Change log for February 20, 2026 Vulkan 1.4.344 spec update:

Github Issues

  * Clarify the NOTE describing how the <<limits-viewportboundsrange,
    viewportBoundsRange>> limit is defined (public issue 2634).
  * Remove collision between type name and structure member / parameter
    names in VK_SEC_ubm_surface, disallow this in style guide as it affects
    some downstream language bindings, and add a check to xml_consistency.py
    (public issue 2679).
  * Remove extra double-quote at the end of a comment (public PR 2680).

Internal Issues

  * Relax vkGetCalibratedTimestampsKHR VU 09246 to allow
    VK_TIME_DOMAIN_PRESENT_STAGE_LOCAL_EXT time domain calibrations to not
    be unique, and the VkSwapchainCalibratedTimestampInfoEXT::swapchain
    external synchronization requirement, allowing calibration of multiple
    present stages and determining timestamps for a swapchain-local
    time-domain in the same thread as vkGetPastPresentationTimingEXT
    (internal issue 4596).
  * Fix VkVideoSessionParametersKHR `parent` attribute to correct
    parent-child relationship in implicit VUs (internal issue 4615).
  * Add a NOTE to <<decode-av1-picture-info, AV1 Std Picture Information>>
    regarding a known decode app bug and driver workaround (internal issue
    4639).
  * Generate new "`Structure Chaining`" documentation linking structures
    back to the base structures the extend (internal issue 4670).
  * Move VU 04745 from standalone SPIR-V to runtime SPIR-V section (internal
    issue 4682).
  * Reorder XML entries to not need dependency tracking for enabling modular
    headers (internal MR 7994).
  * Remove unnecessary parenthesis in complex XML dependencies (internal MR
    7995).
  * Add boolean expression support to XML `protect` attribute (internal MR
    7996).
  * Update legacy header comments and registry schema documentation to match
    specification language for "`ignored`" legacy entries (internal MR
    7998).
  * Unify description of VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT in
    VkAccessFlagBits and the corresponding bit in VkAccessFlagBits2
    (internal MR 8062).
  * Fix Vulkan Object Flags/Handles (internal MR 8067).

New Extensions

  * VK_VALVE_shader_mixed_float_dot_product

Author: oddhack

ChangeLog.adoc

@@ -14,6 +14,56 @@ appears frequently in the change log.
 
 '''
 
+Change log for February 20, 2026 Vulkan 1.4.344 spec update:
+
+Github Issues
+
+  * Clarify the NOTE describing how the <<limits-viewportboundsrange,
+    viewportBoundsRange>> limit is defined (public issue 2634).
+  * Remove collision between type name and structure member / parameter
+    names in VK_SEC_ubm_surface, disallow this in style guide as it affects
+    some downstream language bindings, and add a check to xml_consistency.py
+    (public issue 2679).
+  * Remove extra double-quote at the end of a comment (public PR 2680).
+
+Internal Issues
+
+  * Relax vkGetCalibratedTimestampsKHR VU 09246 to allow
+    VK_TIME_DOMAIN_PRESENT_STAGE_LOCAL_EXT time domain calibrations to not
+    be unique, and the VkSwapchainCalibratedTimestampInfoEXT::swapchain
+    external synchronization requirement, allowing calibration of multiple
+    present stages and determining timestamps for a swapchain-local
+    time-domain in the same thread as vkGetPastPresentationTimingEXT
+    (internal issue 4596).
+  * Fix VkVideoSessionParametersKHR `parent` attribute to correct
+    parent-child relationship in implicit VUs (internal issue 4615).
+  * Add a NOTE to <<decode-av1-picture-info, AV1 Std Picture Information>>
+    regarding a known decode app bug and driver workaround (internal issue
+    4639).
+  * Generate new "`Structure Chaining`" documentation linking structures
+    back to the base structures the extend (internal issue 4670).
+  * Move VU 04745 from standalone SPIR-V to runtime SPIR-V section (internal
+    issue 4682).
+  * Reorder XML entries to not need dependency tracking for enabling modular
+    headers (internal MR 7994).
+  * Remove unnecessary parenthesis in complex XML dependencies (internal MR
+    7995).
+  * Add boolean expression support to XML `protect` attribute (internal MR
+    7996).
+  * Update legacy header comments and registry schema documentation to match
+    specification language for "`ignored`" legacy entries (internal MR
+    7998).
+  * Unify description of VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT in
+    VkAccessFlagBits and the corresponding bit in VkAccessFlagBits2
+    (internal MR 8062).
+  * Fix Vulkan Object Flags/Handles (internal MR 8067).
+
+New Extensions
+
+  * VK_VALVE_shader_mixed_float_dot_product
+
+'''
+
 Change log for February 6, 2026 Vulkan 1.4.343 spec update:
 
 Github Issues

Makefile

@@ -151,7 +151,7 @@ VERBOSE =
 # ADOCOPTS options for asciidoc->HTML5 output
 
 NOTEOPTS     = -a editing-notes -a implementation-guide
-PATCHVERSION = 343
+PATCHVERSION = 344
 BASEOPTS     =
 
 ifneq (,$(findstring VKSC_VERSION_1_0,$(VERSIONS)))

appendices/VK_VALVE_shader_mixed_float_dot_product.adoc

@@ -0,0 +1,36 @@
+// Copyright 2020-2026 The Khronos Group Inc.
+//
+// SPDX-License-Identifier: CC-BY-4.0
+
+include::{generated}/meta/{refprefix}VK_VALVE_shader_mixed_float_dot_product.adoc[]
+
+=== Other Extension Metadata
+
+*Last Modified Date*::
+    2026-02-04
+*IP Status*::
+    No known IP claims.
+*Contributors*::
+  - Georg Lehmann, Valve
+  - Mike Blumenkrantz, Valve
+
+=== Description
+
+This extension enables support for mixed precision dot product accumulate
+operations in shaders as defined in `SPV_VALVE_mixed_float_dot_product`.
+
+
+include::{generated}/interfaces/VK_VALVE_shader_mixed_float_dot_product.adoc[]
+
+=== New SPIR-V Capabilities
+
+  * <<spirvenv-capabilities-table-DotProductFloat16AccFloat32VALVE,DotProductFloat16AccFloat32VALVE>>
+  * <<spirvenv-capabilities-table-DotProductFloat16AccFloat16VALVE,DotProductFloat16AccFloat16VALVE>>
+  * <<spirvenv-capabilities-table-DotProductBFloat16AccVALVE,DotProductBFloat16AccVALVE>>
+  * <<spirvenv-capabilities-table-DotProductFloat8AccFloat32VALVE,DotProductFloat8AccFloat32VALVE>>
+
+
+=== Version History
+
+  * Revision 1, 2026-02-04 (Georg Lehmann)
+  ** Initial specification

appendices/spirvenv.adoc

@@ -730,10 +730,6 @@ or knowledge of runtime information, such as enabled features.
   * [[VUID-StandaloneSpirv-OpTypeForwardPointer-04711]]
     code:OpTypeForwardPointer must: have a {StorageClass} of
     code:PhysicalStorageBuffer
-  * [[VUID-StandaloneSpirv-None-04745]]
-    All block members in a variable with a {StorageClass} of
-    code:PushConstant declared as an array must: only be accessed by
-    dynamically uniform indices
   * [[VUID-StandaloneSpirv-OpVariable-06673]]
     There must: be at most one variable in the code:PushConstant
     {StorageClass} listed in the code:Interface for each code:OpEntryPoint
@@ -1187,6 +1183,10 @@ ifdef::VK_COMPUTE_VERSION_1_1[]
     texel buffer, the storage texel buffer through which that memory is
     accessed must: be dynamically uniform within the invocation group
 endif::VK_COMPUTE_VERSION_1_1[]
+  * [[VUID-RuntimeSpirv-None-04745]]
+    All block members in a variable with a {StorageClass} of
+    code:PushConstant declared as an array must: only be accessed by
+    dynamically uniform indices
   * [[VUID-RuntimeSpirv-None-10148]]
     If an instruction accesses memory through any resource,
 ifdef::VK_COMPUTE_VERSION_1_1[]
@@ -3322,7 +3322,7 @@ ifdef::VK_EXT_shader_long_vector[]
     than four components (or an aggregate containing such a type) must: have
     {StorageClass} of code:Function, code:Private, code:Uniform,
     code:Workgroup, code:StorageBuffer, code:PhysicalStorageBuffer,
-    code:PushConstant, or code:ShaderRecordBufferKHR.
+    code:PushConstant, or code:ShaderRecordBufferKHR
 endif::VK_EXT_shader_long_vector[]
 ifdef::VK_EXT_descriptor_heap[]
   * [[VUID-RuntimeSpirv-samplerDescriptorAlignment-11348]]

chapters/VK_KHR_surface/wsi.adoc

@@ -1218,17 +1218,17 @@ include::{generated}/api/structs/VkUbmSurfaceCreateInfoSEC.adoc[]
   * pname:pNext is `NULL` or a pointer to a structure extending this
     structure.
   * pname:flags is reserved for future use.
-  * pname:ubm_device is a pointer to a code:ubm_device to associate the
-    surface with.
-  * pname:ubm_surface is a pointer to a code:ubm_surface to associate the
+  * pname:device is a pointer to a code:ubm_device to associate the surface
+    with.
+  * pname:surface is a pointer to a code:ubm_surface to associate the
     surface with.
 
 .Valid Usage
 ****
-  * [[VUID-VkUbmSurfaceCreateInfoSEC-ubm_device-12366]]
-    pname:ubm_device must: point to a valid UBM code:ubm_device
-  * [[VUID-VkUbmSurfaceCreateInfoSEC-ubm_surface-12367]]
-    pname:ubm_surface must: point to a valid UBM code:ubm_surface
+  * [[VUID-VkUbmSurfaceCreateInfoSEC-device-12366]]
+    pname:device must: point to a valid UBM code:ubm_device
+  * [[VUID-VkUbmSurfaceCreateInfoSEC-surface-12367]]
+    pname:surface must: point to a valid UBM code:ubm_surface
 ****
 
 include::{generated}/validity/structs/VkUbmSurfaceCreateInfoSEC.adoc[]
@@ -2836,15 +2836,15 @@ include::{generated}/api/protos/vkGetPhysicalDeviceUbmPresentationSupportSEC.ado
 
   * pname:physicalDevice is the physical device.
   * pname:queueFamilyIndex is the queue family index.
-  * pname:ubm_device is a pointer to the code:ubm_device associated with a
-    UBM compositor.
+  * pname:device is a pointer to the code:ubm_device associated with a UBM
+    compositor.
 
 This platform-specific function can: be called prior to creating a surface.
 
 .Valid Usage
 ****
-  * [[VUID-vkGetPhysicalDeviceUbmPresentationSupportSEC-ubm_device-12368]]
-    pname:ubm_device must: point to a valid UBM code:ubm_device
+  * [[VUID-vkGetPhysicalDeviceUbmPresentationSupportSEC-device-12368]]
+    pname:device must: point to a valid UBM code:ubm_device
   * [[VUID-vkGetPhysicalDeviceUbmPresentationSupportSEC-queueFamilyIndex-12369]]
     pname:queueFamilyIndex must: be less than
     pname:pQueueFamilyPropertyCount returned by

chapters/features.adoc

@@ -9475,6 +9475,41 @@ include::{generated}/validity/structs/VkPhysicalDeviceDescriptorHeapFeaturesEXT.
 --
 endif::VK_EXT_descriptor_heap[]
 
+ifdef::VK_VALVE_shader_mixed_float_dot_product[]
+[open,refpage='VkPhysicalDeviceShaderMixedFloatDotProductFeaturesVALVE',desc='Structure describing support for mixed float dot products',type='structs']
+--
+The sname:VkPhysicalDeviceShaderMixedFloatDotProductFeaturesVALVE structure
+is defined as:
+
+include::{generated}/api/structs/VkPhysicalDeviceShaderMixedFloatDotProductFeaturesVALVE.adoc[]
+
+  * pname:sType is a elink:VkStructureType value identifying this structure.
+  * pname:pNext is `NULL` or a pointer to a structure extending this
+    structure.
+  * [[features-shaderMixedFloatDotProductFloat16AccFloat32]]
+    pname:shaderMixedFloatDotProductFloat16AccFloat32 indicates whether the
+    implementation supports shaders with the
+    code:DotProductFloat16AccFloat32VALVE capability.
+  * [[features-shaderMixedFloatDotProductFloat16AccFloat16]]
+    pname:shaderMixedFloatDotProductFloat16AccFloat16 indicates whether the
+    implementation supports shaders with the
+    code:DotProductFloat16AccFloat16VALVE capability.
+  * [[features-shaderMixedFloatDotProductBFloat16Acc]]
+    pname:shaderMixedFloatDotProductBFloat16Acc indicates whether the
+    implementation supports shaders with the code:DotProductBFloat16AccVALVE
+    capability.
+  * [[features-shaderMixedFloatDotProductFloat8AccFloat32]]
+    pname:shaderMixedFloatDotProductFloat8AccFloat32 indicates whether the
+    implementation supports shaders with the
+    code:DotProductFloat8AccFloat32VALVE capability.
+
+:refpage: VkPhysicalDeviceShaderMixedFloatDotProductFeaturesVALVE
+include::{chapters}/features.adoc[tag=features]
+
+include::{generated}/validity/structs/VkPhysicalDeviceShaderMixedFloatDotProductFeaturesVALVE.adoc[]
+--
+endif::VK_VALVE_shader_mixed_float_dot_product[]
+
 
 [[features-requirements]]
 == Feature Requirements

chapters/limits.adoc

@@ -591,15 +591,14 @@ ifdef::VK_COMPUTE_VERSION_1_3,VK_KHR_maintenance4[or code:LocalSizeId]
 +
 [NOTE]
 ====
-The intent of the pname:viewportBoundsRange limit is to allow a maximum
-sized viewport to be arbitrarily shifted relative to the output target as
-long as at least some portion intersects.
-This would give a bounds limit of [eq]#[-pname:size {plus} 1, 2 {times}
-pname:size - 1]# which would allow all possible non-empty-set intersections
-of the output target and the viewport.
-Since these numbers are typically powers of two, picking the signed number
-range using the smallest possible number of bits ends up with the specified
-range.
+The wide range of values required for pname:viewportBoundsRange allows the
+viewport to be arbitrarily shifted relative to the output render target
+while still partially overlapping.
+However, the minimum range required to achieve this would actually be
+[eq]#[-pname:size {plus} 1, 2 {times} pname:size - 1]#.
+As these limits in implementations are typically simple power-of-two values,
+the specification reflects this convention, rounding the lower bound
+accordingly.
 ====
   * [[limits-viewportSubPixelBits]] pname:viewportSubPixelBits is the number
     of bits of subpixel precision for viewport bounds.

chapters/synchronization.adoc

@@ -1083,6 +1083,10 @@ endif::VK_EXT_shader_tile_image[]
     via certain render pass <<renderpass-load-operations, load>>,
     <<renderpass-store-operations, store>>, and
     <<renderpass-resolve-operations, multisample resolve>> operations.
+ifdef::VK_GRAPHICS_VERSION_1_2,VK_KHR_depth_stencil_resolve[]
+    This includes <<renderpass-resolve-operations, multisample resolve>>
+    operations for depth/stencil resolve attachments.
+endif::VK_GRAPHICS_VERSION_1_2,VK_KHR_depth_stencil_resolve[]
     Such access occurs in the
     ename:VK_PIPELINE_STAGE_2_COLOR_ATTACHMENT_OUTPUT_BIT pipeline stage.
   * ename:VK_ACCESS_2_DEPTH_STENCIL_ATTACHMENT_READ_BIT specifies read
@@ -1522,15 +1526,14 @@ ifdef::VK_EXT_shader_tile_image[]
     in the ename:VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT pipeline stage.
 endif::VK_EXT_shader_tile_image[]
   * ename:VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT specifies write access to a
-ifndef::VK_GRAPHICS_VERSION_1_2,VK_KHR_depth_stencil_resolve[]
-    <<renderpass, color or resolve attachment>>
-endif::VK_GRAPHICS_VERSION_1_2,VK_KHR_depth_stencil_resolve[]
+    <<renderpass, color attachment>> during a <<renderpass, render pass>> or
+    via certain render pass <<renderpass-load-operations, load>>,
+    <<renderpass-store-operations, store>>, and
+    <<renderpass-resolve-operations, multisample resolve>> operations.
 ifdef::VK_GRAPHICS_VERSION_1_2,VK_KHR_depth_stencil_resolve[]
-    <<renderpass, color, resolve, or depth/stencil resolve attachment>>
+    This includes <<renderpass-resolve-operations, multisample resolve>>
+    operations for depth/stencil resolve attachments.
 endif::VK_GRAPHICS_VERSION_1_2,VK_KHR_depth_stencil_resolve[]
-    during a <<renderpass, render pass>> or via certain render pass
-    <<renderpass-load-operations, load>> and <<renderpass-store-operations,
-    store>> operations.
     Such access occurs in the
     ename:VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT pipeline stage.
   * ename:VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT specifies read access
@@ -8322,6 +8325,10 @@ timestamps on a regular basis.
   * [[VUID-vkGetCalibratedTimestampsKHR-timeDomain-09246]]
     The pname:timeDomain value of each slink:VkCalibratedTimestampInfoKHR in
     pname:pTimestampInfos must: be unique
+ifdef::VK_EXT_present_timing[]
+    except for those with a value of
+    ename:VK_TIME_DOMAIN_PRESENT_STAGE_LOCAL_EXT
+endif::VK_EXT_present_timing[]
 ****
 
 include::{chapters}/commonvalidity/no_dynamic_allocations_common.adoc[]

chapters/video/av1_decode.adoc

@@ -387,6 +387,19 @@ pname:pStdPictureInfo are interpreted as follows:
   ** code:pMiRowStarts is a pointer to an array of code:TileRows number of
      unsigned integers that corresponds to code:MiRowStarts defined in
      section 6.8.14 of the <<aomedia-av1,AV1 Specification>>;
++
+[NOTE]
+====
+Historically some applications incorrectly specified the values in
+code:pMiColStarts and code:pMiRowStarts in terms of superblocks instead of
+mode info blocks, therefore it is recommended for implementations to ignore
+the values specified in code:pMiColStarts and code:pMiRowStarts.
+Instead, the values should be calculated based on
+code:flags.uniform_tile_spacing_flag, code:pWidthInSbsMinus1,
+code:pHeightInSbsMinus1, the <<decode-av1-sequence-header,sequence header>>
+parameter code:flags.use_128x128_superblock, and the coded extent of the
+decoded picture.
+====
   ** code:pWidthInSbsMinus1 is a pointer to an array of code:TileCols number
      of unsigned integers that corresponds to code:width_in_sbs_minus_1
      defined in section 6.8.14 of the <<aomedia-av1,AV1 Specification>>;