Getting asm macros expanded

[Philipp Marek]

On Thursday 19 March 2009 Nick Clifton wrote:
> > Some background information is available at
> > 	http://gcc.gnu.org/ml/gcc/2009-02/msg00497.html
>
> Interesting.  You know you could probably achieve the same effect in the
> linker...
But the linker sees no assembler statements - it would have to deassemble the 
code, right?

> It already has mechanisms in place for changing the code 
> (search for "relax") 
Yes, thank you.

> and if you made the transformation in the linker 
> you would not have to worry about assembler macros. 
That's true.

But for me as an outsider it's much easier to provide some scripts working via 
STDIN/STDOUT than looking in the C sources of some big project.

Of course, if that's seen as valuable it might be re-implemented in the linker 
or assembler - which would provide great speed benefits, too.

> Plus you would be 
> able to transform code in libraries as well...
Libraries can be done - they just have to be re-compiled, like normal 
binaries. Or do I misunderstand you here?

I even thought about exporting such often-used sequences via glibc or 
whatever, so that programs can use these directly - but I think that the jump 
into another object wouldn't provide so much benefit (space of 
exports/imports, speed of double jump, dependency on exact identical code 
sequences, etc.)

I nearly got a kernel to compile (only some symbols couldn't be resolved 
in .tmp_vmlinux1), and got 30500 bytes off the binary - that's 0.73%.
Other things like libgcj.so.90 could get 1.5% (rouch estimate, didn't try to 
compile), because this one is nearly *only* functions with common endings.


Regards,

Phil

-- 
Versioning your /etc, /home or even your whole installation?
             Try fsvs (fsvs.tigris.org)!