http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59420
--- Comment #4 from Alexander Holler <holler at ahsoftware dot de> --- Thanks a lot for the very fast solution. I wasn't aware that gcc does such optimizations and maybe posted the bug too fast without really having examined the assembler. Using -fno-tree-loop-distribute-patterns indeed fixes the problem. Again, thanks a lot.