Gcc builtin functions used by glibc not available on mips

[H . J . Lu]

On Thu, Feb 07, 2002 at 12:00:55PM +0100, Jakub Jelinek wrote:
> On Wed, Feb 06, 2002 at 10:25:54PM -0800, H . J . Lu wrote:
> > In glibc, there are codes
> > 
> >   /* Partly clean the `bootstrap_map' structure up.  Don't use
> >      `memset' since it might not be built in or inlined and we cannot
> >      make function calls at this point.  Use '__builtin_memset' if we
> >      know it is available.  */
> > #if __GNUC_PREREQ (2, 96) 
> >   __builtin_memset (bootstrap_map.l_info, '\0', sizeof (bootstrap_map.l_info));
> > #else
> >   for (cnt = 0;
> >        cnt < sizeof (bootstrap_map.l_info) / sizeof (bootstrap_map.l_info[0]);
> >        ++cnt)
> >     bootstrap_map.l_info[cnt] = 0;
> > #endif
> > 
> > used in ld.so bootstrap. The problem is even gcc 3.1 doesn't have a
> > real __builtin_memset for mips. I am very reluctant to call it a gcc
> > bug. Maybe we should check it during the gcc configure instead of
> > depending on "#if __GNUC_PREREQ (2, 96)".
> 
> Well, I think on most arches this __builtin_memset will actually expand to
> real memset, since sizeof (bootstrap_map.l_info) is more than 64 * sizeof (void *),
> ie. really big. Guess just ia-32 will expand it inline.

Shouldn't we test __builtin_memset (x, 0, 64 * sizeof (void *)) before
using it?

> 
> But, memset is local symbol in ld.so, do even calls to local functions have
> to go through plt or got on mips?

I think so. It goes through local GOT. The dynamic linker still has
to adjust it.


H.J.