#sega #prs #scheme #debugging #profile #high

no-std prs-rs

High performance compressor/decompressor for the SEGA PRS Compression scheme

6 releases (stable)

1.0.4 Jun 29, 2024
1.0.3 Jun 23, 2024
1.0.2 Feb 29, 2024
1.0.1 Feb 11, 2024
0.1.0 Feb 11, 2024

#75 in Compression

Custom license

63KB
908 lines

prs-rs

Crates.io Docs.rs CI codecov

About

Rust port of the SEGA PRS Compression scheme.

You can learn more about this project in the dedicated documentation page.

Development

How to develop this project.

Clone this Repository:

# When cloning, make sure symlinks are enabled
git clone -c core.symlinks=true https://github.com/Sewer56/prs-rs.git

Install Rust:

Setup IDE:

Visual Studio Code Integration

Code/VSCode is the de-facto Rust development environment.

The following extensions are required:

The VSCode configuration in Reloaded projects (.vscode) contain the following:

  • Run Rust linter clippy on Save.
  • Run code format rustfmt on Save.
  • Tasks for common operations (generate documentation, active CI/CD etc.).

These configurations are in the .vscode folder; and the tasks can be ran via Ctrl+Shift+P -> Run Task.

Test Coverage

To run Coverage, run task (Ctrl+Shift+P -> Run Task), you should see:

Task Description
Cargo Watch Tarpaulin Automatically runs tests and updates coverage on save.
Generate Code Coverage Manually generate code coverage (cobertura.xml, tarpaulin-report.html)

The tarpaulin-report.html file can be opened in VSCode (Show Preview) for a live view.

For GUI integration, run action Coverage Gutter: Watch (in Ctrl+Shift+P actions menu).

Debugging Benchmarks

If you wish to debug benchmarks in VSCode, go to Run and Debug Menu and generate the launch profiles, you should get one for debugging benchmarks.

Profiling Benchmarks

Linux/OSX

Execute the following:

cargo bench --bench my_benchmark --profile profile -- --profile-time 10

This should give you a flamegraph in target/criterion/<method_name>/profile. You can open that flamegraph in a web browser.

Windows

Execute the following:

cargo bench --bench my_benchmark --no-run --profile profile

Navigate to the executable listed in the commandline:

target/profile/deps/my_benchmark-eced832ac8f31257.exe

And run with command my_benchmark-eced832ac8f31257.exe --bench --profile-time 10 under an external profiler, such as Visual Studio.

example

Optimizing for Size when Creating C Libraries

  1. Add "cdylib" crate type to Cargo.toml (if not already present)
[lib]
crate-type = ["cdylib"]

Install cargo-bloat, nightly toolchain and build-std:

cargo install cargo-bloat
rustup toolchain install nightly
rustup component add rust-src --toolchain nightly

Run cargo-bloat the following command to calculate package size:

RUSTFLAGS="-C panic=abort -C lto=fat -C embed-bitcode=yes" cargo +nightly bloat -Z build-std=std,panic_abort -Z build-std-features=panic_immediate_abort --target x86_64-pc-windows-gnu --profile profile --crate-type cdylib -n 100 --features c-exports

Change --target if needed for your platform.
This should produce binaries more appropriate for dynamic linking from C.

PGO (Profile Guided Optimization) on C Libraries

This Reloaded-based library is built with Profile Guided Optimization (PGO).

PGO is a compiler optimization technique that uses data from profiling runs to improve the quality of the generated code.

Details of the PGO implementation in this project are as follows:

  • We collect PGO data by running the benchmarks with the pgo feature enabled.
  • This is done in CI, before building the final C library.

You should ensure that only realistic representative workloads are used to collect the PGO data.

For example, if this was a compression library, you should run the 'compress' and 'decompress' methods on real files (not random data) as part of your benchmarks.

Non-realistic/representative workloads in benchmarks should be excluded through the 'pgo' feature flag, for example an unrealistic benchmark can be excluded like this:

#[cfg(not(feature = "pgo"))]
{
    bench_create_dict(c);
}

Testing PGO

PGO isn't guaranteed to always provide an improvement, after adding representative workloads, always test.

We will test with cargo pgo.

First, install the following:

cargo install cargo-pgo
rustup toolchain install nightly
rustup component add llvm-tools-preview

Then run an 'instrumented' benchmark, this will run your code in pgo_benchmark and collect some data:

cargo +nightly pgo instrument bench

After that run a regular benchmark to create a 'baseline' number:

cargo +nightly bench

And run the PGO optimized build:

cargo +nightly pgo optimize bench

If most of the results are equal or show an improvement, PGO has helped. Otherwise disable PGO from the library by editing the rust.yml workflow.

File Layout

The following is the expected file layout for your project:

.vscode/
docs/
src/
Cargo.toml
mkdocs.yml

The docs folder, and mkdocs.yml contain MkDocs Material documentation for your project.
The src folder should contains all source code for your project.

Cargo.toml should be in the root of the project.

C# Bindings for prs_rs

This Reloaded-based project provides C# bindings, as prs_rs.Net.Sys.

These are the raw bindings to the C exports of this Rust library, and are automatically generated.

The project is inside bindings/csharp folder. It shouldn't be modified.

Instead, if you want to make a 'friendlier' API, make a separate project with prs_rs.Net.Sys as a dependency, and provide high level bindings.

Contributing

See CONTRIBUTING for guidance on how to contribute to this project.

License

Licensed under GPL v3 (with Reloaded FAQ).

Learn more about Reloaded's general choice of licensing for projects..

No runtime deps

~0–385KB