TASK FORCE - [AMD] to missing SSE4.1 Instruction set - older CPUs on macOS Sierra (10.12.x)

[Mr.Darkwraith 0]

Wouldn't this kernel work on 65nm Core 2 processors?

[ITzTravelInTime 622]

Wouldn't this kernel work on 65nm Core 2 processors?

unfortunately we haven't managed to create a working kernel yet, because of the complexity of doing this job, emulates this instructions is very difficult

[AXAXAXAXAXAX 52]

yes it boots, but does not olow that's what it has become. A new art

that's el capitan not sierra

After booting with the old El Capitan kernel, my desktop picture looks like this.(Pic.4)

Probably there is an issue in roundd_ and  rounds_. With ydeng's latest code the system boots with OPEMU: ud2 error and never reaches blank screen, but if the code for roundd_ and  rounds_ is replaced with ydeng's previous one for them, the system will boot to the blank screen and without OPEMU: ud2 error.

 

[spakk 1,806]

Probably there is an issue in roundd_ and  rounds_. With ydeng's latest code the system boots with OPEMU: ud2 error and never reaches blank screen, but if the code for roundd_ and  rounds_ is replaced with ydeng's previous one for them, the system will boot to the blank screen and without OPEMU: ud2 error.

 

I will test now the new OPCODE from ydeng

(the pictures are from the experiment with the first OPEMU from yedeng)

[spakk 1,806]

the same here with sierra kernel, Opemu ud2 error loop

[AXAXAXAXAXAX 52]

the same here with sierra kernel, Opemu ud2 error loop

I get OPEMU ud2 error loop with the latest code from ydeng as well.

 

Can you test this Sierra kernel?

kernel.zip

[spakk 1,806]

I get OPEMU ud2 error loop with the latest code from ydeng as well.

 

Can you test this Sierra kernel?

I'll check this

and here is my last kernel

I get OPEMU ud2 error loop with the latest code from ydeng as well.

 

Can you test this Sierra kernel?

it stops at this point

AMD10.12-SSE4.1-Ydeng.zip

[Shaneee 1,456]

Trying to build a kernel for you to test but having an issue with the OPEMU posted. There's an undefined symbol,

undef: _op_sse41_run

Where is this getting defined?

[spakk 1,806]

is it possible that you "ve not listed sse4.1 in osfmk/conf/files.x86_64   ?

[Shaneee 1,456]

And there is the half asleep mistake lol Forgot to add the addition of sse41.c to the conf file 

 

Edit: Give this a go guys. 

xnu-3248.60.10_OLD_CPU.zip

[AXAXAXAXAXAX 52]

And there is the half asleep mistake lol Forgot to add the addition of sse41.c to the conf file 

 

Edit: Give this a go guys. 

it doesn't boot at all. just showing black screen

[spakk 1,806]

Note that Shanee's kernel is a kernel for El Capitan ☺

[AXAXAXAXAXAX 52]

Note that Shanee's kernel is a kernel for El Capitan ☺

El Capitan:

[spakk 1,806]

Hey Ydeng, is it possible to integrate the AVX2 into OPEMU, so that to emulate AVX2 features for older CPUs ?

The problem may be caused by a unavailable AVX2 features. Please look at the content of this new file, since 10.12.0 : "osfmk/x86_64/lz4_decode_x86_64.s"

 

Edit:

Maybe I'm wrong with my guess. This assumption exists since apple has published the first Sierra source. Maybe someone has a viable solution to this problem.

 

that comes very close to the note you've made here:
(unfortunately I had other thoughts at that time)

http://www.insanelymac.com/forum/topic/322370-task-force-amd-to-missing-sse41-instruction-set-older-cpus-on-macos-sierra-1012x/page-4?do=findComment&comment=2497520

[Shaneee 1,456]

Strange thing is in theory that should still boot on El Capitan. I'll play about with the opemu and try on my old Athlon as it's setup now.

 

GPU decided not to work in old PC so can't test. Give this a go on Sierra when you get a chance,

 

 

Sierra_Test.zip

[AXAXAXAXAXAX 52]

Strange thing is in theory that should still boot on El Capitan. I'll play about with the opemu and try on my old Athlon as it's setup now.

 

GPU decided not to work in old PC so can't test. Give this a go on Sierra when you get a chance,

It gets kernel panic after something like "OPEMU: pinsrd xmm1, esi, 0x1".

You can comment these lines to disable debug prints so I will be able to show that in pic

https://github.com/Shaneee92/SierraAMD-XNU/blob/34b926dc342319403feb392575dc6505fb6aa1e5/osfmk/i386/AT386/model_dep.c#L1468

https://github.com/Shaneee92/SierraAMD-XNU/blob/34b926dc342319403feb392575dc6505fb6aa1e5/osfmk/i386/AT386/model_dep.c#L1471

 

 

[Gigamaxx 566]

It gets kernel panic after something like "OPEMU: pinsrd xmm1, esi, 0x1".

You can comment these lines to disable debug prints so I will be able to show that in pic

https://github.com/Shaneee92/SierraAMD-XNU/blob/34b926dc342319403feb392575dc6505fb6aa1e5/osfmk/i386/AT386/model_dep.c#L1468

https://github.com/Shaneee92/SierraAMD-XNU/blob/34b926dc342319403feb392575dc6505fb6aa1e5/osfmk/i386/AT386/model_dep.c#L1471

 

I've noticed on your previous pic you are using MacMini and this one 14.2, have you tried MacPro 6.1 or 5.1?  

[Shaneee 1,456]

These SSE4.1 instructions are also present and need added to the opemu. I've attempted to add to the opemu but too many errors on building,

PCMPEQB
PCMPEQW
PCMPEQD
PCMPEQQ

[ITzTravelInTime 622]

Hy guys, any news? The project seems to be dead, i hope, that there is someone with new ideas to unlock the situation, i have also seen that maybe creating some assembly function into the c code to perform the sse4.1 tasks could be helfoul and also to chech the compiled assembly for thing like worg opcodes or instructions operands will be also usefould, but i understand that it's a lot complicated

[ITzTravelInTime 622]

Reding apple's documentation another usefoul thing is to listen to kernel debug prints from a serial connection or network connection from another machine, so you won't miss any debug print and you can read all the prints from the kernel and system processes, just use the kdp_match_name boot arg as in this examples

 

kdp_match_name=en1

kdp_match_name=com1

 

Edited May 20, 2018 by ITzTravelInTime