In some cases the MTLContext was being destroyed before all GPU work was completed causing the (outstanding) command buffer completion event handler to update a command buffer that had already been freed. This behaviour was introduced by [this](https://projects.blender.org/blender/blender/commit/6da42e9c951b) change which updated the event handler to track the number of outstanding command buffers per context as well as system-wide.
Reproduced the issue with ASAN enabled and confirmed that waiting for the GPU to complete fixes the issue.
Also contains a minor fix for unitiiliased values in MTLAttachments identified by ASAN.
Authored by Apple: James McCarthy"
Co-authored-by: James McCarthy <jamesmccarthy@apple.com>
Pull Request: https://projects.blender.org/blender/blender/pulls/129686
Printf buffer read needs to be inside render boundaries
to work. Since render boundaries can be nested, use a stack.
Fixes assert when quitting blender.
The list of vulkan devices is empty. This regression was introduced by
91960b07db. Vulkan needs to initialize the
platform twice. Once when the backend starts, and the second time when
the device is selected. The device list is added when the backend
starts, but was cleared when the device was initialized, making the list
empty.
Pull Request: https://projects.blender.org/blender/blender/pulls/130020
When gl_FragDepth is defined in the fragment shader and the depth isn't
written to the driver would reset the value to 0. This PR fixes this by
not putting the gl_FragDepth in the shader when it uses the
`depth_unchanged` qualifier or doesn't find any usage in the GLSL.
Pull Request: https://projects.blender.org/blender/blender/pulls/130016
For Metal we can change the texture usage flags to get more optimal
behaviour - one example is adding the attachment flag so we can utilise
renders to do texture clears. However these usage flags are used as the
part of the match-criteria when trying to reuse released textures in
the texture pool.
The modifications means a request for the same type of texture will
fail causing a cache miss. When we render to an
image-view the texture pool is not released until the final sample has
been rendered as we consider the entire render to be a single frame
(as opposed to normal viewport rendering when we are presenting the
intermediate results).
This causes the texture pool to grow and grow and grow hence the large
memory usage. This fix splits the usage flags
into two sets, the internal ones we use to create the MTLTexture (which
we may modify) and the originally requested ones. The originally requested
ones are used for the texture pool matching.
This fix also improves memory efficiency for normal viewport rendering.
Mr Elephant Scene
Before -> After
Load scene in viewport: 13.04Gb -> 9.15 Gb
Viewport Render Image: 78.69Gb -> 16.61Gb
Authored by Apple: James McCarthy
Pull Request: https://projects.blender.org/blender/blender/pulls/129951
When running test cases
`test_texture_roundtrip__GPU_DATA_10_11_11_REV__GPU_R11F_G11F_B10F`
would read and write outside of allocated memory. This is an error in
the test case itself the GPU API doesn't have a public function to get
the desired byte and component size.
Pull Request: https://projects.blender.org/blender/blender/pulls/129958
Drivers should perform a limit check when creating images and return
`VK_ERROR_OUT_OF_DEVICE_MEMORY`. However there are drivers where this
check is a pass-through and leads to `VK_ERROR_DEVICE_LOST`.
This issue was introduced !128877 and only shows up on official NVIDIA
drivers.
Pull Request: https://projects.blender.org/blender/blender/pulls/129939
This allows to use `assert()` directly inside shader source.
The current implementation is just replacing it with a printf
that gives some information about the location of the assert.
It is not that much more helpful than a printf with a condition.
What is useful is that they are disabled by default. So they
can be sprinkle around and only turned on during debugging.
Adding shader name inside the assert report is not trivial and
could be added later (the shader name is only known at compile
time which is too late for string parsing).
Adding which drawcall produced the assert isn't trivial either.
It would require flushing the printf buffer for each drawcall
which might force synchronization and remove bugs.
Pull Request: https://projects.blender.org/blender/blender/pulls/129728
This was caused by `drw_ResourceID` taking one vertex input
(at slot 15) which was then also used by material shaders.
Starting material shaders at 14 in this case avoid the overlap.
Note that this reduces the amount of supported attribute when
using the workarounds by one.
Crash manifested after the inclusion of #128877.
The very tall 3D texture tested by the new code
were not supported / tested by the Metal Backend.
Simply adding the appropriate upfront checks fixes
the issue.
Needs to be backported to 4.2
Adding a dummy storage buffer to the classification shader
seems to fix the issue on Qualcomm drivers (WoA).
The workaround is added to the force workaround option to
allow other platforms to test the fix.
Rel #122837
Pull Request: https://projects.blender.org/blender/blender/pulls/129857
When using AMD official driver clearing depth attachments can fail.
I assume it is related to previous pipeline states and dynamic rendering
that don't require the correct pipeline state for clearing depth
attachments.
Co-authored-by: jeroen@blender.org <Jeroen Bakker>
Pull Request: https://projects.blender.org/blender/blender/pulls/129852
Qualcomm driver can not handle the optimized SPIR-V that Blender
creates. According to their driver engineers this is an issue in
shaderc. As we are late in the release process they asked us to disable
the optimizations.
In Blender 4.4 we are planning to update shaderc what might fix the
issue. This should be retested after the update.
Pull Request: https://projects.blender.org/blender/blender/pulls/129775
