* Bump API patch number and header version number to 60 for this update.
Github Issues:
* Document that <<queries-timestamps, Timestamp Queries>> can only be
meaningfully compared when they are written from the same queue (public
issue 216).
* Document that the `<extension>` tag `type` attribute is required for
non-disabled extensions (derived from, but does not close public issue
354).
* Clean up registry schema length attribute descriptions to be consistent
and correct (public issue 555).
Internal Issues:
* Replace as much of the hand-written extension appendix metadata as
possible with asciidoc includes generated from corresponding attributes
of +vk.xml+, and enhance the style guide to match. This avoids
inconsistencies between +vk.xml+ and the appendices, and produces a more
uniform style (internal issue 137).
* Remove the generated extDependency.{py,sh} files from the tree and
create them dynamically on demand instead, reducing merge conflicts
(internal issue 713).
* Add a prototype tool for generating in-place difference markup for
sections guarded by asciidoc conditionals, and new syntax for open
blocks to support it (internal issue 833).
* Remove unnecessary restriction of etext:*SYNC_FD_BIT_KHR external handle
types to the same physical device in the
slink:VkPhysicalDeviceIDPropertiesKHR,
flink:VkImportMemoryWin32HandleInfoKHR,
slink:VkImportFenceWin32HandleInfoKHR, slink:VkImportFenceFdInfoKHR,
slink:VkImportSemaphoreWin32HandleInfoKHR,
slink:VkImportSemaphoreFdInfoKHR
<<external-memory-handle-types-compatibility, External memory handle
types compatibility>>, <<external-semaphore-handle-types-compatibility,
External semaphore handle types compatibility>>, and
<<external-fence-handle-types-compatibility, External fence handle types
compatibility>> sections (internal issue 956).
Other Issues:
* Remove dependency of +VK_KHX_device_group+ on +VK_KHR_swapchain+ (there
is an interaction, but not a strict dependency), and add a new
`extension` attribute to the `<require` XML tag to allow classifying a
subset of interfaces of an extension as requiring another extension.
Update the registry schema and documentation accordingly.
New Extensions:
* `VK_AMD_shader_fragment_mask` (and related `GL_AMD_shader_fragment_mask`
GLSL extension)
* `VK_EXT_sample_locations`
* `VK_EXT_validation_cache`
* Bump API patch number and header version number to 56 for this update.
Github Issues:
* Add valid usage statements for commands introduced by
`VK_EXT_debug_report` and `VK_EXT_debug_marker` extensions, regarding
the valid pname:object and pname:objectType values (public issue 495).
* Modify `GL_KHR_vulkan_glsl` specification to document that uniform and
buffer block arrays each take only a single binding (public issue 514).
* Add `KHX` author tag to +vk.xml+ (public issue 526).
Internal Issues:
* Document use of code: macro for non-Vulkan APIs in the style guide
(internal issue 863).
* Document that reference page open block delimiters must not contain
asciidoc section markup in the style guide (internal issue 898).
* Fix <<spirvenv,SPIR-V appendix>> to say
code:VariablePointersStorageBuffer instead of
code:VariablePointersUniformBufferBlock (internal issue 928).
Other Commits:
* Add missing extension structures to dependency attributes in +vk.xml+.
New Extenions:
* `VK_EXT_depth_range_unrestricted`
* Bump API patch number and header version number to 53 for this update.
Github Issues:
Internal Issues:
* Clarify mappings of coordinates for mutable, compatible image views in
slink:VkImageViewCreateInfo (internal issue 815).
* Make ename:VK_BIND_SFR_BIT require a logical device with multiple
physical devices, so that standard sparse image block dimensions are
only required on systems that support multi-GPU (internal issue 835).
* Convert all files from use of // refBegin .. // refEnd comments to
delimit ref pages, to use of open blocks, and update style guide
accordingly (internal issue 839).
* Add valid usage for slink:VkWriteDescriptorSet when performing updates
to a ename:VK_STORAGE_IMAGE descriptor with layout
ename:VK_IMAGE_LAYOUT_GENERAL.
* Add a hack to the validity generator script to support an odd
interaction between flink:vkCmdFillBuffer and an extension (internal
issue 853).
* Remove redundant text describing slink:VkBufferCreateInfo::pname:usage,
which was already covered by implicit valid usage (internal issue 854).
* Update implicit validity generator script to properly handle the
pname:sType and pname:pNext members of "returnedonly" structures
(internal issue 874).
* Note that slink:VkApplicationInfo::pname:pApplicationName &
slink:VkApplicationInfo::pname:pEngineName are optional, and add missing
implicit valid usage statements for flink:vkDestroyInstance.
* Added missing valid usage for flink:vkCmdWriteTimestamp to require a
timestamp query pool.
* Simplify and/or split "`non-atomic`" valid usage statements.
New Extensions:
* `VK_AMD_gpu_shader_int16`
* `VK_EXT_blend_operation_advanced`
* `VK_EXT_sampler_filter_minmax`
* `VK_NV_framebuffer_mixed_samples`
-----------------------------------------------------
Note: the 1.0.52 spec wasn't published on github, so the 1.0.53 release
combines both change sets.
-----------------------------------------------------
Change log for June 13, 2017 Vulkan 1.0.52 spec update:
* Bump API patch number and header version number to 52 for this update.
Github Issues:
Internal Issues:
* Clarify behavior when non-coherent memory has
<<memory-device-unmap-does-not-flush, not been flushed before being
unmapped>> (internal issue 819).
* Fix description of code:WorkgroupSize builtin to note it decorates an
object, not a variable (internal issue 836).
* Fix asciidoc attributes so that trailing '{plus}' symbols in [eq] style
equations are rendered properly (internal issue 845).
* Add language to the "`Extension Handles, Objects, Enums, and Typedefs`"
section of the Procedures and Conventions document stating that any new
handle type requires a corresponding entry in the elink:VkObjectType
enumerated type (internal issue 856).
* Update style guide to use slink macro for Vulkan handle type names, and
define narrow conditions under which to use the *name and *text macros
instead of *link (internal issue 886).
* Add a dependency of the <<VK_KHX_device_group,VK_KHX_device_group>>
extension on VK_KHX_device_group_creation to +vk.xml+ and the extension
appendix.
* Change the copyright on Vulkan specification asciidoc *source* files to
CC-BY 4.0, and update the proprietary Khronos copyright applied to the
generated *output* formats (internal issue 327). This enables broader
re-use and modification of the Vulkan specification sources, while not
affecting the Reciprocal IP License between Vulkan Adopters and Working
Group Members.
New Extensions:
* `VK_NV_fill_rectangle`
* `VK_NV_fragment_coverage_to_color`
* Bump API patch number and header version number to 51 for this update.
Github Issues:
* Add Valid Usage statement to flink:vkCmdResolveImage to require that
source and destination image formats match (public issue 492).
* Specify that a code:char* parameter must: be a valid null-terminated
string in the <<fundamentals-implicit-validity, implicit valid usage>>
section (public issue 494).
* Removed unnecessary VU for slink:VkPhysicalDeviceFeatures which is
covered by ename:VK_ERROR_FEATURE_NOT_PRESENT already (public issue
496).
* Clarify valid usage of pname:pQueueFamilyIndices in
slink:VkBufferCreateInfo, slink:VkImageCreateInfo, and
slink:VkSwapchainCreateInfoKHR (public issue 501).
* Document that dependencies of enabled extensions must also be enabled in
the <<extended-functionality-extensions-dependencies, Extension
Dependencies>> section (public issue 507).
Internal Issues:
* Change slink:VkMappedMemoryRange valid usage to allow pname:offset +
pname:size == size of the allocation. Also, if ename:VK_WHOLE_SIZE is
used, require the end of the mapping to be aligned to a multiple of
pname:nonCoherentAtomSize (internal issue 611).
* Add issue to `VK_KHR_win32_surface` about reusing window objects from a
different graphics API or Vulkan ICD (internal issue 639).
* Require locations on user in/out in `GL_KHR_vulkan_glsl` (internal issue
783).
* Added version info to the json validation output, and updated the schema
to match (internal issue 838).
* Restructure enumerated type descriptions separately from the command or
structure they are used in, allowing better reference page generation
(internal issue 841).
* Re-sort extension appendices to be in alphabetical order within each
author ID section.
* Fix enum naming and clarify behavior for
`VK_NVX_device_generated_commands` extension.
New Extensions:
* Bump API patch number and header version number to 50 for this update.
Github Issues:
* Fix numerous minor issues with the VK_EXT_debug_report extension (public
issues 478, 483, 486, 489, 490).
Internal Issues:
* Update flink:vkAllocateDescriptorSets to specify conditions under which
to return ename:VK_ERROR_FRAGMENTED_POOL or
ename:VK_ERROR_OUT_OF_POOL_MEMORY instead of
out-of-host/out-of-device-memory, and improve the
<<fundamentals-errorcodes, description of those errors (internal issue
654).
* Add a NOTE documenting that flink:vkAcquireNextImageKHR can only signal
a single semaphore, and how to deal with that when multiple physical
devices in a logical device need to wait on it (internal issue 730).
* Improve description of pname:pNext chains of
slink:VkPhysicalDeviceImageFormatInfo2KHR and
slink:VkImageFormatProperties2KHR (internal issue 814).
* Clean up math markup issues in the <<textures, Image Operations>>
chapter (internal issue 818).
* Update validusage target to use more robust code for preprocessing, by
making direct use of Asciidoctor's preprocessor. Added uniqueItems check
to JSON vu schema and add clean_validusage target (internal issue 826).
* Update style guide to prohibit writing non-self-contained (on a single
bullet point) Valid Usage statements, and modify offending Valid Usage
statements in the Specification to match, to assist with automatic
extraction for the validation layers (internal issue 828).
* Add ename:VK_VALIDATION_CHECK_SHADERS_EXT to elink:VkValidationCheckEXT
of the `VK_EXT_validation_flags` extension, to selectively disable
shader validation.
* Remove duplicate valid usage statement for slink:VkImageMemoryBarrier.
* Modify reflow.py script to place VUID tag anchors standalone on a line
following their corresponding bullet point, and reflow the spec text
accordingly (this had been pending since the initial tag deployment).
New Extensions:
* `VK_AMD_texture_gather_bias_lod`
* Bump API patch number and header version number to 49 for this update.
Github Issues:
* Modify reference page extraction script to make internal links to spec
anchors refer to the core specification instead of being dangling links
(public issue 455).
* Fix GL_KHR_vulkan_glsl typo and add a nor-normative mapping to the newly
published StorageBuffer class (public issue 466).
* Both flink:vkEnumerateInstanceExtensionProperties and
flink:vkEnumerateDeviceExtensionProperties return
ename:VK_ERROR_LAYER_NOT_PRESENT, which covers the error case of an
application providing a layer name that wasn't returned by
ftext:vkEnumerate{Instance|Device}LayerProperties (public issue 487).
* The specification for flink:VkApplicationInfo::apiVersion says that the
driver must return ename:VK_ERROR_INCOMPATIBLE_DRIVER in the case that
pname:apiVersion specifies a non-supported version. That means that the
valid usage should not also state that, and so the VU statement is
removed. The VU had language about "`an effective substitute`" that
would have been lost, and so it was moved to the pname:apiVersion
description (public issue 488).
Internal Issues:
* Modify implicit validity generator script to assign asciidoc anchors to
all valid usage statements it generates, and reflow.py script to insert
Valid Usage ID (VUID) tags into the specification source files for
explicit valid usage statements. This has no semantic effects on the
specification, but will support the validation layer's detection of
valid usage violations and allow it to link into the corresponding part
of the specification (internal issue 583).
* Assign VUID tags to all explicit VU statements and document
the process and tag format in the style guide (internal issue 583).
* Clarify the rules of whether to structure new functionality as instance
extensions, device extensions, or both in the
<<extended-functionality-instance-extensions-and-devices, Instance
Extensions and Device Extensions>> section (internal issue 749).
* Require that SPIR-V run-time arrays are only used with the
code:BufferBlock decoration (internal issue 750).
* Fix implicit and explicit valid usage statements for
slink:VkWriteDescriptorSet::pname:dstSet (internal issue 767)
* Fix SPIR-V code sample for ename:VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER
in the <<descriptorsets-uniformtexelbuffer, Uniform Texel Buffer>>
section (internal issue 770).
* Clarify that disabling depth testing also disables depth writes in the
<<fragops-ds-state, Depth and Stencil Operations>> section (internal
issue 775).
* flink:VkDescriptorImageInfo::pname:imageLayout must match the actual
imageLayout at the time the image is accessed. This was in the spec
text, but needed an associated valid usage statement.
* Note that only 32-bit atomic operations are supported in the
<<spirvenv-module-validation, Validation Rules within a Module>>
section.
* Note that code:UniformConstant variables must not have initializers in
the <<spirvenv-module-validation, Validation Rules within a Module>>
section.
* Add a new elink:VkObjectType enumeration to the core API, promoted from
elink:VkDebugObjectTypeEXT, since it is used for much more than just the
debug_report extension.
New Extensions:
* `VK_KHR_get_surface_capabilities2`
* `VK_KHR_shared_presentable_image`
* Bump API patch number and header version number to 28 for this update.
Github Issues:
* Minor spelling and typography cleanup, add definitions of
ename:VK_FALSE and ename:VK_TRUE as just what their names say
(public issues 220, 318, 325, 365; internal issues 451, 496)
* Clarify that the pname:maxDescriptorSet limits in the
<<features-limits-required,Required Limits>> table are n *
maxPerStage limit (where n=number of supported stages) (public issue
254).
* Minor cleanup to <<boilerplate-platform-macros,Platform-Specific
Macro Definitions>> appendix (public issue 314).
* Add valid usage statement to slink:VkPipelineLayoutCreateInfo
disallowing multiple push constant ranges for the same shader stage
(public issue 340).
* Clarify the elink:VkSharingMode description of what executing the
"same" barriers means in case of ownership transfer (public issue
347).
* Rename copyright.txt and add COPYING.md to try and reduce confusion
about applicable copyrights (public issue 350).
* Extend the table in the <<boilerplate-wsi-header, Window System-Specific
Header Control>> section to describe the external headers included when
each etext:VK_USE_PLATFORM_* macro is defined (public issue 376).
Internal Issues:
* Add "Revision History" to the PDF outputs following the table of
contents, to match HTML outputs (internal issue 43).
* Clarified that flink:vkMapMemory may fail due to virtual address
space limitations (internal issue 346).
* Add +refBody+ comment markup for ref page autoextraction when required
(internal issue 400).
* Document proper use of "mipmap" and "mip" in the style guide API
naming rules, matching the spelling rules (internal issue 471).
* Tweak the <<extensions,Layers and Extensions>> appendix to note that
the Specification may be built with arbitrary combinations of
extensions (internal issue 483).
* Remove incorrect statement allowing
slink:VkClearAttachment::pname:colorAttachment to be >=
slink:VkSubpassDescription::pname:colorAttachmentCount (internal
issue 488).
* The <<features-limits-viewportboundsrange,viewportBoundsRange>> is
expressed in terms of the pname:maxViewportDimensions but this is
actually two values. Clarify that it's based on the larger of the two
(if they differ) (internal issue 499).
Other Issues:
* Reflowed text of the entire spec using the 'reflow' Makefile gater,
to (hopefully) reduce future internal git churn as edits are made
and extensions added in return for one-time pain. This has no
perceptible change on the spec outputs but considerable changes on
the asciidoc source (internal issue 367).
Jon Leech