Re: Fix i386 memcpy/memset expanders

[Jan Hubicka <jh at suse dot cz>]

> > On Wed, Dec 31, 2003 at 08:19:27PM +0100, Jan Hubicka wrote:
> > > 	* i386.md (movstr expander): Enable when optimizing for size.
> > 
> > I presume this came from Roger's work that showed that rep movs
> > plus associated setup code is larger than the libcall.
> 
> Then we should honnor -minline-all-stringops and document it.
> I see that for small testcase containing the rep;movs produces 2 bytes
> longer code, but on the other hand it don't have problems with call
> clobbering and producing more register spills.  Did we tested this more
> seriously?

Changing this back saves 4 bytes on Gerald's testcase and 300 bytes on
GCC components [a-d]*.c

Honza
> 
> Honza
> > 
> > 
> > r~