You're here for our Next Generation Arithmetic (NGA) work with Hastlayer because you've attended one of our talks on the topic or heard of it in person. Welcome! We use this page to give you guidance, since this project of ours is a work in progress and development is ongoing in branches in all the relevant repositories.
- If you don't know what NGA is check out its website. But in an overly simplified nutshell it's about proposals for new floating point number types that computers can use instead of the standard floats and doubles (for example), making numeric results more accurate and possibly faster. This is most interesting for scientific and engineering use-cases.
- If you're unsure about what Hastlayer is check out the FAQ.
- .NET unum type I proof of concept and fully working type III (posit) implementation: https://github.com/Lombiq/Arithmetics/tree/issue/HAST-136. (The repository's name is in plural, as we'll add multiple different arithmetic implementations there, including moving our 64b fixed point number type there.)
- The Arithmetics code is a standard .NET library usable in any .NET software, and is also fully compatible with Hastlayer, thus can be converted into FPGA hardware. How it can be used is also included as a sample in the Hastlayer SDK's branch. Specifically the whole code that executes as FPGA hardware is here, and it's started from the host PC here (this also includes some basic software benchmarks too).
- You can chime in to the development under this GitHub issue.
- We've written a detailed whitepaper about our implementation and its results for CoNGA. You can download the whitepaper for free from the ACM Digital Library (this link will only work if you click it on this page).
- We've also presented it as a short talk, the recording of which you can watch here.
- If you'd like to keep up to date with Hastlayer and our NGA work then follow Hastlayer on Twitter or subscribe to our newsletter.
- Anything else? Just drop us an e-mail!