Vulkan-Docs/appendices/VK_NV_cooperative_matrix.txt
Jon Leech 521e98405f Change log for August 17, 2019 Vulkan 1.1.120 spec update:
* Update release number to 120.

Github Issues:

  * Add slink:VkAccelerationStructureTypeNV explicitly to extension XML for
    `<<VK_NV_ray_tracing>>` (public issue 848).
  * Add missing valid usage statements for feature flags in
    slink:VkCommandBufferInheritanceInfo (public pull request 1017).

Internal Issues:

  * Clarify behavior of non-premultiplied destination colors for
    `<<VK_EXT_blend_operation_advanced>>` prior to the definition of
    slink:VkBlendOverlapEXT (internal issue 1766).
  * Fix the confusing phrasing "`no other queue must: be (doing something)`"
    for flink:vkQueuePresentKHR, flink:vkQueueSubmit, and
    flink:vkQueueBindSparse (internal issue 1774).
  * Add `<<VK_EXT_validation_features>>` flag to enable best practices
    checks, which will soon be available in the validation layer (internal
    issue 1779).
  * Specify allowed characters for VUID tag name components in the style
    guide (internal issue 1788).
  * Update links to SPIR-V extension specifications, and parameterize their
    markup in case the URLs change in the future (internal issue 1797).
  * Fix an off-by-one error in the valid usage statement for
    slink:VkPipelineExecutableInfoKHR (internal merge request 3303).
  * Clean up markup indentation not matching the style guide (internal merge
    request 3314).
  * Minor script updates to allow refpage aliases, generate a dynamic TOC
    for refpages, generate Apache rewrite rules for aliases, open external
    links from refpages in a new window, and synchronize with the OpenCL
    scripts. This will shortly enable a paned navigation setup for refpages,
    similar to the OpenCL 2.2 refpages (internal merge request 3322).
  * Script updates to add tests to the checker, refactor and reformat code,
    generate better text for some valid usage statements, use more Pythonic
    idioms, and synchronize with the OpenXR scripts (internal merge request
    3239).
  * Script updates and minor fixes in spec language to not raise checker
    errors for refpage markup of pages not existing in the API, such as
    VKAPI_NO_STDINT_H. Remove corresponding suppression of some
    check_spec_links.py tests from .gitlab-ci.yml and 'allchecks' target
    (internal merge request 3315).
2019-08-17 15:33:21 -07:00

79 lines
2.3 KiB
Plaintext

include::meta/VK_NV_cooperative_matrix.txt[]
*Last Modified Data*::
2019-02-05
*Contributors*::
- Jeff Bolz, NVIDIA
- Markus Tavenrath, NVIDIA
- Daniel Koch, NVIDIA
This extension adds support for using cooperative matrix types in SPIR-V.
Cooperative matrix types are medium-sized matrices that are primarily
supported in compute shaders, where the storage for the matrix is spread
across all invocations in some scope (usually a subgroup) and those
invocations cooperate to efficiently perform matrix multiplies.
Cooperative matrix types are defined by the
{spirv}/NV/SPV_NV_cooperative_matrix.html[`SPV_NV_cooperative_matrix`]
SPIR-V extension and can be used with the
https://github.com/KhronosGroup/GLSL/blob/master/extensions/nv/GLSL_NV_cooperative_matrix.txt[`GL_NV_cooperative_matrix`]
GLSL extension.
This extension includes support for enumerating the matrix types and
dimensions that are supported by the implementation.
=== New Object Types
None.
=== New Enum Constants
* Extending elink:VkStructureType:
** ename:VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_COOPERATIVE_MATRIX_FEATURES_NV
** ename:VK_STRUCTURE_TYPE_COOPERATIVE_MATRIX_PROPERTIES_NV
** ename:VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_COOPERATIVE_MATRIX_PROPERTIES_NV
=== New Enums
* elink:VkScopeNV
* elink:VkComponentTypeNV
=== New Structures
* Extending slink:VkPhysicalDeviceFeatures2:
** slink:VkPhysicalDeviceCooperativeMatrixFeaturesNV
* Extending slink:VkPhysicalDeviceProperties2:
** slink:VkPhysicalDeviceCooperativeMatrixPropertiesNV
* slink:VkCooperativeMatrixPropertiesNV
=== New Functions
* flink:vkGetPhysicalDeviceCooperativeMatrixPropertiesNV
=== New SPIR-V Capabilities
* <<spirvenv-capabilities-table-cooperativeMatrix,CooperativeMatrixNV>>
=== Issues
(1) What matrix properties will be supported in practice?
RESOLVED: In NVIDIA's initial implementation, we will support:
* AType = BType = fp16 CType = DType = fp16 MxNxK = 16x8x16 scope =
Subgroup
* AType = BType = fp16 CType = DType = fp16 MxNxK = 16x8x8 scope =
Subgroup
* AType = BType = fp16 CType = DType = fp32 MxNxK = 16x8x16 scope =
Subgroup
* AType = BType = fp16 CType = DType = fp32 MxNxK = 16x8x8 scope =
Subgroup
=== Version History
* Revision 1, 2019-02-05 (Jeff Bolz)
- Internal revisions