>>>> After compiling binutils with "--enable-gold --enable-plugins" >>>> (ld.gold fails without --enable-plugins) and gcc with >>>> "--enable-gold", the following seems to work fine: >>>> >>>> CC="gcc -fuse-ld=gold -flto -fuse-linker-plugin.." configure ... >>> >>> Just a note, GCC doesn't need "--enable-gold" passed during >>> configure. GCC shares its top level configure script with other GNU >>> projects, such as binutils. When people build toolchains with >>> binutils, gmp, mpfr, mpc and many others in GCC's source tree, then it'd be >>> needed. >>> >>> When configuring binutils and using "--enable-gold=yes --enable-plugins" >>> then ld.bfd is the default with ld, and then there is ld.gold. >> >> I'd be interested in some statistics. Is there are difference in size >> or speed of applications that use this technology? I'd like to see >> some quantification of the results. >> >> My initial reaction is that using this would be fairly invasive for BLFS. >> How many packages would need to be changed? >> >> On the other hand, there are a lot of small packages where it wouldn't >> make any difference. For example, would it make a difference for >> something like firefox? I dowbt it would make a difference for >> something like xfce for speed purposes as I can detect no delay in an >> function now. > > From what I saw when last building firefox, the configure stuff looks for > ld.gold itself, > so there would be no change needed. Don't know about other packages. > I compiled qt5 with and without gold (it auto detects gold) and the compile time with and without was the same almost to the second.
I also tried to compile gcc itself, but it wouldn't compile with gold - not entirely surprising, the same was true of lto in the early days. -- http://lists.linuxfromscratch.org/listinfo/lfs-dev FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page
