaller binaries because it can extremely aggressively internalize functions. It can also produce slightly faster binaries since total knowledge in a compiler is typically better than fragmented knowledge. ThinLTO's primary purpose is figure out what to inline across codegen units, while Fat LTO basically says "eh just do your normal thing but here's all the code in the world" There was a problem hiding this comment. That makes sense. Is there a way we can rephrase this to make this point more clear? The way it stands now makes it seem like thin is always better. "Similar performance while taking much less time" There was a problem hiding this comment. I've tweaked the wording a little to try to make it clearer. I think as with any optimization settings, every project will need to experiment with different settings to see what works best for them. |
||
| * Thin-local LTO. By default, `rustc` will split a crate into multiple "codegen units" so that they can be processed in parallel by LLVM. But this prevents some optimizations as code is separated into different codegen units, and is handled independently. Thin-local LTO will perform thin LTO across the codegen units within a single crate, bringing back some optimizations that would otherwise be lost by the separation. This is `rustc`'s default behavior if opt-level is greater than 0. | ||
|
|
||
| ## What has changed | ||
|
|
||
| Changes have been made to both `rustc` and Cargo to control which libraries should include object code and which should include bitcode based on the project's [profile] LTO settings. If the project is not using LTO, then Cargo will instruct `rustc` to not place bitcode in the rlib files, which should reduce the amount of disk space used. This may have a small improvement in performance since `rustc` no longer needs to compress and write out the bitcode. | ||
|
|
||
| If the project is using LTO, then Cargo will instruct `rustc` to not place object code in the rlib files, avoiding the expensive code generation step. This should improve the build time when building from scratch, and reduce the amount of disk space used. | ||
|
|
||
| Two `rustc` flags are now available to control how the rlib is constructed: | ||
|
|
||
| * [`-C linker-plugin-lto`] causes `rustc` to only place bitcode in the `.o` files, and skips code generation. Cargo uses this when the rlib is only intended for use with LTO. This can also be used when doing cross-language LTO. | ||
|
There was a problem hiding this comment. It might be worth nothing that this flag is being repurposed for Cargo's use since it was first added for what the name is on the tin, it just so happens to work for Cargo's use case too! |
||
| * [`-C embed-bitcode=no`] causes `rustc` to avoid placing bitcode in the rlib altogether. Cargo uses this when LTO is not being used, which reduces some disk space usage. | ||
|
|
||
| Additionally, the method in which bitcode is embedded in the rlib has changed. Previously, `rustc` would place compressed bitcode as a `.bc.z` file in the rlib archive. Now, the bitcode is placed as an uncompressed section within each `.o` [object file] in the rlib archive. This can sometimes be a small performance benefit, because it avoids cost of compressing the bitcode, and sometimes can be slower due to needing to write more data to disk. This change helped simplify the implementation, and also matches the behavior of clang's `-fembed-bitcode` option (typically used with Apple's iOS-based operating systems). | ||
|
|
||
| ## Improvements | ||
|
|
||
| The following is a summary of improvements observed on a small number of real-world projects of small and medium size. Improvements of a project will depend heavily on the code, optimization settings, operating system, environment, and hardware. These were recorded with the 2020-06-21 nightly release on Linux with parallel job settings between 2 and 32. | ||
|
|
||
| The performance wins for debug builds were anywhere from 0% to 4.7% faster. Larger binary crates tended to fair better than smaller library crates. | ||
|
|
||
| LTO builds were recorded anywhere from 4% to 20% faster. Thin LTO faired consistently better than fat LTO. | ||
|
|
||
| The number of parallel jobs also had a large impact on the amount of improvement. Lower parallel job counts saw substantially more benefit than higher ones. A project built with `-j2` can be 20% faster, whereas the same project at `-j32` would only be 1% faster. Presumably this is because the code-generation phase benefits from higher concurrency, so it was taking a relatively smaller total percentage of time. | ||
|
There was a problem hiding this comment. "Lower parallel job counts saw substantially more benefit than higher ones" -- oh man that's pipelining in action I think! |
||
|
|
||
| The overall target directory size is typically 20% to 30% smaller for debug builds. LTO builds did not see as much of an improvement, ranging from 11% to 19% smaller. | ||
|
|
||
| ## More details | ||
|
|
||
| Nicholas Nethercote wrote about the journey to implement these changes at <https://blog.mozilla.org/nnethercote/2020/04/24/how-to-speed-up-the-rust-compiler-in-2020/>. It took several PRs across `rustc` and Cargo to make this happen: | ||
|
|
||
| - [#66598](https://github.com/rust-lang/rust/pull/66598) — The original approach, that was decided to be too simplistic. | ||
| - [#66961](https://github.com/rust-lang/rust/issues/66961) — The issue outlining the strategy that was employed. | ||
| - [#70289](https://github.com/rust-lang/rust/pull/70289) | ||
| [#70297](https://github.com/rust-lang/rust/pull/70297) | ||
| [#70345](https://github.com/rust-lang/rust/pull/70345) | ||
| [#70384](https://github.com/rust-lang/rust/pull/70384) | ||
| [#70644](https://github.com/rust-lang/rust/pull/70644) | ||
| [#70729](https://github.com/rust-lang/rust/pull/70729) | ||
| [#71374](https://github.com/rust-lang/rust/pull/71374) | ||
| [#71716](https://github.com/rust-lang/rust/pull/71716) | ||
| [#71754](https://github.com/rust-lang/rust/pull/71754) — A series of refactorings to prepare for the new behavior and do some cleanup. | ||
| - [#71323](https://github.com/rust-lang/rust/pull/71323) — Introduced a new flag to control whether or not bitcode is embedded. | ||
| - [#70458](https://github.com/rust-lang/rust/pull/70458) [#71528](https://github.com/rust-lang/rust/pull/71528) — Switched how LLVM bitcode is embedded. | ||
| - [#8066](https://github.com/rust-lang/cargo/pull/8066) | ||
| [#8192](https://github.com/rust-lang/cargo/pull/8192) | ||
| [#8204](https://github.com/rust-lang/cargo/pull/8204) | ||
| [#8226](https://github.com/rust-lang/cargo/pull/8226) | ||
| [#8254](https://github.com/rust-lang/cargo/pull/8254) | ||
| [#8349](https://github.com/rust-lang/cargo/pull/8349) — The series of Cargo changes to implement the new functionality. | ||
|
|
||
| ## Conclusion | ||
|
|
||
| Although this is a conceptually simple change (LTO=bitcode, non-LTO=object code), it took quite a bit of preparation and work to make it happen. There were many edge cases and platform-specific behaviors to consider, and testing to perform. And, of course, the obligatory bike-shedding over the names of new command-line flags. This resulted in quite a substantial improvement in performance, particularly for LTO builds, and a huge improvement in disk space usage. Thanks to all of those that helped to make this happen! | ||
|
|
||
| [archive file]: https://en.wikipedia.org/wiki/Ar_(Unix) | ||
| [LLVM bitcode]: https://llvm.org/docs/BitCodeFormat.html | ||
| [Link Time Optimization]: https://llvm.org/docs/LinkTimeOptimization.html | ||
| [Thin LTO]: http://blog.llvm.org/2016/06/thinlto-scalable-and-incremental-lto.html | ||
| [profile]: https://doc.rust-lang.org/cargo/reference/profiles.html | ||
| [object file]: https://en.wikipedia.org/wiki/Object_file | ||
| [`-C linker-plugin-lto`]: https://doc.rust-lang.org/nightly/rustc/codegen-options/#linker-plugin-lto | ||
| [`-C embed-bitcode=no`]: https://doc.rust-lang.org/nightly/rustc/codegen-options/#embed-bitcode | ||
| [metadata]: https://github.com/rust-lang/rust/blob/0b66a89735305ebac93894461559576495ab920e/src/librustc_metadata/rmeta/mod.rs#L172-L214 | ||
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.
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 great to include a concrete date (or even commit) from which support is fully available on nightly, just so that people testing can be certain they're in good shape. (In theory they can just rustup update, I guess, but if there's regressions or whatever would be useful).
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.
Sounds good, added the date for the most recent fixes.