On Thu, Jan 15, 2015 at 9:15 AM, Lubomir I. Ivanov <[email protected]> wrote:
> On 15 January 2015 at 02:56, Tomaz Canabrava <[email protected]> wrote: > > Lubomir, > > > > I'v removed almost every allocation that I could find. can you try this 8 > > patches and tell me if anything worked to reduce the cpu usage? > > > > > > We can still do some stuff to not remove the animations ( yup, I like > them > > ) > > like using a square instead of a rounded - rect notification panel. > > > > > > unfortunately the patches did not improve the performance much for me > - it's still above 20%. > but i suggest you skip any more fixes for this ATM, because we have no > user complains. > > the allocations boosted it, but setText(), setPixmap() and the little > graph drawing still require high CPU usage. > We can also try to move the Profile to OpenGL ( I'v always wanted to learn that and that's a pretty damn good excuse. ) > > it started to smell to me as a FPU denormal issue, so i tested that > because i've noticed the GCC uses FPU instructions for this particular > part and the control word flag was unset, but it ain't the cause, > unless Qt updates the FPU between calls. > > -O3 doesn't helps which means that calls to another library are heavy > (e.g. Qt). > > i've also tried using S&H on refresh() to reduce the amounts of calls > on power of two intervals and that is the only thing that helps, but > reduces the smoothness of the tooltip itself and isn't pretty; a > factor of 4 is tolerable but still caps at more than 16% CPU usage. > > perhaps there are other tricks to try, but i don't have the time. > But I do have. :) I'll try to do other things: - use raster graphics - do not use paths - do not use translucency - try to test the profile in other ways ( OpenGL, perhaps ) > > overall i'm very surprised that such as simple UI element is so slow > in Qt. i've written *much more* complicated UI in other languages and > frameworks and it was blazing fast compared to this. > > but like i said, leave it be for now IMHO. > > lubomir > -- >
_______________________________________________ subsurface mailing list [email protected] http://lists.subsurface-divelog.org/cgi-bin/mailman/listinfo/subsurface
