parng

parng is not ready yet. It still has serious bugs and needs benchmarks and a test suite. The following statements are likely lies. Please do not publicize it widely until ready.

Introduction

parng is a advanced, parallel PNG decoder written in Rust. It has several features that make it especially suitable for client software:

• Very fast.

  • parng combines both multicore parallelism (taking advantage of dual core if available) and SIMD parallelism (with AVX on x86-64 and NEON on ARM if available).

  • In all benchmarks against popular PNG decoding libraries, parng is the fastest PNG decoder by a significant margin, reaching speedups of up to 2x over the next fastest library.

  • The AVX SIMD routines were generated using the Stanford STOKE superoptimizer and manually refactored for clarity.

• Secure.

  • parng is written almost entirely in safe Rust and memory-safety-verified assembly.

  • The implementation uses no unsafe code other than the system libraries (including zlib), a tiny driver for the accelerated assembly code, and a helper routine that avoids zeroing memory (but see below).

  • The SIMD assembly code is verified for memory safety using a custom static analysis written in Ruby.

  • parng can be made to zero out buffers if desired with a Cargo feature (zero-out-buffers).

• Flexible.

  • parng has a C API, so it can be used from C, C++, or any language that can call out to C.

  • The input and output are completely customizable. With the data provider API, you can store and the image data however you like and display the image whenever you wish.

  • parng has a scalar fallback for systems that do not support AVX or NEON.

Building

parng is a standard Rust package built using Cargo. After installing Rust, use cargo build --release to build the package. Inside target/release you can find both the Rust library and a C library.

Usage

parng has two APIs: the simple API and the image loader API. The simple API is in the simple module and consists of the Image structure and its associated load method. It is useful if all you need to do is to decode the image into memory. If you need more fine-grained control—for example, if you want to display the image as it is loading, or you want to downscale individual scanlines as you decode them to save memory—then you will want to use the ImageLoader API. That API can be found in imageloader.rs.

The C API mirrors the Rust API. When called from C, parng has two additional convenience methods: parng_image_load_from_file and parng_image_load_from_memory. These APIs allow you to load PNGs from C FILE * or byte arrays respectively with one function call.

License

parng is licensed under the same terms as Rust itself. See the LICENSE-APACHE and LICENSE-MIT files.

Acknowledgements

• The prefix sum technique used to implement the Left filter is taken from https://github.com/kobalicek/simdtests/blob/master/depng/depng_sse2.cpp -- thanks!