Intel speeds up the Athlon64

[samwichse]

Hilarious but true. The author used the intel compiler to compile a program that showed an especially high boost when "optimized for the P4" and removed the flags that would run it only on a P4 (it wasn't checking for just SSE or SSE2, but that it was being run on an intel processor).

The result? The Athlon64 showed the same speed increase the P4 did. What does this mean? Well, for one thing that the intel compiler speeds up all x86 code, but prevents anything but one of their own processors from running it.

The link:

http://groups.google.ca/groups?dq=&h...ing.google.com

Sam

[ArcticFox]

That's really dumb. Not the hack that the guy did, but that Intel checks your proc before enabling the flag. Someone needs to write a utility that will do that to every program written with the compiler.

[nukes]

But if they did that, what would happen if it was run on a 32 bit Athlon or PIII?
Maybe extend the check to support Athlon64s, but removing it altogether would be "really dumb" as you put it.
Also, I doubt intel would do that just yet, the P4 is its main plug in the high-end market. Unless the GNU project come up with an equally performing compiler (it may just happen in the next few point releases i.e. 3.4/3.5-4.0) Athlon64s aren't going to be directly supported by the intel compiler.