On 6/26/2012 6:39 AM, Michael Zintakis wrote:
Personally, I think using "-O2 -march=atom -msssse3 -mfpmath=sse
-fomit-frame-pointer -pipe" is decent enough. Any other opinions? I
am using gcc 4.4.5.
-msssse3 -> -mssse3
Yeah, it was a typo, thanks!
As I know, -mfpmath=sse may slow some code speed in some
architechures, but I don't know if it also is on ATOM.
No idea! Hopefully, some of the more knowledgeable on this ML could
provide some insights.
Do you refer to the pentium-m? Last I knew, gcc -m32 still had
available -march=pentium-m, so you could experiment with that if you are
interested. IIRC, there was just one decoder for SSE2, so it was
possible to gain by mixing 387 and sse when vectorization was not
successful. I've never heard of much interest in comparing atom with
pentium-m, even though each might have been a reasonable choice for
personal laptop in its time.
--
Tim Prince
