Working on a clang patch for taking advantage of native cpu instructions. So, far -march=native seems to be fine for Haswell E. Using optimizations beyond the default cpu instructions might improve performance. Is there any reason this is not turned on? In the editor which actions should I take to check if it breaks anything during runtime?
Couple of years back I’ve compiled engine compiler flags. Engine not fully stable with it but usable and there is no or little performance improvement doing so. It also adds compatibility issue when are packing project / plugin.
There is more advantage if you can use simd in code without causing architecture / platform/ compiler specific issue.
If you need examples see the some of the newest cinematics plugins in the by ep.ic.
It’s not used because epic typically build the engine once for distribution, rather than locally. And the people who do build it locally, are going to be debugging something in the engine.