svn commit: r236137 - head/contrib/gcc/config/i386

[Konstantin Belousov]

On Mon, May 28, 2012 at 09:17:23PM +0100, David Chisnall wrote:
> On 28 May 2012, at 20:33, Dimitry Andric wrote:
> 
> > On the other hand, it's really platform-dependent: I've checked several
> > Linux distributions, and it is fairly unpredictable whether their gcc
> > passes --hash-style to the linker, or if they do, which option they use.
> 
> Can we make it dependent on the triple? i.e. if the triple is
> arch-whatever-freebsd9 or greater, make it pass the flag, otherwise
> don't bother? Or is it not worth caring about older FreeBSD? There's
> no real disadvantage in passing it unconditionally (marginally longer
> link times) and potentially a big benefit. I don't see a problem with
> committing it upstream, but it would be nice to pull that change in
> locally before 9.1 and not have to wait for LLVM 3.2 before we got to
> make use of it.
Having unused hash in the image causes on-disk bloat. Since hashes are
merged into text segments (they are needed by rtld and so are mapped),
this also causes memory use bloat.
It is definitely not huge, at least not for normal binaries with reasonably
sized symbol table.

Also, it is more work for the linker to generate both tables.
>
> Misleading and poorly designed benchmarks on Phoronix are at stake!
>
> David
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 196 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/svn-src-head/attachments/20120528/31a0d51f/attachment.pgp