Analyzing Optimization: What's New in 3.4?

[Nathanael Nerode]

>What about -fnew-ra, -ftracer, and other options rarely discussed (and 
>probably unknown to most gcc users)?

-fnew-ra is still experimental because it still fails on code which the 
old register allocator allocated Just Fine.  It also doesn't usually 
give better performance at the moments.

Its purpose is mainly to improve the future maintainability of GCC, 
since the current regalloc/reload mishmash is very, very, very hard to 
maintain.

You might like to try Geoff Keating's intermodule optimizations.
They are enabled in some funny way which I can't remember.  :-)

There's a profiling-directed branch prediction option which I can never 
remember how to use, but which seems to really do wonders.

Be sure to mess around with -fschedule-insns and -fschedule-insns2.
It seems that on many machines -fschedule-insns creates substantially 
*worse* code (so use -fno-schedule-insns) and -fschedule-insns2 creates 
slightly better code, but all four combinations should be tried.  ;-)
 
-- 
Nathanael Nerode  <neroden at gcc.gnu.org>
http://home.twcny.rr.com/nerode/neroden/fdl.html