`Vector::remove_if` allows certain elements to be removed based on a predicate.
However `std::remove_if` only shifts the elements that will not be deleted
to the beginning of the container and then `Vector::remove_if` only
updates the pointer to the new last element after using `std::remove_if`.
This works well if `Vector` is used with trivial types that don't have a destructor.
Having a `Vector<std::unique_ptr>` for example, can generate undefined behavior
if the predicate gives `true` to elements that are contiguous at the end, if
`Vector::remove_if` only updates the end pointer in these specific cases, these
makes these smart pointers useless because they will not be freed by themselves.
To prevent that, also destruct the elements being removed.
Pull Request: https://projects.blender.org/blender/blender/pulls/115914
This PR enables shader validation testing on buildbot for Metal.
OpenGL isn't enabled as OpenGL requires an actual driver and GPU
attached to the build bot infrastructure. Also the OpenGL backend
caches data (glsl_patch) globally and requires a restart in order to
create the correct one.
Vulkan isn't enabled as it requires some changes:
* For windows it requires to install more recent vulkan software versions as
part of the buildbot windows configuration
* For Linux it requires to start a GHOST System without any X11/Wayland
This currently fails on the buildbot. We should check if we can use
`GHOST_SystemHeadless` with `GHOST_ContextVK`
Each shaders are compiled twice. Once based on the
actual features of the installed GPU/backend. And once with all
the work-arounds enabled, simulating a platform close to the
minimum requirements of Blender.
Pull Request: https://projects.blender.org/blender/blender/pulls/116040
Due to recent changes a cached patch string in GLShader grew out of
its bounds. This resulted in incorrect shader generation on selected
platforms (Reported was Windows/NVIDIA). The patch string can differ
based on the features that the GPU supports.
This PR replaces the old C-style string generation with CPP-style
string stream, making sure that the allocated memory grows with the
size of the string.
Pull Request: https://projects.blender.org/blender/blender/pulls/116085
The debug assert condition was inverted.
It is expected that we do not find a `FileHandlerType` with the given
`idname` during add. Like Operators being added twice from c code, this
will now assert if one is actually found. Python duplicates are handled
through `register_class` already.
Pull Request: https://projects.blender.org/blender/blender/pulls/116084
In the property context menu, the operator to View in Graph Editor can
apply to a single value or all values of a vector property.
Currently, these two options are presented in this order (Single, then
All), while all other operators (replace, delete, clear, add to keying
set, reset, copy to selected) follow the opposite order (All, then
Single).
This commit inverts this order for consistency's sake.
Pull Request: https://projects.blender.org/blender/blender/pulls/115962
This reverts #110582 - the workaround is no longer needed, as MS have updated
the version of mesa they ship as part of their compatibility pack, which contains
a workaround on that side instead.
This means that the supported version of OpenGL based on the host GPU is now
correct, bar `GL_ARB_texture_view`.
`GL_ARB_texture_view` is faked for ARM64 devices, as the part that is not
implemented (channel casting) is not used in blender.
This means that ARM64 devices (at least) now report compatibility up to OpenGL
4.6 in the system info export.
See: 698344b93c
Pull Request: https://projects.blender.org/blender/blender/pulls/116052
Report suspicious time-stamps as they could cause bugs which are
difficult to troubleshoot.
Currently the time-stamps are only used for double-click detection but
we may want to use them in the future, so ensure they're valid.
Print an error when two events are handled in the last second and the
current event has a time-stamp 5 seconds before or after the current
time.
Update code examples to current C++ style. Change formatting
slightly to use more columns. Rewrite secions to be more explicit
and precise. Avoid cryptic / uncommon code in examples. Finding
real edges in particular is never inlined, we just use the function
for it.
This patch adds an alternative path for devices/OSs
which do not support native texture atomics in Metal.
Support is encapsulated within the backend, ensuring
any allocated texture with the USAGE_ATOMIC flag is
allocated with a backing buffer, upon which atomic
operations happen.
The shader generation is also changed for the atomic
case, which instructs the backend to insert additional
buffer bind-points for the buffer resource. As Metal
also only supports buffer-backed textures for
textureBuffers or 2D textures, TextureArrays and
3D textures are emulated within a 2D texture, with
sample locations being indirected.
All usage of atomic textures MUST now utilise the
correct atomic texture types in the high level shader
and GPUShaderCreateInfo declarations.
Authored by Apple: Michael Parkin-White
Pull Request: https://projects.blender.org/blender/blender/pulls/115956
The logic can be much simpler when curves are selected rather than points,
because then we just copy all of the points in each curve. Like some other
operators, implement both cases.
Content cut off by `anim_startofs` is as if it does not exist for
sequencer. But Audaspace seeking relies on having animation buffer
initialized for whole sequence. Initialize pitch of cut content to 1.
This adds a new hybrid mode where the light and lightprobe
textures are bound to the gbuffer pass and used by
shaders that use ShaderToRGBA.
The counterpart is that these shaders have less texture
slots available but still two more than the Blended
materials.
These materials will correctly write to the AOVs and
render passes.
This is not too invasive in terms of implementation. The
only quirk was that we needed another fragment shader to
simplify the dependencies on the resources. But if we ever
get pre-processor support in our include / require system
this could be cleaned up.
Adds new curves support to "Draw Curve" operator. This is a direct copy
and replace of the old code to the new editors directly, with a change to
create the new data structure instead of the old one. There is no attempt
at code deduplication, since that would complicate removing the old
curve type when it comes time.
To replace the "2D curve" option a new operator property is added
for projecting to the Z=0 plane.
Pull Request: https://projects.blender.org/blender/blender/pulls/115864
This patch creates a static cached resource from bokeh kernel images,
for better performance and reusability, since it will be used by the
Defocus node as well.
This fixes a case where the source was set but the
internal pointers for binding were not updated.
This also tag for update internally.
Cleanup and remove the redundant tagging.
This also use the planar depth buffer as layer view.
During normal execution it isn't possible to switch workarounds.
However when running test cases we want to check if shaders and
other tests work when work arounds are enabled. Currently shader
patches are cached globally.
This PR moves the cached shader patch to the device level which
is rebuild every time a device needs to be reinitialized.
For OpenGL this is also an issue, but harder to solve as the concept
device doesn't exist there.
Pull Request: https://projects.blender.org/blender/blender/pulls/116042
