Created attachment 333 [details]
Merged twin vec64 reverse calls with single vec128 reverse call
Replaced separate NEON vrev64_f32 calls for the low/high sections of each vector with the combined vrev64q_f32 version which does both at once.
Thank you for the patch:
User: Simon Pilgrim
Date: 2013-06-10 15:59:03
Summary: Fix bug 591: minor optimization in NEON vectorization support