[PATCH] analysis of global statics and removal associated vdefs and vuses
Jeffrey A Law
law@redhat.com
Fri Sep 3 17:18:00 GMT 2004
On Fri, 2004-09-03 at 09:00, Kenneth Zadeck wrote:
> This difference in timings between the first and second patch indicates
> to me that we have a real problem in the implementation of our splay
> trees. I do not know how much this data structure is used but there was
> one splay tree lookup for dozens of bit map operations and it appears
> that the single splay tree lookup to takes hundreds of times as long as
> a bitmap operation. This seems wrong.
We don't use splay trees that heavily, so I wouldn't be surprise if
there are some inefficiencies in that code. Ignoring targets with
conditional execution capabilities, I think the most time critical
use of splay trees in the past has been the type based alias analysis
code.
Jeff
More information about the Gcc-patches
mailing list