Comparing the Polyhedron 2005 benchmarks from current gcc 4.4 branch (r157587)
with those of gcc trunk (r157600) on x86_64-apple-darwin10 shows significant
regressions in the compile time compared to the previous release. The results
for gcc 4.4 branch are...

================================================================================
Date & Time     : 20 Mar 2010 20:47:42
Test Name       : gfortran_lin_O3
Compile Command : gfortran -ffast-math -funroll-loops -msse3 -O3 %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      0.88       10000     11.24      10  0.0070
      aermod     36.51       10000     21.49      10  0.0117
         air      2.10       10000      5.69      15  0.0463
    capacita      1.41       10000     35.28      10  0.0351
     channel      0.74       10000      1.75      10  0.0255
       doduc      4.31       10000     27.49      10  0.0062
     fatigue      1.64       10000      8.45      12  0.0358
     gas_dyn      2.38       10000      5.07      14  0.0926
      induct      3.46       10000     12.81      10  0.0055
       linpk      0.73       10000     15.43      10  0.0633
        mdbx      1.39       10000     11.27      10  0.0052
          nf      2.99       10000     29.19      16  0.0987
     protein      3.42       10000     35.32      10  0.0253
      rnflow      4.45       10000     23.38      10  0.0106
    test_fpu      4.13       10000      8.74      10  0.0136
        tfft      0.45       10000      1.90      10  0.0170

Geometric Mean Execution Time =      11.46 seconds

================================================================================

whereas the results for gcc trunk are...

================================================================================
Date & Time     : 20 Mar 2010 22:44:44
Test Name       : gfortran_lin_O3
Compile Command : gfortran -ffast-math -funroll-loops -msse3 -O3 %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      1.84       10000     10.59      10  0.0061
      aermod     60.90       10000     20.96      10  0.0076
         air      4.13       10000      5.68      10  0.0880
    capacita      2.76       10000     33.16      10  0.0113
     channel      0.98       10000      1.83      12  0.0089
       doduc      8.23       10000     27.75      10  0.0085
     fatigue      3.09       10000      8.05      10  0.0934
     gas_dyn      5.39       10000      4.38      15  0.0621
      induct      6.96       10000     12.65      10  0.0073
       linpk      1.03       10000     15.40      10  0.0651
        mdbx      2.41       10000     11.33      10  0.0059
          nf      3.38       10000     27.76      13  0.0973
     protein      7.79       10000     37.02      10  0.0041
      rnflow      9.70       10000     23.64      10  0.0224
    test_fpu      7.25       10000      8.67      12  0.0331
        tfft      0.75       10000      1.88      10  0.0514

Geometric Mean Execution Time =      11.25 seconds

================================================================================

The compile times have almost doubled which seem excessive for the optimization
improvements obtained.


-- 
           Summary: large compile time regressions for Polyhedron 2005
                    benchmarks
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: howarth at nitro dot med dot uc dot edu
 GCC build triplet: x86_64-apple-darwin10
  GCC host triplet: x86_64-apple-darwin10
GCC target triplet: x86_64-apple-darwin10


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43463

Reply via email to