I was wondering if anyone had any hard numbers on ARM vs Thumb code performance on iPhone 3GS. Specifically for non-floating point (VFP or NEON) code - I\'m aware of the issues
Thumb code will essentially always be slower than equivalent ARM. The one case where Thumb code can be a big performance win is if it makes the difference between your code fitting into on-chip memory or cache.
It's hard to give exact numbers on performance differences, because it's entirely dependent on what your code actually does.
You can set per-architecture compiler flags in XCode, which would avoid breaking the simulator build. See the XCode build setting documentation.