Made Texture compositor input node single-threaded since
texture trees are not thread-safe.
Also fixed texture being flipped horizontally and vertically.
Why nobody noticed this for 3 releases already??
Added compatibility option "Straight Alpha Output" to image input node
When this option is enabled, image input node will convert float buffer
to straight alpha.
This is not what you'll usually want with new alpha pipeline, nit this
is needed to preserve compatibility with older files saved in 2.65.
In that version byte image are resulting with straight alpha passing
to the compositor and alpha-overing required extra premultiplication
of inputs.
So, that's why Straight Alpha Output is needed -- it's set in versioning
code for byte node images so they'll still output straight alpha.
This option is currently only available in N-panel.
Additional change: added Alpha Mode for image input node to N-panel.
Calculate dependent area based on distortion model rather than
using 15% of image resolution for this.
Some assumptions here:
- We're assuming real-life camera calibration is used here
- Maximal undistortion delta would be achieved on frame boundary
- Distortion fully goes inside frame
This makes it possible to approximate margin for distortion by
checking undistortion delta across frame boundary and use it
for dependent area of interest.
We do not use any formula-based equation here because we're likely
support other distortion models and in that case it'll be stupid
to try detecting formula here.
* added area of interest based on a radius around the input tile.
If someone can implement a correct formula (based on K1, K2 and K3) and send it to me, I will be happy to apply it!
- At Mind -
Justa cluster did not have enough memory to handle all Mango 4k scenes.
Option is default disabled and can be enabled in the performance panel.
- At Mind -
is straight or not (premultiplied is default).
This is useful in cases when you want to check on output of such nodes
as keying which does have straight alpha output.
Also added missing do_version code to previous compo do_versions.
makes it possible to specify an offset relative to the render resolution (so 0.5
is half the image rather than giving the number of pixels). It's a bit late but
it's a trivial change and needed for 4k mango render.
Added additional buffers - new subtree - for groupnodes.
One needs to be aware of how groupnodes should be created.
Having translate & scale nodes, with the translate inside the groupnode and the scale node outside, causes artefacts.
Both should be inside or outside the groupnode. Same holds for other distort nodes.
This modules does not depend on any blender-specific data
structures or algorithms and due to our policy better be
placed to intern/
Shall be no functional changes, tested CMake and SCons on
Linux, hopefully other platforms will work as well.
P.S. SVN history shall be preserved for the files.
this patch enables the translate node to wrap around the image borders. This is especially needed if the translate node is not used to position elements on a layer but when it is used instead for seamless backgrounds like mountains or clouds that should be repeated over time (by animating the x/y values).
No trunk without docs! So here is my documentation: http://wiki.blender.org/index.php/User:Plasmasolutions/TranslateNodeExtension
The code is properly documented and should be easy to read and understand. When there are any problems or issues, please comment, I'll tackle them right away!
Greetings, Thomas Beck
* optimized determination dependant areas
* fixed some issues with scale node
There are still some issues when scaling very small values (x=0.0001)
- At Mind -
another patch for the dilate/erode step method, still without any functional changes.
This time it keeps the general algorithm but uses the tile system to make it
multithreaded. I could not measure a speedup on my 2-core laptop, but hope that
it will be faster for more cores. The immediate speedup that is very visible though is
that tiles come in as soon as they are calculated and a dilate/erode node does not
block the whole image to be calculated.
till then, David.
this patch optimizes the dilate/erode step method (hopefully without any functional change),
making its speed not depend on the distance anymore.
Couldn't detect funtional changes so committing. Haven't tested for speed gain.
* credits to erwin94 David M
It was caused by image threading safe commit and it was noticeable
only on really multi-core CPU (like dual-socket Xeon stations), was
not visible on core i7 machine.
The reason of slowdown was spinlock around image buffer referencing,
which lead to lots of cores waiting for single core and using image
buffer after it was referenced was not so much longer than doing
reference itself.
The most clear solution here seemed to be introducing Image Pool
which will contain list of loaded and referenced image buffers, so
all threads could skip lock if the pool is used for reading only.
Lock only needed in cases when buffer for requested image user is
missing in the pool. This lock will happen only once per image so
overall amount of locks is much less that it was before.
To operate with pool:
- BKE_image_pool_new() creates new pool
- BKE_image_pool_free() destroys pool and dereferences all image
buffers which were loaded to it
- BKE_image_pool_acquire_ibuf() returns image buffer for given
image and user. Pool could be NULL and in this case fallback to
BKE_image_acquire_ibuf will happen.
This helps to avoid lots to if(poll) checks in image sampling
code.
- BKE_image_pool_release_ibuf releases image buffer. In fact, it
will only do something if pool is NULL, in all other case it'll
equal to DoNothing operation.
- Drawing masks in image editor requires LOCK_DRAW_IMAGE around
ED_space_image_get* functions since they'll acquire image buffer.
Lock is needed because viewers will be modified directly in
compositor (see commend in draw_image_main)
- Seems that was wrong order of invalidating render result and
viewer image invalidation happened in Composite node, which
could easily lead to thread lock.
Issue was caused by resolution detecting which assumed zero resolution is
undefined one and should be re-evaluated. It doesn't work in cases when
there's a missing input, causing lots of unneeded resolution re-calculation.
It wasn't so much issue in average sized node trees, but it was a real
problem in generated tree from the report.
Currently used pretty simple solution which added a boolean flag to the
node operation which signal whether resolution was ever set or not.
There're probably smarter solutions here but can not think about them.
This codec is absolutely needed to generate DCP using OpenDCP,
before that external application to convert JP2 to J2K was used
which slowed down export a lot.
New codec is exposed to image format settings panel and called
Codec. Default one is JP2 which creates files with .jp2 extension,
new one is called J2K which creates with .j2c extension.
Other changes:
- Fixed avi jpeg warning which was treating as error here.
- Made it so extension is detecting from ImageFormatData instead
of image file type, which makes it possible to have different
extension for the same file type depending on it's settings.
IRIS format should still be changed (depending on number of
channels it'll be .bw, .rgb or .rgba extension)
- Default image format settings would be set from image buffer
when re-saving it. Makes it possible to easily open .j2c file
and save it using J2K codec (without this change it'll save as
.jp2 using JP2 codec)
There was a missing image reload signal in node creation by drag-dropping,
which lead to incorrectly set image type.
Also fixed misusage of IMB_freeImBuf used to release buffer acquired by
BKE_image_acquire_ibuf.
Think should be pretty much harmless since if this node was used for buffers
with infinities it already showed artifacts. Now it should be more useful for
mapping Z buffers.
from main thread using job update callback.
Added new execution-time callback to bNodeTree which marks job to be updated.
The code here could be a bit not so obvious because in some cases job update
callback need to merge local tree, but it's only needed for old compositor
system which is gonna to be removed soon, so decided not to bother with
cleanup now. Removing old compositor system will also allow to drop stats_draw
callback from bNodeTree.
This should fix following bugs:
