Replies to post #3266 on Advanced Micro Devices Inc (AMD)

[dougSF30]

EP- Re: It could be intentional and I wouldn't have a problem with that as long as it was disclosed and understood.

Sure, they could do that. They have every right to. But,

(1) It looks kinda pathetic-- "oh, you're *scared* of your complier output running on the Opteron? Why is that?"

(2) I'll guess it wouldn't take long for someone to post a hack that short-circuits any "check" to return whatever you want it to.

Doug

[gollem]

Elmer,

We can hardly expect intel to produce compilers that create code that runs well on amd chips, that's true. Intel will put some effort into trying to prevent this. There's nothing wrong with the workaround, however, it indicates that speed improvements are possible. I hope AMD will (help) create some compilers that optimize for Opteron but they do have limited means.

[Haddock]

"If the compiler checked the CPU ID and chose optimized code only for genuine Intel CPUs then I wouldn't have a problem with that."

How about if the generated code just didn't work at all on AMD CPUs? Presumably you wouldn't have a problem with that, since AMD deserves no free ride. On the other hand I don't think users of Intel's compilers would like that feature too much. I don't think they would like it if Intel-produced code failed to use SSE2 on AMD CPUs, but they might not know it's happening.

On the other hand I don't know that Intel-compiler-produced code really fails to use SSE2 on AMD chips, so the discussion is a bit hypothetical.