On Wed, Mar 02, 2022 at 04:26:11PM -0500, Ken Dreyer wrote:
> Some of the Ceph developers were investigating mold, since Ceph takes
> so long to build. Linking time is a problem for us with Ceph. But I
> don't know if those interested Ceph developers have done benchmarks
> yet.
>
> And the lack of s390x and ppc64le is problematic for us in Ceph too.
> But overall I'm amazed at how fast mold development is progressing.
I think a linker that can be trusted is more important than speed,
especially when part of that speed comes from not performing any checking
and just blindly assuming things look the way it expects.
I've only spent 15 minutes on it to find that mold (1.1) does significantly
less effective SHF_MERGE | SHF_STRINGS constant merging (so ends up with
significantly larger .debug_str or the .rodata string sections), or
supports only small portion of TLS transitions ld.bfd supports and does no
checking. Small testcase found during that time:
a.c:
__thread int a = 42;
__attribute__((noipa)) int
foo (void)
{
return a;
}
int
main ()
{
return foo () != 42;
}
gcc -O2 -mcmodel=large -fpic -c a.c -o a.o
gcc -o a_ a.o
./a_; echo $?
yields expected 0,
gcc -o a a.o -fuse-ld=mold
./a
Segmentation fault (core dumped)
Jakub
_______________________________________________
devel mailing list -- devel@xxxxxxxxxxxxxxxxxxxxxxx
To unsubscribe send an email to devel-leave@xxxxxxxxxxxxxxxxxxxxxxx
Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: https://lists.fedoraproject.org/archives/list/devel@xxxxxxxxxxxxxxxxxxxxxxx
Do not reply to spam on the list, report it: https://pagure.io/fedora-infrastructure
