Optimize Workbench performance so it's on par with the previous
implementation.
Most of these changes are barely noticeable on powerful GPUs,
but can cause a notable performance improvement on old or low-end
hardware.
* Avoid unnecessary texture copies and draw directly to the viewport
textures.
* Optimize-out depth/stencil reads, using stencil testing instead.
* Avoid using `Texture::clear` and use framebuffer clears instead.
* Avoid framebuffer state changes (always use the same attachments).
* Avoid constant variation of acquired `TextureFromPool`s.
Fix#113010
Pull Request: https://projects.blender.org/blender/blender/pulls/113251
Fixed version of #112544 (reverted).
`DRW_drawdata_get` reuses the same `DrawData` for all duplis,
so they all end up using the same `ObjectHandle` and `ObjectKey`,
which breaks motion vectors.
* Don't rely on `DRW_drawdata_get` for storing `ObjectKey`s.
* Simplify `ObjectKey`.
This also solves the issue of objects created "on the fly" always having
the `ID_RECALC_ALL` flag.
Pull Request: https://projects.blender.org/blender/blender/pulls/113252
This PR enables vulkan backend as experimental option.
It will only be available in alpha builds on Linux and Windows.
This option is highly experimental and enabled to get some insight
on supported platforms. Don't expect a fully working Blender
yet. Also don't expect it to have usable performance.
**What is known to not work?**
* OCIO textures are not supported on Intel and AMD GPUs. sRGB/Standard is supported
on those platforms.
* AMD Polaris based GPUs on Linux will generate a crash when drawing the 3d cursor as it
doesn't support the needed vertex format. Comment out `DRW_draw_cursor` in `DRW_draw_region_info`.
* The colors in the node editor and sequencer are of as sRGB viewports aren't detected correctly.
* The image / UV editor isn't working as many texture formats haven't been tested yet. Some
tweaks are also needed to do correct depth testing.
* 3D Viewport is known to be flickering. Sometimes workbench doesn't display anything.
* 3D Viewport wireframe will crash as it uses a framebuffer with gaps between color attachments,
which isn't supported yet. (#113141)
* Rotate the view widget is partially drawn due to incompatible depth clipping.
* GPU Selection isn't working. It is expected to be solved when Overlay-Next will become the
default engine. For now disable GPU depth picking in the preferences.
* Cycles/EEVEE are known to not work with Vulkan yet. Cycles requires Vulkan Pixel Buffer.
Cuda <-> Vulkan interop might require a different approach than OpenGL as Vulkan doesn't allow
importing memory from a Cuda context. EEVEE uses features that aren't available yet in the backend
* Workbench is working, except Workbench shadows.
* EEVEE-Next basics are working. Shadows, lights are known to be not working. Materials/Shading
works in simple scenes. Changes are expected in EEVEE-Next that will break Vulkan compatibility
in the near future.
* Systems with multiple GPUs is not expected to work.
* Wayland support is in development and requires some iterations. You can start Blender, but
the protocols are not aligned yet.
* OpenXR hasn't been modified and is expected to fail.
* The backend is very strict when mis-using the GPU module. In debug builds it may crash
on asserts.
* Older drivers/GPUs might not have all the features that we require. The workarounds
for the missing features still need to be implemented.
**A word about performance**
In the project planning we focus first on stability and platform support. The performance of Vulkan is
around 20% of what we want to achieve. The reason is that each command sent to the
GPU is done one at a time. The implementation even waits until we have feedback that the GPU
is idle again.
Geometry is currently stored in System RAM. The GPU will read and cache the data when
accessing geometry. This slows down when using objects with much geometry.
Some performance features like MDI (Multi-Draw-Indirect) hasn't been implemented and
falls back to Single Draw Indirect.
**Why enable it is an experimental option?**
* Ensures that new features are being tested with Vulkan
* Ensure that building with Vulkan is possible on supported platforms
* Get feedback from developers if Vulkan can run on their system or that
there are special cases that we are not aware of. Main development
environment has been Linux/X11 with occasionally testing using Windows.
* Validate Add-ons that use the `gpu` module.
* Possible to enable GLSL validation on the buildbot. (Needs more work).
* Does it compile on all machines or does it require more changes to cmake
config. We expect it to be able to compile without installing the Vulkan SDK.
The Vulkan SDK is a very powerful tool, but only when actually doing GPU
development. Otherwise it is an overhead which slows down other
activities.
**How can the backend be enabled?**
Currently the Vulkan backend can be enabled per Blender session by starting
using the command line argument `--gpu-backend vulkan`. In the future, when
the backend is more mature, we will add a user preference to switch between
OpenGL and Vulkan.
Pull Request: https://projects.blender.org/blender/blender/pulls/113057
Wayland WSI would crash the system when used. The reason is that the
wayland vulkan WSI doesn't provide windowing support. Vulkan gets full access
to the desktop of the OS and it is the responsibilty of the application to
do the right thing.
For OpenGL Wayland proved basic windowing support using `wayland-egl.h`.
Which essentially is a tiny wrapper that keeps track of the window position and
size.
This PR changes a few things to make the Wayland surface usable:
- Do not load debug extensions when blender isn't started with
`--debug-gpu`.
- Recreate swapchain images when surface size changes.
Pull Request: https://projects.blender.org/blender/blender/pulls/113007
This fixes several issues related to using reflection probes in EEVEE-Next.
- When using a single sample, the reflection probes weren't always updated.
- Composite world background in reflection probes
- Removing reflection probes wasn't working
- Update UBO when world and reflection probes are active in the scene.
Pull Request: https://projects.blender.org/blender/blender/pulls/113347
Custom node trees may not suppor the default NodeSocketFloat socket
type. In case this default type is not supported, search all registered
socket types and pick the first one that is supported by the custom
node tree.
Pull Request: https://projects.blender.org/blender/blender/pulls/113330
Previously the colors were just uploaded as white, the default value.
Even if they aren't interpolated properly, it is still helpful to see
the colors. At worst, the unaffected parts of the mesh will still look
right.
A previous commit made vertex colors interpolate properly, but
face corner colors will still reset to their default value.
As a reminder, only color and byte color attributes are currently
supported for the specialized PBVH drawing.
Pull Request: https://projects.blender.org/blender/blender/pulls/113333
This change makes it so the newly added vertices have properly
interpolated attributes. This includes things like vertex colors.
New vertices are created by splitting edges, so the interpolation
mixes between the edge's two vertices equally.
Co-Authored-By: Hans Goudey <hans@blender.org>
This assert triggers whenever dyntopo is used, even when all the
objects and environment is pristine. The semantic of the assert
is not very clear either.
Avoid having a false-positive trigger which gets in the way of any
development in the area.
The issue was that the code filtered for selected channels,
while the expectation was that it would only filter for selected keys.
This PR changes the behavior of the operator in the following way:
* when "Clean Channels" is **disabled**, it will clean only selected keyframes, regardless of the channel selection
* when "Clean Channels" is **enabled**, it will clean selected channels regardless of keyframe selection
The same logic was applied to the Graph Editor code.
It only makes a difference in the case when "Clean Channels" is enabled.
That is because channels were automatically selected when a key was selected.
In addition to that I moved the menu entry for "Clean Channels" to the channel menu
to reduce confusion.
Another solution would have been to make the Dope Sheet select channels
when keys are selected. This might still be done in the future, but I think the
only correct fix is to change the actual operator behavior.
Pull Request: https://projects.blender.org/blender/blender/pulls/113335
Unify the way the different state's of a cache are shown in the timeline:
* Baked: fully opaque
* Cached: slightly transparent
* Invalid cache: slightly transparent, dark diagonal stripes
This improves accessibility since patterns are easier to recognize
for colorblind or otherwise visually impaired people.
The slight transparency is done with an alpha of 0.7 and the diagonal
stripes use the cache's color at 50% value.
Implements #108196.
Pull Request: https://projects.blender.org/blender/blender/pulls/108481
Speckles and missing lights were experienced in scenes with Nishita Sky
Texture and a Sun Size smaller than 1.5°, such as in Lone Monk and Attic
scenes.
We previously worked around these by using a more precise
software implementation of cosine.
After recent changes in Cycles, it turns out this workaround isn't
currently needed.
This fails to differentiate between active buttons and disabled buttons
for some custom themes (and also in blender light theme)
Instead use text color with 0.5 alpha value for disabled item's text.
(Don't blend between text and inner color)
Pull Request: https://projects.blender.org/blender/blender/pulls/113082
Previously this was the double the CPU count because:
- Modern CPU's from AMD & Intel support SMT/hyper-threading which
present twice as many cores, doubling again has little to no benefit.
- Using 2x or 4x the number of physical cores number can use a lot of
memory on systems with many cores which are becoming more common.
Update the label of the "Bone Pose" theme setting to "Bone Pose Selected".
That is now consistent with the already-existing "Bone Pose Active"
label.
Also add tooltips that clarify what these theme colors are used for.
This PR implements an initial drawing tool that can already be used for testing.
While this is not fully feature complete (compared to the current grease pencil draw tool) the following is already implemented:
* Pressure support for radius and opacity.
* Material color and vertex color support.
* New active smoothing algorithm based on curve fitting.
* Simplify algorithm as a post-process step.
Some deliberate limitations include:
* The drawing plane is always the front plane. Drawing on surfaces is also not supported.
*
The current approach has not been optimized for performance yet. The goal was to have a straightforward implementation
first and then focus on performance later.
There are numerous parameters in the code that are hard-coded for now. These should be exposed at some point, potentially as user settings.
Pull Request: https://projects.blender.org/blender/blender/pulls/110093
AgX exhibited some banding-like artifacts that were due to being
approximated with a 3D LUT. This commit resolves that by increasing
the LUT resolution enough to mitigate the artifacts and make them
unnoticeable.
Additionally:
- The previous LUTs were written in a space-inefficient way, using
e.g. "0.000000" instead of "0". The new LUTs are written more
efficiently, avoiding quite as dramatic a file size increase as
usually accompanies 3D LUT resolution increases.
- The previous LUTs included output values greater than 1.0, which was
both incorrect for a tone mapper, and also pointless since Blender
immediately clips them anyway. The new LUTs clip to 1.0. This also
allows the more efficient writing to squeeze even more space savings
out of the LUTs.
- The previous inverse AgX LUT contained NaNs. Those have been
replaced with 0.0 in the new inverse LUT.
Note that due to discrepancies between the LUTs previously provided
to Blender and the AgX scripts that were later published, the color
transform in these LUTs are slightly different. But they are close, and
equivalently good.
Pull Request: https://projects.blender.org/blender/blender/pulls/113253
The window contents and the window boarders were noticeably out of sync
when resizing the window quickly.
Resolve by keeping the current size as-is, rely on deferred handling
of the pending window size to apply the new size along with the contents.
Any window state change (resizing for e.g.) triggered
activation/deactivation events. Resolve by only sending events on state
change. The activation caused cursor motion events from #107594.
Resizing a window in Wayland caused cursor motion events in the window
which could be seen as buttons flashing when the cursor was detected
as hovering over buttons.
This was caused by two bugs:
- Missing checks for failure to access the cursor location before
converting the coordinates from GHOST to screen-space meant the
wmWindow::eventstate location would move each time the location
was updated.
- Resizing the window wasn't detecting state changes and would
continuously send window activation events. Window activation set
wmWindpw::addmousemove which triggered the previous bug, making the
cursor flicker during resize.
This commit only addresses the first issue, where failure to access
the cursor location wasn't accounted for
(window activation will be fixed separately).
All GHOST_GetCursorPosition & wm_cursor_position_get calls now account
for failure, resolving uninitialized stack memory use in some cases.
This resolves similar issues for macOS, WIN32 & X11 although it seems
likely these platforms rarely fail to access the cursor location.
BGL is deprecated and will not work on Metal devices. Although the
inital plan was to remove it in Blender 4.0, We don't see any harm
to still have it in the code-base until OpenGL itself is deprecated.
Add-on developers are warned when using the BGL module that the
add-on/script will not work on all platforms.
There are still some limitations inside the GPU module that needs
a more friendly API. This API isn't clear at this time.
Pull Request: https://projects.blender.org/blender/blender/pulls/112579
When units were initially defined having each on their own line was
compact, since them more fields have been added, making the lines
overly long and the difference between each field non-obvious.
Further, the conversion from C to C++ [0], wrapped definitions onto the
same line (for some reason), resulting in lines over 700 wide.
Use clang-format & add struct ID's for clarity.
[0]: 129f78eee7
