Katja:

Attached is a flat profile made by xprofiler from an interactive
run of tiegcm on blackforest. This run used MPI with 4 tasks
(MP_PROCS 4, MP_TASKS_PER_NODE 2), and xprofiler was invoked
with the executable and all 4 gmon.out files.

The top 2 functions were _xlmrr8.f and _xlmri4.f (libxlf90.a),
which totaled almost 65% of my total cpu time! What are these
and why are they taking so much time? I am using 8-byte reals
and 4-byte ints. Below is from the makefile:

FC = mpxlf_r
# Use -pg for gprof or xprofiler (must also be in LDFLAGS)
FFLAGS= -qfixed -qarch=pwr3 -qtune=pwr3 -qrealsize=8 \
        -qalias=noaryovrlp -O3 -qstrict -WF,-DMPI,-DAIX \
        -I/usr/local/include -pg
LDFLAGS = -bloadmap:loadmap -pg
LIBS    = -L /usr/local/lib32/r4i4 -lmass -lnetcdf -lmss

Thanks,

--Ben