Compiler Benchmarks

We'd like to have some idea of how different compilers affect APBS runtime, and this is where that information will be documented.

As an initial attempt, below are first the first round comparisons, which may be useful, if taken with great skepticism.

To generate the numbers in the table below, I ran the APBS test suite using the Intel and Apple compilers. The run times are aggregated numbers for each test: a folder under apbs/examples, which is composed of several APBS runs. Warning: the reported times are rounded to the second.

For the Intel compiler, I used the APBS 1.4.1 OS X binary, which was compiled with icc 14.0.4. I then built the 1.4.1 source with Apple LLVM 7.0.2 (clang-700.1.82). It should be noted that the Intel compiler implements OpenMP, whereas the Apple compiler does not.

Test Suite	icc	clang
born	18s	18s
actin-dimer	558s	645s
alkanes	4s	12s
FKBP	40s	45s
hca-bind	17s	16s
ionize	6s	6s
ion-pmf	3s	7s
ion-protein	44s	62s
pka-lig	43s	43s
point-pmt	2s	2s
solv	2s	2s
geoflow	4s	3s