Hacker News Reader

Comment by Skunkleton | Hacker News Reader

Has anyone seen a detailed writeup on how they are performing the x86->ARM64 recompilation? I would imagine it works something like a software hypervisor once the individual instructions are translated?

View on HN

I'll wager that there is still some of the old FX!32 technology in there, that was used back in the day to translate between x86 and AXP.

Remember NT was always designed to be cross platform - it shipped on x86, PPC, MIPS and Alpha originally. I ran SQL Server on Alpha for a while. Good times.

Does DEC's corpse still own it? I remember they were the ones who developed the thing. Quite remarkable.

No, Compaq bought Digital, then, when HP bought Compaq, they killed Alpha in favor of the instruction set of the future, Itanium, and, adding insult to injury, sold the Alpha IP to Intel to ensure it’d never see the light of day again.

The Sunway line of processors developed by the Chinese military are rumored to use a modernized version of the Alpha instruction set (unauthorized obviously): https://en.wikipedia.org/wiki/Sunway

Still, they may own FX!32 and its patents. This may prove profitable.

It those went to Intel... Well... It'll be fun to watch the fireworks.

Ah, I think I misunderstood you (thought you were referring to Alpha, not FX!32).

It's based on Virtual PC for Mac, MS's x86 to PPC JIT back in the times.

It's WOW64, like Itanium and AMD64. The former also used an emulator, whereas the latter is hardware of course.

Actually, ARM64 Windows uses double-WoW64. Two WoW64s are running at the same time - one running ARM32 code - and another x86_32.

Yep! I was aware of that but didn't mention it.

I wonder if WoW64 had to be modified to support having two architectures.

Though they maybe planned ahead: it's "Program Files (x86)", not "Program Files (32-bit)" :)

Program Files (x86) and SysWoW64 + Program Files (ARM) and SysArm32

The instruction set support is quite recent tho :) No AVX there.

MMX/SSE/SSE2/SSE3/SSSE3/SSE4.1/SSE4.2/POPCNT/AESNI/CTMUL

In this case I would say the lack of details is a good thing. It must work so well that they can keep it a secret from major vendors. If it wasn't near native speed or had bad x86/x64 ISA support they would be giving a heads up to the bigger software shops.

From the 68K to PPC migration at Apple, I remember software spent much more time running system code than application code (listening for events, redrawing stuff, and so on). It's true PPC was much faster than 68K and that made the move easier, but, still, just not having to emulate anything beyond the system library calls is great.

It has no x64 support...

It hasn't because it's based on a Microsoft product which already existed - Virtual PC for PowerPC Macs.

> Has anyone seen a detailed writeup on how they are performing the x86->ARM64 recompilation

It was somewhat outlined in this channel 9 video: https://channel9.msdn.com/Events/Build/2017/P4171

Reply on news.ycombinator.com