flow speed regression [Re: How long should -O1 compiles take?]
Marc Feeley
feeley@IRO.UMontreal.CA
Sun Oct 31 23:33:00 GMT 1999
> On Wed, Oct 06, 1999 at 08:14:25AM -0500, Brad Lucier wrote:
> > Some applications really need a "pretty good" compiler that runs
> > relatively quickly. For example, I believe that the fastest way
> > to do genetic programming on image data is to compile the evolving
> > programs before testing them.
>
> If you really care about compile speed, the scheme output should be
> restructured to not use computed gotos. Flow graphs with an edge/block
> ratio this high (463:1) simply _cannot_ be compiled quickly.
>
> In fact, I bet it could be restructured so that the resulting code
> runs _faster_ than with computed gotos.
What do you have in mind here? I've been looking for solutions for
the past 5 years and the computed gotos of gcc are currently the
fastest solution. The best alternative, which has the advantage of
being portable to other C compilers, is a large switch statement but
that can be up to 3 times slower on a Pentium.
Marc Feeley
More information about the Gcc-patches
mailing list