Name Strings

SPV_KHR_fma

Contact

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

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

Contributors

  • Graeme Leese, Broadcom

  • Alan Baker, Google

  • Georg Lehmann, Valve

  • Jeff Bolz, NVIDIA

  • Tobias Hector, AMD

Notice

Copyright (c) 2025 The Khronos Group Inc. Copyright terms at http://www.khronos.org/registry/speccopyright.html

Status

  • Approved by the SPIR-V Working Group: 2025-08-06

  • Approved by the Khronos Board of Promoters: 2025-09-19

Version

Last Modified Date

2026-06-29

Revision

2

Dependencies

This extension is written against the SPIR-V Specification, Version 1.6 Revision 1, Unified

This extension requires SPIR-V 1.0.

Overview

This extension provides a core operation for fused-multiply-add (fma) to replace the different variants that have existed in extended instruction sets. It is intended to give greater consistency among consuming APIs and give access to the higher-accuracy operation where this has not been supported in the past.

Extension Name

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

OpExtension "SPV_KHR_fma"

Modifications to the SPIR-V Specification, Version 1.6

FP Fast Math Mode

In section 3.15 "FP Fast Math Mode", following "Only valid on …​", add OpFmaKHR.

Capability

In section 3.31 "Capability" add the following row to the capability table:

Capability Implicitly Declares

6030

FMAKHR
Uses OpFmaKHR.

Arithmetic Instructions

In section 3.49.13 "Arithmetic Instructions" add the following instruction

OpFmaKHR

Floating-point fused multiply add, (Operand 1 * Operand 2) + Operand 3.

Result Type must be a scalar or vector of floating-point type.

The types of Operand 1, Operand 2, and Operand 3 must be the same as Result Type.

Results are computed per component.

Capability:
FMAKHR

6

4427

<id>
Result Type

Result <id>

<id>
Operand 1

<id>
Operand 2

<id>
Operand 3

Deprecation

This extension deprecates the following features:

  • The OpenCL.std instruction fma.

Issues

  1. Should unfused multiply-accumulate be included here?

    RESOLVED: SPIR-V’s fast-math flags already allow representing multiply-add which the compiler is allowed to fuse (using the Contract flag), so adding operation variants for these cases seems unnecessary. The first version of this extension does not add these operations.

Revision History

Rev Date Author Changes

1

2024-05-09

Graeme Leese

Initial public release.

2

2026-06-29

Arseniy Obolenskiy

Clarify that OpFmaKHR results may be decorated with FPFastMathMode.