Nicolas Neuss <[EMAIL PROTECTED]> writes:

> I recently tried to accelerate mesh refinement in my PDE toolbox Femlisp by
> auto-generating the refinement code.  This worked quite well, but I
> observed that compilation times increase dramatically with the size of the
> code (increasing dimension of the cell).
>
> Dimension   Size of code (#atoms in tree)   Time (secs)
>   2             384                            0.4
>   3            1524                            4.0
>   4            5736                           61.3

It looks like somewhere in the compiler there is an O(n^2) algorithm
which is the bottleneck in the compilation for your program.  This
isn't entirely surprising to me: many of the phases in a compiler have
problems with solutions easily expressed in O(n^2) time or worse.
Have you tried profiling bits of the compiler to see where most of the
time is spent?

Cheers,

Christophe


Reply via email to