Months ago when Haswell was brand new I had already run some compiler optimization benchmarks looking to see the impact of the new CPU instruction set extensions added by Haswell provided. The results weren't too dramatic over the Ivy Bridge optimization level (core-avx-i), but this weekend I have some fresh results using a GCC 4.9 compiler development snapshot.
The -march= levels tested today on GCC 4.9 included nocona, core2, corei7, corei7-avx, core-avx-i, and core-avx2. For those unfamiliar with these compiler optimizations/tuning, refer to the online GCC documentation.
Overall there weren't any huge surprises, but the compiler benchmark results in full are available via clicking 1308319-SO-GCC49SNAP71. Embedded below are just a few of the more interesting results.