This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Don't emit cmpstrsi and movstr* insns with -Os
- From: Roger Sayle <roger at eyesopen dot com>
- To: Richard Henderson <rth at redhat dot com>
- Cc: gcc-patches at gcc dot gnu dot org, Ulrich Weigand <weigand at informatik dot uni-erlangen dot de>
- Date: Sun, 7 Sep 2003 17:23:38 -0600 (MDT)
- Subject: Re: [PATCH] Don't emit cmpstrsi and movstr* insns with -Os
On Sun, 7 Sep 2003, Richard Henderson wrote:
> On Sat, Sep 06, 2003 at 04:58:40PM -0600, Roger Sayle wrote:
> > Disabling cmpstrsi with -Os shouldn't be controversial, the only backend
> > .md file that defines a pattern for it is i386.md. My belief is that
> > disabling movstr* should also be a space win on other platforms, most of
> > which don't have the CISC advantages of x86's variable instruction length
> > encoding, complex memory move instructions and unaligned memory accesses.
>
> Any chance you could get this tested on s390? If you get
> similar results there, then this is ok.
At the risk of upsetting all of the embedded s390 developers out there :>,
I've just done some experimentation with a s390-ibm-linux cross-compiler.
Unfortunately, s390 isn't supported by newlib, so I'm unable to generate
CSiBE's object files myself. Perhaps CSiBE should contain preprocessed
source files?
However, I have been able to try the following minimal benchmark:
void foo(char *dst, char *src, int len)
{
memcpy(dst,src,len);
}
with mainline CVS versions of gcc, gas and gld (20030907), the cross
compiler to s390-ibm-linux-gnu produced the following object sizes
(as reported by binutils' size).
Options Text Data Total
-c 64 52 116
-c -O2 56 32 88
-c -Os 56 32 88
-c -Os -fno-builtin-memcpy 36 44 80
-c -Os [with this patch] 36 44 80
Hence it looks like disabling movstr* when the user specifies -Os
is also a "size" win on IBM's s390, for this single test atleast.
Checking the -S output, the call to memcpy isn't being sibcalled,
which would bias/invalidate this particular test.
Ulrich, if you can, please could you evaluate my patch on your s390?
If you haven't tried it before, the CSiBE benchmark is remarkably
easy to use. Very many thanks in advance.
I'm happy to wait for Ulrich (or some other kind volunteer?) to
provide the definitive CSiBE results. Is this single data-point
conclusive enough?
Roger
--