Name Strings

SPV_INTEL_rounded_divide_sqrt

Contact

To report problems with this extension, please open a new issue at:

https://github.com/KhronosGroup/SPIRV-Registry

Contributors

  • Ben Ashbaugh, Intel

  • Greg Lueck, Intel

  • Piotr Kanclerz, Intel

Notice

Copyright (c) 2026 Intel Corporation. All rights reserved.

Status

  • Complete

Version

Last Modified Date

2026-04-20

Revision

1

Dependencies

This extension is written against the SPIR-V Specification, Version 1.6 Revision 7.

This extension requires SPIR-V 1.0.

Overview

This extension adds a new SPIR-V capability to allow floating-point divide and square root instructions to be decorated with an optional FPRoundingMode decoration. When an FPRoundingMode decoration is applied to a divide or square root instruction, it indicates that the result of the instruction will be correctly rounded using the specified rounding mode.

Extension Name

To use this extension within a SPIR-V module, the following OpExtension must be present in the module:

OpExtension "SPV_INTEL_rounded_divide_sqrt"

Modifications to the SPIR-V Specification, Version 1.6

Validation Rules

Modify Section 2.16.2, "Validation Rules for Shader Capabilities", adding to the conditions where the FPRoundingMode decoration is valid:

  • The FPRoundingMode decoration must be applied only to […​], or to instructions enabled by the RoundedDivideSqrtINTEL capability.

Capabilities

Modify Section 3.2.30, "Capability", adding these rows to the Capability table:

Capability Implicitly Declares

6265

RoundedDivideSqrtINTEL
Allows an FPRoundingMode decoration to be applied to the following instructions when they are operating on a floating-point type using the IEEE 754 encoding: OpFDiv, the sqrt instruction from the OpenCL.std extended instruction set, and the Sqrt instruction from the GLSL.std.450 extended instruction set.

When an FPRoundingMode decoration is applied to these instructions, it indicates that the result of the instruction will be correctly rounded using the specified rounding mode.

Issues

  1. How many capabilities do we need? Do we need different capabilities for different instructions? Do we need different capabilities for different floating-point types?

    RESOLVED: We will have a single capability for both divide and square root. This single capability applies to all floating-point types using IEEE 754 encodings, so it will work for 16-bit half, 32-bit float, and 64-bit double.

    This extension does not currently support rounding modes on floating-point types with non-IEEE 754 encodings, such as bfloat16.

  2. Should this extension support Shader SPIR-V?

    RESOLVED: There is nothing in this extension that would prohibit usage in Shader SPIR-V.

Revision History

Rev Date Author Changes

1

2026-04-20

Ben Ashbaugh

Initial revision for publication