Commit Graph

156 Commits

Author SHA1 Message Date
1faffa1d14 'fill_dptr' instead of 'zero_dptr' 2011-10-27 10:35:01 -04:00
3c1dac530b Updates to run_job.py 2011-10-27 10:26:30 -04:00
5368a9254a Clamp DE radius further.
The maximum standard deviation pushes far too hard into the limits of
the filter width, giving discrete points a weird boxy blur. The filter
slice width needs to be expanded, but that's a whole lot of coeffecient
debugging, and I'm putting it off by just reducing the maximum DE width
for now.
2011-10-27 08:58:51 -04:00
9049902b4f Add a crap gradient detect to make DE less bad.
Use the vertical and horizontal gradients to "detect" when a pixel is
part of an edge that has been softened by grid-shift AA, and avoid
blurring it further. This causes occasional 1px artifacts in stills, but
fixes the truly grotesque DE bleed-out for a net win. A better edge
detector is still needed.
2011-10-27 08:51:40 -04:00
7c84c6a7a9 Final xform color *is* used after all 2011-10-27 08:46:55 -04:00
f650844cb9 Fix two variations 2011-10-26 08:11:10 -04:00
a8528a9e1d Fix rgb2hsv 2011-10-26 08:10:57 -04:00
376cd752d6 Palette interpolation on device 2011-10-25 22:56:19 -04:00
e793527c29 A few harmless const modifiers 2011-10-25 22:49:26 -04:00
3436291eb6 Improve spline loading 2011-10-25 19:03:35 -04:00
fb5bdc2a9f Remove now-unused pyflam3_hacks 2011-10-25 19:03:10 -04:00
8939a6343a New genome representation, and device interp. 2011-10-25 15:44:39 -04:00
be31708c09 Fix memory corruption bug (overshoot in colorclip) 2011-10-25 15:43:05 -04:00
efc2ac23e2 Fix rendering at insane resolutions 2011-10-19 14:17:01 -04:00
20520d2f69 Open primes.bin in binary mode. 2011-10-17 19:31:09 -04:00
6c2df777b0 Remove a TODO 2011-10-16 13:52:01 -04:00
8ce2470dfb Relax FUSE a little (no visible impact so far) 2011-10-16 13:45:27 -04:00
c4ce3cf4c2 Don't crash on empty render(times) 2011-10-16 13:44:22 -04:00
0cc904c4f1 Do post affine transforms. How did I miss this? 2011-10-16 13:43:46 -04:00
5111a0f05c Eliminate needless pre_ var separation 2011-10-16 13:42:37 -04:00
9bafbda81a Refactor host rendering code for better load 2011-10-15 22:22:43 -04:00
9ff018de87 Actually fix dithering. (I've seen this before...) 2011-10-15 19:08:16 -04:00
63483480d0 Bias the radius to avoid very large dither offsets 2011-10-15 00:50:24 -04:00
3be14547ea Use 3*256 instead of 2*512 blocks; faster on GF104 2011-10-15 00:33:37 -04:00
c7728d3507 Add faster no-L1 accum 2011-10-15 00:32:30 -04:00
dd645bcbf6 Use one dither offset per block. 2011-10-15 00:29:22 -04:00
83670df2c7 Fix random seeds. 2011-10-14 11:56:58 -04:00
b168a2431e 32-bit compatibility (I think?) 2011-10-13 16:56:20 -04:00
14872ee6ed Add --sleep for slightly more usable system 2011-10-13 16:55:26 -04:00
e6e2c4a8d7 Add --sync option. 2011-10-13 07:53:55 -04:00
4834c9fdfa Change synchronization model. 2011-10-12 14:08:13 -04:00
81f61d4d5d Improve asynchrony; improve palette interp perf. 2011-10-12 14:07:28 -04:00
7b9bb165ac Change the way compile options are handled 2011-10-12 14:02:32 -04:00
f04ad7ab68 Performance improvements in Genome.__init__() 2011-10-12 13:57:43 -04:00
0f615bd98b Performance improvements in affine helpers 2011-10-12 13:56:34 -04:00
d409f02e4a Precompile accessors.
This improves packing speed by 8x, which is visible on small or
low-quality frames.
2011-10-12 11:50:07 -04:00
a12714f4c4 Fix MWC test 2011-10-12 07:36:07 -04:00
9b03f557c2 Fix missing control points in async version.
The allocation pool was reallocating the same frame as soon as it had
left the current scope, before it had been copied. We just reallocate
the same chunks. I don't think this has any real performance impact but
this can be verified.
2011-10-11 20:54:33 -04:00
b081bc9378 Remove a sync from iter.
A small but consistent improvement.
2011-10-11 14:56:23 -04:00
095936666e Actually asynchronous rendering.
This change didn't affect GPU performance at all, but it did improve CPU
startup time, and should also improve time for long-running animations.
2011-10-11 11:27:40 -04:00
8c7e86c7c7 Fixed fraction to not exceed range 2011-10-11 11:26:38 -04:00
618b51b1b1 Speed enhancement: alpha packing.
When the alpha channel is used in a color palette, the code now replaces
the blue channel in the accumulation buffer with a pair of two U16s,
which encode the values of the blue and alpha channels as a fraction of
the value of the density. When the alpha channel is always 1.0, the blue
channel works as normal. Density is now always the last element in the
accumulation buffer.

Eliminating the separate IO operations improved total runtime by more
than 30% on my card, while the extra calculations reduced that to 20%
when alpha was present (though that can be optimized further).
2011-10-11 09:57:37 -04:00
46c6074b92 Use C++ pass-by-reference to explicitly share. 2011-10-03 16:53:29 -04:00
851980f682 mobius d params were missing 2011-07-06 12:47:03 -06:00
18a60ec066 Major bugfix. Also include thread-swapping that works. 2011-06-25 20:37:08 -04:00
44f897f28e fixed enable/disable chaos 'if chain' in iter kernel function 2011-06-24 09:59:14 -06:00
b732a3c244 now the chaos 'if chain' is only used if there are non-unity chaos entries in the genome. 2011-06-24 08:18:08 -06:00
50b664b1f9 chaos support \0/ 2011-06-24 06:09:04 -06:00
bc2aa00e2a removed stray debug printfs 2011-06-21 11:24:06 -06:00
addad052b1 closes 11 - all black pixels with non-zero density were not being handled properly. 2011-06-21 11:22:20 -06:00