Re: Some (small) c++ compilation profiling data (oprofile)

[Richard Henderson <rth at redhat dot com>]

On Sun, May 19, 2002 at 01:54:59PM +0100, Neil Booth wrote:
> That worries me then.  If 1% of compile time is simply looping back
> to the start of the for loop that's an awful lot of collisions.

I see a lot of our hash tables get initially sized to 37 entries.
This is going to cause quite a bit of re-hashing as we expand the hash.

I think for quite a few of these we should measure their sizes 
across a set of "reasonable" sized compilations, and start with
that.  The average size of the hashes while bootstraping would
seem to be a reasonable first cut.


r~