- User Since
- Dec 26 2013, 7:53 PM (298 w, 6 d)
Update on the per-block compression: It works, but the compression ratio suffers from the small blocks - the file from above is now 544MB on level 3. In theory dictionaries could help here, but I'm not sure how to implement that practically.
Mon, Sep 16
Comparison with LZ4 here (average of three runs):
Sun, Sep 15
Oh, well, I didn't know about that patch. I'll retest tomorrow with LZ4 included.
Fri, Sep 13
The invisible light in front of the background isn't a problem in Cycles - if you check the pixel values of the rendered image, you'll see that the lamp is actually there, but they also have an alpha of zero.
This is consistent with the lamps being transparent - after all, the point of transparent background is that you can composite it over your own background, and since the lamp is transparent, you're supposed to get both the background and the lamp.
Wed, Sep 11
The current code samples the CDF as a piecewise linear function, but the underlying PDF is piecewise constant since it's the derivative of the CDF.
In the past, I've considered tweaking the sampling algorithm to sample from a piecewise linear function instead of piecewise constant, but that would only help if the CDF map exactly aligns with the background image.
Tue, Sep 10
Sun, Sep 8
Sat, Sep 7
Fri, Sep 6
Fri, Aug 30
Wed, Aug 28
Mon, Aug 26
Fri, Aug 23
Aug 19 2019
I have an updated version of this patch that I could upload here, sure.
However, I'd like to investigate the updates that Stefan linked above first.
Aug 13 2019
Aug 10 2019
Aug 8 2019
Aug 7 2019
Aug 6 2019
Aug 5 2019
Aug 4 2019
Rebase (actually more of a rewrite) to the current master.
Aug 2 2019
Jul 31 2019
Memory spike shouldn't be too bad, even with the 1400 copies of the fairly complex shader it's only 10MB (and that's before the mesh and image sync).
Note that the merging step could also be done multithreaded by using atomic fetch_and_add to count the total nodes while getting the own per-thread offset, then allocating the memory in the main thread and then copying the nodes to the position from before in each thread. However, this would require some way of synchronizing the threads and I doubt that the additional complexity is worth it.
Fix a typo in a comment.
Jul 29 2019
Update to current master and changed from Groups to Collections.
Update to latest master and D3203
Rebase to latest master and two crash fixes.
Jul 24 2019
Jul 22 2019
Some info on what I have found so far:
Jul 20 2019
Jul 18 2019
Jul 7 2019
Jul 5 2019
Jun 13 2019
Jun 12 2019
As far as I can tell, this issue is somewhat misleading - at first glance it appears to be a color management issue, but what actually happens is that enabling the Gradient option disables the Color Picker controls. Note that by toggling the Gradient checkbox, you can swap which of the two colors appears dull and that the Color Swap button changes along with the colors.
Already fixed by rBfb03f50e069d.
Jun 7 2019
Jun 6 2019
Jun 5 2019
Jun 1 2019
Add a missing call to initialize the CurveMapping and change the random parameter to be a percentage.
May 29 2019
May 23 2019
May 20 2019
May 19 2019
Update for formatting.
May 17 2019
Rebase to current master, added support for automatically translating picked values into names
May 11 2019
Another rebase to get to the newwest master, also squashed to a single commit now.