src/dawn/common/ColorSpace.h - dawn.git - Git at Google

 // Copyright 2026 The Dawn & Tint Authors
 //
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are met:
 //
 // 1. Redistributions of source code must retain the above copyright notice, this
 //    list of conditions and the following disclaimer.
 //
 // 2. Redistributions in binary form must reproduce the above copyright notice,
 //    this list of conditions and the following disclaimer in the documentation
 //    and/or other materials provided with the distribution.
 //
 // 3. Neither the name of the copyright holder nor the names of its
 //    contributors may be used to endorse or promote products derived from
 //    this software without specific prior written permission.
 //
 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 // DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
 // FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 // DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
 // SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
 // CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 // OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

 #ifndef SRC_DAWN_COMMON_COLORSPACE_H_
 #define SRC_DAWN_COMMON_COLORSPACE_H_

 #include "src/dawn/common/Algebra.h"

 namespace dawn {

 // Colorspaces define how numbers used to encode a color are related to physical quantities of
 // light (for example captured by a camera, or emitted by a display). Dawn needs some handling of
 // colorspace because ExternalTexture allows importing data in a source colorspace and sampling it
 // in shaders with a different colorspace called the destination colorspace.
 //
 // Dawn also needs to tell the OS how to display values in a wgpu::Surface texture to the user, but
 // that's done by passing the correct enum to the backend API's swapchain/surface configuration so
 // Dawn doesn't need to do any math for that path.
 //
 // # RGB color spaces
 //
 // Colorspaces are defined with respect to a reference colorspace, usually the CIE XYZ one but the
 // exact reference doesn't matter much in Dawn as long as all conversion are defined from/to it (so
 // combining a transform from colorspace from A to XYZ and from XYZ to B gives a transform from A to
 // B). Numbers in an RGB colorspace (as opposed to YCbCr, see below) can be converted to XYZ by
 // performing two steps.
 //
 // The electro-optical transfer function (EOTF) that's used to transform colors in "perceptual"
 // space to colors in "physical" space. This is because the human eye has a logarithmic response
 // to light intensity (such as doubling the amount of light "feels" like a single step increase,
 // the same mechanism as decibels for audition). Storing values in perceptual space is done to have
 // as much perceptual precision for low and high color values and allows for smooth perceptual
 // color gradients. (Read up on sRGB, that's the most used and well known perceptual color space).
 // The inverse transfer function is called the opto-electronic transfer function (OETF).
 //
 // Note that a weirdness of HLG is that the EOTF is not the inverse of its OETF but of OETF(OOTF)
 // (for unfortunate and historical reasons). The Annex 1 "The relationship between the OETF, the
 // EOTF and the OOTF" of
 // https://www.itu.int/dms_pubrec/itu-r/rec/bt/R-REC-BT.2100-1-201706-S!!PDF-E.pdf
 //
 // The color primaries, what it means to say "red of 1" in that colorspace. This is where CIE XYZ is
 // used as the primaries for the colorspace are mapped to colors in the XYZ colorspace. XYZ itself
 // is a physical standard developed through experimentation on the human perception of color. The
 // mapping from a colorspace to XYZ is a linear mapping represented by a matrix whose column vectors
 // are the XYZ colors corresponding to the colorspace's primaries.
 //
 // Thus the transformation between colorspace A and B is:
 //
 //  - Apply A's EOTF.
 //  - Transform the primaries, this is done with a single matrix multiply that's a combination or:
 //    - Apply the matrix from A to XYZ
 //    - Apply the matrix from XYZ to B
 //  - Apply B's OETF (the inverse of the EOTF)
 //
 // # YCbCr color spaces
 //
 // The human vision is more sensitive to variation in light intensity (luminance) than color
 // (chroma) so video (and some images) are stored in a YCbCr representation with the luminance Y and
 // two dimensions for the chroma: Cr "the difference in red" and "Cb the difference in blue". Green
 // is usually the luminance as it has the most response from human vision. Then the chroma data can
 // be stored with lower frequency than luminance data, which gives storage and memory traffic gains.
 //
 // YCbCr colorspaces define how to convert YCbCr data to RGB, and how that RGB data converts to XYZ
 // (with an EOTF and primaries). YCbCr data is converted to RGB by performing two steps.
 //
 // The transfer function defines how to map numbers to values with Y in [0, 1] and both Cb and Cr in
 // [-0.5, 0.5] (assuming we are not using some wide gamut). The "full range" converts by normalizing
 // values almost as one would expect (a uint8_t Y by dividing by 255 and a uint8_t Cb/Cr by dividing
 // by 128 and offsetting by -0.5). However historically some values where used as control signal and
 // a "narrow range" of values was used (for Y this is [16, 235]).
 //
 // The YCbCr to RGB linear transformation that defines how Y, Cb and Cr map to RGB..
 //
 // Thus the transformation between YCbCr colorspace A and RGB colorspace B is:
 //
 // - Transform numbers representing YCbCr to RGB with a single 4x3 matrix multiply that's a
 //   combination of:
 //   - Apply the 4x3 matrix in homogeneous coordinate space that computes the range (the homogeneous
 //     coordinate space is used to allow for translation to be encoded in the matrix).
 //   - Apply the 3x3 YCbCr to RGB linear transformation.
 //
 //  (same steps as for the RGB case)
 //
 //  - Apply A's EOTF.
 //  - Transform the primaries, this is done with a single matrix multiply that's a combination or:
 //    - Apply the matrix from A to XYZ
 //    - Apply the matrix from XYZ to B
 //  - Apply B's OETF (the inverse of the EOTF)
 //
 // # Resources
 //
 // The Khronos Data Format Specification that makes the ITU standards more approachable (it is
 // linked to many times below):
 // https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html
 //
 // The CSS color module are another great reference:
 //
 //  - CSS Color Module Level 4 for the definition of many SDR color spaces:
 //    https://www.w3.org/TR/css-color-4/
 //  - CSS Color Module HDR Level 1 for the definition of HDR color spaces:
 //    https://www.w3.org/TR/css-color-hdr-1/
 //
 // The ITU specifications for the "source of truth" on many of the YCbCr colorspaces.
 //
 // Chromium's gfx::Colorspace.

 // YCbCr range transforms, assuming that the Y and Cb Cr values are already normalized to [0, 1]
 // since we get them via texture sampling.

 // https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#QUANTIZATION_FULL
 inline constexpr math::Mat4x3f kYCbCrRange_Full = {
     {1.0f, 0.0f, 0.0f},
     {0.0f, 1.0f, 0.0f},
     {0.0f, 0.0f, 1.0f},
     {0.0f, -128.0f / 255.0f, -128.0f / 255.0f},
 };

 // https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#QUANTIZATION_NARROW
 constexpr float kYCbCrRange_NarrowYFactor = 255.0f / 219.0f;
 constexpr float kYCbCrRange_NarrowChromaFactor = 255.0f / 224.0f;
 constexpr auto kYCbCrRange_Narrow = math::Mat4x3f::FromRows({
     {kYCbCrRange_NarrowYFactor, 0.0f, 0.0f, -16.0f / 255.0f * kYCbCrRange_NarrowYFactor},
     {0.0f, kYCbCrRange_NarrowChromaFactor, 0.0f, -128.0f / 255.0f * kYCbCrRange_NarrowChromaFactor},
     {0.0f, 0.0f, kYCbCrRange_NarrowChromaFactor, -128.0f / 255.0f * kYCbCrRange_NarrowChromaFactor},
 });

 // YCbCr to RGB matrices for various standards

 // https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#MODEL_BT601
 inline constexpr auto kYCbCrToRGB_Rec601 = math::Mat3x3f::FromRows({
     {1.0f, 0.0f, 1.402f},
     {1.0f, -(0.202008f / 0.587f), -(0.419198f / 0.587f)},
     {1.0f, 1.772f, 0.0f},
 });

 // https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#MODEL_BT709
 inline constexpr auto kYCbCrToRGB_Rec709 = math::Mat3x3f::FromRows({
     {1.0f, 0.0f, 1.5748f},
     {1.0f, -(0.133974f / 0.7152f), -(0.334802f / 0.7152f)},
     {1.0f, 1.8556f, 0.0f},
 });

 // https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#MODEL_BT2020
 inline constexpr auto kYCbCrToRGB_Rec2020 = math::Mat3x3f::FromRows({
     {1.0f, 0.0f, 1.4746f},
     {1.0f, -(0.111567f / 0.678f), -(0.387377f / 0.678f)},
     {1.0f, 1.8814f, 0.0f},
 });

 // RGB to XYZ (D65) for various standards.
 // TODO(https://crbug.com/468988322): Define an XYZPrimaries structure and compute the matrices from
 // it. The wx and wy give the coordinates of the white point and would allow adapting to D50 if
 // needed.

 // https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#PRIMARIES_BT601_EBU
 // Rec 601 625-line.
 inline constexpr auto kRGBToXYZ_Rec601 = math::Mat3x3f::FromRows({
     {0.430554f, 0.341550f, 0.178352f},
     {0.222004f, 0.706655f, 0.071341f},
     {0.020182f, 0.129553f, 0.939322f},
 });
 inline constexpr math::Mat3x3f kXYZToRGB_Rec601 = kRGBToXYZ_Rec601.Inverse();

 // https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#PRIMARIES_BT709
 inline constexpr auto kRGBToXYZ_Rec709 = math::Mat3x3f::FromRows({
     {0.412391f, 0.357584f, 0.180481f},
     {0.212639f, 0.715169f, 0.072192f},
     {0.019331f, 0.119195f, 0.950532f},
 });
 inline constexpr math::Mat3x3f kXYZToRGB_Rec709 = kRGBToXYZ_Rec709.Inverse();

 // sRGB is the same as Rec709.
 inline constexpr math::Mat3x3f kRGBToXYZ_sRGB = kRGBToXYZ_Rec709;
 inline constexpr math::Mat3x3f kXYZToRGB_sRGB = kXYZToRGB_Rec709;

 // https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#PRIMARIES_BT2020
 inline constexpr auto kRGBToXYZ_Rec2020 = math::Mat3x3f::FromRows({
     {0.636958f, 0.144617f, 0.168881f},
     {0.262700f, 0.677998f, 0.059302f},
     {0.000000f, 0.028073f, 1.060985f},
 });
 inline constexpr math::Mat3x3f kXYZToRGB_Rec2020 = kRGBToXYZ_Rec2020.Inverse();

 // https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#PRIMARIES_BT2020
 inline constexpr auto kRGBToXYZ_DisplayP3 = math::Mat3x3f::FromRows({
     {0.4865709486f, 0.2656676932f, 0.1982172852f},
     {0.2289745641f, 0.6917385218f, 0.0792869141f},
     {0.0000000000f, 0.0451133819f, 1.0439443689f},
 });
 inline constexpr math::Mat3x3f kXYZToRGB_DisplayP3 = kRGBToXYZ_DisplayP3.Inverse();

 // Transfer functions for various standards, in a format.

 // Returns an EOTF that performs the following transformation.
 //
 //    if (abs(v) < D) {
 //        return sign(v) * (C * abs(v) + F);
 //    }
 //    return pow(A * x + B, G) + E
 //
 // TODO(https://crbug.com/497571469): Add support for PQ and HLG.
 struct TransferFunction {
     float g = 1;
     float a = 1;
     float b = 0;
     float c = 0;
     float d = 0;
     float e = 0;
     float f = 0;
 };

 inline constexpr TransferFunction kEOTF_Identity = {};

 // https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#TRANSFER_SRGB
 inline constexpr TransferFunction kEOTF_sRGB = {
     .g = 2.4,
     .a = 1.0 / 1.055,
     .b = 0.055 / 1.055,
     .c = 1.0 / 12.92,
     .d = 0.04045,
     .e = 0,
     .f = 0,
 };
 inline constexpr TransferFunction kEOTFInverse_sRGB = {
     .g = 1.0 / 2.4,
     .a = 1.13711,  // 1.055 ^ 2.4
     .b = 0,
     .c = 12.92,
     .d = 0.0031308,
     .e = -0.055,
     .f = 0,
 };

 // https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#TRANSFER_DCIP3
 inline constexpr TransferFunction kEOTF_DisplayP3 = {
     .g = 2.6,
     .a = 1.0,
     .b = 0,
     .c = 0,
     .d = 0,
     .e = 0,
     .f = 0,
 };
 inline constexpr TransferFunction kEOTFInverse_DisplayP3 = {
     .g = 1.0 / 2.6,
     .a = 1.0,
     .b = 0,
     .c = 0,
     .d = 0,
     .e = 0,
     .f = 0,
 };

 // https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#TRANSFER_ITU
 inline constexpr TransferFunction kEOTF_SMPTE_170M = {
     .g = 1.0 / 0.45,
     .a = 1.0 / 1.099,
     .b = 0.099 / 1.099,
     .c = 1.0 / 4.5,
     .d = 0.0812,
     .e = 0,
     .f = 0,
 };

 // Luminance values of various color spaces for the white of value "1" after the transfer function.

 // When SDR content is composited in HDR, it should be assumed to have a white at 203 nits, even if
 // the sRGB spec defines it as 80 nits. Note that this constant is only informative and Dawn allows
 // configuring the reference white luminance with wgpu::ColorSpaceDawn::hdrReferenceWhiteLuminance.
 // See https://www.w3.org/TR/css-color-hdr-1/#Compositing-SDR-HDR and Section 2.1 "HDR Reference
 // White" of https://www.itu.int/dms_pub/itu-r/opb/rep/R-REP-BT.2408-9-2026-PDF-E.pdf
 inline constexpr float kDefaultHDRReferenceWhiteLuminance = 203;

 // https://www.w3.org/TR/css-color-hdr-1/#valdef-color-rec2100-pq
 inline constexpr float kPQLuminanceOf1 = 10000;

 // https://www.itu.int/dms_pubrec/itu-r/rec/bt/R-REC-BT.2100-3-202502-I!!PDF-E.pdf table 5
 // "Hybrid Log-Gamma (HLG) system reference non-linear transfer functions". The peak luminance of
 // HLG is 1 (assuming an OOTF of 1.2).
 inline constexpr float kHLGLuminanceOf1 = 1000;

 }  // namespace dawn

 #endif  // SRC_DAWN_COMMON_COLORSPACE_H_
	// Copyright 2026 The Dawn & Tint Authors
	//
	// Redistribution and use in source and binary forms, with or without
	// modification, are permitted provided that the following conditions are met:
	//
	// 1. Redistributions of source code must retain the above copyright notice, this
	// list of conditions and the following disclaimer.
	//
	// 2. Redistributions in binary form must reproduce the above copyright notice,
	// this list of conditions and the following disclaimer in the documentation
	// and/or other materials provided with the distribution.
	//
	// 3. Neither the name of the copyright holder nor the names of its
	// contributors may be used to endorse or promote products derived from
	// this software without specific prior written permission.
	//
	// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
	// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
	// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
	// DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
	// FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
	// DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
	// SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
	// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
	// OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
	// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

	#ifndef SRC_DAWN_COMMON_COLORSPACE_H_
	#define SRC_DAWN_COMMON_COLORSPACE_H_

	#include "src/dawn/common/Algebra.h"

	namespace dawn {

	// Colorspaces define how numbers used to encode a color are related to physical quantities of
	// light (for example captured by a camera, or emitted by a display). Dawn needs some handling of
	// colorspace because ExternalTexture allows importing data in a source colorspace and sampling it
	// in shaders with a different colorspace called the destination colorspace.
	//
	// Dawn also needs to tell the OS how to display values in a wgpu::Surface texture to the user, but
	// that's done by passing the correct enum to the backend API's swapchain/surface configuration so
	// Dawn doesn't need to do any math for that path.
	//
	// # RGB color spaces
	//
	// Colorspaces are defined with respect to a reference colorspace, usually the CIE XYZ one but the
	// exact reference doesn't matter much in Dawn as long as all conversion are defined from/to it (so
	// combining a transform from colorspace from A to XYZ and from XYZ to B gives a transform from A to
	// B). Numbers in an RGB colorspace (as opposed to YCbCr, see below) can be converted to XYZ by
	// performing two steps.
	//
	// The electro-optical transfer function (EOTF) that's used to transform colors in "perceptual"
	// space to colors in "physical" space. This is because the human eye has a logarithmic response
	// to light intensity (such as doubling the amount of light "feels" like a single step increase,
	// the same mechanism as decibels for audition). Storing values in perceptual space is done to have
	// as much perceptual precision for low and high color values and allows for smooth perceptual
	// color gradients. (Read up on sRGB, that's the most used and well known perceptual color space).
	// The inverse transfer function is called the opto-electronic transfer function (OETF).
	//
	// Note that a weirdness of HLG is that the EOTF is not the inverse of its OETF but of OETF(OOTF)
	// (for unfortunate and historical reasons). The Annex 1 "The relationship between the OETF, the
	// EOTF and the OOTF" of
	// https://www.itu.int/dms_pubrec/itu-r/rec/bt/R-REC-BT.2100-1-201706-S!!PDF-E.pdf
	//
	// The color primaries, what it means to say "red of 1" in that colorspace. This is where CIE XYZ is
	// used as the primaries for the colorspace are mapped to colors in the XYZ colorspace. XYZ itself
	// is a physical standard developed through experimentation on the human perception of color. The
	// mapping from a colorspace to XYZ is a linear mapping represented by a matrix whose column vectors
	// are the XYZ colors corresponding to the colorspace's primaries.
	//
	// Thus the transformation between colorspace A and B is:
	//
	// - Apply A's EOTF.
	// - Transform the primaries, this is done with a single matrix multiply that's a combination or:
	// - Apply the matrix from A to XYZ
	// - Apply the matrix from XYZ to B
	// - Apply B's OETF (the inverse of the EOTF)
	//
	// # YCbCr color spaces
	//
	// The human vision is more sensitive to variation in light intensity (luminance) than color
	// (chroma) so video (and some images) are stored in a YCbCr representation with the luminance Y and
	// two dimensions for the chroma: Cr "the difference in red" and "Cb the difference in blue". Green
	// is usually the luminance as it has the most response from human vision. Then the chroma data can
	// be stored with lower frequency than luminance data, which gives storage and memory traffic gains.
	//
	// YCbCr colorspaces define how to convert YCbCr data to RGB, and how that RGB data converts to XYZ
	// (with an EOTF and primaries). YCbCr data is converted to RGB by performing two steps.
	//
	// The transfer function defines how to map numbers to values with Y in [0, 1] and both Cb and Cr in
	// [-0.5, 0.5] (assuming we are not using some wide gamut). The "full range" converts by normalizing
	// values almost as one would expect (a uint8_t Y by dividing by 255 and a uint8_t Cb/Cr by dividing
	// by 128 and offsetting by -0.5). However historically some values where used as control signal and
	// a "narrow range" of values was used (for Y this is [16, 235]).
	//
	// The YCbCr to RGB linear transformation that defines how Y, Cb and Cr map to RGB..
	//
	// Thus the transformation between YCbCr colorspace A and RGB colorspace B is:
	//
	// - Transform numbers representing YCbCr to RGB with a single 4x3 matrix multiply that's a
	// combination of:
	// - Apply the 4x3 matrix in homogeneous coordinate space that computes the range (the homogeneous
	// coordinate space is used to allow for translation to be encoded in the matrix).
	// - Apply the 3x3 YCbCr to RGB linear transformation.
	//
	// (same steps as for the RGB case)
	//
	// - Apply A's EOTF.
	// - Transform the primaries, this is done with a single matrix multiply that's a combination or:
	// - Apply the matrix from A to XYZ
	// - Apply the matrix from XYZ to B
	// - Apply B's OETF (the inverse of the EOTF)
	//
	// # Resources
	//
	// The Khronos Data Format Specification that makes the ITU standards more approachable (it is
	// linked to many times below):
	// https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html
	//
	// The CSS color module are another great reference:
	//
	// - CSS Color Module Level 4 for the definition of many SDR color spaces:
	// https://www.w3.org/TR/css-color-4/
	// - CSS Color Module HDR Level 1 for the definition of HDR color spaces:
	// https://www.w3.org/TR/css-color-hdr-1/
	//
	// The ITU specifications for the "source of truth" on many of the YCbCr colorspaces.
	//
	// Chromium's gfx::Colorspace.

	// YCbCr range transforms, assuming that the Y and Cb Cr values are already normalized to [0, 1]
	// since we get them via texture sampling.

	// https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#QUANTIZATION_FULL
	inline constexpr math::Mat4x3f kYCbCrRange_Full = {
	{1.0f, 0.0f, 0.0f},
	{0.0f, 1.0f, 0.0f},
	{0.0f, 0.0f, 1.0f},
	{0.0f, -128.0f / 255.0f, -128.0f / 255.0f},
	};

	// https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#QUANTIZATION_NARROW
	constexpr float kYCbCrRange_NarrowYFactor = 255.0f / 219.0f;
	constexpr float kYCbCrRange_NarrowChromaFactor = 255.0f / 224.0f;
	constexpr auto kYCbCrRange_Narrow = math::Mat4x3f::FromRows({
	{kYCbCrRange_NarrowYFactor, 0.0f, 0.0f, -16.0f / 255.0f * kYCbCrRange_NarrowYFactor},
	{0.0f, kYCbCrRange_NarrowChromaFactor, 0.0f, -128.0f / 255.0f * kYCbCrRange_NarrowChromaFactor},
	{0.0f, 0.0f, kYCbCrRange_NarrowChromaFactor, -128.0f / 255.0f * kYCbCrRange_NarrowChromaFactor},
	});

	// YCbCr to RGB matrices for various standards

	// https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#MODEL_BT601
	inline constexpr auto kYCbCrToRGB_Rec601 = math::Mat3x3f::FromRows({
	{1.0f, 0.0f, 1.402f},
	{1.0f, -(0.202008f / 0.587f), -(0.419198f / 0.587f)},
	{1.0f, 1.772f, 0.0f},
	});

	// https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#MODEL_BT709
	inline constexpr auto kYCbCrToRGB_Rec709 = math::Mat3x3f::FromRows({
	{1.0f, 0.0f, 1.5748f},
	{1.0f, -(0.133974f / 0.7152f), -(0.334802f / 0.7152f)},
	{1.0f, 1.8556f, 0.0f},
	});

	// https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#MODEL_BT2020
	inline constexpr auto kYCbCrToRGB_Rec2020 = math::Mat3x3f::FromRows({
	{1.0f, 0.0f, 1.4746f},
	{1.0f, -(0.111567f / 0.678f), -(0.387377f / 0.678f)},
	{1.0f, 1.8814f, 0.0f},
	});

	// RGB to XYZ (D65) for various standards.
	// TODO(https://crbug.com/468988322): Define an XYZPrimaries structure and compute the matrices from
	// it. The wx and wy give the coordinates of the white point and would allow adapting to D50 if
	// needed.

	// https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#PRIMARIES_BT601_EBU
	// Rec 601 625-line.
	inline constexpr auto kRGBToXYZ_Rec601 = math::Mat3x3f::FromRows({
	{0.430554f, 0.341550f, 0.178352f},
	{0.222004f, 0.706655f, 0.071341f},
	{0.020182f, 0.129553f, 0.939322f},
	});
	inline constexpr math::Mat3x3f kXYZToRGB_Rec601 = kRGBToXYZ_Rec601.Inverse();

	// https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#PRIMARIES_BT709
	inline constexpr auto kRGBToXYZ_Rec709 = math::Mat3x3f::FromRows({
	{0.412391f, 0.357584f, 0.180481f},
	{0.212639f, 0.715169f, 0.072192f},
	{0.019331f, 0.119195f, 0.950532f},
	});
	inline constexpr math::Mat3x3f kXYZToRGB_Rec709 = kRGBToXYZ_Rec709.Inverse();

	// sRGB is the same as Rec709.
	inline constexpr math::Mat3x3f kRGBToXYZ_sRGB = kRGBToXYZ_Rec709;
	inline constexpr math::Mat3x3f kXYZToRGB_sRGB = kXYZToRGB_Rec709;

	// https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#PRIMARIES_BT2020
	inline constexpr auto kRGBToXYZ_Rec2020 = math::Mat3x3f::FromRows({
	{0.636958f, 0.144617f, 0.168881f},
	{0.262700f, 0.677998f, 0.059302f},
	{0.000000f, 0.028073f, 1.060985f},
	});
	inline constexpr math::Mat3x3f kXYZToRGB_Rec2020 = kRGBToXYZ_Rec2020.Inverse();

	// https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#PRIMARIES_BT2020
	inline constexpr auto kRGBToXYZ_DisplayP3 = math::Mat3x3f::FromRows({
	{0.4865709486f, 0.2656676932f, 0.1982172852f},
	{0.2289745641f, 0.6917385218f, 0.0792869141f},
	{0.0000000000f, 0.0451133819f, 1.0439443689f},
	});
	inline constexpr math::Mat3x3f kXYZToRGB_DisplayP3 = kRGBToXYZ_DisplayP3.Inverse();

	// Transfer functions for various standards, in a format.

	// Returns an EOTF that performs the following transformation.
	//
	// if (abs(v) < D) {
	// return sign(v) * (C * abs(v) + F);
	// }
	// return pow(A * x + B, G) + E
	//
	// TODO(https://crbug.com/497571469): Add support for PQ and HLG.
	struct TransferFunction {
	float g = 1;
	float a = 1;
	float b = 0;
	float c = 0;
	float d = 0;
	float e = 0;
	float f = 0;
	};

	inline constexpr TransferFunction kEOTF_Identity = {};

	// https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#TRANSFER_SRGB
	inline constexpr TransferFunction kEOTF_sRGB = {
	.g = 2.4,
	.a = 1.0 / 1.055,
	.b = 0.055 / 1.055,
	.c = 1.0 / 12.92,
	.d = 0.04045,
	.e = 0,
	.f = 0,
	};
	inline constexpr TransferFunction kEOTFInverse_sRGB = {
	.g = 1.0 / 2.4,
	.a = 1.13711, // 1.055 ^ 2.4
	.b = 0,
	.c = 12.92,
	.d = 0.0031308,
	.e = -0.055,
	.f = 0,
	};

	// https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#TRANSFER_DCIP3
	inline constexpr TransferFunction kEOTF_DisplayP3 = {
	.g = 2.6,
	.a = 1.0,
	.b = 0,
	.c = 0,
	.d = 0,
	.e = 0,
	.f = 0,
	};
	inline constexpr TransferFunction kEOTFInverse_DisplayP3 = {
	.g = 1.0 / 2.6,
	.a = 1.0,
	.b = 0,
	.c = 0,
	.d = 0,
	.e = 0,
	.f = 0,
	};

	// https://registry.khronos.org/DataFormat/specs/1.4/dataformat.1.4.html#TRANSFER_ITU
	inline constexpr TransferFunction kEOTF_SMPTE_170M = {
	.g = 1.0 / 0.45,
	.a = 1.0 / 1.099,
	.b = 0.099 / 1.099,
	.c = 1.0 / 4.5,
	.d = 0.0812,
	.e = 0,
	.f = 0,
	};

	// Luminance values of various color spaces for the white of value "1" after the transfer function.

	// When SDR content is composited in HDR, it should be assumed to have a white at 203 nits, even if
	// the sRGB spec defines it as 80 nits. Note that this constant is only informative and Dawn allows
	// configuring the reference white luminance with wgpu::ColorSpaceDawn::hdrReferenceWhiteLuminance.
	// See https://www.w3.org/TR/css-color-hdr-1/#Compositing-SDR-HDR and Section 2.1 "HDR Reference
	// White" of https://www.itu.int/dms_pub/itu-r/opb/rep/R-REP-BT.2408-9-2026-PDF-E.pdf
	inline constexpr float kDefaultHDRReferenceWhiteLuminance = 203;

	// https://www.w3.org/TR/css-color-hdr-1/#valdef-color-rec2100-pq
	inline constexpr float kPQLuminanceOf1 = 10000;

	// https://www.itu.int/dms_pubrec/itu-r/rec/bt/R-REC-BT.2100-3-202502-I!!PDF-E.pdf table 5
	// "Hybrid Log-Gamma (HLG) system reference non-linear transfer functions". The peak luminance of
	// HLG is 1 (assuming an OOTF of 1.2).
	inline constexpr float kHLGLuminanceOf1 = 1000;

	} // namespace dawn

	#endif // SRC_DAWN_COMMON_COLORSPACE_H_