Le 18/04/2019 à 16:54, Pavel Sanda a écrit :
On Thu, Apr 18, 2019 at 04:42:59PM +0200, Jean-Marc Lasgouttes wrote:
Le 18/04/2019 ?? 15:49, Jean-Marc Lasgouttes a écrit :
commit c6754ed548deecdc37c5a019fb1e5a705dc44da5
Author: Jean-Marc Lasgouttes <lasgout...@lyx.org>
Date: Thu Apr 18 11:53:36 2019 +0200
Do not redefine a stripped-down LyXRC object.
I am not sure why it was done this way, but it is not really
necessary.
Moreover gcc LTO does not like it.
So it turns out that we can compile with LTO. Is it worth it? Here is what
I get, compared to a release build.
tanuki: size profdbuild/src/lyx ltodbuild/src/lyx
text data bss dec hex filename
17586230 376496 95364 18058090 1138b6a profdbuild/src/lyx
15979625 374656 94060 16448341 fafb55 ltodbuild/src/lyx
tanuki: ll --si profdbuild/src/lyx ltodbuild/src/lyx
-rwxrwxr-x 1 lasgoutt lasgoutt 17M avril 18 16:41 ltodbuild/src/lyx
-rwxrwxr-x 1 lasgoutt lasgoutt 18M avril 18 16:40 profdbuild/src/lyx
(note that the binaries are stripped).
It used to be that LTO could quite increase linking time, is there some visible
compil time difference for lyx?
Yes, the link time is longer, but it seems that one can use -flto=4 to
use 4 CPUS.
For reference, the configure line I used was:
../master/configure --enable-build-type=rel
--with-version-suffix=-master CXX='g++ -flto' AR='gcc-ar'
RANLIB=gcc-ranlib LD='g++ -flto'
I could try to include that in configure (and test it with clang too).
JMarc
