osx, x86_64:
-O3 -mfpmath=sse -mtune=nocona -msse3 -ftree-vectorize -funroll-loops
...

Yeah, the OS X flags look fine to me (where I specialize ;-) ).

Please use -march=nocona instead of -mtune=nocona.
-mtune will tune the code for specific processor, but -march will actually use the processor specific instructions (and implicitly set -mtune for the target architecture)

hm this might be why the x86_64 version for os x isn't running as fast as it should. (i bumped it up from sse2 to sse3).

---

osx, x86_64:
-O3 -mfpmath=sse -mtune=nocona -msse3 -ftree-vectorize -funroll-loops
...

Yeah, the OS X flags look fine to me (where I specialize ;-) ).

Please use -march=nocona instead of -mtune=nocona.
-mtune will tune the code for specific processor, but -march will actually use the processor specific instructions (and implicitly set -mtune for the target architecture)

hm this might be why the x86_64 version for os x isn't running as fast as it should. (i bumped it up from sse2 to sse3).

I updated the x86_64 os x app, so hopefully it should be running as fast as it should.

---

Please use -march=nocona instead of -mtune=nocona.
-mtune will tune the code for specific processor, but -march will actually use the processor specific instructions (and implicitly set -mtune for the target architecture)

hm this might be why the x86_64 version for os x isn't running as fast as it should. (i bumped it up from sse2 to sse3).

As I remember, SSE2 is automatically enabled for x86_64 processors, so you don't need it. As for the -mtune and -march arguments, I keep forgetting the difference between the two, but that makes sense why -march would give better results.

---