On 4/10/12 6:49 AM, Jack Howarth wrote: > In case package maintainers are wondering if the new gcc47 package > contains > code generation improvements making it worthwhile to upgrade their packages > from > a dependency on gcc46 to gcc47, here are the results from the Polyhedron 2005 > benchmarks with the most exhaustive optimizations of -Ofast -unroll-loops > -flto > -fwhole-program. Note that the -Ofast option includes -fstack-arrays and > -fno-protect-parens so any code compiled with that should be checked with its > testsuite. > > ================================================================================ > Date& Time : 9 Apr 2012 21:27:00 > Test Name : gfortran46_lin_O3_Ofast_wholeprogram > Compile Command : gfortran-fsf-4.6 -Ofast -funroll-loops -flto > -fwhole-program %n.f90 -o %n > Benchmarks : ac aermod air capacita channel doduc fatigue gas_dyn induct > linpk mdbx nf protein rnflow test_fpu tfft > Maximum Times : 2000.0 > Target Error % : 0.100 > Minimum Repeats : 10 > Maximum Repeats : 100 > > Benchmark Compile Executable Ave Run Number Estim > Name (secs) (bytes) (secs) Repeats Err % > --------- ------- ---------- ------- ------- ------ > ac 6.77 50896 10.68 13 0.0272 > aermod 85.90 1172288 22.34 14 0.0888 > air 5.83 74176 7.57 20 0.0975 > capacita 4.79 73104 39.74 14 0.0796 > channel 1.87 34864 2.24 13 0.0603 > doduc 14.28 188152 38.03 14 0.0256 > fatigue 4.73 81480 10.78 13 0.0570 > gas_dyn 8.67 115784 6.84 18 0.0751 > induct 10.61 148888 32.22 14 0.0429 > linpk 5.97 26040 26.65 12 0.0441 > mdbx 4.94 76952 13.49 12 0.0536 > nf 5.02 71768 36.35 56 0.1085 > protein 11.26 131328 39.30 10 0.0878 > rnflow 14.82 167800 33.35 10 0.0160 > test_fpu 11.21 142288 10.39 13 0.0384 > tfft 1.27 22408 2.53 21 0.0915 > > Geometric Mean Execution Time = 15.00 seconds > > ================================================================================ > Date& Time : 9 Apr 2012 23:59:59 > Test Name : gfortran47_lin_O3_Ofast_wholeprogram > Compile Command : gfortran-fsf-4.7 -Ofast -funroll-loops -flto > -fwhole-program %n.f90 -o %n > Benchmarks : ac aermod air capacita channel doduc fatigue gas_dyn induct > linpk mdbx nf protein rnflow test_fpu tfft > Maximum Times : 2000.0 > Target Error % : 0.100 > Minimum Repeats : 10 > Maximum Repeats : 100 > > Benchmark Compile Executable Ave Run Number Estim > Name (secs) (bytes) (secs) Repeats Err % > --------- ------- ---------- ------- ------- ------ > ac 7.38 59040 9.98 12 0.0757 > aermod 93.14 1240960 21.68 10 0.0883 > air 10.78 106904 7.93 12 0.0959 > capacita 5.10 77192 39.31 12 0.0285 > channel 1.91 34840 2.22 13 0.0959 > doduc 15.30 196288 37.14 11 0.0945 > fatigue 5.12 77472 6.06 10 0.0535 > gas_dyn 8.49 115680 4.87 16 0.0906 > induct 9.62 153240 30.67 10 0.0293 > linpk 1.63 21952 26.26 12 0.0980 > mdbx 5.66 81048 13.34 10 0.0166 > nf 5.11 71816 23.20 10 0.0546 > protein 11.75 127208 38.35 10 0.0319 > rnflow 14.11 135096 33.21 10 0.0100 > test_fpu 10.85 105408 9.95 10 0.0620 > tfft 1.28 22416 2.53 22 0.0988 > > Geometric Mean Execution Time = 13.57 seconds > > ================================================================================ > > With this in mind, does anybody have any objections to my updating atlas to use gcc47 ? This would be for purely selfish reasons: I'd like the Octave-family dep tree only to have one gcc4N. :-)
-- Alexander Hansen, Ph.D. Fink User Liaison http://finkakh.wordpress.com/2012/02/21/got-job/ ------------------------------------------------------------------------------ For Developers, A Lot Can Happen In A Second. Boundary is the first to Know...and Tell You. Monitor Your Applications in Ultra-Fine Resolution. Try it FREE! http://p.sf.net/sfu/Boundary-d2dvs2 _______________________________________________ Fink-devel mailing list Fink-devel@lists.sourceforge.net List archive: http://news.gmane.org/gmane.os.apple.fink.devel Subscription management: https://lists.sourceforge.net/lists/listinfo/fink-devel
