Interesting news from OSnews

[TransformX]

Intel Forced to Remove "Cripple AMD" Function from Compiler?

Activate Windows 7's Hidden 'God Mode' (works for Vista32 as well)

Enna - A media center application (check the screen shots!)

A brand new AMIGA is coming..

[Umfriend]

The issue of the Intel compilers providing suboptimal code for non-Intel processors is old. Years old. I remember reading about in the register and it has been ages since I last ventured there.

[Chucky Cheese]

Good Tip

WOW...I did that "GodMode" thingy and I like that!! I put the folder in the root of the C drive, put a shortcut to that folder on the desktop and called it "All Settings"...sweet!

Good Tip, thanks

[VJ]

Strange that Intel has to remove the cripple amd function... I don't agree with the presence of the cripple function, but after all, it is the Intel compiler, and there are alternatives available that do optimize for amd. And nobody is preventing amd with supplying its own compiler with cripple-intel function.

To have them remove this is a bit borderline I think. What if amd claims afterwards that the generated code could still be better by using some specific amd instructions? Will Intel the be forced to create an amd-optimal compiler?

Jörg

[Umfriend]

Years ago, the issue was that the compiler would ask for certain capability bits from the processor, such as the "I support SSE2". If an Intel processor had the bit set, optimal code would be generated, if a non-Intel processor had the bit set, non-SSE2 code would be generated.

IIRC, at the time, the Intel compilers were sold/marketed as top-of-the-bill compilers for x86 computers.

[Jammrock]

What Umfriend said.

Programmers are very picky about their compilers, and Intel makes a very good one (they do have particularly good insight into the assembly language of the processors) which makes it very popular. It is also very fast and efficient. Those enhancements, or lack thereof, end up translating down into the product that is being compiled.

Say you are making a C++ compiler and you use Intel's compiler to build your C++ engine. Intel's compiler cripples AMD, which makes whatever you compile in your C++ engine work slower on AMD versus Intel.

You see this a lot in the gaming arena. Where a couple of particular games will always run significantly better on Intel chips while the majority of games run within a certain average percent of frames per second of difference.

With a compiler that no longer discriminates you should see the playing field even out in business and entertainment benchmarks. Good for AMD, not as good for Intel.

[VJ]

Yes, but is it up to Intel to make a compiler that performs well? Intel's programmers have a good insight the assemlylanguage of the processor, but can you expect them to have this insight for all processors? Sure, deliberatly crippling should not be allowed, but I just ask myself where it should stop?

I understand that your compiler built on Intel will cripple amd code, but there are other compilers you could use to build your compiler that won't cripple (gcc). I agree that a good compiler is a must, but I don't really agree Intel is the one to have to make it. Now, I see their compiler more as a service towards their (development) users. Anyone writing software were performance is of the importance should be aware of compiler issues, and not blindly go for the Intel one.

[Umfriend]

VJ, the point is that (a) Intel produces a x86 compiler, not an Intel Processor opnly compiler (if they sold iot as such you'd have a point) and (b) it behaves differently for processors based on brand as opposed to defined capability. If a CPU states it supports SSE2 then produce SSE2 code, don't go out of your way to do so for one but not for the other.

So, no, nobody is asking Intel to tune their compiler for AMD, just not purposefully against it.

[VJ]

Yes, and I also said that deliberate crippling should not be allowed.

But even if it is removed, then it still won't solve the problem Jammrock mentions in his last paragraph. Compiled code will run better on amd than before, but still not be optimized to the full extent.

[Umfriend]

We agree then? I thought initially you found it strange that Intel had to remove it.

[Jammrock]

Yes, and I also said that deliberate crippling should not be allowed.

But even if it is removed, then it still won't solve the problem Jammrock mentions in his last paragraph. Compiled code will run better on amd than before, but still not be optimized to the full extent.

It's up to AMD to optimize their core design to make the x86 instructions and extensions (SSE) optimal and faster. Blocking extensions from the compiler for AMD/VIA chips will only level the playing field when the Intel compiler is in play, not make AMD suddenly superior. Though I like the bulldozer design. That could seriously put AMD in the lead if it works as advertised.

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3683

Two int APU's per "core" with one big fat FPU.

[Elie]

Enna looks nice (Enna in Greek means the number one) but they should make it available for Windows 7 as well not just Linux.

[VJ]

We agree then? I thought initially you found it strange that Intel had to remove it.

I agree they have to remove it, but think it should stop there.

It's up to AMD to optimize their core design to make the x86 instructions and extensions (SSE) optimal and faster. Blocking extensions from the compiler for AMD/VIA chips will only level the playing field when the Intel compiler is in play, not make AMD suddenly superior.

True. But even if architectures are the same, processor specific optimizations are always possible, so the Intel compiler will always generate code that makes more optimal use of an Intel than of an AMD processor. So IMO, to get a fair comparison, you either need to a compiler that doesn't optimize towards processors, but only towards architecture (and Intel would be stupid not to optimize their compiler for their processors, so while the removal of the crippling feature is a good thing, the generated code for an amd processor would still not yield a fair comparison); or a compiler that optimizes both processors (and you can't expect Intel to be the one making it).

Enna looks nice (Enna in Greek means the number one) but they should make it available for Windows 7 as well not just Linux.

Looks very nice! But a lot of these things are pictures and user interface... I know this is what attracts people, but features are important. There are a number of alternatives (MediaPortal, LinuxMCE, ...) out there that mostly have similar functionality, and still lack easy access to interesting functionality (e.g. accessing tuners in one computer from computers in the network - MediaPortal allows it, not sure about LinuxMCE).

Jörg