Morgan Wesström wrote:
-march=prescott -O2 -fomit-frame-pointer seems to be a general consensus
when googling but I'm unable to find the reasoning for it.
-march=core2 because the Atom is claimed to
be "merom ISA compliant"
The -mfpmath=sse confuses me too since more googling claims that the SSE
instructions on the Atom use several times the number of clock cycles
they use on core2 or prescott
-march=prescott is the usual option for 32-bit SSE3 support, so seems a
reasonable stop-gap. One would think that it implied -msse3. Even if
SSE instructions are relatively slow, you might expect 387 to be even
slower, so no reason to avoid sse. If you're into experimentation, you
could try -march=pentium-m, for the best of sse2 and the best of 387,
but no sse3. If you were able to bring up the referred patch page,
didn't it shed any light?
I don't know why -march=core2 would be useful, except that it presumably
enables you to use SSSE3 intrinsics, should you want them. Yes, the
atom presumably supports the same instruction set as Core 2.
No one can make a blanket recommendation, certainly not without the
familiarity you should have with your own code. -mtune=barcelona might
even work sometimes.
