RE: empty function optimizations

["Ken Clark" <ken at gesn dot com>]

> From: Martin v. Loewis [mailto:martin@loewis.home.cs.tu-berlin.de]
> Sent: Wednesday, July 12, 2000 11:58 AM
>
> "erase__t4list2Z3fooZt24__default_alloc_template2b1i0Gt15__list_it
erator3Z3f
> > ooZR3fooZP3fooT1:" and its baz counterpart.  They are line for line
> > identical.
>
> Ok. But I understood your proposal was to have the linker look at the
> object files, so I assembled your file. Doing objdump -h duplicate.o
> |grep erase, I got


Damn.  I screwed up:  I see now there is one line different in the asm too.
Sorry.  Honestly I wasn't trying to pull a fast one.

> Looks similar, right? Unfortunately, it's not identical. Looking at
> the fourth line (offset 0030), you'll see a1040000 in one case and
> a1080000. So your strategy for combining them would fail here.


Conceded.  Of course this is the part where I say "this was a bad example".
I can make up some examples if you insist, but really the only way to see if
this is worth while is to write a util to look for duplicates in real code.

I don't know enough about ld internals to actually do a real merge, but I
should be able to write something to parse through objdump output and
generate some stats.

Maybe I am wrong.  It will be interesting to run vmlinux through it.  There
were three copies of zlib in the kernel source last time I checked :-).

Ken