This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: speedup collect2 (by not using it)


Mike Stump <mrs@apple.com> writes:

> These patches speed up compiles by around 2.8% on Finder_FE by not
> trying to using collect2 when it isn't necessary.  This is about 62ms
> per compile.  If collect2 is in use for collecting (USE_COLLECT2 is
> defined), then these changes don't have any effect, otherwise, if
> -frepo is given, collect2 is used, otherwise, we use ld as normal.
>
> So, the question is, does collect2 ever do anything interesting when
> USE_COLLECT2 isn't defined and -frepo isn't given?  Problem areas
> might be IBM (COLLECT_EXPORT_LIST).  If aix is a problem, we can
> always do collect2 when COLLECT_EXPORT_LIST is defined.

I think this is a fine idea in principle.  We need to nail down
precisely what collect2 is or is not used for, of course, and make
sure it's invoked when necessary.

> Another question would be, do we trust gcc.c to be able to find ld as
> well as collect2.c can find ld?  :-(  Answer should be yes, but in
> practice, there could be bugs.

Given that the exact same code is relied on to find the assembler, and
given that GCC used to invoke the linker directly if !USE_COLLECT2
(before there was -frepo) I think the risk of this is low.  Better
check unified tree builds and the like, though.

> 	* tlink.c (scan_linker_output): Add support for darwin linker, as it
> 	  emits unresolved symbols one per line, consuming entire line.

Unrelated change, please submit separately.

> ! 
> ! #ifndef USE_COLLECT2
> ! 	  /* On a system that doesn't define USE_COLLECT2, we don't run collect2
> ! 	     for any reason other than -frepo.  */
> ! 	  /* If we have no other reason to run collect2, don't.  */
> ! 	  if (repo_flag)
> ! #endif
> ! 	    s = find_a_file (&exec_prefixes, "collect2", X_OK, 0);

Would you mind restructuring things such that we don't have an #ifdef
here?  Put, say,

#ifndef USE_COLLECT2
#define USE_COLLECT2 0
#endif

at the top of the file, and then you can write

   if (USE_COLLECT2 || repo_flag)
     s = find_a_file (...);

which is neater.

zw


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]