[Bug target/55803] SSE2 double negation less efficient than explicit xor

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55803



--- Comment #1 from Uros Bizjak <ubizjak at gmail dot com> 2012-12-24 13:47:54 
UTC ---

movsd reads 64bit value from memory, but xorps reads 128bit value.