http://gcc.gnu.org/bugzilla/show_bug.cgi?id=47769
--- Comment #2 from Matthias Kretz <kretz at kde dot org> 2011-02-17 10:00:55 UTC --- Created attachment 23375 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=23375 Test code to see whether btr gets used automatically and to compare speed compile with g++ -O3 -march=core2 -msse4 -Wall -funroll-loops -funroll-loops is not required to see the speedup, but it shows that a higher instruction level parallelism can be achieved with the use of btr.