| <<<Back 1 day (to 2014/01/03) | 2014/01/04 |
ray_laptop | Robin_Watts: I see. so that's why the relatviely small DeviceN color cache helped so much (23->12 sec) ? | 00:01.45 |
Robin_Watts | I think so. | 00:03.59 |
ray_laptop | I during work on the clist for transparency for devices that don't ordinairily support a clist, I found that the pdf14 optimization was never really enabled. Getting that close now. | 00:04.26 |
Robin_Watts | We could possibly convert to RGB at the top, and then linearly interpolate that - but the problem with doing that is that the functions might have local minima/maxima in etc. | 00:05.03 |
ray_laptop | It was (sort of) collecting the transparency bbox for bands. but then never using the data, so the pdf14 device was always being used. | 00:05.34 |
Robin_Watts | Ah, that could be a big win in some cases. | 00:05.51 |
ray_laptop | Robin_Watts: with tint transforms, that's not too likely | 00:06.13 |
| Robin_Watts: for the pdf14 clist optimization, I had it down to 6 cases that were bad. I fixed 3 of them and decided to re-run the bmpcmp. We'll see... | 00:07.55 |
| All right! Down to a single fie that's problematic: tests_private/pdf/sumatra/x_-_blend_modes.pdf | 00:25.11 |
| This file is a *real* mess. mudraw and gs get totally different results from Acrobat Foxit and GS are reasonably close | 01:41.15 |
| Forward 1 day (to 2014/01/05)>>> | |