[Kst] [PATCH] Make plugins faster
netterfield at astro.utoronto.ca
Sat Nov 6 03:39:08 CET 2004
OK. Commit it, but watch it carefully over the next few days.
On November 5, 2004 05:21 pm, George Staikos wrote:
> On Friday 05 November 2004 16:04, Barth Netterfield wrote:
> > *If* we can show that it is a significant speedup (>~30%) in at least
> > some cases, then, given the long RC1 time scale, I think it can go in.
> > Planck-HFI is by far the most intense plugin user right now, and if they
> > need it, they are going to use it, whether we branch before or after the
> > change. And if they use it, then they are certainly going to test it
> > over the next couple weeks. The changes are well isolated and easily
> > revertable if we find troubles.
> Here's the best performance information I can gain:
> 1) At the beginning, performance gain is about 35%.
> 2) Over time, performance gain goes to 0% on realtime data, naturally,
> since the computation time becomes most significant.
> 3) The real benefits are in situations of multiple plugins since an update
> will do loads of malloc/free.
> 4) It also helps keep the UI responsive since malloc lock contention drops
> and UI events cause loads of mallocs.
> I see both sides to the argument equally. I'm hoping someone will make
> the call for me. :)
More information about the Kst