On 1/17/2018 7:22 PM, deadalnix wrote:
Reading this again, I think there is a bit of a misunderstanding. Only
cent/ucent took me ~1h to implement. The rest is more complex. That being said,
having cent/ucent would unlock a great deal of performance for crypto libraries,
and this is where the bottleneck is as far as CPU is concerned in this type of
application.
I don't remember how long, but it took me a fair while to do the divide:
https://github.com/dlang/druntime/blob/master/src/rt/llmath.d
It could be upscaled by rote to 128 bits, but even that would take me much
longer than an hour. And it would still leave the issue of making ucent work
with 32 bit code gen.
It could also be translated to D, but I doubt the generated code would be as
good.
Nevertheless, we do have the technology, we just need someone to put it
together.
