On Sun, 2005-07-10 at 16:46 +0400, Serge Belyshev wrote: > Hi, > > I have measured compilation time and memory usage of mainline on x86_64 > from 2005-02-25 to 2005-07-10 with one day interval. So does Jan, and to be honest, his tester is better than your graphs, because it knows the difference between ggc and non-ggc memory, etc. Looking at Jan's tester shows that Diego's patch to make the new pt solver the default wiped out some of my improvements from the decl hierarchy
See http://gcc.gnu.org/ml/gcc-regression/2005-07/msg00031.html vs http://gcc.gnu.org/ml/gcc-regression/2005-07/msg00030.html A lot of this is probably because the constraint graph is currently ggc alloced. I'll fix that in a moment.
