2009-08-04, 12:40
No, you can also load folder.jpg/poster.jpg inside the list... I tried it with my 1080p poster's and it's awfully slow. ^^
2009-08-04, 13:10
Ah, that's right but you can't (or at least couldn't when I tried) do it for TV Shows.
2009-08-04, 21:16
Hitcher: I believe that thumbsize is only for autogenerated thumbs and bookmarks. RESAMPLE_CACHED_IMAGES has to be defined for XBMC to do that and as far as I can tell it is not.
All the rest of you fruitcakes with 1080p resolution images complaining about how long it takes to decompress and load them, here's a tip: use uncompressed 24bit bitmaps for your mastodonic-sized art. This may seem counter-intuitive because the files will be 10x larger, but I/O throughput isn't as big a problem as decompression CPU time.
1080p PNG - 300ms
1080p JPG - 200ms
1080p BMP - 70ms (this should actually be much lower but I'm not about to dig into why it isn't)
You could probably hack the background image loader to use like 4 threads instead of one too.
All the rest of you fruitcakes with 1080p resolution images complaining about how long it takes to decompress and load them, here's a tip: use uncompressed 24bit bitmaps for your mastodonic-sized art. This may seem counter-intuitive because the files will be 10x larger, but I/O throughput isn't as big a problem as decompression CPU time.
1080p PNG - 300ms
1080p JPG - 200ms
1080p BMP - 70ms (this should actually be much lower but I'm not about to dig into why it isn't)
You could probably hack the background image loader to use like 4 threads instead of one too.
2009-08-04, 21:38
CapnBry Wrote:You could probably hack the background image loader to use like 4 threads instead of one too.go for it
2009-08-04, 22:38
I originally did not have the delay when I was using the orginal Aeon, Auriga, or Visim. I have kept all the settings the same. I have the posters all saved in the movie folders as <movie>.tbn.
Now when I use Aeon Hitched or andyblac's Aeon I get the delay.
Hitcher are you suggesting I lower that value for thumbnail size in advanced settings or raise it?
And, as always, any help is appreciated.
Now when I use Aeon Hitched or andyblac's Aeon I get the delay.
Hitcher are you suggesting I lower that value for thumbnail size in advanced settings or raise it?
And, as always, any help is appreciated.
2009-08-04, 22:53
watzen Wrote:go for itNoooo! I said you. you could hack it. Quit trying to make me work.
2009-08-08, 23:35
Just for fun this afternoon I hacked in DXT1 support into my win32 build. First the downside:
-- The files are bigger. A 1080p texture is 1.31MB regardless of content, unlike PNGs whose size varies based on the entropy of the image.
-- No alpha. 1-bit alpha is ok but the image quality may suffer to compensate. If we want full alpha we're talking 1.62MB per 1080p texture
-- Can't resample image if original is non-power-of-two and the target doesn't support NPOT textures.
-- May not be as high quality as the original depending on content.
Now the good parts:
-- My test 1080p PNG takes 180ms to load on average. The DDS version takes 20-30ms.
-- The DDS takes 1.31MB of VRAM, the PNG takes 7.91MB of VRAM.
-- There's also the amount of time it takes to upload to GL, which is probably significantly less, but is difficult to measure.
Oh back to the bad news, a good chunk of code would have to change to support it.
-- The files are bigger. A 1080p texture is 1.31MB regardless of content, unlike PNGs whose size varies based on the entropy of the image.
-- No alpha. 1-bit alpha is ok but the image quality may suffer to compensate. If we want full alpha we're talking 1.62MB per 1080p texture
-- Can't resample image if original is non-power-of-two and the target doesn't support NPOT textures.
-- May not be as high quality as the original depending on content.
Now the good parts:
-- My test 1080p PNG takes 180ms to load on average. The DDS version takes 20-30ms.
-- The DDS takes 1.31MB of VRAM, the PNG takes 7.91MB of VRAM.
-- There's also the amount of time it takes to upload to GL, which is probably significantly less, but is difficult to measure.
Oh back to the bad news, a good chunk of code would have to change to support it.
