On 1/9/07, Daniel Berlin <dberlin@dberlin.org> wrote:
> On 1/9/07, Kenneth Zadeck <zadeck@naturalbridge.com> wrote:
> > Daniel Berlin wrote:
> > > On 1/9/07, Seongbae Park <seongbae.park@gmail.com> wrote:
> > >> This patch cuts the time spent in dataflow solver in half, when
> > >> compiling top 5 largest files for building cc1. For one analysis
> > >> (reaching store problem used during dse),
> > >> traversing in the order of the reverse postorder of inverted CFG
> > >> for the forward problems reduces the block visit count by 90%.
> > >> The worklist based algorithm reduces it by couple % further.
> > >
> > > That's pretty funny. We originally changed the algorithm to what it is
> > > now because it was faster for the problems it was solving.
> > > Oh well.
> > trust but verify.
> I did.
> I ran the numbers myself back then.
> Of course, i may have misread them what with having the lantern
> running out of oil and all that.