Go to file
bors 32d3276561 Auto merge of #83357 - saethlin:vec-reserve-inlining, r=dtolnay
Reduce the impact of Vec::reserve calls that do not cause any allocation

I think a lot of callers expect `Vec::reserve` to be nearly free when no resizing is required, but unfortunately that isn't the case. LLVM makes remarkably poor inlining choices (along the path from `Vec::reserve` to `RawVec::grow_amortized`), so depending on the surrounding context you either get a huge blob of `RawVec`'s resizing logic inlined into some seemingly-unrelated function, or not enough inlining happens and/or the actual check in `needs_to_grow` ends up behind a function call. My goal is to make the codegen for `Vec::reserve` match the mental that callers seem to have: It's reliably just a `sub cmp ja` if there is already sufficient capacity.

This patch has the following impact on the serde_json benchmarks: ca3efde8a5 run with `cargo +stage1 run --release -- -n 1024`

Before:
```
                                DOM                  STRUCT
======= serde_json ======= parse|stringify ===== parse|stringify ====
data/canada.json         340 MB/s   490 MB/s   630 MB/s   370 MB/s
data/citm_catalog.json   460 MB/s   540 MB/s  1010 MB/s   550 MB/s
data/twitter.json        330 MB/s   840 MB/s   640 MB/s   630 MB/s

======= json-rust ======== parse|stringify ===== parse|stringify ====
data/canada.json         580 MB/s   990 MB/s
data/citm_catalog.json   720 MB/s   660 MB/s
data/twitter.json        570 MB/s   960 MB/s
```

After:
```
                                DOM                  STRUCT
======= serde_json ======= parse|stringify ===== parse|stringify ====
data/canada.json         330 MB/s   510 MB/s   610 MB/s   380 MB/s
data/citm_catalog.json   450 MB/s   640 MB/s   970 MB/s   830 MB/s
data/twitter.json        330 MB/s   880 MB/s   670 MB/s   960 MB/s

======= json-rust ======== parse|stringify ===== parse|stringify ====
data/canada.json         560 MB/s  1130 MB/s
data/citm_catalog.json   710 MB/s   880 MB/s
data/twitter.json        530 MB/s  1230 MB/s

```

That's approximately a one-third increase in throughput on two of the benchmarks, and no effect on one (The benchmark suite has sufficient jitter that I could pick a run where there are no regressions, so I'm not convinced they're meaningful here).

This also produces perf increases on the order of 3-5% in a few other microbenchmarks that I'm tracking. It might be useful to see if this has a cascading effect on inlining choices in some large codebases.

Compiling this simple program demonstrates the change in codegen that causes the perf impact:
```rust
fn main() {
    reserve(&mut Vec::new());
}

#[inline(never)]
fn reserve(v: &mut Vec<u8>) {
    v.reserve(1234);
}
```

Before:
```rust
00000000000069b0 <scratch::reserve>:
    69b0:       53                      push   %rbx
    69b1:       48 83 ec 30             sub    $0x30,%rsp
    69b5:       48 8b 47 08             mov    0x8(%rdi),%rax
    69b9:       48 8b 4f 10             mov    0x10(%rdi),%rcx
    69bd:       48 89 c2                mov    %rax,%rdx
    69c0:       48 29 ca                sub    %rcx,%rdx
    69c3:       48 81 fa d1 04 00 00    cmp    $0x4d1,%rdx
    69ca:       77 73                   ja     6a3f <scratch::reserve+0x8f>
    69cc:       48 81 c1 d2 04 00 00    add    $0x4d2,%rcx
    69d3:       72 75                   jb     6a4a <scratch::reserve+0x9a>
    69d5:       48 89 fb                mov    %rdi,%rbx
    69d8:       48 8d 14 00             lea    (%rax,%rax,1),%rdx
    69dc:       48 39 ca                cmp    %rcx,%rdx
    69df:       48 0f 47 ca             cmova  %rdx,%rcx
    69e3:       48 83 f9 08             cmp    $0x8,%rcx
    69e7:       be 08 00 00 00          mov    $0x8,%esi
    69ec:       48 0f 47 f1             cmova  %rcx,%rsi
    69f0:       48 85 c0                test   %rax,%rax
    69f3:       74 17                   je     6a0c <scratch::reserve+0x5c>
    69f5:       48 8b 0b                mov    (%rbx),%rcx
    69f8:       48 89 0c 24             mov    %rcx,(%rsp)
    69fc:       48 89 44 24 08          mov    %rax,0x8(%rsp)
    6a01:       48 c7 44 24 10 01 00    movq   $0x1,0x10(%rsp)
    6a08:       00 00
    6a0a:       eb 08                   jmp    6a14 <scratch::reserve+0x64>
    6a0c:       48 c7 04 24 00 00 00    movq   $0x0,(%rsp)
    6a13:       00
    6a14:       48 8d 7c 24 18          lea    0x18(%rsp),%rdi
    6a19:       48 89 e1                mov    %rsp,%rcx
    6a1c:       ba 01 00 00 00          mov    $0x1,%edx
    6a21:       e8 9a fe ff ff          call   68c0 <alloc::raw_vec::finish_grow>
    6a26:       48 8b 7c 24 20          mov    0x20(%rsp),%rdi
    6a2b:       48 8b 74 24 28          mov    0x28(%rsp),%rsi
    6a30:       48 83 7c 24 18 01       cmpq   $0x1,0x18(%rsp)
    6a36:       74 0d                   je     6a45 <scratch::reserve+0x95>
    6a38:       48 89 3b                mov    %rdi,(%rbx)
    6a3b:       48 89 73 08             mov    %rsi,0x8(%rbx)
    6a3f:       48 83 c4 30             add    $0x30,%rsp
    6a43:       5b                      pop    %rbx
    6a44:       c3                      ret
    6a45:       48 85 f6                test   %rsi,%rsi
    6a48:       75 08                   jne    6a52 <scratch::reserve+0xa2>
    6a4a:       ff 15 38 c4 03 00       call   *0x3c438(%rip)        # 42e88 <_GLOBAL_OFFSET_TABLE_+0x490>
    6a50:       0f 0b                   ud2
    6a52:       ff 15 f0 c4 03 00       call   *0x3c4f0(%rip)        # 42f48 <_GLOBAL_OFFSET_TABLE_+0x550>
    6a58:       0f 0b                   ud2
    6a5a:       66 0f 1f 44 00 00       nopw   0x0(%rax,%rax,1)
```

After:
```asm
0000000000006910 <scratch::reserve>:
    6910:       48 8b 47 08             mov    0x8(%rdi),%rax
    6914:       48 8b 77 10             mov    0x10(%rdi),%rsi
    6918:       48 29 f0                sub    %rsi,%rax
    691b:       48 3d d1 04 00 00       cmp    $0x4d1,%rax
    6921:       77 05                   ja     6928 <scratch::reserve+0x18>
    6923:       e9 e8 fe ff ff          jmp    6810 <alloc::raw_vec::RawVec<T,A>::reserve::do_reserve_and_handle>
    6928:       c3                      ret
    6929:       0f 1f 80 00 00 00 00    nopl   0x0(%rax)
```
2021-03-30 03:41:14 +00:00
.github
compiler Rollup merge of #83643 - JohnTitor:is-freeze-no-longer-uses-span, r=RalfJung 2021-03-30 00:32:24 +02:00
library Auto merge of #83357 - saethlin:vec-reserve-inlining, r=dtolnay 2021-03-30 03:41:14 +00:00
src Auto merge of #83664 - Dylan-DPC:rollup-wx6idpd, r=Dylan-DPC 2021-03-30 01:16:08 +00:00
.editorconfig
.gitattributes
.gitignore
.gitmodules
.mailmap
Cargo.lock Rollup merge of #83239 - JohnTitor:reduce-deps, r=Mark-Simulacrum 2021-03-27 12:37:18 +09:00
Cargo.toml
CODE_OF_CONDUCT.md
config.toml.example
configure
CONTRIBUTING.md
COPYRIGHT
LICENSE-APACHE
LICENSE-MIT
README.md
RELEASES.md Rollup merge of #83508 - JohnTitor:platform-support-link, r=joshtriplett 2021-03-27 12:37:22 +09:00
rustfmt.toml
triagebot.toml
x.py

The Rust Programming Language

This is the main source code repository for Rust. It contains the compiler, standard library, and documentation.

Note: this README is for users rather than contributors. If you wish to contribute to the compiler, you should read the Getting Started section of the rustc-dev-guide instead.

Quick Start

Read "Installation" from The Book.

Installing from Source

The Rust build system uses a Python script called x.py to build the compiler, which manages the bootstrapping process. More information about it can be found by running ./x.py --help or reading the rustc dev guide.

Building on a Unix-like system

  1. Make sure you have installed the dependencies:

    • g++ 5.1 or later or clang++ 3.5 or later
    • python 3 or 2.7
    • GNU make 3.81 or later
    • cmake 3.13.4 or later
    • ninja
    • curl
    • git
    • ssl which comes in libssl-dev or openssl-devel
    • pkg-config if you are compiling on Linux and targeting Linux
  2. Clone the source with git:

    git clone https://github.com/rust-lang/rust.git
    cd rust
    
  1. Configure the build settings:

    The Rust build system uses a file named config.toml in the root of the source tree to determine various configuration settings for the build. Copy the default config.toml.example to config.toml to get started.

    cp config.toml.example config.toml
    

    If you plan to use x.py install to create an installation, it is recommended that you set the prefix value in the [install] section to a directory.

    Create install directory if you are not installing in default directory

  2. Build and install:

    ./x.py build && ./x.py install
    

    When complete, ./x.py install will place several programs into $PREFIX/bin: rustc, the Rust compiler, and rustdoc, the API-documentation tool. This install does not include Cargo, Rust's package manager. To build and install Cargo, you may run ./x.py install cargo or set the build.extended key in config.toml to true to build and install all tools.

Building on Windows

There are two prominent ABIs in use on Windows: the native (MSVC) ABI used by Visual Studio, and the GNU ABI used by the GCC toolchain. Which version of Rust you need depends largely on what C/C++ libraries you want to interoperate with: for interop with software produced by Visual Studio use the MSVC build of Rust; for interop with GNU software built using the MinGW/MSYS2 toolchain use the GNU build.

MinGW

MSYS2 can be used to easily build Rust on Windows:

  1. Grab the latest MSYS2 installer and go through the installer.

  2. Run mingw32_shell.bat or mingw64_shell.bat from wherever you installed MSYS2 (i.e. C:\msys64), depending on whether you want 32-bit or 64-bit Rust. (As of the latest version of MSYS2 you have to run msys2_shell.cmd -mingw32 or msys2_shell.cmd -mingw64 from the command line instead)

  3. From this terminal, install the required tools:

    # Update package mirrors (may be needed if you have a fresh install of MSYS2)
    pacman -Sy pacman-mirrors
    
    # Install build tools needed for Rust. If you're building a 32-bit compiler,
    # then replace "x86_64" below with "i686". If you've already got git, python,
    # or CMake installed and in PATH you can remove them from this list. Note
    # that it is important that you do **not** use the 'python2', 'cmake' and 'ninja'
    # packages from the 'msys2' subsystem. The build has historically been known
    # to fail with these packages.
    pacman -S git \
                make \
                diffutils \
                tar \
                mingw-w64-x86_64-python \
                mingw-w64-x86_64-cmake \
                mingw-w64-x86_64-gcc \
                mingw-w64-x86_64-ninja
    
  4. Navigate to Rust's source code (or clone it), then build it:

    ./x.py build && ./x.py install
    

MSVC

MSVC builds of Rust additionally require an installation of Visual Studio 2017 (or later) so rustc can use its linker. The simplest way is to get the Visual Studio, check the “C++ build tools” and “Windows 10 SDK” workload.

(If you're installing cmake yourself, be careful that “C++ CMake tools for Windows” doesn't get included under “Individual components”.)

With these dependencies installed, you can build the compiler in a cmd.exe shell with:

python x.py build

Currently, building Rust only works with some known versions of Visual Studio. If you have a more recent version installed and the build system doesn't understand, you may need to force rustbuild to use an older version. This can be done by manually calling the appropriate vcvars file before running the bootstrap.

CALL "C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Auxiliary\Build\vcvars64.bat"
python x.py build

Specifying an ABI

Each specific ABI can also be used from either environment (for example, using the GNU ABI in PowerShell) by using an explicit build triple. The available Windows build triples are:

  • GNU ABI (using GCC)
    • i686-pc-windows-gnu
    • x86_64-pc-windows-gnu
  • The MSVC ABI
    • i686-pc-windows-msvc
    • x86_64-pc-windows-msvc

The build triple can be specified by either specifying --build=<triple> when invoking x.py commands, or by copying the config.toml file (as described in Installing From Source), and modifying the build option under the [build] section.

Configure and Make

While it's not the recommended build system, this project also provides a configure script and makefile (the latter of which just invokes x.py).

./configure
make && sudo make install

When using the configure script, the generated config.mk file may override the config.toml file. To go back to the config.toml file, delete the generated config.mk file.

Building Documentation

If youd like to build the documentation, its almost the same:

./x.py doc

The generated documentation will appear under doc in the build directory for the ABI used. I.e., if the ABI was x86_64-pc-windows-msvc, the directory will be build\x86_64-pc-windows-msvc\doc.

Notes

Since the Rust compiler is written in Rust, it must be built by a precompiled "snapshot" version of itself (made in an earlier stage of development). As such, source builds require a connection to the Internet, to fetch snapshots, and an OS that can execute the available snapshot binaries.

Snapshot binaries are currently built and tested on several platforms:

Platform / Architecture x86 x86_64
Windows (7, 8, 10, ...)
Linux (kernel 2.6.32, glibc 2.11 or later)
macOS (10.7 Lion or later) (*)

(*): Apple dropped support for running 32-bit binaries starting from macOS 10.15 and iOS 11. Due to this decision from Apple, the targets are no longer useful to our users. Please read our blog post for more info.

You may find that other platforms work, but these are our officially supported build environments that are most likely to work.

Getting Help

The Rust community congregates in a few places:

Contributing

If you are interested in contributing to the Rust project, please take a look at the Getting Started guide in the rustc-dev-guide.

License

Rust is primarily distributed under the terms of both the MIT license and the Apache License (Version 2.0), with portions covered by various BSD-like licenses.

See LICENSE-APACHE, LICENSE-MIT, and COPYRIGHT for details.

Trademark

The Rust programming language is an open source, community project governed by a core team. It is also sponsored by the Mozilla Foundation (“Mozilla”), which owns and protects the Rust and Cargo trademarks and logos (the “Rust Trademarks”).

If you want to use these names or brands, please read the media guide.

Third-party logos may be subject to third-party copyrights and trademarks. See Licenses for details.