416 replies to this topic

[Dragon]

it emulates SSE2 and SSE3.
and i will give OS X a try this weekend when i get time to move the p3 out of my sisters room

[munky]

Rock on

[Baltazar]

Yeah, go crazy!

[Cruxado]

I've been out for awhile... TESTs are flying everywhere here at the Univ.

I'll be testing using the SLAX version and OS X 10.4.6 x86

[mac-mini]

no im not using the modded QEMU

Edited by mac-mini, 09 February 2007 - 02:48 AM.

[Dragon]

i've made my own slax cd with qemu and pearpc modules and am just about to burn so i can test it on the p3.

This is the pearpc module required for SLAX.

Attached Files

•  pearpc.mo.gz   374.46K   10 downloads

[Dragon]

Setting up pearpc on the p3 tomorrow morning, also try qemu with kqemu/qvm86.

[munky]

good luck dude - cant wait

[Dragon]

Kqemu module.

Attached Files

•  kqemu.mo.gz   25.44K   3 downloads

[Dragon]

This is qmv86... i'm just converting all these from rpms and storing them here.
Sorry for the "spam".

Attached Files

•  qvm86.mo.gz   8.56K   3 downloads

[Pu7o]

The thing is, kqemu and qvm86 change qemu from an emulator into a virtualizer. So, if you run qemu on a P3 with Kqemu/qvm86, you don't get SSE2/3, and therefore you don't run Mac OS X.

[Dragon]

You are right... we have to use both pearpc and qemu (with no accelerator).

[mac-mini]

so thats my problem in windows im using kqemu

[munky]

so there's no way to virtualise non-SSE instructions and emulate the rest?  

maxxuss used to trap the exception raised by the lack of an SSE3 instruction and trapped it to call emulation code, i think.  is there any way we could do something similar?

[Dragon]

qvm86 says :
Virtualisation allows "emulated" code to be run natively on the host cpu, using the CPU protection mechanisms to intercept and emulate priveleged events.

i'm wondering if it's possible to set which events are "priveleged".
maybe we could set all SSE2 calls to be emulated and the rest virtualized then.

[Embio]

surely emulating instructions  on something that runs them natively would be fast anyway, because it isnt going to take much effort to do? thats the nature of computers - it has to be faster than x86 on PPC or the other way round - no?

[munky]

Embio - dont bet on it.  Emulating a full x86 processor doesnt get any faster just because your executing your emulation code on an x86 processor.  You have to specifically code to take advantage of x86 features if you want to optimise.  Thats why QEMU on x86 is slow without the accelerator module, which does offload the work to the real processor.

What we want is a hybrid approach where instructions supported by the processor are run 'natively' or at least executed in a virtualisation-style way, and instructions which are not supported (eg SSE, SSE2, SSE3) are emulated in software.

[JaS]

I wonder if the same sse2 emulation that is used in Qemu can be ported to the kernel and be done with it. Is the Qemu sse2 emulation avalable in source code somewhere ...

[Embio]

Qemu is open source, thats going back to my original suggestion of integration. Obviously its the easiest way of doing things, and the fastest but we need someone with a lot of experience in this area

Edited by Embio, 12 February 2007 - 09:14 PM.

[munky]

that would be a great solution, and would negate the need for qemu and linux.  any volunteers?

EDIT: I think ops_sse.h in target_i386 of the qemu source will provide interesting reading...

EDIT2: If this is possible, it would be HUGE!  Imagine all the 'disenfranchised' machines which could be supported...