Sadly, at least in the embedded space some of us still deal with platforms where the proprietary core vendor's compiler routinely beats open source compiler cyc

Sadly, at least in the embedded space some of us still deal with platforms where the proprietary core vendor's compiler routinely beats open source compiler cycle counts by a factor of 1.5 to 3. The big weak region seems to be in-order machines with smaller numbers of general purpose registers. GCC at least seems to do its basic block planning entirely before register allocation with no feedback between phases.