From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 10857 invoked by alias); 9 Sep 2008 16:02:44 -0000 Received: (qmail 10419 invoked by uid 48); 9 Sep 2008 16:01:17 -0000 Date: Tue, 09 Sep 2008 16:02:00 -0000 Message-ID: <20080909160117.10418.qmail@sourceware.org> X-Bugzilla-Reason: CC References: Subject: [Bug target/37364] [4.4 Regression] IRA generates inefficient code due to missing regmove pass In-Reply-To: Reply-To: gcc-bugzilla@gcc.gnu.org To: gcc-bugs@gcc.gnu.org From: "hjl dot tools at gmail dot com" Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Archive: List-Post: List-Help: Sender: gcc-bugs-owner@gcc.gnu.org X-SW-Source: 2008-09/txt/msg01072.txt.bz2 ------- Comment #14 from hjl dot tools at gmail dot com 2008-09-09 16:01 ------- I re-enabled the regmove pass on ira-merge branch at revision 140065 and ran SPEC CPU 2K/2006 with -O2 -msse2 -mfpmath=sse -ffast-math for both 32bit and 64bit on Intel Core 2. Here are the performance impacts of the regmove pass: 32bit: 164.gzip -0.597213% 175.vpr 0.42898% 176.gcc 2.66576% 181.mcf -0.913938% 186.crafty -0.514019% 197.parser 0.185644% 252.eon 1.10181% 253.perlbmk -2.17957% 254.gap -0.993521% 255.vortex 0.209644% 256.bzip2 -0.294551% 300.twolf 0.131363% SPECint_base2000 -0.0744109% 168.wupwise 0.45045% 171.swim 0.653595% 172.mgrid 4.23729% 173.applu 1.31265% 177.mesa -2.31596% 178.galgel -0.410633% 179.art 0.504994% 183.equake -0.744153% 187.facerec 4.28781% 188.ammp 1.28707% 189.lucas 13.7548% 191.fma3d -0.53528% 200.sixtrack 1.49551% 301.apsi 0.27088% SPECfp_base2000 1.66845% 400.perlbench 1.52284% 401.bzip2 -1.36986% 403.gcc 0.497512% 429.mcf 0% 445.gobmk -1.08696% 456.hmmer 0% 458.sjeng 0% 462.libquantum -0.490196% 464.h264ref -20.1581% 471.omnetpp 0.689655% 473.astar 0.826446% 483.xalancbmk 0% SPECint(R)_base2006 -2.1978% 410.bwaves 0% 416.gamess 1.37931% 433.milc 0% 434.zeusmp -3.0303% 435.gromacs -6.19469% 436.cactusADM -8.57143% 437.leslie3d 0% 444.namd 1.5625% 447.dealII -0.446429% 450.soplex 0% 453.povray 1.26582% 454.calculix 1.59884% 459.GemsFDTD 1.76991% 465.tonto -0.793651% 470.lbm -5.64103% 481.wrf 0.917431% 482.sphinx3 4.69484% SPECfp(R)_base2006 -1.41844% 64bit: 164.gzip -0.124611% 175.vpr -0.145208% 176.gcc 0.172831% 181.mcf 0.143833% 186.crafty -0.934579% 197.parser 0.122549% 252.eon -0.465839% 253.perlbmk 0.565348% 254.gap -0.158983% 255.vortex 0.0648298% 256.bzip2 0.0863931% 300.twolf 0.346566% SPECint_base2000 -0.0274812% 68.wupwise 1.79766% 171.swim 0.389105% 172.mgrid -1.01574% 173.applu 4.83871% 177.mesa -0.634026% 178.galgel 0.239704% 179.art 0.71311% 183.equake 1.14286% 187.facerec 2.05011% 188.ammp -1.99076% 189.lucas 2.137% 191.fma3d -0.20816% 200.sixtrack 0.303951% 301.apsi 0.0847099% SPECfp_base2000 0.690346% 400.perlbench -0.440529% 401.bzip2 -0.564972% 403.gcc -0.5% 429.mcf -0.510204% 445.gobmk -1.05263% 456.hmmer 2.7027% 458.sjeng 1.5544% 462.libquantum 0.45045% 464.h264ref 0.374532% 471.omnetpp -0.70922% 473.astar 0% 483.xalancbmk 0.507614% SPECint(R)_base2006 0% 410.bwaves 0% 416.gamess -0.568182% 433.milc 1.45985% 434.zeusmp 0% 435.gromacs 3.22581% 436.cactusADM -6.66667% 437.leslie3d 0% 444.namd 0% 447.dealII 1.13636% 450.soplex 0.546448% 453.povray -0.900901% 454.calculix 1.98598% 459.GemsFDTD 2.27273% 465.tonto 0% 470.lbm -2.06186% 481.wrf 0.8% 482.sphinx3 0.881057% SPECfp(R)_base2006 0% At -O2, the impact of the regmove pass is mixed. However, for -mtune=core2 is used, the results could be different. See PR 37437. -- hjl dot tools at gmail dot com changed: What |Removed |Added ---------------------------------------------------------------------------- OtherBugsDependingO| |37437 nThis| | http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37364