WebGL EXT_color_buffer_half_float Extension Specification

Name

EXT_color_buffer_half_float

Contact

WebGL working group (public_webgl 'at' khronos.org)

Contributors

Mark Callow, HI Corporation

Members of the WebGL working group

Version

Last modified date: December 02, 2022
Revision: 9

Number

WebGL extension #14

Dependencies

Written against the WebGL API 1.0 specification.

In WebGL 1.0 contexts, implementations must also support the OES_texture_half_float extension.

Overview

This extension exposes the EXT_color_buffer_half_float functionality to WebGL.

The following WebGL-specific behavioral changes apply:

In WebGL 1.0 contexts, references to the R16F and RG16F types are ignored.
WebGL implementations supporting this extension are required to support rendering to the RGBA16F format.

Consult the above extension for documentation, issues and new functions and enumerants.

When this extension is enabled:

In WebGL 1.0 contexts, the 16-bit floating-point types RGB16F and RGBA16F become available as color-renderable formats. Renderbuffers can be created in these formats. These and textures created with type = HALF_FLOAT_OES, which will have one of these internal formats, can be attached to framebuffer object color attachments for rendering. Implementations supporting this extension are required to support rendering to the RGBA16F format. Applications must check framebuffer completeness to determine if RGB16F is supported.
In WebGL 2.0 contexts, the 16-bit floating-point types RGBA16F, RG16F, and R16F become available as color-renderable formats. Renderbuffers can be created in these formats. These and textures created with type = HALF_FLOAT, which will have one of these internal formats, can be attached to framebuffer object color attachments for rendering. Implementations supporting this extension are required to support rendering to RGBA16F format. The RGB16F format is not color-renderable.
NOTE: fragment shaders' outputs gl_FragColor and gl_FragData[0] will only be clamped and converted when the color buffer is fixed-point, and clearColor() will no longer clamp its parameter values on input. Clamping will be applied as necessary at draw time according to the type of color buffer in use.
If the implementation supports unclamped blend color, the blendColor() will no longer clamp its parameter values on input. Clamping will be applied as necessary at draw time according to the type of color buffer in use. Applications may check the effective blend color with getParameter.
The format and type combination RGBA and FLOAT becomes valid for reading from a floating-point rendering buffer. Note: RGBA and UNSIGNED_BYTE cannot be used for reading from a floating-point rendering buffer.
The component types of framebuffer object attachments can be queried.

IDL

[Exposed=(Window,Worker), LegacyNoInterfaceObject]
interface EXT_color_buffer_half_float {
  const GLenum RGBA16F_EXT = 0x881A;
  const GLenum RGB16F_EXT = 0x881B;
  const GLenum FRAMEBUFFER_ATTACHMENT_COMPONENT_TYPE_EXT = 0x8211;
  const GLenum UNSIGNED_NORMALIZED_EXT = 0x8C17;
}; // interface EXT_color_buffer_half_float

New Tokens

undefined renderbufferStorage(GLenum target, GLenum internalformat, GLsizei width, GLsizei height)

In WebGL 1.0 contexts, RGB16F_EXT and RGBA16F_EXT are accepted as the internalformat parameter.

In WebGL 2.0 contexts, RGBA16F, RG16F, and R16F are accepted as the internalformat parameter.

any getFramebufferAttachmentParameter(GLenum target, GLenum attachment, GLenum pname)

In WebGL 1.0 contexts, FRAMEBUFFER_ATTACHMENT_COMPONENT_TYPE_EXT is accepted as the pname parameter.

An INVALID_OPERATION error is generated if attachment is DEPTH_STENCIL_ATTACHMENT and pname is FRAMEBUFFER_ATTACHMENT_COMPONENT_TYPE_EXT.

Additions to the WebGL Specification

In section 5.13.12 Reading back pixels, change the allowed format and types table to:

frame buffer type	`format`	`type`
normalized fixed-point	RGBA	UNSIGNED_BYTE
floating-point	RGBA	FLOAT

Change the paragraph beginning "If pixels is null ..." to

If frame buffer type is not that indicated in the table for the format and type combination, an INVALID_OPERATON error is generated. If pixels is null ...

Revision History

Revision 1, 2012/11/08

Initial revision.

Revision 2, 2012/11/13

"Add reading-pixels-as-FLOAT feature to the Overview and related changes to WebGL section 5.13.12.

Revision 3, 2012/11/26

Move to draft.

Revision 4, 2014/07/15

Removed webgl module. Added NoInterfaceObject extended attribute.

Revision 5, 2014/11/24

Move to community approved.

Revision 6, 2016/05/05

Subsumed in WebGL 2.0 by EXT_color_buffer_float.

Revision 7, 2017/09/14

Require RGBA16F to be color-renderable and RGB16F to be optionally color-renderable.

Revision 8, 2020/08/31

Expose to WebGL 2.0 contexts to support devices which can render only to 16-bit floating point targets, and therefore can not support EXT_color_buffer_float.

Revision 9, 2022/12/02

Clarified framebuffer object attachment component type queries.