* Bump API patch number and header version number to 66 for this update.
Github Issues:
* Clarified how and when ename:VK_ERROR_TOO_MANY_OBJECTS is generated in
flink:vkAllocate Memory, and remove incorrect valid usage statement
about exceeding the API limit (public issue 356).
* Minor clarification of the description of
flink:vkUpdateDescriptorSetWithTemplateKHR::pname:descriptorUpdateTemplate
(public issue 564).
* Minor fixes for flink:vkCmdSetViewportWScalingNV (public pull request
588).
* Fix random name markup issues (public pull request 603).
* Fix code:BuiltIn decoration typo in the <<fxvertex-attrib, Vertex
Attributes>> section (public pull request 606).
* Fix synchronization language following the definition of
flink:vkAcquireNextImageKHR (public issue 607).
* Restore descriptions of several commands and structures missing from the
generated spec due to a mistyped asciidoctor conditional (public issue
612).
* Fix 1.0.41 changelog to refer to public issues 403/404 (public issue
618).
Internal Issues:
* Refactor valid usage statements with internal conditionals in
`copies.txt`, `pipelines.txt`, `renderpass.txt`, and `resources.txt` so
each branch of the conditional appears as a standalone statement which
can contain a separate VUID. This should have no impact on the generated
specs, but is necessary given the present state of the VU extractor and
the validation layer code that consumes them (internal issue 1043).
* Fix VkQueueGlobalPriorityEXT enum values missing _EXT suffix (internal
issue 1045).
* Clarified initial ownership of resources bound to shared memory objects,
(internal issue 1068).
* Fix duplicated valid usage ID tag for flink:vkCmdCopyImage, and make the
required layouts include ename:VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIONAL in
both cases (internal issue 1084).
Other Issues:
* Remove the noise functions from GLSL for SPIR-V for Vulkan in the
`GL_KHR_vulkan_glsl.txt` extension.
New Extensions:
* `VK_EXT_external_memory_host`
* `VK_EXT_external_memory_dma_buf`
* `VK_EXT_queue_family_foreign`
- add missing `*name` and `*link` and `code`
- change where wrong macro was used
- swap `*name` and `*link` where appropriate
- correct misspelled names
- that sort of thing...
* Bump API patch number and header version number to 65 for this update.
Github Issues:
* Replaced inaccurate "`pixel`" with "`texel`" or "`compressed texel
block`" as appropriate in the <<sparsememory, Sparse Resources>> chapter
(public issue 86).
* Attempt to clarify security/integrity guarantees in the
<<fundamentals-errors, Errors>> section (public issue 147).
* Update the <<memory-device,Device Memory>> section with clarifications
and markup fixes (public pull request 194).
* Fix typo VkDeviceCreateInfo -> slink:VkDebugMarkerObjectNameInfoEXT in
sample code for `VK_EXT_debug_marker` extension (public pull request
227).
* Clarified slink:VkFramebufferCreateInfo language regarding concurrent
use of attachment resources during a render pass instance (public issue
299).
* Added overlap rules for destination regions in <<copies,copy commands>>.
Also unified the sparse and non-sparse source-destination overlap rules,
since the non-sparse rules were technically inaccurate in the face of
aliasing in flink:vkBindMemory2 - the new rules are true regardless
(public issue 317).
* Clarified the <<features-features-samplerAnisotropy,
pname:samplerAnisotropy feature>> to only affect the
slink:VkSamplerCreateInfo::pname:anisotropyEnable value, and that
pname:maxAnisotropy is ignored when pname:anisotropyEnable is VK_FALSE
(public issue 503).
* Clarify pointer valid usage statements to use "`valid pointer to valid
_object_`" terminology and update the
<<fundamentals-validusage-pointers,Valid Usage for Pointers>> section
accordingly (public pull request 547).
* Some operations that use integer coordinates can also accept a LOD to
sample from. Add a description of that selection and the validity
conditions in the new <<textures-integer-coordinate-operations, Integer
Texel Coordinate Operations>> section (public issue 548).
* Update stext:VkImageSubresource* valid usage statements (public pull
request 550).
* Added text tying ename:VK_OUT_OF_POOL_MEMORY error for
flink:vkAllocateDescriptorSets to the number of descriptor types in the
allocating pool. Removed redundant "`length`" text about number of
descriptors returned (public issue 582).
* Update slink:VkSwapchainCreateInfoKHR descriptions (public pull request
585).
* Update slink:VkPipelineViewportWScalingStateCreateInfoNV and related
structures' valid usage statements (public pull request 587).
* Change some dates to conform to ISO 8601 as specified in the style guide
(public pull request 601).
* Fix some math markup problems and be more consistent in use of asciidoc
math markup (public pull request 602).
Internal Issues:
* Clarified that attribute reads from incomplete vertex buffer elements
are considered out of bounds accesses, in the
slink:VkPhysicalDeviceFeatures and flink:vkCmdBindVertexBuffers.txt
sections (internal issue 842).
Other Issues:
New Extensions:
- fixes rest of #416 (`VkImageSubresourceLayers`)
- migrate `VkImageSubresource` VUs where the state is known
- Simplify `VkImageSubresource` (`*count` always > 0)
1) Make explicit there is a <=> relationship between `DEVICE_LOCAL` heap
and `DEVICE_LOCAL` memory type
2) Make >4GB runtime error report mandatory. Otherwise it basically says
"do not pass >= 4GB, because possible undefined behavior".
3) Removed `VUID-vkAllocateMemory-device-00636`. Behavior is
deterministic returning `VK_ERROR_TOO_MANY_OBJECTS`
4) Moved `allocationSize` VU (depends on non-local state).
5) Changed `allocationSize`. Not sure what the original means, so
improvized a new one saying alloc size must be <= the whole size of the
heap.
6) Added VU restricting `memoryTypeIndex` to only the memory types of
the device.
1) Random grammar and formatting
2) Make example in first paragraph more
straightforward with elementary
Vulkan primitives (the original also
seemed factually wrong in places;
e.g. does UBO really have to be in
"on-device" memory?)
3) Removed unnecessary brandishing of obscure math
describing the memory
type list order. Tried something equally formal,
but more readable.
4) Made a note into a Note
5) Updated example.
Function names seem to be lower-case in the spec.
Hopefully made it bit
more readable. Turned the acompanying note into
the example content and
inline comments.
6) Removed redundant `VkDeviceMemory` intro
7) Break
out specifying language out of a Note
8) More straightforward description of `memoryTypeIndex`
* Bump API patch number and header version number to 64 for this update.
Github Issues:
* Add chapter name to the PDF page footer (public pull request 458).
* Fix several mistaken references to the nonexistent etext:VK_DEVICE_LOST
status to etext:VK_ERROR_DEVICE_LOST (public pull request 502).
* Fix description of the tlink:PFN_vkDebugReportCallbackEXT debug report
callback function pointer to match the validation layer behavior (public
issue 534).
* Document experimental KHX extensions and alternate vendor author IDs
also ending in X in more detail in the <<extensions, Layers &
Extensions>> appendix, the extensions section of the style guide, and
the registry schema description document (public issues 536, 580).
* Fix references to ptext:pDepthStencil to properly refer to
pname:pDepthStencilState or pname:pRasterizationState as appropriate in
the slink:VkGraphicsPipelineCreateInfo description (public issue 542).
* Fix wrong parameter name in slink:VkPipelineMultisampleStateCreateInfo
valid usage (public pull request 571).
Internal Issues:
* Update the style guide to describe how to write LaTeX math expressions
in table cells (internal issue 908).
* Define how framebuffer-local dependencies work between subpasses with
the same or different numbers of samples, in the
slink:VkSubpassDescription and <<synchronization-framebuffer-regions,
Framebuffer Region Dependencies>> sections. This clarifies which samples
in an input attachment you are allowed to access after a
framebuffer-local dependency (internal issue 915).
* Specify which storage classes can have an initializer in the
<<spirvenv-module-validation, Validation Rules within a Module>> section
(internal issue 1023).
* Use "LOD" consistently for "level-of-detail", to eliminate spelling
inconsistencies. The term is already standardized in the Glossary
(internal issue 1027).
Other Issues:
* Fix false positives in Makefile dependencies when rules fail, by
deleting partially-made targets.
New Extensions:
* `VK_AMD_shader_info`
- fix `pretransform` and `compositeAlpha`; they are values not bitmasks
- update `imageFormat`, `imageColorSpace`, `imageUsage` and
`oldSwapchain` to be specific to behavior of this command
- fix broken unintended code block
- move `VK_TRUE` clipping text to correct bullet point
- random tweaks...
- fix bad description (VU like) block + some tweaks
- remove redundant VUs (redundant due to required equality to
`VkPipelineViewportStateCreateInfo::viewportCount`)
- move `viewportCount` equality VU to
`VkPipelineViewportStateCreateInfo` where it belongs
- make a dynamic state related VU in `VkGraphicsPipelineCreateInfo`
1) State more explicitly in implicit VUs if a "valid" pointer is required
2) For some time Implicit validity is quite explicit in implicit VU
sections. Update pointer implicit validity chapter section as such.
3) Sometimes pointer parameter can be ignored. Update implicit pointer
validity section as such.
4) Changeing to more forceful language, to clarify memory behind
pointers need synchronizing too
* Bump API patch number and header version number to 63 for this update.
Github Issues:
* Add missing valid usage statements for ptext:maxDescriptorSets*,
ptext:maxPerStageDescriptorInputAttachments, and
ptext:maxPerStageResources to slink:VkPipelineLayoutCreateInfo.txt,
flink:VkComputePipelineCreateInfo, and
flink:VkGraphicsPipelineCreateInfo (public issue 546).
* Fix typos in ftext:vkCmdDraw*AMD descriptions (public pull request 549).
* Fixed flink:vkCmdWriteTimestamp so it is not unnecessarily restricted to
queues supporting graphics or compute operations (public issue 558).
* Improvements to valid usage generator for output `*Flags` pointer
parameters and for some `void *` parameters (public pull requests 560,
562).
* Document `altlen` attribute in XML schema as valid C99 syntax and tweak
`vk.xml` to match (public pull request 566).
* Clarify when pname:fence is signaled by flink:vkQueueSubmit in a more
obvious place (public issue 577).
Internal Issues:
* Specify a whitelist of supported SPIR-V Storage Classes in the
<<spirvenv-module-validation, Validation Rules within a Module>>
appendix (internal SPIR-V issue 166).
* Relax the shared semaphore wait timeout requirement in the
<<synchronization-semaphores-importing, Importing Semaphore Payloads>>
section (internal issue 820).
* Update the <<textures-image-level-selection, Image Level(s) Selection>>
equations so that the parameters returned by the level-of-detail query
appear explicitly, also fixing the issue that linear filtering would
select a level below the base level for magnification (internal issue
926).
* Disallow creation of a swapchain with zero pname:imageExtent in
slink:VkSurfaceCapabilitiesKHR and slink:VkSwapchainCreateInfoKHR
(internal issue 1020).
Other Issues:
* Clarify in <<textures-operation-validation,Image View Validation>> that
the layout of subresources in an image view must have a layout that
matches that written into the descriptor, and that this section is about
validating image views, not images.
New Extensions:
* `VK_EXT_global_priority`
1) Fix `pSampleMask `s `altlen` to have C99 semantics equivalent to len
2) Fix `registry.rnc` to also have altlen for `param`s
3) Document `altlen` in `registry.rnc` comments
4) Clarify `altlen` in schema doc that beside C syntax it also follows C
semantics
5) Clarify `altlen` in schema doc in case there ever is `latexmath` in
comma separated `len`
- Make implicit validity for optional `void*` mean "nothing" in the
script
- Make `optional` mean non-NULL for `void*` (not implemented)
- Uses of `void*` that point to something should have `noautovalidity`
(adding it)
- Add missing explicit VUs of `void*` parameters
(`vkCmdPushDescriptorSetWithTemplateKHR` and
`vkUpdateDescriptorSetWithTemplateKHR`)
- Also indirectly fixes `vkMapMemory`, `VkAllocationCallbacks`,
`VkDebugReportCallbackCreateInfoEXT`, and Apple and NN surface create
- Remove ugly text hack from script
Previously
(`vkGetDeviceGroupSurfacePresentModesKHX::pModes` and
`vkGetDeviceGroupPeerMemoryFeaturesKHX::pPeerMemoryFeatures`)
required output flags pointer to be valid + non-0
* Bump API patch number and header version number to 62 for this update.
Github Issues:
* Move asciidoc conditionals for `VK_KHR_maintenance1` in
slink:VkDescriptorSetAllocateInfo so valid usage statements for
`VK_KHR_push_descriptor` aren't accidentally removed when the first
extension isn't enabled (public issue 573).
* Fixed errors in API example code for
flink:vkUpdateDescriptorSetWithTemplateKHR and
flink:vkCmdPushDescriptorSetWithTemplateKHR (public issue 577).
Internal Issues:
* Specify constraints on concurrent access to fences that share payload in
the <<synchronization-fences-importing, Importing Fence Payloads>> and
<<synchronization-semaphores-waiting-state, Semaphore State Requirements
For Wait Operations>> sections (internal issue 820).
* Define the term "`retired swapchain`", reorganize some swapchain
language, and improve language for pname:oldSwapchain in
flink:VkSwapchainCreateInfoKHR, the <<swapchain-wsi-image-create-info>>
table, flink:vkDestroySwapchainKHR, and flink:vkAcquireNextImage2KHX
(internal issue 869).
* Describe in the <<writing-arrays, Describing Properties of Array
Elements>> section of the style guide how and when to use "`each`" and
"`any`" to refer to properties of array elements, and make those uses in
the specification consistent (internal issue 884).
* Clarified that events cannot be used for cross-queue synchronization in
the <<synchronization-events, Events>> section and for
flink:vkCmdWaitEvents (internal issue 970).
* Add success and error codes to +vk.xml+ for
flink:vkCreateValidationCacheEXT (internal issue 995).
* Clarify aspect mask usage for image memory barriers of multi-plane
images in slink:VkImageSubresourceRange, slink:VkImageMemoryBarrier, and
the <<textures-layout-validation, Layout Validation>> section (internal
issue 996).
Other Issues:
* Fixed typo in flink:VkRenderPassSampleLocationsBeginInfoEXT (renamed
field pname:pSubpassSampleLocations to
pname:pPostSubpassSampleLocations).
* Add missing buffer usage requirements for indirect draws in
flink:vkCmdDrawIndirect, flink:vkCmdDrawIndirectCountAMD,
flink:vkCmdDrawIndexedIndirect, and
flink:vlCmdDrawIndexedIndirectCountAMD.
* Modify Makefile to allow specification to build in git "`detached HEAD`"
state.
* Update valid usage ID generation script to allow recursively operating
on all `.txt` files in a specified directory, and move the `startVUID`
tracking information into a separate python file that is automatically
updated by the script.
New Extensions:
* Add +vk.xml+ entries for pending `VK_ANDROID_native_buffer` extension
(note, this extension is not yet enabled).
* `VK_AMD_shader_image_load_store_lod`
* Bump API patch number and header version number to 61 for this update.
Github Issues:
* Provide alternate length attributes (altlen=) in the XML schema, for
those using length attributes to generate code instead of documentation
(public issue 555).
* Fix erroneous references to `latex:` being used for asciidoc math
markup, rather than `latexmath:` (public pull request 556).
* Add author ID to XML for Kazan software renderer (public pull request
557).
Internal Issues:
* Add the <<fundamentals-abi,Application Binary Interface>> section
describing platform ABI requirements and recommendations, add examples
of function and function pointer declarations to the
<<boilerplate-platform-specific-calling-conventions, Platform-Specific
Calling Conventions>> section, and remove related language that existed
elsewhere in the specification (internal issue 64).
* Describe where to document valid usage interactions of chained
structures in the style guide, and fix one case now appearing in
slink:VkBufferCreateInfo instead of the child
slink:VkDedicatedAllocationBufferCreateInfoNV structure (internal issue
715).
* Add example to the style guide of describing enumerated types which are
empty when the spec is built without relevant extensions enabled, and
apply it to existing examples for
elink:VkDescriptorSetLayoutCreateFlagBits and
elink:VkSubpassDescriptionFlagBits (internal issue 864).
* Add a note to the <<fundamentals-validusage-enums, Valid Usage for
Enumerated Types>> section that the special values suffixed with
etext:_BEGIN_RANGE, etext:_END_RANGE, etext:_RANGE_SIZE and
etext:_MAX_ENUM are not part of the API and should: not be used by
applications (internal issue 872).
* Added note to flink:vkCmdUpdateBuffers explaining the performance
penalty for copies done in this way, and why the upper copy limit is
what it is (internal issue 952).
* Update `VK_KHX_device_group` to split some functionality into the new
`VK_KHR_bind_memory2` extension, and rename that functionality (internal
issue 969).
* Remove *Status* fields from extension appendices, since they are by
definition published and complete by the time they reach the public
github repository (internal issue 973).
Other Issues:
* Update Data Format specification dependency to version 1.2 and change
references to DF sections accordingly.
* Update XML to make the pname:pAllocator parameter of
flink:vkRegisterDeviceEventEXT and flink:vkRegisterDisplayEventEXT in
the `VK_EXT_display_control` extension as optional.
New Extensions:
* `VK_KHR_bind_memory2`
* `VK_KHR_image_format_list`
* `VK_KHR_maintenance2`
* `VK_KHR_sampler_ycbcr_conversion`
Petr Kraus
Jon Leech
Jon Leech
Lei Zhang
tomaka
Markus Tavenrath