? (0xe2.0x9a.0x9b@gmail.com) on 11/13/09 wrote:
---------------------------
>Gabriele Svelto (gabriele.svelto@gmail.com) on 11/13/09 wrote:
>---------------------------
>>rwessel (robertwessel@yahoo.com) on 11/12/09 wrote:
>>---------------------------
>>>Now that doesn't mean that real projects can't use parallelism during the build
>>>process. Any parallel make tool will do that for you, and is probably already being
>>>used on large projects. Unfortunately Amdahl bites us at this point and often one
>>>of the biggest consumers of (wall clock) time is then the link step (and frankly
>>>linker performance is often astonishingly slow).
>>
>>That's a problem that is being worked on fortunately (at least in the FOSS world):
>>
>>http://sourceware.org/ml/binutils/2008-03/msg00162.html
>
>I am using the Gold linker myself, and it is definitely faster than the normal
>one. However, I think rwessel was referring to LTCG - and the Gold linker is not
>going to be any help here. Furthermore, LTCG on Linux is in its infancy right now,
>since GCC does not support it (there will be support in GCC 4.5).


I did mean both. Plenty of linkers are dog slow even before LTCG. That's most visible in partial builds (where the make tool only has to compile a small portion of the tree) and in parallel/distributed builds.

I think it's mostly because the linker is not typically owned by any of the compiler groups (often it seems to be the OS group), and that actual owners don't see that much direct pressure to improve performance. Obviously it is possible - gold, as Gabriele mentioned, is an excellent example.

Note that LTCG performance is basically not an issue for the linker owners, since the back end that does the code generation is invariably from the compiler (LTCG usually just collects the partially compiled objects and passes them, en mass, to the appropriate compiler back end – then there’s usually a convention link process following that).