Closed brhillman closed 5 years ago
Results of regression test:
############################################################################################
Treating relative differences less than 0.0010000000% as insignificant
atb532_perp: 16.89 % of values differ, relative range: -5.90e-05 to 7.03e-05
atb532: 16.93 % of values differ, relative range: -5.89e-05 to 7.02e-05
cfadLidarsr532: 0.00 % of values differ, relative range: -1.00e+00 to 1.00e+00
lidarBetaMol532: 19.35 % of values differ, relative range: -5.91e-05 to 7.02e-05
clthinemis: 0.65 % of values differ, relative range: -1.19e-05 to -1.19e-05
lidarBetaMol532gr: 18.42 % of values differ, relative range: -5.89e-05 to 6.97e-05
atb532gr: 13.71 % of values differ, relative range: -5.89e-05 to 6.97e-05
lidarBetaMol355: 3.97 % of values differ, relative range: -1.53e-05 to 2.19e-05
atb355: 3.72 % of values differ, relative range: -1.53e-05 to 2.21e-05
dbze94: 0.24 % of values differ, relative range: -2.78e-04 to 1.05e-04
boxtauisccp: 1.93 % of values differ, relative range: -2.37e-05 to 2.40e-05
boxptopisccp: 2.22 % of values differ, relative range: -4.67e-05 to 4.10e-05
All other fields match.
Diffs seem to be result of machine precision. Comparison against outputs from master built and run on the same machine with the same compiler and flags reveals no diffs:
############################################################################################
Treating relative differences less than 0.0010000000% as insignificant
All fields match
(python2) [bhillma@ghost-login2 driver (fix-quickbeam-optics-types)]$
Thanks for doing this, @brhillman.
Floating point literals that include a decimal without explicit precision, i.e.
1.
, are dangerous because we do not know what precision the literal will be interpreted as, and expressions like1.+bpm
, wherebpm
has explicit precisionwp
caused Intel to throw errors when compiled with the-r8
flag. Adding precision to the literals fixes this problem. That is, replacing1.+bpm
with1._wp+bpm
. Fixes #28.