PIII crashes with an invalid instruction

[Michael_Ljunggren]

Hi,
I am writing this on the forum instead of calling Support, as I always appreciate anyone revealing their problems & solutions in public. Anyways...

A customer recently reported a strange error for a product of ours. The program was compiled using Intel C++ 9.1 034 with the switches /QaxTPBNWK. As it was compiled with /Qax... it should run on just about any processor, but the program crashes with an invalid instruction (an SSE2 instruction) on a PIII.

Is build 034 known to have this problem? Is upgrading the way to go? Have I done something wrong?

Thanks for reading. Thanks for a great compiler!

/Michael

[TimP]

My guess is you have combined too many options for anyone to say whether that compiler version would have this specific problem, unless you care to submit a working example. Usually, the overhead for building so many code paths and choosing among them would outweigh slight performance advantages. For example, /QxKW would run on any CPU since P-III, and should give good performance for any code which does not actually need SSE3. The older options aKB haven't been getting as much attention as the newer ones. B has an advantage only for scalar floating point on Banias and Dothan.

[Dale_S_Intel]

Appreciate the info, but it doesn't ring a bell. If you really want a fix for it, I'd recommend that you file a reort at premier.intel.com and provide a test case and we can look into it.

Thanks!

Dale

[levicki]

Michael, do you perhaps have any hand-written inline assembler code in that project?

I am asking because I have just noticed that for functions containing inline assembler, Intel compiler creates an auto-dispatch prolog targeting different CPUs with the same code, and it obviously doesn't check whether that inline assembler code block can actually be executed on any of those targets.