* Update release number to 107.
Public Issues:
* Fix revision date for the `<<VK_AMD_gpu_shader_half_float>>` appendix
(public issue 617).
* Make <<synchronization-pipeline-barriers-subpass-self-dependencies,
subpass self-dependencies>> less restrictive (public issue 777).
* Fix the `<<VK_EXT_full_screen_exclusive>>` dependency on
`<<VK_KHR_win32_surface>>` in `vk.xml` (public pull request 849).
* Remove single-page (`apispec.html`) refpage sub-targets from the
Makefile `allman` target and the build instructions. The target is still
present in the Makefile, but we have not been actively maintaining the
single-page document and do not promise it will work. The full
Specification and the individual API reference pages are what we support
and publish at present (public issue 949).
Internal Issues:
* De-duplicate common valid usage statements shared by multiple commands
or structures by using asciidoctor includes and dynamically assigning
part of the valid usage ID based on which command or structure they're
being applied to (internal issue 779).
* Add reference pages for constructs not part of the formal API, such as
platform calling convention macros, and script changes supporting them
This required suppressing some check_spec_links warning classes in order
to pass CI, until a more sophisticated fix can be done (internal issue
888).
* Change math notation for the elink:VkPrimitiveTopology descriptions to
use short forms `v` and `p` instead of `vertex` and `primitive`,
increasing legibility (internal issue 1611).
* Rewrite generated file includes relative to a globally specified path,
fixing some issues with refpage generation (internal issue 1630).
* Update contributor list for `<<VK_EXT_calibrated_timestamps>>`.
* Fix use of pathlin in `scripts/generator.py` so the script will work on
Windows under Python 3.5 (internal merge request 3107).
* Add missing conditionals around the
<<descriptorsets-accelerationstructure, Acceleration Structure>>
section (internal merge request 3108).
* More script synchronization with OpenXR spec repository (internal merge
request 3109).
* Mark the `<<VK_AMD_gpu_shader_half_float>>` and
`<<VK_AMD_gpu_shader_int16>>` extensions as deprecated in `vk.xml` and
the corresponding extension appendices (internal merge request 3112).
New Extensions:
* `<<VK_EXT_headless_surface>>`
* Update release number to 103.
Public Issues:
* Remove (unnecessary) scoped modification order case from the memory
model <<memory-model-location-ordered, location-ordered>> definition
(public pull request 924).
* Add an <<memory-model-acyclicity, acyclicity>> axiom to the memory model
(public pull request 927).
Internal Issues:
* Fix reversed logic of slink:VkFormatProperties discussion of multi-plane
formats and ename:VK_FORMAT_FEATURE_DISJOINT_BIT (internal issue 1493).
* Clarify how slink:VkImageStencilUsageCreateInfoEXT works, including new
valid usage statements for flink:vkCmdClearDepthStencilImage, and the
<<copies, Common Operation>> section of the Copy Commands chapter
(internal issue 1565).
* Update <<spirvenv-precision-operation, Precision and Operation of SPIR-V
Instructions>> section to require that denorms be preserved by several
instructions that don't perform any mathematical operations (internal
issue 1584).
* Remove duplicate valid usage statement from flink:vkAcquireNextImageKHR
(internal merge request 3062).
* Update release number to 102.
Public Issues:
* Simplify flink:vkGetImageMemoryRequirements constraint for
ename:VK_MEMORY_PROPERTY_LAZILY_ALLOCATED_BIT (public pull request 817).
* Fix typo in markup of the <<textures-texel-coordinate-systems-diagrams,
Texel Coordinate Systems, Corner Sampling>> image that was generating
complaints from chunked HTML output generation (public pull request
928).
Internal Issues:
* Split the old <<features, Features>> chapter into four chapters:
<<features, Features>>, <<limits, Limits>>, <<formats, Formats>>, and
<<capabilities, Capabilities>>, with minor edits to the introductory
paragraph of each chapter. Anchor names in these chapters were changed,
with corresponding effects to xrefs to these anchors elsewhere in spec
markup . The purpose is to make the chunked HTML spec output load faster
on what was previously a single, gigantic chapter (internal issue 1554).
* Add ename:VK_PIPELINE_STAGE_RAY_TRACING_SHADER_BIT_NV, to the supported
pipeline stages for ename:VK_ACCESS_UNIFORM_READ,
ename:VK_ACCESS_SHADER_READ, and ename:VK_ACCESS_SHADER_WRITE in the
<<synchronization-access-types-supported>> table.
* Correct legal name of Google, LLC in vk.xml \<tags> section and a
copyright statement.
* Clarify that Vulkan treats the Android
code:AHARDWAREBUFFER_FORMAT_R8G8B8X8_UNORM format as RGBA, and the
application is responsible for forcing the X/A component to be read as
1.0, in the <<memory-external-android-hardware-buffer-formats>> table.
* Clarify the vertex order of various primitive topologies, and define the
order of transform feedback vertex capture based on that. This involves
a lot of refactoring and cleanup in the <<drawing-primitive-topologies,
Primitive Topologies>>, <<geometry-input, Geometry Shader Input
Primitives>> sections, and <<vertexpostproc-transform-feedback Transform
Feedback>> sections, and numerous places in the <<tessellation>>
chapter,
New Extensions:
* `VK_EXT_metal_surface`
* `VK_EXT_ycbcr_image_arrays` (internal issue 1361).
* `VK_NVX_image_view_handle`
Jon Leech