This code was being generated in both Vulkan-ValidationLayers
and Vulkan-ExtensionLayer. Further uses are on the horizon so
lets stop the copypasta.
Also, add functions to manipulate extension lists and pNext chains,
since many client layers have been doing that themselves.
- Update known-good
- Generate source
- Remove vk_layer_settings_ext.h
- Change all usage of VkLayerSettingEXT::count to use the new name
valueCount from the public header.
Generated code must macro-guard platform specific code, but did it in a
naive fashion where consecutive guards for the same platform repeated the
macro. With the help of PlatformGuardHelper, the code generation will elide
redundant macro guards.
Non Dispatchable handles are not typed in 32 bit builds, so the
vku::GetObjectHandle() helper fails to compile. The solution is to just
not define the helper in 32 bit builds.
Move the vk_typemap_helper.h file in Vulkan-ValidationLayers over to this
repo with a different name to reflect the use of it
Changes from that include:
* Rename the functions to be more obvious as to what they do
* Placing them in the `vku` namespace
* Adding the `InitStructHelper` class which deduces the type based on the
variable that is being initialized
* Compiler error if there is no corresponding sType
The library originates from Vulkan-ValidationLayers, but is being moved
into this repo to make it easier for others to use it.
The library has also been modified to be header only and C compatible,
which allows more developers to be able to use it. This does require some
changes, but only affects the vkuFormatElementSize and vkuFormatTexelSize
functions which used default parameters. Two new functions,
vkuFormatElementSizeWithAspect and vkuFormatTexelSizeWithAspect have been
added to handle the non-default image aspect case (the default was COLOR_BIT).
Renaming was done using the following convention:
* public header files begin with `vk_`
* enums begin with VKU_FORMAT_
* functions begin with vku
CI will fail now if we ship a .h / .hpp file that doesn't use
the correct prefix.
I also documented the precedent for this prefix and the existing
integration testing.
