This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [REVISED][PATCH/RFT] Fix PR middle-end/PR28690, modify swap_commutative_operands_p
- From: "H. J. Lu" <hjl at lucon dot org>
- To: Peter Bergner <bergner at vnet dot ibm dot com>
- Cc: "gcc-patches at gcc dot gnu dot org" <gcc-patches at gcc dot gnu dot org>, Paolo Bonzini <paolo dot bonzini at lu dot unisi dot ch>
- Date: Sun, 10 Jun 2007 21:01:38 -0700
- Subject: Re: [REVISED][PATCH/RFT] Fix PR middle-end/PR28690, modify swap_commutative_operands_p
- References: <1177614080.20709.34.camel@otta> <20070428153118.GA515@lucon.org> <1177795225.20709.42.camel@otta> <20070429235551.GA12511@lucon.org> <1177935400.20709.44.camel@otta> <20070506202116.GA12898@lucon.org> <20070604194005.GA19014@vnet.ibm.com> <20070608222617.GA18020@lucon.org> <20070609035348.GA5335@vervain.rchland.ibm.com>
On Fri, Jun 08, 2007 at 10:53:48PM -0500, Peter Bergner wrote:
> On Mon, Jun 04, 2007 at 02:40:05PM -0500, Peter Bergner wrote:
> > On Sun, May 06, 2007 at 01:21:16PM -0700, H. J. Lu wrote:
> > > Here are what I got on Core 2 Duo with -O2 -ffast-math:
> > >
> > > 2. SPEC CPU 2000 cores
> > [snip]
> > > 168.wupwise -6.56276%
> >
> > Well, we found a Core 2 Duo box here and we're not seeing as big
> > a degradation as you are (-3.9%), but we seem to have narrowed
> > down one slowdown and created a C testcase that seems to mimic the
> > slowdown. Can you confirm that the testcase below runs slower
> > for you too. Also, we don't know the x86/x86_64 arch that well,
> > so can you tell us why the code is running slower so I can attempt
> > to modify the patch so it doesn't? Thanks.
>
>
> On Fri, Jun 08, 2007 at 03:26:17PM -0700, H. J. Lu wrote:
> > With revision 125027, I got the following on Core 2 Duo 64bit:
> >
> > (r125027 + patch - r125027)/125027
>
> Strange how the numbers are jumping around from your previous
> numbers given the newer mainline revision. Anyway, having
> looked at the wupwise slowdown from before, it seems the
> (set X (X Y)) swapping in swap_commutative_operands_with_target
> is still important, so I reverted that part of the patch and
> early indications are that it doesn't seem to affect our POWER6
> numbers and it seems to fix the wupwise slowdown you were seeing.
> So can I ask you to try the updated patch below and see how that
> compares with the previous one? Thanks.
>
On 64bit Core 2 Duo, using -O2 -ffast-math, for SPEC CPU 2000, I
got
(r125577 + patch - r125577)/r125577
164.gzip -0.0645995%
175.vpr -0.0541126%
176.gcc -2.00373%
181.mcf -0.380022%
186.crafty -0.248668%
197.parser 0.78853%
252.eon 1.41844%
253.perlbmk 0.0720981%
254.gap 1.0159%
255.vortex 1.49547%
256.bzip2 -0.826446%
300.twolf -0.248227%
SPECint_base2000 0.0912825%
168.wupwise 0.670017%
171.swim 0.0420168%
172.mgrid 0.567537%
173.applu 0.44843%
177.mesa 0.146735%
178.galgel -0.724802%
179.art -4.48607%
183.equake -0.12848%
187.facerec -0.0367242%
188.ammp -1.29745%
189.lucas 0%
191.fma3d -3.53122%
200.sixtrack -3.96476%
301.apsi -0.561224%
SPECfp_base2000 -0.927097%
For SPEC CPU 2006,
400.perlbench 1.48515%
401.bzip2 -0.613497%
403.gcc 0%
429.mcf 0%
445.gobmk 0%
456.hmmer 0.8%
458.sjeng 0.571429%
462.libquantum 0%
464.h264ref 0.395257%
471.omnetpp 0%
473.astar -0.869565%
483.xalancbmk -0.420168%
SPECint(R)_base2006 0%
410.bwaves 0%
416.gamess -1.15607%
433.milc 0%
434.zeusmp 0.657895%
435.gromacs -0.862069%
436.cactusADM -2.88462%
437.leslie3d -0.216216%
444.namd -0.675676%
447.dealII -4.65116%
450.soplex 0%
453.povray 1.5%
454.calculix -8.0207%
459.GemsFDTD 0%
465.tonto 0%
470.lbm 0%
481.wrf 0.934579%
482.sphinx3 0.510204%
SPECfp(R)_base2006 -1.50376%
H.J.