While Cycles already supports using both CPU and GPU at the same time, there
currently is a large problem with it: Since the CPU grabs one tile per thread,
at the end of the render the GPU runs out of new work but the CPU still needs
quite some time to finish its current times.
Having smaller tiles helps somewhat, but especially OpenCL rendering tends to
lose performance with smaller tiles.
Therefore, this commit adds support for tile stealing: When a GPU device runs
out of new tiles, it can signal the CPU to release one of its tiles.
This way, at the end of the render, the GPU quickly finishes the remaining
tiles instead of having to wait for the CPU.
Thanks to AMD for sponsoring this work!
Differential Revision: https://developer.blender.org/D9324
Previously Wintab was handled by saving the most recent tablet pressure and tilt information and deferred appending tablet infromation to Windows mouse events. This caused synchronization issues evident at the beginning and ending of strokes where pressure and tilt were either ahead or behind in time from mouse button up or down events. This also dicarded swaths of data which resulted in blockly grease pencil lines most apparent when a context switch resulted in several coalesced mouse events.
This patch changes the behavior of Wintab to instead rely entirely on Wintab information for pressure, tilt, position, and button input.
Wintab has several design decisions and conventions which complicate relying soley on it's input while retaining current workflows reliant on non-API behavior. For example, many device optionally allow the user to map barrel buttons to non-mouse actions. These mappings may or may not modify the intended behavior when touching the stylus down, such as scroll vs alt mappings. This behavior is not exposed in the Wintab API, but Wintab will continue updating button state sans this necessary context.
To work around the problem, this refactor synchronizations tablet input to Windows mouse down and up events, this captures events which should result in pen input while allowing events such as pen scrolling. Until a Windows mouse down event fires Wintab input is left unprocessed; when a Windows up event occurs Wintab is processed until a corresponding button up event is found.
Wintab allows for either button state or changes to be reported, but not both. An earlier refactor tried to use button changes to let state to be managed by Wintab. This was replaced when it was found that button change events were unreliable at corner cases such as switching windows. It was also found that with Wacom drivers Wintab peek functions would modify events in the queue causing errant and loss of button events.
For the latter stated reason this patch opts to read all Wintab events into a queue as they arrive, removing events as they become stale. This was chosen over using Wintab peek functions due to the afformentioned issue. As a bonus this seems to work better as it prevents the queue in Wintab from filling, thus neither a flood of events need to be handled when Wintab processing begins and a Wintab implementation need not be trusted to overwrite old events in it's queue.
Maniphest Tasks: T75566
Differential Revision: https://developer.blender.org/D7840
event to the button down location as this should be a more accurate point
of contact than the last mouse move event.
Signed-off-by: Nicholas Rishel <rishel.nick@gmail.com>
events until one is found. This prevents errant cursor moves that occur
before the Wintab button event is reported. We need to skip these events
because if no button event exists, we generate one assuming it will either
arrive later in the Wintab queue or that the button was from a non-Wintab
device. For the case that this was generated by a non-wintab device, such
as buttons mapped to mouse on the tablet pad, these cursor move events can
significantly move the cursor from the intended click position.
Signed-off-by: Nicholas Rishel <rishel.nick@gmail.com>
the issue with Wintab button events are more significant than simply
setting what buttons should receive button up/down events during context
initialization.
Signed-off-by: Nicholas Rishel <rishel.nick@gmail.com>
handling mouse input. This Wintab to mouse synchronization issues, and
likely prevents queue exhaustion for some Wintab implmenetations.
Signed-off-by: Nicholas Rishel <rishel.nick@gmail.com>
window intitialization can specify whether it will be visible regardless
of whether it is yet visible.
Signed-off-by: Nicholas Rishel <rishel.nick@gmail.com>
Button events now include tabletdata, so move is unnecessary.
Generate mouse button events when the system has an event but Wintab did not find a correlated event.
Only filter mouse button events, not Win32 Pointer events.
Signed-off-by: Nicholas Rishel <rishel.nick@gmail.com>
Maniphest Tasks: T75566
Differential Revision: https://developer.blender.org/D7404
Basic support for velocity updates with the APIC method.
This commit adds APIC to the already existing dropdown menu for the simulation method. The APIC plugin within Mantaflow has been updated to the latest version.
Don't use the current mouse position at the time the event is handled, but
rather the position at the time of the event. This should make e.g. brush
stroke paths more accurate.
In addition, this may solve issues with other software that does mouse
position smoothing. Ref T82143.
Use of the current mouse position was added in 12b642062c as part of a
large commit that also made continuous grab work. But it appears to still
work getting the mouse position from the event.
The issue stems from the fact that scene arrays are not cleared when rendering is done. This was not really an issue before the introduction of the ownership system (rB429afe0c626a) as the id_map would recreate scene data arrays based on their new content. However, now that the id_maps do not have access to the scene data anymore the arrays are never created.
Another related issue is that the BlenderSync instance is never freed when the persistent data option is activated.
To fix this, we delete nodes created by the id_maps in their destructors, and delete the BlenderSync instance before creating a new one, so the id_maps destructors are actually called.
Reviewed By: brecht
Maniphest Tasks: T82129
Differential Revision: https://developer.blender.org/D9378
Rather than just printing a message and falling back to the CPU. For render
farms it's better to avoid a potentially slow render on the CPU if the intent
was to render on the GPU.
Ref T82193, D9086
The existing code for this was incomplete. Each instance can now have a set
of attributes stored separately from geometry attributes. Geometry attributes
take precedence over instance attributes.
Ref D2057
This avoids OpenCL inlining heavy volume interpolation code once for every
data type, which could cause a performance regression when we add a float4
data type in the next commit.
Ref D2057
Previously, only predefined and limited set of intrinsics combinations
could have been refined. This was caused by a bundle adjustment library
used in the early days of the solver.
Now it is possible to fully customize which intrinsics are to be refined
during camera solving. Internally solver supports per-parameter settings
but in the interface they are grouped as following:
* Focal length
* Optical center
* Radial distortion coefficients (which includes k1, k2, k3, k4)
* Tangential distortion coefficients (which includes p1, p2)
Differential Revision: https://developer.blender.org/D9294
