This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [Fortran] Yet more array dependence improvements


Paul Thomas wrote:
Steve Kargl wrote:

On Thu, Mar 09, 2006 at 04:18:57PM -0700, Roger Sayle wrote:


Ok for mainline?

2003-03-09 Roger Sayle <roger@eyesopen.com>

    * dependency.c (gfc_dep_compare_expr) <EXPR_OP>: Allow unary and
    binary operators to compare equal if their operands are equal.
    <EXPR_FUNCTION>: Allow "constant" intrinsic conversion functions
    to compare equal, if their operands are equal.

    * gfortran.dg/dependency_10.f90: New test case.
    * gfortran.dg/dependency_11.f90: Likewise.




OK.

Here are Polyhedron results.  Note the nf.f90 results.
Also, these are single runs, not some average.

gfc4x -w -o z -funroll-loops -O2

Pre Roger Patch:

ac.f90       161.48 real        51.90 user         0.01 sys
aermod.f90   437.58 real       137.06 user         0.07 sys
air.f90      163.37 real        63.35 user         0.10 sys
capacita.f90 664.77 real       276.91 user         4.71 sys
channel.f90  224.35 real        79.38 user        15.40 sys
doduc.f90    126.41 real       120.16 user         0.00 sys
gas_dyn.f90   83.48 real        73.32 user         7.63 sys
induct.f90   110.17 real       108.56 user         0.06 sys
linpk.f90    249.92 real       110.71 user         0.18 sys
mdbx.f90     174.16 real        73.04 user         0.02 sys
nf.f90       522.41 real       171.63 user        38.20 sys
protein.f90  310.24 real       138.41 user         6.24 sys
rnflow.f90   141.73 real       123.03 user         9.25 sys
test_fpu.f90  95.91 real        89.35 user         1.35 sys
tfft.f90      34.67 real        34.25 user         0.01 sys
fatigue.f90   68.99 real        38.61 user         0.03 sys


Post Roger Patch:


ac.f90        73.28 real        51.75 user         0.00 sys
aermod.f90   136.54 real       134.94 user         0.02 sys
air.f90       63.79 real        61.71 user         0.03 sys
capacita.f90 289.56 real       274.84 user         4.65 sys
channel.f90   97.02 real        78.37 user        15.37 sys
doduc.f90    123.77 real       120.27 user         0.00 sys
gas_dyn.f90   83.36 real        72.74 user         7.95 sys
induct.f90   111.40 real       108.72 user         0.08 sys
linpk.f90    116.94 real       111.37 user         0.21 sys
mdbx.f90      75.30 real        72.18 user         0.05 sys
nf.f90       162.29 real       139.19 user        16.74 sys
protein.f90  147.55 real       137.72 user         5.85 sys
rnflow.f90   137.58 real       122.81 user         9.36 sys
test_fpu.f90  94.60 real        89.55 user         1.31 sys
tfft.f90      35.47 real        34.11 user         0.00 sys
fatigue.f90   38.78 real        38.34 user         0.00 sys

__ Steve




As suspected from Richard Guenther's figures, nf.f90 is the real beneficiary, so for. Does the number of alloc's and malloc's go down?

Paul


Looks to me like significant improvement on almost every test

Jerry


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]