It would be fine if the regression tests would
also report resource usage like:
Summary for 'mcs51': 0 failures, 2479 tests, 480 test cases,
xxx cpu cycles, yyy byte code
This would help to see advances / avoid regressions
in code generation.
One could f.e. notice if a peephole does not
apply any more and would have to be adapted.
(Stuff like mov a,#0x00 versus clr a, ordering
of registers, label not there anymore and the like.)
Furthermore there finally is an objective argument in
the MyCpuIsBiggerThanYours debate;)