This was actually added in iOS 13, but it wasn't present in the betas.
Since the betas also didn't support family 6, this leads me to suspect
that `min_lod_clamp()` requires family 6. So to be safe, only enable the
feature on family 6.
Update SPIRV-Cross to pull in the changes needed for this.
Apple family 7 GPUs (A14) on iOS support multisample layered rendering,
as well as sampler border colors and the mirror clamp to edge sampler
address mode.
We know the actual threadgroup size, because it is declared in the
shader. Therefore, we know the total count of threads per threadgroup,
which is simply the product of the threadgroup size in all three
dimensions. This is necessary if Metal picks a size lower than the app
is expecting. At least one game (NieR: Automata) needs this to work
correctly.
Metal's validation layer doesn't like it when we render to both a 2D and
a 3D texture with layered rendering. Work around this by not setting the
`renderTargetArrayLength` if it would be one and we are rendering to
mixed 2D and 3D textures.
If the pipeline sets `[[render_target_array_index]]`, we're hosed either
way. Perhaps this is why Vulkan has a special flag,
`VK_IMAGE_CREATE_2D_ARRAY_COMPATIBLE_BIT`, intended for rendering to 3D
textures.
This is now supported in MSL 2.3. Support varies by device; devices that
support this return `YES` from `supportsPullModelInterpolation`. Based
on my testing, AMD devices do not yet support this, and Intel devices
do. Apple GPUs probably also support this, in order to support OpenGL on
top.
Update SPIRV-Cross to pull in the changes needed for this.
They can now be used as blit destinations and be cleared with a render
pass.
This also fixes a bug where we were incorrectly validating that the
format supported shader writes if the image were linear, even on
iOS/tvOS.
Add macros to set MTLPixelFormatASTC_*_HDR to MTLPixelFormatInvalid on tvOS.
Use MTLRenderPipelineDescriptor::inputPrimitiveTopologyMVK instead of native method.
Starting in macOS 10.15, all desktop GPUs support 256 kiB offsets. Apple
GPUs, however, do not support this until family 7. It isn't clear which
one is the supported limit on Apple Silicon Macs with family 5 or 6
GPUs, so I have left the Apple Silicon limit at 64 kiB for now.
Don't pass descriptor type to MVKDescriptor functions, which already know their
descriptor type, and remove redundant switch tests in each subclass.
Move sanity tests for correct descriptor type to callers.
MVKBufferDescriptor increment dynamic offset index as consumed during binding.
Remove need to track dynamic offset indexes in MVKDescriptorSetLayoutBinding.
