[maemo-developers] [maemo-developers] Strange behavior with -O3 CFLAGS on ARM
From: Giuseppe Zompatori giuseppe_zompatori at yahoo.itDate: Tue Aug 9 01:23:21 EEST 2005
- Previous message: [maemo-developers] Strange behavior with -O3 CFLAGS on ARM
- Next message: [maemo-developers] Strange behavior with -O3 CFLAGS on ARM
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Il giorno lun, 08/08/2005 alle 18.02 -0300, Rudá Moura ha scritto: > Well, in resume: pygame was never the problem, the problem was in > compiling Python 2.4 with -O3 CFLAGS and then compiling PyGame with -O3, > this just make bugged code for ARM. So I recompiled everything with -O2 > CFLAGS to be more conservative. > Going "-O3" instead of "-O2" for production builds tends to produce bigger executables which also tend to be buggier, bloated & slower as opposed to faster. This is something everyone with good gcc knowledge knows ;) And the only really tested configuration from the FSS is "-O2 -g". That said "-O2 -march=<insert your CPU here> -fomit-frame-pointer -ffast-math" usually produces good code which is usually faster and not as bloated as with "-O3" on most platforms. The "-fomit-frame-pointer" bit can be left out on some platforms (as it doesn't give any extra gain) and for debugging because it makes debbugging impposible. No go try to figure out how to tell this to people that put "-O6" in their Makefiles and expect it to be faster than any other option. Cheers, -Giuseppe ___________________________________ Yahoo! Mail: gratis 1GB per i messaggi e allegati da 10MB http://mail.yahoo.it
- Previous message: [maemo-developers] Strange behavior with -O3 CFLAGS on ARM
- Next message: [maemo-developers] Strange behavior with -O3 CFLAGS on ARM
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]