This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
RE: (RFA): Enable first scheduling pass for SH4
- From: "Sanjiv Kumar Gupta, Noida" <sanjivg at noida dot hcltech dot com>
- To: "Vladimir N. Makarov" <vmakarov at redhat dot com>
- Cc: Richard Henderson <rth at redhat dot com>, gcc-patches at gcc dot gnu dot org, Joern Rennecke <joern dot rennecke at superh dot com>, "Naveen Sharma, Noida" <naveens at noida dot hcltech dot com>
- Date: Mon, 8 Sep 2003 16:02:26 +0530
- Subject: RE: (RFA): Enable first scheduling pass for SH4
Vladimir N. Makarov Wrote
> "Sanjiv Kumar Gupta, Noida" wrote:
>
> > > On Wed, Sep 03, 2003 at 01:56:26PM +0530, Sanjiv Kumar Gupta,
> > > Noida wrote:
> > > > The overall idea is to keep count of SImode and SFmode regs
> > > required by
> > > > already scheduled insns.
> > >
> > > I don't think this should at all be done via magic hooks in
> > > the sh backend.
> >
> > The heuristics are pretty target specific. Do you think,
> > I should change the generic scheduler for this?
> >
>
> I think Richard is right. You could make it more general (register
> pressure calculation and reordering ready queue) because it could be
> profitable for other ports (like arm or sparc) too. Then the most of
> code could be in the scheduler itself. And may be one hook
> to switch on
> (off) the heuristic for ports which could have benefit from this
> heuristic. If there is no harm from this for all ports, the
> hook could
> be removed.
It is good idea to make this patch target independent. However,
significant effort is required to fine-tune heuristics for all
targets. I would like to do it, but it is out of scope of my
current project; and a more severe constraint is that
I don't have resources to test it on multiple targets.
> Another question, I see results only with and without the 1st insn
> scheduling. They are impressive. But having results with and without
> register pressure heuristics is more appropriate (may be they will be
> more impressive).
I understand your point.
For SH4, the comparison (with and without heuristics) will look
very impressive (something like 30-40% improvement).
The current comparison is to show that first
scheduling pass can increase performance. The fact that very
bad code is generated without any heuristics, is known, and had
led to its "turning off" for the SH.
Anyway, I will try to make the report with and without
heuristics too.
> It is also interesting to look at the SPEC (95 or
> 2000) results too. I saw many times when small benchmarks (with few
> loops) got the performance improvement but SPEC benchmarks (like gcc)
> got the degradation. If it is not possible that is ok (unfortunately
> many people have no access to SPEC).
I don't have access to SPEC benchmarks. So I had to test with "free"
benchmarks.
Regards,
Sanjiv.