This commit makes use of new taskpool feature (instead of allocating own tasks),
and removes the spinlock used to generate chunks (using atomic ops instead).
In best cases (dynamic scheduled loop with light processing func callback), we
get a few percents of speedup, in most cases there is no sensible enhancement.
This is what old dependency graph was doing and apparently this avoids some updates,
however it's not really clear why those nodes are considering done. Maybe just because
of the way how relations are working. But needs a closer investigation.
Appears mutex was guarateeing number of tasks is not modified at moments
when it's not expected. Removing those mutexes resulted in some hard-to-catch
locks where worker thread were waiting for work by all the tasks were already
done.
This reverts commit a1d8fe052c.
Brain melt here, intention was to reduce number of tasks in case we have not much chunks of data to loop over,
not to increase it!
Note that this only affected dynamic scheduling.
It uses some additional compute power and the evaluation priority is
not even used.
This brings fps 88.2 with blenrig_for_debugging.blend on this desktop.
Simple thing, and apparently fps goes up to 80 with the demo file from jpbouza.
Not sure why at this point fps is so much higher than the old dependency graph
here now. And it's definitely something what others should verify as well.
This reduces stress on the task scheduler and avoids some unwanted overhead
caused by all the threading business in the cases when there's only one
children node. We try to immediately switch to it's evaluation now, keeping
active thread up and running.
This bumps FPS from 58 to 64 on the blenrig test file from jpbouza.
This commit implements new function BLI_task_pool_push_from_thread()
who's main goal is to have less parasitic load on the CPU bu avoiding
memory allocations as much as possible, making taks pushing cheaper.
This function expects thread ID, which must be 0 for the thread from
which pool is created from (and from which wait_work() is called) and
for other threads it mush be the ID which was sent to the thread working
function.
This reduces allocations quite a bit in the new dependency graph,
hopefully gaining some visible speedup on a fewzillion core machines
(on my own machine can only see benefit in profiler, which shows
significant reduce of time wasted in the memory allocation).
Many keyboard layouts (italian, spanish, german...) have direct access to '+' key on main
keyboard area (not the numpad one), ans x11 has own define for this key, so use it instead
of generating an unkown key event.
Note that we most likely have much more missing 'specific' keycodes for non-US keyboard layout,
but think since we already had a 'minus' keyevent, supporting 'plus' one is totally consistent.
And we had a spare space in our defined values just for it even!
This keyevent is only supported/generated by x11 and cocoa Ghost backends for now,
neither SDL nor win32 seem to have matching key events...
This commit:
* Removes most of all dirty internal details from public atomi_ops.h file, and move them into /intern private subdir.
* Removes unused 'architectures' (__apple__ and jemalloc).
* Split each implementation into its own file.
* Makes use of C99's limits.h system header to determine pointer and int size, instead of using fix hardcoded list of architectures.
* Introduces new 'faked' atomics ops for floats.
Note that we may add a lot more real and 'faked' atomic operations over integers and floats
(multiplication, division, bitshift, bitwise booleans, etc.), as needs arise.
Reviewers: sergey, campbellbarton
Differential Revision: https://developer.blender.org/D1982
This was originally done for T46320 in order to re-store depsgraph state
after all the constraint modifications.
However, it relations were updated if there's any IK chain the the pose,
which is a bit too annoying.
Set is much slower to iterate through (due to cache misses and such) and
the only advantage of using set is faster removal of link. However, we are
iterating links much much more often than removing them, and even when we
are removing links we don't really need to remove link from nodes which it
connects -- we don't support partial depsgraph updates, so removing links
from nodes on destruction is a waste of time.
If we ever want to support partial updates we can have dedicated function
to remove link from nodes it connects.
This gives a surprising increase of fps from 42 to 56 with test file from
Mr. J.P.Bouza (blenrig_for_debugging.blend). Surprising because old DEG is
actually slower here (52 fps). Didn't see any regressions (and don't see
why they will happen), so let's ask our riggers and animators to perform
further speed tests ;)
Use atomic operations instead, should in theory improve timing of
scheduling. However, probably not so visible yet because actual
task scheduling still have some locks and memory allocations.
Baby steps, what would i say.
This was causing alignment issues which were only visible on a platform
of particular bitness, so it was easy to break stuff for 32bit when
working on 64bit platform and vice versa.
* Fix "Attempt to free NULL pointer" when copying strokes for the first time
* Fix poll callback on "paste" operator, so that it is possible to paste
strokes when there are no editable strokes visible.
Commit 86abddc9, caused an error when the face-region included boundary edges.
Since removing the faces first, caused the edges along the boundaries to be removed.
Add support for deleting faces and internal edges, that keeps boundaries.
Now, when creating new keyframes in the graph editor by ctrl-clicking, only
the newly created keyframes will be selected. This is a little workflow tweak
to make it faster to work, as you no longer have to deselect all, and then
re-select the newly added keys in order to manipulate them.
The old behaviour (not modifying the selection status of the old keys) has been
kept, but is now available via Shift-Ctrl-Click.
Feature request from @Shhlife
This is an experimental option that I found would have been useful to have,
when making my earlier anim test. The only contentious point is whether it's
better to use a separate boolean flag (perhaps bound to the labels) instead
of overloading this setting (and describing the usage in the tooltip).
In the nodes editor, Ctrl-LMB-drag is taken by "Cut", and it seems impossible
for any other keymap to override that. Instead, the default lasso select hotkey
there is Ctrl-Alt-LMB-drag instead.
To get Grease Pencil lasso select (in GP Editmode) to work in the Nodes Editor too,
this commit adds the Ctrl-Alt-LMB-drag binding here too. However, to make things easier,
this extra binding will be part of the Grease Pencil keymap everywhere. There doesn't
seem to be any conflicts with having this in place - until we find them, this should
be ok to have.
This commit adds some of the initial support for a properties region in the
Action Editor. There are currently no panels to display, as there is still
a lot of work required to port over the required internal architecture to
support the panels seen in the Graph Editor.
After some test, a new iteration parameter has been added in order to
apply repetitive smoothing to the stroke. By default 1 iteration is applied,
but can used any number between 1 and 3.
The repetition uses different levels of intensity from 100% of the defined smooth
factor for the first loop, 50% for the second and 25% for the third. We use in each
loop a smaller value in order to avoid deform too much the stroke.
There is still some instability in how the triangulations are happening,
where the triangle count of the resulting triangulation fluctuates resulting
in weird artifacts sometimes.
To reproduce, try drawing some U-shapes, and keep reloading the file.
While trying to track down why I still keep getting some random "extra" triangles
showing up when reloading files with fills, I noticed that there's an index mismatch
here that may cause problems in some rare cases:
There are "gps->totpoints" items in tmp_triangles, but there should only be
"gps->tot_triangles" triangles in the gps->triangles array. If for whatever reason
some of the triangles in tmp_triangles are invalid, this could have meant that
while a tmp_triangles candidate was skipped, a corresponding slot for a valid
triangle also got skipped.
