Can you prove that there is a gain? You can test two executables build using the two schemes and measure the difference in execution speed?
BTW, There is a chance that you'll get better speed/performance if you upgrade your compiler
p.s. also look at Link Time Object generation, or something like that in gcc 4.5
Yes, that is something of similar functionality like link time object generation. However I am not that lucky to have a freedom of changing compilers at will. There are such nasty things like build servers and then other engineers somewhere around the world, who are, so to say, not really into software development, but still need to be able to compile the source code. You should get the picture
We did update to 4.4.0 in the beginning of it all however.
I do not have the proof for you. Is adding this functionality worth the effort? That was a part of my question - a codeblocks developer could quickly outline, how much effort there is to have that implemented.
For instance, SQLite is distributing it's source code primarily as an amalgamation - all code put into a single translation unit. They claim a 5%-10% performance gain from that alone. A db engine is quite a special case, I agree. Most of user applications would probably gain much less from it. Still, it is a feature supported by gcc and I guess all of the advanced optimizing compilers, so it would be good to have an option to quickly switch to this kind of configuration for making release builds.