1. Introduction

Many APIs operate on bulk data — buffers, images, volumes, etc. — each composed of many elements with a fixed and often simple representation. Frequently, multiple alternative representations of data are supported: vertices can be represented with different numbers of dimensions, textures may have different bit depths and channel orders, and so on. Sometimes the representation of the data is highly specific to the application, but there are many types of data that are common to multiple APIs — and these can reasonably be described in a portable manner. In this standard, the term data format describes the representation of data.

It is typical for each API to define its own enumeration of the data formats on which it can operate. This causes a problem when multiple APIs are in use: the representations are likely to be incompatible, even where the capabilities intersect. When additional format-specific capabilities are added to an API which was designed without them, the description of the data representation often becomes inconsistent and disjoint. Concepts that are unimportant to the core design of an API may be represented simplistically or inaccurately, which can be a problem as the API is enhanced or when data is shared.

Some APIs do not have a strict definition of how to interpret their data. For example, a rendering API may treat all color channels of a texture identically, leaving the interpretation of each channel to the user’s choice of convention. This may be true even if color channels are given names that are associated with actual colors — in some APIs, nothing stops the user from storing the blue quantity in the red channel and the red quantity in the blue channel. Without enforcing a single data interpretation on such APIs, it is nonetheless often useful to offer a clear definition of the color interpretation convention that is in force, both for code maintenance and for communication with external APIs which do have a defined interpretation. Should the user wish to use an unconventional interpretation of the data, an appropriate descriptor can be defined that is specific to this choice, in order to simplify automated interpretation of the chosen representation and to provide concise documentation.

Where multiple APIs are in use, relying on an API-specific representation as an intermediary can cause loss of important information. For example, a camera API may associate color space information with a captured image, and a printer API may be able to operate with that color space, but if the data is passed through an intermediate compute API for processing and that API has no concept of a color space, the useful information may be discarded.

The intent of this standard is to provide a common, consistent, machine-readable way to describe those data formats which are amenable to non-proprietary representation. This standard provides a portable means of storing the most common descriptive information associated with data formats, and an extension mechanism that can be used when this common functionality must be supplemented.

While this standard is intended to support the description of many kinds of data, the most common class of bulk data used in Khronos standards represents color information. For this reason, the range of standard color representations used in Khronos standards is diverse, and a significant portion of this specification is devoted to color formats.