[Libreoffice] [REVIEW] 4 minutes 27 seconds -> 11 seconds

2011-07-12 Thread Caolán McNamara
cmc->thb: I had some .doc documents which were awesomely slow to load up, apparently because some polygons had to be split up into large polypolygons due to a custom dashed line in use to draw them or something of that nature. Could you double check my change to basegfx2 http://cgit.freedesktop.or

Re: [Libreoffice] [REVIEW] 4 minutes 27 seconds -> 11 seconds

2011-07-12 Thread Michael Meeks
Hi Caolan, On Tue, 2011-07-12 at 09:27 +0100, Caolán McNamara wrote: > I'm *sure* its good, but the scale of speed-up from 4 minutes 27 seconds > to 11 seconds is massively more than I expected. Oooh - nice :-) > Especially as the insert point is at the end, rather than at the > beginni

Re: [Libreoffice] [REVIEW] 4 minutes 27 seconds -> 11 seconds

2011-07-12 Thread Radek Doulik
Hi Caolán, the patch looks good to me. I am also surprised by the speedup (given that it always inserts at the end of the vector array in your case). Hard to check without the actual document. Might be problem in std::vector not detecting that the rest of reserved space doesn't need to be moved o

Re: [Libreoffice] [REVIEW] 4 minutes 27 seconds -> 11 seconds

2011-07-12 Thread Caolán McNamara
On Tue, 2011-07-12 at 12:51 +0200, Radek Doulik wrote: > Hi Caolán, > > the patch looks good to me. > > I am also surprised by the speedup (given that it always inserts at the > end of the vector array in your case). Hard to check without the actual > document. Might be problem in std::vector not

Re: [Libreoffice] [REVIEW] 4 minutes 27 seconds -> 11 seconds

2011-08-16 Thread Thorsten Behrens
Caolan McNamara wrote: > I'm *sure* its good, but the scale of speed-up from 4 minutes 27 seconds > to 11 seconds is massively more than I expected. Especially as the > insert point is at the end, rather than at the beginning or middle which > would have required a lot of memmove type operations. >

Re: [Libreoffice] [REVIEW] 4 minutes 27 seconds -> 11 seconds

2011-08-16 Thread Caolán McNamara
On Tue, 2011-08-16 at 16:14 +0200, Thorsten Behrens wrote: > Played with it today, surely ::append is on a critical path, but > beats me why exactly this is so much faster. Surely it saves some > temporary B2DPolygon instances... Were you able to reproduce the difference yourself ? IIRC I sent a s

Re: [Libreoffice] [REVIEW] 4 minutes 27 seconds -> 11 seconds

2011-08-16 Thread Thorsten Behrens
Caolan McNamara wrote: > Were you able to reproduce the difference yourself ? IIRC I sent a > sample document which, while horribly busted itself, was useful to > trigger it ? > Nope, on a dbgutil build currently - anyhow, the change is definitely good, knowing the why tickles my curiosity, but no