Re: The case of directly-executable parsers

On 2006-01-30, momchil.velikov@gmail.com <momchil.velikov@gmail.com> wrote:
> I have also implemented a kind of a "switch optimizer". After removing
> the most frequent transition, the keys are separated in "dense" ranges,
> where density threshold is a parameter (say, 0.33 or 0.5). Then a linear
> or a binary search is performed depending on the number of ranges (say >=
> 8 binary search, otherwise - linear).
>
> This did not result in any performance gain, although it did reduce the
> code size by a decent amount.


The authors of re2c did fiddle around with the binary search in some
circumstances, and they discuss this in the accompanying LOPLAS
article that comes with the source. However, their conclusions might
be somewhat dated. I seem to recall they had different results using
if/else blocks vs. switch statements. Re2c appears to have a new
maintainer:


http://re2c.org


-Clint

Return to the comp.compilers page.
Search the comp.compilers archives again.