Faster floating point math with Rust's new API

pythonspeed.com

6 pointsbirdculture5 comments

Probably slightly offtopic, how many generations of vector math instructions will there be in the next 20 years, when the bit count increases yet again ?

Intel already has several different generations starting at SIMD and ARM also has a few. X86 will be a complete horror in another 20 years with all these generations.

"when the bit count increases yet again".

Maybe I lack imagination, but I doubt we'll be needing to address more than 18EB in personal computers for a while longer than 20 years.

perhaps they mean of vector instructions specifically

afaik we currently in the wild have 64, 128, 256, 512 bit extensions?

Exactly and in a few years when theres another generation of new bit width extensions, all those old programs will of course require a new release, which is precisely the problem my suggestion might help with.

Bit width of vector instructions is not related to address size or register size.

Even today x86 have 512 bit instructions when clearly addresses and non vector registers are not that wide.

Reply on news.ycombinator.com