This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: Recent polyhedron regressions.
Richard,
I'll take a look see if I can figure out which it is.
I am a bit flummoxed initially - there are no calls to dot_product or
where in nf.f90.....
but blow me if there are not calls to DOT_PRODUCT!
There are 3 of these per iteration, each of which takes ~20-30ms for
vectors of length ~10**6. Thus:
gcc-4.1 spends 7.8s out of 139 doing DOT_PRODUCT and
gcc-4.2 spends 8.2 out of 155.
'tain't DOT_PRODUCT or dt_product.
However, the number of allocations has gone up from 4.1 to 4.2:
12 _gfortran_allocate and 10 _gfortran_internal_malloc in 4.1 versus
12 _gfortran_allocate_array and 15 _gfortran_internal_malloc in 4.2
Given that one each of the mallocs is taken up by an allocate, we are
seeing a ratio in the code of 21:26 calls to malloc. For some reason, I
am getting complete nonsense out of gprof, so I wil have to go through
the code to find where the increase has occurred.
Best regards
Paul