>> > should we support -march=pentiumpro ?
>> 
>> What do you mean?
>
>I guess there are many codes in our tree
>that produce bad code with -march=pentiumpro.
>so, I wonder we should fix them or just give up
>using -march=xxx...
>

They should be fixed. Almost all of them are due to the compiler doing whats
allowed there per it's docs and optimizing things. If you don't want items
potentially re-ordered you have to add the volatile attributes.

I'd rather be able to run (at my choice) with a compiler that will actually
optmiize for my current processor rather than the original x86 model only..

James