wx257osn2 / qoixx

Single Header Quite Fast QOI(Quite OK Image Format) Implementation written in C++20

I (@wx257osn2)

I'm I, a Japanese Virtual C++ Programmer / IさんはC++が好きなインターネットパーソナリティです

Major Repositories

Libraries

	C++ header-only library contains C++11 constexpr lambda State machine expected Parser Combinator based on Parsing Expression Grammar and so on
	Header-only WinAPI wrapper library Error handling with expected RAII
	Single header readline library for C++17
	Single header high performance QOI implementation Especially the encoder is one of the fastest implementation
	Header-only library for getting information of Japanese hoildays C++ implementation of repositories on holiday-jp
	Single header library of strong typedef for bool
	Single header implementation of Clockwork Base32 Good performance

Applications

	REPL (interactive environment) of C Preprocessor It can show macro replacement steps
	Cross-language benchmark for QOI implementations Reference (phoboslab/qoi) wx257osn2/qoixx aldanor/qoi-rust zakarumych/rapid-qoi pfusik/qoi-fu
	Generic plugin loader on Windows
	Symboli plugins for some applications
	mklink command looks like ln for Windows

Major Contributions

• cpprefjp/site
  • misc fixes(#261, #372, #760, #787, #808, #817, #835, #942, #951, 7662dc08, #1015, e3233e7d, 1d666c61, e92dbda8, c0e60600, 663009ed, 52813130, 1117ef22)
  • fix typo and add description about type param Args
  • add description about [associative.reqmts]
  • add example of std::has_virtual_destructor
  • advice about code simplification
  • add descriptions about relaxing constraints on constexpr in C++14
  • correct a description about literal type
  • change description about deprecation of std::allocator
• cpprefjp/kunai
  • fix font size of <code> tag on page title
• facebookresearch/faiss
  • fix typo and bump up version forgotten to update
  • fix to conform C++11 and test the architecture specific code correctly
  • add NEON implementations(#2390, #2392, #2447)
  • update document
  • fix simdlib, the 128x2bit-SIMD abstraction layer internally used on faiss(#2448, #2885)
  • fix CI(#2442, #2889)
  • help some people(#2386, #2463, #2465)
• poacpm/poac
  • update document(#81, #100)
  • fix build/CI(#108, #109, #112, #377)
  • refactor regexp on cmake
  • improve subcommand suggestion
  • fix .gitignore generated with poac create
  • add --profile option to build and run
  • add poac clean subcommand
  • fix linker error
  • add conan v1 support
• reki2000/langs-bench-dijkstra
  • refactor and improve performance of C++ implementation(#4, #9, #27)
  • fix Rust dijkstra algorithm implementation
  • improve Makefile (add missing target, use auto-defined variables, and so on)
  • remove trailing white spaces
  • improve usability of run.sh
  • add extremely fast C++20 implementation
  • improve visibility of plot image
  • add note about haskell prerequisites on README
  • make script for exec the benchmark
• simdutf/simdutf
  • fix aarch64 implementations which fell into buffer overrun
• melpon/wandbox
  • improve UI
• melpon/wandbox-builder
  • add -std=c++2b
  • add Clang 11.1.0
  • update compilers(#123, and fix that on #124)
• p-ranav/structopt
  • fix help message in some situations for special options, --version and --help
• fimbul/Shiro
  • support empty tuple as a class member
  • improve comparison (comparison without forward_as_tuple)
• zakarumych/rapid-qoi
  • report issue about incorrect outputs
  • update document
  • fix typo, apply clippy
• pfusik/qoi-fu
  • report issue about incorrect outputs
• hayamdk/ryzen_segv_test
  • fix resource leak
• KinoMyu/FastHCADecoder
  • fix and refactoring
• EzoeRyou/cpp17book
  • fix typos(#140, #141, #142)
• cppmap/cppmap.docs
  • add info about online compiler
• Siv3D/OpenQuestion1
  • propose fast implementation
    • finally it doesn't be adopted
• Others
  • provide the implementation to investigate the cause of a report that qoixx, my implementation, is slow
    • qoixx, one of the fastest implementation (by me)
    • the report (by awesomebytes)
    • the implementation to investigate the problem (by me)
    • the result of the investigation (by awesomebytes)
      • finally, it has been concluded that qoixx is not slow

qoixx's Issues

Add more tests

• #11 adds tiny test, but it should be added more cases
  • especially, we should confirm qoixx throws exceptions correctly with invalid inputs

ans1 entropy coder

As stated here, there is a very real possibility of turning qoi into a strong replacement for png. Coupled with a very fast ANS1 entropy coder, its ratio it's on par with heavily optimized png files that costs 2 orders of magnitude more to encode.

Would you like to look at that as an option for qoixx or perhaps a derivative program?

Add SVE CI

related #18

Add aarch64 CI

Add support for 32bit arm

At the moment 32bit android arm is not supported, so the fallback without SIMD is required.

Bug: qoibench shows wrong 'size kb' totals

## Benchmarking ./vlcSnaps/*.png -- 10 runs

## Total for "./vlcSnaps"
        decode ms   encode ms   decode mpps   encode mpps   size kb    rate
qoi:       5.5211      7.1469       159.510       123.226       820   31.8%
qoixx:     4.5902      5.3635       191.860       164.198       820   31.8%

vlcSnaps is a folder of 58.045.051 bytes of optimized pngs. the qoi converted files for that are 109.163.275 bytes total.

Range support