Port it to Alpha and let's see. I'll loan you my 21164 over the weekend.

dfox · on Nov 1, 2016

The article mentions that alpha was weak on bit-twiddling, which is something that almost certainly got fixed by introduction of BWX instructions on 21164.

EvanAnderson · on Nov 1, 2016

Thanks for mentioning that. You sent me down a brief rabbit-hole of nostalgia looking for the EV4/EV5 differences. If anybody else wants to go down that rabbit-hole here's where I ended-up: http://alasir.com/articles/alpha_history/alpha_21164_21164pc...

dchest · on Nov 1, 2016

Interestingly, LZ4 works mostly on nibble and byte level without much bit-twiddling.

krylon · on Nov 1, 2016

Also, when was the last Alpha built? About ten years ago?

It's fine for OpenBSD or NetBSD to support it, but Microsoft has not supported it in a long time.

honkhonkpants · on Nov 1, 2016

Was axp even the weakest of the four windows architectures for this task? PPC and MIPS can leave the bit twiddler scratching their heads, too.

dfox · on Nov 1, 2016

Problem with original Alpha is that accessing memory as anything other than aligned 32b or 64b words (yes, including bytes) counts as bit-twiddling because only load/store instructions that it has operate on these two word sizes.

I can't remember any other non-niche architecture with 8*2^n word size that shares this (mis-)feature. (I suspect that Cray 1 and it's derivates also share this, but that probably counts as niche architecture)

toast0 · on Nov 1, 2016

Since the last released build of Windows for Alpha was Windows 2000 RC1 (although it looks like there's an RC2 build floating around), I don't think we need to support that processor for a new NTFS extension.