Code Bloat g++

Neelakanth Nadgir NEELAKANTH.NADGIR@SUN.COM
Tue Feb 15 14:28:00 GMT 2000


> The place is right; the information is wrong. What system?  On
> i586-pc-linux-gnu, glibc 2.1.3, shared libstdc++, I cannot at all
> reproduce this effect I'm compiling oopack_v1p8.C with -O0..4, with
> and without -g. The resulting executables are
> 
> -rwxrwxr-x   1 martin   martin      37767 Feb 15 23:07 o0
> -rwxrwxr-x   1 martin   martin      21937 Feb 15 23:07 o1
> -rwxrwxr-x   1 martin   martin      21969 Feb 15 23:07 o2
> -rwxrwxr-x   1 martin   martin      22641 Feb 15 23:08 o3
> 

For my system (solaris 8/ ultra 10, 2.95.2 shared libstdc++, glib
1.2.6, ) I get the following results.
-----------
153k Feb 15 14:23 O*
153k Feb 15 14:23 O2*
153k Feb 15 14:23 O3*
153k Feb 15 14:23 O3__mtune=ultrasparc*
153k Feb 15 14:23 O3__mtune=ultrasparc__mcpu=v8*
153k Feb 15 14:23 O3__mtune=ultrasparc__mcpu=v8__ffast_math*
167k Feb 15 14:23 Os*
205k Feb 15 11:48 noflags*

After stripping
48k Feb 15 14:25 O*
48k Feb 15 14:26 O2*
48k Feb 15 14:26 O3*
49k Feb 15 14:26 O3__mtune=ultrasparc*
49k Feb 15 14:26 O3__mtune=ultrasparc__mcpu=v8*
49k Feb 15 14:26 O3__mtune=ultrasparc__mcpu=v8__ffast_math*
49k Feb 15 14:26 Os*
55k Feb 15 14:25 noflags*


That is 3x more size before stripping!
-----------

The reason I brought this up is because this is one of the few cases
that gcc generates bigger binaries than sun's 5.0 compiler.

thanks,
-neelakanth


More information about the Gcc mailing list