Use cases for LibQ

[piotr-semenov]

Dear stargazers,

@kasandko, @dvolosnykh, @GerHobbelt, @Eoin-ONeill-Yokai, @timostrating, @milliyang, @JensGrabner, @nicoulaj, @d-led, and @brotherofken.

I really appreciate that you found this repository interesting and maybe even useful some time or years ago. I hope you felt it positive :)
If you had applications for this library, can you please kindly share your use cases with me? I will appreciate any feedback you'll give.

I discovered that easy-to-go fixed-point arithmetics is still demanded in 2024.
So I decided to resume this project and upgrade it up to C++20 within modern compilers like LLVM/CLang/GCC13/etc and make it to work properly with different runtimes.

I have ideas how to evolve this repository. Some of them are here:

• Move under Boost library.
• Wrap for Python in order to use it for fixed-precision Deep Learning & LLM quantification/optimisation. This topic is very hot (e.g., look at 0-1 LLMs from Microsoft).
• Interoperate with Rust/Zig programming languages getting permanently increasing popularity.

My guess is that reducing of memory footprint, making computations energy efficient, and suitable for deployment on resource-constrained devices like edge/IoT will be very hot in next decade. So I believe this repository still have potential!

[kasandko]

Dear Piotr Semenov,

Thank you for your library. I am interested in this library because I encountered one problem. I wrote trading robot using TinkoffSDK. This API used fixed type with 64 bit in integer part and 32 bit in fractional part. I didn't find any fixed point library for C++ which works with integral type more than 64 bit. I started to write own fixed point library which use any integer value (include types are provided different bignum arithmetic libraries) as storage type. But this process isn't fast: I get different problems with algorithms for precise calculations result of trigonometry functions, sqrt, gamma and other. Your library was interesting for me because you use CORDIC.

Opinion on your ideas:

• Moving under Boost doesn't look good idea. I saw such fixed point libraries earlier. This libraries where highly dependent from Boost. But Boost is big library and it is not always needed. We get big and unnecessary dependence. This may be good if this dependence will be optional.
• The rest of the ideas look good.

[piotr-semenov]

Thank you very much for response!
Your case looks very interesting. It can be tricky to support fixed-point integers with word length > 64 bits for arbitrary target platform. I should check if boost::multiprecision's integer types can be an option here for words.

Move under Boost library

Sorry for confusion. My idea was to follow Boost Library Requirements and Guidelines in order to become a part of Boost library further. It will give great credibility to this repository.
I understand your argument that Boost is heavy dependency. Turning to Boost can be on extra development track, not main.

[kasandko]

In that case, the idea really doesn't look bad.

I didn't know about boost::multiprecision. I will look it. Thank you!

[d-led]

I’ve saved it for the time if I ever find myself talking about computational geometry. No particular use case in mind, I guess

[piotr-semenov]

Thank you very much for response! Was your idea for game industry or just academic research?