Compiler effect on FPU of P4 and AXP/Thunderbird

I hardly understood what you said juin but anyway:


No actually, the Athlon XP's FPU is around 2.5X faster than the P4's according to SiSoft Sandra.


Not because of the FPU...



AMD technology + Intel technology = Intel/AMD Pentathlon IV; the <b>ULTIMATE</b> PC processor

Basically the MS compiler does not perform automatic vectorization, while the Intel compiler will analyze the loop structures and produce SSE/SSE2 compatible instructions.

With out a doubt, Athlon’s x86 FPU is the fastest often producing 2 to 3 parallel FPU instructions per clock. i.e. the 2.5x number. Add the XP SSE and 3dnow compatible FPU system that may under certain cache situations outperform Intel’s.

Then the final card in the deck SSE2, twice the pack size twice the precision. On a high throughput bus, it just can’t be beat.

It is as fair to say AMD has a 2.5x FPU, as it is to claim that Intel wears a FPU crown that puts it leaps and bounds ahead of AMD. I would say the flagships from Intel and AMD perform remarkably close despite their radically different cache and pipeline architectures.

Will Microsoft adopt an auto-vector unit into their compilers? Who knows? The real question is, if a company like Bill’s incorporates such a technology, will it grow to include 3dnow and will the programmers become too lazy to understand and optimize their own data?

What do you mean WILL they become too lazy?  

Most (All) programmers are lazy. To properly use streaming/vectorized instructions it takes more than just calling the appropriate function. You have to design your data to fit your code. There may be some speed up done by auto-vectorizing code, but I guarantee it will never compete with hand coded instructions.

No actually, the Athlon XP's FPU is around 2.5X faster than the P4's according to SiSoft Sandra

Standart test FPU of a northwood/willi is near 3700.
Not because of the FPU...

SSE2



http://service.madonion.com/servlet/Index?pageid=/orb/p...

Basically the MS compiler does not perform automatic vectorization, while the Intel compiler will analyze the loop structures and produce SSE/SSE2 compatible instructions.

If you have read all the test.You have see that Intel compiler make a auto vectorization for SSE2 SSE MMX so not only P4 but all actual CPU except maybe VIA.Also it a plug who can be use with MS compiler who do more parralle instruction good for K7. Everyone gain from new compiler.

So why not use MS with intel plug-ins?

Maybe someone can anserwer this does intel compiler is free and how hard to use?

http://service.madonion.com/servlet/Index?pageid=/orb/p...

The Intel C/C++ Compiler is available for free on Linux and for a free 30-day trial for Windows. The Windows version is a plugin where you check a box in VC++ and it uses the Intel compiler rather than the Microsoft one when you hit the normal Build button(s). The Intel compiler comes only with the compiler. The libraries that are used are still the ones that come with VC++. All VC++ command-line options are supported by the Intel compiler, so it will plug right in without any problems after running the installer.

<A HREF="http://www.intel.com/software/products/eval/" target="_new">http://www.intel.com/software/products/eval/&lt;/A>

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

So there is no reason why all new software come with SSE2.

http://service.madonion.com/servlet/Index?pageid=/orb/p...

So there is no reason for all the new software to come with SSE2/SSE .

http://service.madonion.com/servlet/Index?pageid=/orb/p...

Ummm...I assume you're saying that there's no reason <i>not</i> to compile with Intel's plugin?

Well, for Windows, it does cost money after the 30-day trial. Commercial developers can't get away from that; if they want to sell a product precompiled with the Intel compiler, they have to pay for the compiler. Linux developers can't get away from that either; Intel's Linux C/C++ compiler is only free in Linux for non-commercial use.

<i>If a server crashes in a server farm and no one pings it, does it still cost four figures to fix?

Actually guys the K-7 is being held back by today’s software as much as the P4.
P6 made code just don’t have enough Instruction-Level parallelism to make significant gains from Athlons two FPU units (the FADD and FMUL)… this is also true for the ALU of P4 and K-7.

the K7 was a huge boast in parallel execution of x86/87 code and thus up to today most software don’t know how to take full advantage of both P4 nor K7 execution units.

this us due to lack of p7 compilers...
Visual studio .net is out btw...
----------------

AMD Athlon™ Processor x86 Code Optimization Guide is a very very nice piece of work. you can download it here <A HREF="http://www.amd.com/us-en/assets/content_type/white_pape..." target="_new">PDF</A>



This post is best viewed with common sense enabled