This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] Fix PR18754: add early loop pass, 2nd try
- From: Richard Guenther <rguenth at tat dot physik dot uni-tuebingen dot de>
- To: Giovanni Bajo <giovannibajo at libero dot it>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Thu, 20 Jan 2005 15:05:02 +0100 (CET)
- Subject: Re: [PATCH] Fix PR18754: add early loop pass, 2nd try
On Thu, 20 Jan 2005, Giovanni Bajo wrote:
> > I'll get you numbers for
> > PR8361 and POOMA once a disable-checking bootstrap completed
Comparing -O2 with -O2 -ftree-early-loop-optimize (with Zdeneks
patch for complete unrolling applied, without that or -funroll-loops
the comparison doesn't make any sense - and without the patch we
should conditionalize the early loop on -funroll-loops):
-O2 -O2 -ftree-early-loop-optimize
PR8361: 23.06s 23.86s +3.4%
tramp3d: 111.19s 116.70s +5.0%
PR8361 is not a runtime test, so runtime numbers only for tramp3d.
w/o w/ -ftree-early-loop-optimize
tramp3d: 13.10s 12.40s -5.5%
(though we're memory bandwith limited in this test)
>From inspecting generated assembly code for POOMA, we don't do
very good. It seems the reduced C++ testcase from 18754 does
not contain all difficulties.
Back to experimenting...
Richard Guenther <richard dot guenther at uni-tuebingen dot de>