nemobench5 compiler regression

[teuben]

As of gcc-12 (13 and 14 as well) both orbint and potcode (especially) run a lot slower.

Here's a sample on an i9-12900K, but this was confirmed on laptops as well, as well as an AMD Ryzen 7 5800H, suggestive there's some coding in NEMO that should be improved. potential function pointers?

# gcc-11
CPU_USAGE  directcode  :  3.95  3.95  0.00  0.00  0.00  1395775176
CPU_USAGE  gyrfalcON   :  3.18  3.16  0.00  0.00  0.00  1395775572
CPU_USAGE  hackcode1   :  3.25  3.24  0.00  0.00  0.00  1395775890
CPU_USAGE  orbint      :  3.23  3.22  0.00  0.00  0.00  1395776215
CPU_USAGE  potcode     :  3.59  3.58  0.00  0.00  0.00  1395776539
CPU_USAGE  treecode1   :  3.89  3.88  0.00  0.00  0.00  1395776898
NEMOBENCH5 score:   1422.48

# gcc-13
CPU_USAGE  directcode  :  3.82  3.82  0.00  0.00  0.00  1395837821
CPU_USAGE  gyrfalcON   :  3.21  3.21  0.00  0.00  0.00  1395838204
CPU_USAGE  hackcode1   :  3.26  3.26  0.00  0.00  0.00  1395838526
CPU_USAGE  orbint      :  3.98  3.97  0.00  0.00  0.00  1395838853    <=== bit slower
CPU_USAGE  potcode     :  6.56  6.54  0.00  0.00  0.00  1395839251    <=== much slower
CPU_USAGE  treecode1   :  3.72  3.71  0.00  0.00  0.00  1395839907
NEMOBENCH5 score:   1221.99

[teuben]

It appears the default compilation with -g -O2 was doing the trick. With -O3 most (if not all) the differences in performance with potcode and orbint has gone away.