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.
This commit makes BKE_image_acquire_ibuf referencing result, which means once
some area requested for image buffer, it'll be guaranteed this buffer wouldn't
be freed by image signal.
To de-reference buffer BKE_image_release_ibuf should now always be used.
To make referencing working correct we can not rely on result of
image_get_ibuf_threadsafe called outside from thread lock. This is so because
we need to guarantee getting image buffer from list of loaded buffers and it's
referencing happens atomic. Without lock here it is possible that between call
of image_get_ibuf_threadsafe and referencing the buffer IMA_SIGNAL_FREE would
be called. Image signal handling too is blocking now to prevent such a
situation.
Threads are locking by spinlock, which are faster than mutexes. There were some
slowdown reports in the past about render slowdown when using OSX on Xeon CPU.
It shouldn't happen with spin locks, but more tests on different hardware would
be really welcome. So far can not see speed regressions on own computers.
This commit also removes BKE_image_get_ibuf, because it was not so intuitive
when get_ibuf and acquire_ibuf should be used.
Thanks to Ton and Brecht for discussion/review :)
Storing this list in the node has the advantage of requiring far fewer calls to the potentially expensive internal_connect callback. This was called on every node redraw ...
Also it will allow Cycles to properly use the internal links for muted nodes, which ensures consistent behavior. The previous method was not applicable in Cycles because transient list return values are not supported well in the RNA and particularly the C++ API implementation.
It is not a well-supported feature of the primary node systems (shader, compositor, texture) in Blender. If anybody wants to create a node system that has actual use for loops, they can do so much more elegantly with Python nodes, but it does not have to be a core node type in Blender. Removing this should ease node code maintenance a bit.
The reason is that the per-node updates used for Reroute node type inheritance are not supposed to be looking at connected nodes, they are purely for "local" updates. For this sort of "global" update which requires depth-first search, the update function on the node tree level must be used instead.
Note that I had to script-tag all sokets' names, as they are currently completely unknown from bpy.types (and hence unreachable for our po generating scripts).
By holding shift and "cutting" a node link a new reroute helper node can be inserted. This consists of a single socket that can be used to insert additional connection points into a link. This can be used to keep a connection point in the tree when deleting a node, or to control the path of long connections for layout cleanup.
* Drag'n'drop translation in Outliner
* "Execute" button in file window
* "Labels" of spacing elements, in multi-column enums
* A glitch with nodes "Value to RGB", they where called "ColorRamp" in node_type_base() call. This is not definitive, though, as it appears that UI node names are determined by this call, while it should be by "defines" in rna_nodetrre_types.h, I guess... Anyway, not good to have such things in two different places!
Also moved default context name under BLF_translation.h, much better to have those all in one place, accessible from whole Blender code!
There are a number of features that use a kind of "internal linking" in nodes:
1. muting
2. delete + reconnect (restore link to/from node after delete)
3. the new detach operator (same as 2, but don't delete the node)
The desired behavior in all cases is the same: find a sensible mapping of inputs-to-outputs of a node. In the case of muting these links are displayed in red on the node itself. For the other operators they are used to relink connections, such that one gets the best possible ongoing link between previous up- and downstream nodes.
Muting previously used a complicated callback system to ensure consistent behavior in the editor as well as execution in compositor, shader cpu/gpu and texture nodes. This has been greatly simplified by moving the muting step into the node tree localization functions. Any muted node is now bypassed using the generalized nodeInternalRelink function and then removed from the local tree. This way the internal execution system doesn't have to deal with muted nodes at all, as if they are non-existent.
The same function is also used by the delete_reconnect and the new links_detach operators (which work directly in the editor node tree). Detaching nodes is currently keymapped as a translation variant (macro operator): pressing ALTKEY + moving node first detaches and then continues with regular transform operator. The default key is ALT+DKEY though, instead ALT+GKEY, since the latter is already used for the ungroup operator.
This commit extends limit of ID and objects to 64 (it means 63 meaning
characters and 1 for zero-terminator). CustomData layers names are also
extended.
Changed DNA structures and all places where length constants were hardcoded.
All names which are "generating" from ID block should be limited by MAX_ID_NAME-2,
all non-id names now has got own define called MAX_NAME which should be used all
over for non-id names to make further name migration stuff easier.
All name fields in DNA now have comment with constant which corresponds to
hardcoded numeric value which should make it easier to further update this
limits or even switch to non-hardcoded values in DNA.
Special thanks to Campbell who helped figuring out some issues and helped a lot
in finding all cases where hardcoded valued were still used in code.
Both of forwards and backwards compatibility is stored with blender versions newer
than January 5, 2011. Older versions had issue with placing null-terminator to
DNA strings on file load which will lead to some unpredictable behavior or even
crashes.
Now, compositing, shading and texture nodes have a consistent muting system, with default behaving as previous (for compo), and which can be optionaly customized by each node.
Shader nodes are also GLSL muted.
However, Cycles is currently unaware of muted nodes, will try to address this…
This allows node type init code to have access to the nodetree type object (needed to allow generic muting node initialization). Huge and boring edits...
This prevents access to non-existent typeinfo during type initialization,
when node types have been removed and such nodes are deleted from older files.
All blenkernel functions now only set the node->update flag instead of directly
calling the update function. All operators, etc. calling blenkernel functions
to modify nodes should make a ntreeUpdate call afterward (they already did that
anyway).
Editor/RNA/renderer/etc. high-level functions still can do immediate updates by
using nodeUpdate and nodeUpdateID (replacing NodeTagChanged/NodeTagIDChanged
respectively). These old functions were previously used only for setting
compositor node needexec flags and clearing cached data, but have become generic
update functions that require type-specific functionality (i.e. a valid typeinfo
struct).
This would previously break because begin/end functions for each tree type still have some checks of the ntree->execdata pointer in them, despite the intended use of execdata instances instead of trees themselves for execution data storage. This is an artifact of the old execution system that required these checks to be made in the functions to avoid multiple execution of top-level trees. Now these functions take an additional argument, so group nodes can prevent them from setting and checking the nodetree->execdata pointers.
