[libre-riscv-dev] formal proof of the conditions where smaller fp ops can be emulated by larger fp ops

[Luke Kenneth Casson Leighton]

---
crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68

On Thu, Jan 30, 2020 at 2:57 AM Jacob Lifshay <programmerjake at gmail.com> wrote:
>
> Seems useful.
>
> it's the conditions where you can (for example) emulate fp16 ops by
> converting to fp32, doing fp32 op, then converting back to fp16 before any
> other operations:
>
> let x: f16 = a + b;
>
> becomes:
>
> let x: f16 = f16::from(f32::from(a) + f32::from(b));
>
> https://pdfs.semanticscholar.org/79ea/5587db38a6dd9b771cdf336acfa057afb79f.pdf

always wondered if there were rounding errors that would occur there.

l.