r1618472 | brane | 2014-08-17 11:55:54 +0000 (Sun, 17 Aug 2014)
Remove some code in libsvn_delta that assumes unaligned access is safe
and therefore breaks GCC's vectorizer. The new code also attempts to
make better use of the CPU data cache.
(fast_memcpy): Simplify the short copy loop implementation.
(patterning_copy): Do not blindly use unaligned word-size access.
Keep the data cache hot when repeating a pattern.
Found by: Alan Modra