When creating a texture view, Metal may require that the original
GPUTexture state is modified in some way. This may be a result
of deferred creation, or, to cache the texture view against the
source.
As a result, GPUTexture passed into GPU_texture_create_view
cannot be const.
Small fixes have also been made in the Metal texture
implementation to ensure correct function of texture views.
Authored by Apple: Michael Parkin-White
Pull Request: https://projects.blender.org/blender/blender/pulls/107167
Leverage new API call in Metal to increase the number of threads
dedicated to concurrent shader compilation. First step to improve
parallel compilation times when multiple engines are active.
Would also enable an increase in worker threads for shader
compilation jobs within the DRWManager.
Note that this is only available in the latest
version of macOS Ventura (13.3).
Authored by Apple: Michael Parkin-White
Pull Request: https://projects.blender.org/blender/blender/pulls/106625
Previous fix to resolve GPU hang which could occur in the
Metal backend caused additional flickering to occur as
as side effect, due to removal of required execution
dependencies in certain places.
This patch resolves both problems by only removing the
GPU hang dependency stall when additional synchronization
primitives are used along-side the global sync primitive.
Authored by Apple: Michael Parkin-White
Pull Request: https://projects.blender.org/blender/blender/pulls/106914
Changes to viewport state to resolve texture paint color
selection introduced a side effect wherein the correct
attachment size of a framebuffer was reset. This size is
needed when scissor regions are disabled to return the
state to its correct default. When this default was wrong,
certain operators would have incorrect offsets.
To maintain consistency with the OpenGL backend, the
Metal backend independently tracks the raw attachment
size using default_width/height. This will also reset to zero
when attachments are all removed, unlike other state which
may be retained.
Authored by Apple: Michael Parkin-White
Pull Request: https://projects.blender.org/blender/blender/pulls/106857
Issue caused by inconsistency in GPUFramebuffer viewport state
between Metal and OpenGL. The MTLFramebuffer code has been
modified such that framebuffer viewport/scissor state is retained
and only updated if attachments are modified during bind.
This is consistent with OpenGL. Previously, other updates to the
framebuffer in Metal would reset the viewport region, especially
if attachments were temporarily removed. This caused the color
picker selection to be misaligned.
Authored by Apple: Michael Parkin-White
Pull Request: https://projects.blender.org/blender/blender/pulls/106619
This patch refactors the texture samples code by mainly splitting the
eGPUSamplerState enum into multiple smaller enums and packing them
inside a GPUSamplerState struct. This was done because many members of
the enum were mutually exclusive, which was worked around during setting
up the samplers in the various backends, and additionally made the API
confusing, like the GPU_texture_wrap_mode function, which had two
mutually exclusive parameters.
The new structure also improved and clarified the backend sampler cache,
reducing the cache size from 514 samplers to just 130 samplers, which
also slightly improved the initialization time. Further, the
GPU_SAMPLER_MAX signal value was naturally incorporated into the
structure using the GPU_SAMPLER_STATE_TYPE_INTERNAL type.
The only expected functional change is in the realtime compositor, which
now supports per-axis repetition control, utilizing new API functions
for that purpose.
This patch is loosely based on an older patch D14366 by Ethan Hall.
Pull Request: https://projects.blender.org/blender/blender/pulls/105642
Splitting workload dependency chains such that they
only exist within the context of a single frame.
Dependencies are required to ensure sequential
command buffer submissions execute in order,
but the additional dependencies between frames
could incur GPU timeouts, if a signal was delayed.
This could be triggered by both CPU/GPU cycles
texture updates and Viewport Compositor operations.
Should also resolve#106401
Authored by Apple: Michael Parkin-White
Pull Request: https://projects.blender.org/blender/blender/pulls/106443
Ensure compression is enabled by using shader_read
flag only. Also ensure that MTLTexture contents
remain in optimal layout for reading after any
data modifications.
Authored by Apple: Michael Parkin-White
Pull Request: https://projects.blender.org/blender/blender/pulls/106234
For example
```
OIIOOutputDriver::~OIIOOutputDriver()
{
}
```
becomes
```
OIIOOutputDriver::~OIIOOutputDriver() {}
```
Saves quite some vertical space, which is especially handy for
constructors.
Pull Request: https://projects.blender.org/blender/blender/pulls/105594
Adds two modes of GPU frame capture support for
enhanced debugging. GPU frame capture begin/end
allow instantaneous frame capture of all GPU commands
within the capture boundary.
GPU frame capture scopes allow several user-defined capture
regions which can wrap key parts of code. These scopes are
exposed to connected GPU tools allowing the user to manually
trigger a capture of a known scope at the desired time.
This is currently integrated with the Metal backend for
support with Xcode.
Related to #105591
Pull Request: https://projects.blender.org/blender/blender/pulls/105717
Fix race condition if several competing threads are inserting Metal
buffers into the MTLSafeFreeList simultaneously while a new list
chunk is being created.
Also raise the limit for an MTLSafeFreeListChunk size to optimize
for interactivity when releasing lots of memory simultaneously.
Authored by Apple: Michael Parkin-White
Pull Request: https://projects.blender.org/blender/blender/pulls/105254
Metal buffer binding validation would trigger an error
when a given shader had an empty PushConstantBlock.
This patch removes the default uniform code gen if
no uniforms are present, to avoid any possible issues
with buffers being bound to a shader where the destination
data block is size zero.
Authored by Apple: Michael Parkin-White
Pull Request: https://projects.blender.org/blender/blender/pulls/105796
Crash when selecting objects on AMD platforms running
Metal. This was caused by shader compilation warnings
being treated as errors in macOS 10.15. Wrapping
compilation failure with success check resolves error.
Authored by Apple: Michael Parkin-White
Pull Request: https://projects.blender.org/blender/blender/pulls/105739
immDrawPixels performs significantly slower in Metal
than OpenGL. This was caused by two main factors. Firstly,
the additional overhead of tiled texture update, where all
memory needed to be kept in flight for each update, but
caused update to take a slow path. Avoiding tile update
with Metal is more efficient for both memory pressure
and GPU pipelining.
Secondly, on AMD platforms, the staging buffer used
for temporary texture data was page-faulting when
several texture updates would occur within one frame.
This is due to limitations of allocating one large contiguous
memory chunk. Using the Metal buffer pool for staging
data is more efficient.
Authored by Apple: Michael Parkin-White
Pull Request: https://projects.blender.org/blender/blender/pulls/105794
Fix support for Wireframe and parametric nodes by resolving
compilation failures surrounding barycentric coordinates.
A final missing part of the Metal implementation for barycentric
coordinates was missing.
Feedback also addressed to move barycentric calculation out
of code-gen and into surface_lib.
Authored by Apple: Michael Parkin-White
This also resolves#103606.
Ref #96261
Pull Request: https://projects.blender.org/blender/blender/pulls/105740
Previously, UBO bind locations were linearly incremented and
relied on the correct uniform location being queried. This fix
is a future requirement for EEVEE next, however, pulling forward
due to Issue #105280 highlighting a possible flaw with expected
uniform locations.
Authored by Apple: Michael Parkin-White
Ref #96261
Pull Request #105311
Similar to recent issues with gl_shader_interface,
ShaderInput lists need to be sorted to ensure correct
and efficient uniform lookup by name.
Authored by Apple: Michael Parkin-White
Ref #96261
Pull Request #105239
Metal LineLoop emulation path does not correctly apply
when using SSBO vertex fetch mode alongside 3D line
rendering.
Patch moves line emulation above SSBO
vertex fetch setup to ensure the correct emulation
parameters are passed to the shader.
Authored by Apple: Michael Parkin-White
Ref #96261
Pull Request #105142
Resolves issue with nearest filtering on UI Icons. Note that as
Metal does not support LOD bias as a parameter on a sampler
object, the original code has been modified to perform LOD
biasing at the shader level.
As GPU_SAMPLER_ICON is not widely used, it is more
efficient to apply directly to the affected shaders, rather
than workaround passing in the sampler LOD bias as a
separate value e.g. uniform or push constant.
Original PR feedback addressed to also refactor ICON
shaders to use consistent style for single and multi
Icon rendering.
Authored by Apple: Michael Parkin-White
Ref #96261
Pull Request #105145
This remove default casses from the `switch` statements to catch where
the missing cases are.
Uncomment unimplemented cases for the sake of completeness. Improving the
overall API.
This make the format conversion lists exhaustive and documented.
This replace `validate_data_format_mtl` by the common version as they
don't differ at all now.
For every other texture types this is expected to be implicitly
`GPU_DATA_FLOAT`. There is only one case where this is not the case.
I believe this was previously needed because the data type was
conditionning the texture creation. This is not the case anymore.
The check was triggering the 'this' pointer cannot be null in
well-defined C++ code
We do not check for this pointer in any other areas. If it is
needed due to possible opaque pointer cast to the check prior
to the cast.
Pull Request #104974
This reverts commit 19222627c6.
Something went wrong here, seems like this commit merged the main branch
into the release branch, which should never be done.
