Sun, The SPEC2006 INT and FP results are shown below.
Me: after fix ToT: before fix (after reverting my changes for r3530) Benchmark Me ToT Me/ToT --------------------------------------- 400.perlbench 17.96 18.01 -0.30% 401.bzip2 14.10 14.05 0.36% 403.gcc 15.05 15.06 -0.02% 429.mcf 23.47 23.33 0.60% 445.gobmk 17.84 17.81 0.14% 456.hmmer 28.00 29.96 -6.54% 458.sjeng 16.18 16.17 0.06% 462.libquantum 55.41 55.07 0.62% 464.h264ref 24.54 24.56 -0.05% 471.omnetpp 14.80 14.64 1.10% 473.astar 13.78 13.80 -0.12% 483.xalancbmk 24.66 24.30 1.46% ======================================= SPECint2006: 20.32 20.37 -0.24% Benchmark Me ToT Me/ToT --------------------------------------- 410.bwaves 25.09 24.83 1.04% 416.gamess 19.29 19.30 -0.07% 433.milc 14.73 14.91 -1.17% 434.zeusmp 18.11 18.06 0.29% 435.gromacs 15.49 15.48 0.05% 436.cactusADM 22.65 22.64 0.03% 437.leslie3d 18.43 18.51 -0.46% 444.namd 14.95 14.96 -0.07% 447.dealII 27.09 26.90 0.69% 450.soplex 14.16 14.17 -0.08% 453.povray 21.36 21.36 0.00% 454.calculix 22.67 22.71 -0.20% 459.GemsFDTD 15.19 15.28 -0.62% 465.tonto 21.02 21.16 -0.65% 470.lbm 27.93 28.22 -1.00% 481.wrf 22.74 22.79 -0.25% 482.sphinx3 28.04 27.50 1.96% ======================================= SPECfp2006: 20.01 20.01 -0.03% As you can see above, there are no noticeable performance differences except for 456.hmmer, for which I don't understand what happens. I don't think there should be any difference in performance, especially when none of the benchmarks are using PIC nor CPIC. With IPA, CPIC is set but in both runs for before and after my fix, CPIC is set to false. My fix may make any difference when "-fpic" is given together with "-ipa", which is not the case for any of the benchmarks. Jaewook -----Original Message----- From: Sun Chan [mailto:sun.c...@gmail.com] Sent: Wednesday, April 13, 2011 12:16 AM To: Shin, Jaewook Cc: open64-devel@lists.sourceforge.net Subject: Re: Code review request for bug 721 Sorry I have more questions. Did you check performance differences? If done right, there should be performance gain when IPA is turned on. Sun On Wed, Apr 13, 2011 at 3:10 PM, Shin, Jaewook <jaew...@hp.com> wrote: > Hi Sun, > > > > Attached is a new patch for bug 721. This new patch reverts all changes in > my earlier patch for revision 3530. Instead, the new patch makes the > following modification. > > > > Index: osprey/common/com/config.cxx > > =================================================================== > > --- osprey/common/com/config.cxx (revision 3546) > > +++ osprey/common/com/config.cxx (working copy) > > @@ -1453,8 +1453,15 @@ > > if ( Use_Large_GOT ) Guaranteed_Small_GOT = FALSE; > > - /* if we get both TENV:CPIC and TENV:PIC, use only TENV:CPIC */ > > - if (Gen_PIC_Call_Shared && Gen_PIC_Shared) Gen_PIC_Shared = FALSE; > > + /* If we get both TENV:CPIC and TENV:PIC, use only TENV:PIC. > > + * > > + * To fix bug 721, "use only TENV:CPIC" is changed to "use only > TENV:PIC". > > + * While PIC, represented by Gen_PIC_Shared, is for 'shared objects', > > + * CPIC, represented by Gen_PIC_Call_Shared, is for 'dynamic executables' > > + * that call functions in shared objects. Since CPIC should be the > default, > > + * Gen_PIC_Call_Shared is not very useful. > > + */ > > + if (Gen_PIC_Call_Shared && Gen_PIC_Shared) Gen_PIC_Call_Shared = FALSE; > > /* Select optimization options: */ > > > > The uses of Gen_PIC_Call_Shared seem pretty benign. Mostly, they use > Gen_PIC_Call_Shared together with Gen_PIC_Shared so that either one is > effective. There are a few cases where Gen_PIC_Call_Shared is used alone but > not using Gen_PIC_Call_Shared is not harmful. Gen_PIC_Call_Shared used to be > false and remains false before and after this fix. > > > > This fix passes both SPEC INT and FP benchmarks in the current revision > without any additional flags, and it passes both SPEC INT and FP with "-fpie > -pie" in revision 3522 except for 416.gamess. It also passes the reported > cases in bug 721 and 762. > > > > Jaewook ------------------------------------------------------------------------------ Forrester Wave Report - Recovery time is now measured in hours and minutes not days. Key insights are discussed in the 2010 Forrester Wave Report as part of an in-depth evaluation of disaster recovery service providers. Forrester found the best-in-class provider in terms of services and vision. Read this report now! http://p.sf.net/sfu/ibm-webcastpromo _______________________________________________ Open64-devel mailing list Open64-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/open64-devel
