2469 replies to this topic

[IiTz ShAnE]

ML on AMD would be great. Its a shame I cant code to help but I would happily test on my AMD Athlon II X2 215

[spakk]

IiTz ShAnE, on 28 October 2012 - 04:03 PM, said:

ML on AMD would be great. Its a shame I cant code to help but I would happily test on my AMD Athlon II X2 215

this is good, you can use the latest kernel "connatic" and "Xnu.diff" check on your computer

[IiTz ShAnE]

spakk, on 28 October 2012 - 04:07 PM, said:

this is good, you can use the latest kernel "connatic" and "Xnu.diff" check on your computer

Are they listed in this thread?

[SS01]

yes, look in the previous pages

spakk, is lion on amd relatively stable, or should i stick to snow leopard

[spakk]

SS01, on 28 October 2012 - 04:42 PM, said:

yes, look in the previous pages

spakk, is lion on amd relatively stable, or should i stick to snow leopard

Lion 10.7.2 to 10.7.5

with CPU Phenom II X6 1100T is not stable also only 32-bit DP2 Finder. In between comes kernel panic, it's only for testing purpose.
With CPU FX 8150 8-core, it seems to work very well, with 64-bit

What kind of CPU do you have?

if you do not not have CPU FX8150, then install SNOW 10.6 and then update to 10.6.8.

[SS01]

spakk: okay, was just curious, i have basically your cpu but with 4 cores

[spakk]

I have dug a nice toy from the old forums

"581_AMD_Intel_SSE2_SSE3_Kernel_Patcher" and "OSx86_Socket_Source_Code" checks that, please!

http://www.workupload.com/file/VtYUP7j

SS01, on 28 October 2012 - 06:49 PM, said:

spakk: okay, was just curious, i have basically your cpu but with 4 cores

Snow 10.6.8 works very well

[spakk]

IiTz ShAnE, on 28 October 2012 - 04:14 PM, said:

Are they listed in this thread?

see here #275

[spakk]

Author: Nero24
_{Discrimination by Intel compiler on AMD systems deal}
The theme is basically a few years old and also in our forum discussed sufficiently, so we here simply refer to the relevant threads and messages:

• Still: Intel compiler disadvantaged AMD CPUs [P3D forum]
  
• Antitrust action: boycott the Intel compiler AMD chips?
  
• Futuremark preferred Intel processors?
  
• U.S. anti-trust accuses Intel

Short and concise summary: Intel maintains its own compiler and the high financial cost.  Now you can argue about whether it is "fair" - if that term is applicable at all in a competitive situation on the free market - that Intel will ensure that the optimization of the Intel compiler mostly Intel processors used to benefit and AMD processors - although they would eventually benefit from the optimizations - not.

Already 2 years ago, we have reported in detail the difference it can make in terms of performance, if manufacturer-specific optimizations are used or not.  At that time, the VIA Nano processor was used as a comparison, which makes it possible to change the CPUID Vendor VIA processor and the program - in this case the Futuremark PCMark2005 - An Intel pretend.  Then the value in the memory benchmarks increased from 1845 (CPUID = VIA) to 2721 (CPUID = Intel) points.

With an AMD processor, however, it is not so easy to manipulate the visible Vendor ID for the programs.  Only option here: virtualization.  I set up on VMWare up a virtualized system, it is very possible the guest to miss another CPUID, when the host system actually has.  In Agners CPU blog explains how this works:

By analogy to Andrew's code, I assume that you can make an AMD processor spoof to be "GenuineIntel" with these lines:

cpuid.0.ebx = "0111:0101:0110:1110:0110:0101:0100:0111"
cpuid.0.edx = "0100:1001:0110:0101:0110:1110:0110:1001"
epuid.0.ecx = "0110:1100:0110:0101:0111:0100:0110:1110"

The Intel software checks so the family number, Which should be set to 6:

cpuid.1.eax = "0000:0000:0000:0001:0000:0110:0111:0001"

The software then looks inside the guest an Intel processor, even if the host system has an AMD CPU.  If the performance-critical software to be used, translated with an Intel compiler, which should improve performance significantly.  Provided however, that the software used all the flags correctly polls, because not always dominate current AMD processors and all commands from the comparable Intel processors.  Example: SSSE3, which previously dominated non-AMD processor, or special SMT SIMD instructions older Intel processors, which were not on AMD processors in the absence of SMT implemented.  The software asks all relevant flags from correctly, that's no problem.  However, the software is programmed sloppy and "thinks", "ah, Intel processor, then this and that must indeed be supported", the program crashes when using this virtualization tricks out of hand.

is it possible to implement something in the code? So that the kernel thinks it is such a Intel i7 980 cpu and not a CPU AMD Phenom II X6 1100T is? Something like a generator compared with various Intel and AMD cpus. sorry for my bad english.

here is the download link of cpuidfake:
http://www.agner.org...mize/#cpuidfake

http://www.agner.org/optimize/


Important information for programmers:
http://devgurus.amd.com/thread/95754


look here for if this is useful?
http://code.metager....ochs/cpu/cpudb/

[SS01]

I'm sorry spakk, I'm not a programmer. I think we'll have to wait till connactic gets back.

[spakk]

SS01, on 29 October 2012 - 10:38 AM, said:

I'm sorry spakk, I'm not a programmer. I think we'll have to wait till connactic gets back.

We all hope theconnatic can program an amd bulldozer kernel..

[IiTz ShAnE]

Maybe there is something to be learnt from the Lion AMD kernel thats out?

[spakk]

IiTz ShAnE, on 29 October 2012 - 12:14 PM, said:

Maybe there is something to be learnt from the Lion AMD kernel thats out?

I don't know, probably by Andy Vandijck legacy kernel? I am not a programmer

[spakk]

What do you think ,#310 possible?

[SS01]

spakk, ist Ihr Englisch besser als mein Deutsch!

spakk, on 29 October 2012 - 08:14 AM, said:

Author: Nero24
_{Discrimination by Intel compiler on AMD systems deal}
The theme is basically a few years old and also in our forum discussed sufficiently, so we here simply refer to the relevant threads and messages:

• Still: Intel compiler disadvantaged AMD CPUs [P3D forum]
  
• Antitrust action: boycott the Intel compiler AMD chips?
  
• Futuremark preferred Intel processors?
  
• U.S. anti-trust accuses Intel

Short and concise summary: Intel maintains its own compiler and the high financial cost.  Now you can argue about whether it is "fair" - if that term is applicable at all in a competitive situation on the free market - that Intel will ensure that the optimization of the Intel compiler mostly Intel processors used to benefit and AMD processors - although they would eventually benefit from the optimizations - not.

Already 2 years ago, we have reported in detail the difference it can make in terms of performance, if manufacturer-specific optimizations are used or not.  At that time, the VIA Nano processor was used as a comparison, which makes it possible to change the CPUID Vendor VIA processor and the program - in this case the Futuremark PCMark2005 - An Intel pretend.  Then the value in the memory benchmarks increased from 1845 (CPUID = VIA) to 2721 (CPUID = Intel) points.

With an AMD processor, however, it is not so easy to manipulate the visible Vendor ID for the programs.  Only option here: virtualization.  I set up on VMWare up a virtualized system, it is very possible the guest to miss another CPUID, when the host system actually has.  In Agners CPU blog explains how this works:

By analogy to Andrew's code, I assume that you can make an AMD processor spoof to be "GenuineIntel" with these lines:

cpuid.0.ebx = "0111:0101:0110:1110:0110:0101:0100:0111"
cpuid.0.edx = "0100:1001:0110:0101:0110:1110:0110:1001"
epuid.0.ecx = "0110:1100:0110:0101:0111:0100:0110:1110"

The Intel software checks so the family number, Which should be set to 6:

cpuid.1.eax = "0000:0000:0000:0001:0000:0110:0111:0001"

The software then looks inside the guest an Intel processor, even if the host system has an AMD CPU.  If the performance-critical software to be used, translated with an Intel compiler, which should improve performance significantly.  Provided however, that the software used all the flags correctly polls, because not always dominate current AMD processors and all commands from the comparable Intel processors.  Example: SSSE3, which previously dominated non-AMD processor, or special SMT SIMD instructions older Intel processors, which were not on AMD processors in the absence of SMT implemented.  The software asks all relevant flags from correctly, that's no problem.  However, the software is programmed sloppy and "thinks", "ah, Intel processor, then this and that must indeed be supported", the program crashes when using this virtualization tricks out of hand.

is it possible to implement something in the code? So that the kernel thinks it is such a Intel i7 980 cpu and not a CPU AMD Phenom II X6 1100T is? Something like a generator compared with various Intel and AMD cpus. sorry for my bad english.

here is the download link of cpuidfake:
http://www.agner.org...mize/#cpuidfake

http://www.agner.org/optimize/


Important information for programmers:
http://devgurus.amd.com/thread/95754


look here for if this is useful?
http://code.metager....ochs/cpu/cpudb/

Hmm.. This seems like a rather ambitious idea, but I can actually see it working. I'm learning C, but veeeeeeery slowly - I've just learned how to declare an array, lol. Lets ask connactic about it when he gets back.spakk, ist Ihr Englisch besser als mein Deutsch!

[spakk]

SS01, on 29 October 2012 - 02:47 PM, said:

spakk, ist Ihr Englisch besser als mein Deutsch!

Hmm.. This seems like a rather ambitious idea, but I can actually see it working. I'm learning C, but veeeeeeery slowly - I've just learned how to declare an array, lol. Lets ask connactic about it when he gets back.spakk, ist Ihr Englisch besser als mein Deutsch!

I really hope that someone is able, at the end result, a running kernel.

PS: Dein Deutsch ist gut

AMD Phenom II on Linux

http://duopetalflowe...bit-kernel.html

when running on Linux then it should also be possible to run on Mac OS X?

[IiTz ShAnE]

Here it is trying to boot on my AMD system in my signature,

 IMG_0022.JPG   366.69K   21 downloads

[spakk]

IiTz ShAnE, on 29 October 2012 - 03:54 PM, said:

Here it is trying to boot on my AMD system in my signature,

IMG_0022.JPG

we have to wait until theconnatic builds us a new kernel. This is basically the same kernel panic

you can try with the following boot flags;
-v -x cpus=2 arch=x86_64 maxmem=4096 busratio=20
or
-v -x cpus=2 arch=x86 maxmem=4096 busratio=20

[IiTz ShAnE]

Also tried it on an old P4 CPU and the panic on that is "can't preform kext scan: no kext summary" the rest of the screen is the same as above.

[spakk]

IiTz ShAnE, on 29 October 2012 - 04:16 PM, said:

Also tried it on an old P4 CPU and the panic on that is "can't preform kext scan: no kext summary" the rest of the screen is the same as above.

Problem to load corecrypto.kext