[avr-gcc-list] AVR-GCC 4.3.2 producing larger output than AVR-GCC 4.1.2 with the same source and options (-Os)

[Nicholas Vinen]

This seems rather unfortunate:

$ md5sum *.c
6c66b6bdb0e87a76b146e135b0b0fa0a  charger.c

$ avr-gcc --version
avr-gcc (Gentoo 4.3.2-r2 p1.5, pie-10.1.4) 4.3.2
Copyright (C) 2008 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

$ make
avr-gcc -Os -Wall -mmcu=atmega48   -c -o charger.o charger.c
avr-gcc -Os -Wall -mmcu=atmega48 charger.o -o charger.obj
./check.py charger.obj
text: 3986, data: 128, bss: 52
ERROR: Flash size limit exceeded by 18 bytes.
make: *** [charger.hex] Error 1





$ md5sum *.c
6c66b6bdb0e87a76b146e135b0b0fa0a  charger.c

$ avr-gcc --version
avr-gcc (GCC) 4.1.2 (Gentoo 4.1.2 p1.1)
Copyright (C) 2006 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

$ make
avr-gcc -Os -Wall -mmcu=atmega48   -c -o charger.o charger.c
avr-gcc -Os -Wall -mmcu=atmega48 charger.o -o charger.obj
./check.py charger.obj
text: 3848, data: 128, bss: 52



That's a pretty big difference. If somebody wants to look into it, I've
uploaded the source and object files here:

http://x256.org/~hb/avr-gcc/ <http://x256.org/%7Ehb/avr-gcc/>

Interestingly, avr-gcc 4.3.2 seems to do better on some functions (e.g.
update_buttons, update_state) but functions which contain integer
multiplies by fixed values seem to end up much larger than the older
version. I think if whatever is causing this can be "fixed" then the
output will be even smaller than either case, which would be excellent.



Thanks,
Nicholas.