https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124434
--- Comment #10 from Andrew Pinski <pinskia at gcc dot gnu.org> --- (In reply to Brian M. Sutin from comment #9) > I often develop code using only "-g". Suddenly finding my code running 56x > slower sent me on a wild goose chase. I would say that my experience was not > one of "debuggability". However, it is 100% up to you gcc folk to decide if > this is an issue. All I can do as a user it toss it your way. I am described exactly why it is slower. Basically x87 stack loads don't have a normal stack bypass so it needs to wait for the earlier store to finish. This is unlike sse and gpr loads. That is the quark you are seeing. Since you are compiling at -O0, all variables live in memory for debugability reasons. so things are stored out from the x87 stack to the normal stack and then loaded back.
