FEX-Emu Is Working On Speedy x86/x86_64 Games Support On AArch64, Including Proton

[andreano]

AVX2

For the overlapping set of SIMD instructions, it should be possible to emit 2 NEON instructions for each AVX2 instruction.
Are you seeing a problem that I don't?

[Quackdoc]

best of luck to them, this is one of the projects I find super exciting (along with box86 64bit support)

[brucethemoose]

For the overlapping set of SIMD instructions, it should be possible to emit 2 NEON instructions for each AVX2 instruction.
Are you seeing a problem that I don't?

Or a single SVE2 instruction, some day.

[hotaru]

Not to be a downer, but how is this supposed to handle some of the more complex instructions like AVX2 without totally butchering performance?

https://en.wikipedia.org/wiki/AArch6...Extension_(SVE)

[rabcor]

Hope they do well, RISC PC CPUs are coming, if this project does well, it could be a huge boon for RISC on Linux, particularly chips sharing this 64 bit arm architecture.

Though honestly, the M1 and upcoming M2 notebooks from apple are possibly someo f the most compelling pieces of hardware i've ever seen, if it wasn't bound to that borked macos platform I'd probably buy it, loathe as I am to give money to a company like apple.

[marios]

I think taking an approach like RPCS3 would might work aswell. I.e pass all binaries through LLVM and recompile them for your current hardware ahead of time. Patch library calls to use host libs where applicable and be done with it. JIT will just always result in performance loss.

This is true but AoT looks impossible for x86. The x86 ISA is a pile of crap by itself and things are orders of magnitude worse when you add the fact that lots of apps (especially games) jump to arbitrary addresses for obfuscation reasons.

[brad0]

And the year of linux desktop come: we beat microsoft on arm desktop computing race.

What a useless metric. Still waiting for the year of the Linux desktop 20+ years in the making and still not even close to happening.

[NateHubbard]

Not to be a downer, but how is this supposed to handle some of the more complex instructions like AVX2 without totally butchering performance?

https://www.youtube.com/watch?v=7LRmpx3Rebs&t=1109s

[pWe00Iri3e7Z9lHOX2Qx]

Hope they do well, RISC PC CPUs are coming, if this project does well, it could be a huge boon for RISC on Linux, particularly chips sharing this 64 bit arm architecture.

Though honestly, the M1 and upcoming M2 notebooks from apple are possibly someo f the most compelling pieces of hardware i've ever seen, if it wasn't bound to that borked macos platform I'd probably buy it, loathe as I am to give money to a company like apple.

The hardware and Rosetta 2 certainly are compelling. When I got my base model 14" MBP with the 8 core CPU / 14 core GPU for work, I immediately ran some benchmarks. It ran Unigine Valley at 60fps average at 1920x1080 Ultra 4xAA. That's an old benchmark, but I was impressed considering it's x86 / OGL and that's 80% of the framerate I get in Linux on an old GTX 980 Ti .

[dragon321]

That's how you create a stable Linux API: wrap up everything

Linux system calls are stable.

As for the topic - Rosetta 2 is hardware assisted and that's one of the reasons why it performs so well. Same reason probably won't work for Linux (or at least not on all hardware). Of course it doesn't mean that pure software emulation can't work well.