[Rdkit-devel] Speeding up SMILES parsing

[Greg L. <gre...@gm...>]

Over the past couple of days I've spent some time doing some tuning of
the RDKit's SMILES parser.

I made a couple of minor changes here and there and saw some
improvement before making a change in the YACC grammar used to
generate the parser. This made the parser source a bit more difficult
to read, but had a pretty significant impact on performance.

In order to just measure performance of the SMILES parser, I did a
benchmark using ~560K molecules from ZINC where I generated a molecule
from SMILES without any sanitization.
Here are the timings on my linux box for that benchmark:

RDKit_2011_06_1: 50.6s
RDKit_2012_03_1: 49.6s
RDKit_2012_06_1: 57.6s  [ <- I'm not sure I understand this outlier]
svn: 30.6s

I'm pretty pleased about that last number. :-)

For those who are interested, here's the commit:
https://sourceforge.net/p/rdkit/code/2159/
and the specific grammar changes that made the difference:
https://sourceforge.net/p/rdkit/code/2159/tree//trunk/Code/GraphMol/SmilesParse/smiles.yy?diff=502dda6571b75b41b4b10063:2158

-greg