VK_EXT_shader_ocp_microscaling_types(3) Manual Page

673

1

Ratified

VK_EXT_shader_ocp_microscaling_types

This extension enables support for the floating-point data types defined in the Open Compute Project (OCP) Microscaling Formats specification.

If VK_KHR_cooperative_matrix is supported:

1) How does this extension add support for microscaling?

RESOLVED: It does not directly do so. This extension just introduces basic support for the data types used for microscaling but it does not add any support for microscaled tensors or operations on microscaled data. Reading miscroscaled data and converting it to other formats such as FP16 or FP32 can be implemented in shaders with the new types provided by this extension but these new types cannot for example be used directly in matrix multiplication operations.

2) What conversions to/from the newly-added types should be supported?

RESOLVED: This extension will only add conversions from the newly-added types to a restricted list of existing larger floating-point types. Conversions from existing types to the newly-added types would require complex logic and are not expected to be as useful. Conversions between integer types and the newly-added types are not expected to be generally useful either.

No cross-references are available

For more information, see the Vulkan Specification.

This page is a generated document. Fixes and changes should be made to the generator scripts, not directly.

Copyright 2014-2026 The Khronos Group Inc.

SPDX-License-Identifier: CC-BY-4.0