Mirrors/overte-JulianGro
3
0
Fork 0
mirror of https://github.com/JulianGro/overte.git synced 2025-04-16 22:30:42 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge pull request #11349 from Zvork/hdrcube

Browse source 
Under the hood support for HDR cube maps
This commit is contained in:
Sam Gateau 2017-10-10 11:43:41 -07:00 committed by GitHub
commit c44efda978
11 changed files with 501 additions and 89 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

23
libraries/gpu-gl/src/gpu/gl/GLTexelFormat.cpp
View file

@ -19,6 +19,7 @@ bool GLTexelFormat::isCompressed() const {
case GL_COMPRESSED_RED_RGTC1:
case GL_COMPRESSED_RG_RGTC2:
case GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM:
case GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT:
return true;
default:
return false;
@ -94,6 +95,11 @@ GLenum GLTexelFormat::evalGLTexelFormatInternal(const gpu::Element& dstFormat) {
result = GL_R11F_G11F_B10F;
break;
case gpu::RGB9E5:
// the type should be float
result = GL_RGB9_E5;
break;
case gpu::DEPTH:
result = GL_DEPTH_COMPONENT32;
switch (dstFormat.getType()) {
@ -244,6 +250,9 @@ GLenum GLTexelFormat::evalGLTexelFormatInternal(const gpu::Element& dstFormat) {
case gpu::COMPRESSED_BC5_XY:
result = GL_COMPRESSED_RG_RGTC2;
break;
case gpu::COMPRESSED_BC6_RGB:
result = GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT;
break;
case gpu::COMPRESSED_BC7_SRGBA:
result = GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM;
break;
@ -396,6 +405,9 @@ GLTexelFormat GLTexelFormat::evalGLTexelFormat(const Element& dstFormat, const E
case gpu::COMPRESSED_BC5_XY:
texel.internalFormat = GL_COMPRESSED_RG_RGTC2;
break;
case gpu::COMPRESSED_BC6_RGB:
texel.internalFormat = GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT;
break;
case gpu::COMPRESSED_BC7_SRGBA:
texel.internalFormat = GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM;
break;
@ -495,10 +507,16 @@ GLTexelFormat GLTexelFormat::evalGLTexelFormat(const Element& dstFormat, const E
case gpu::R11G11B10:
texel.format = GL_RGB;
// the type should be float
texel.type = GL_UNSIGNED_INT_10F_11F_11F_REV;
texel.internalFormat = GL_R11F_G11F_B10F;
break;
case gpu::RGB9E5:
texel.format = GL_RGB;
texel.type = GL_UNSIGNED_INT_5_9_9_9_REV;
texel.internalFormat = GL_RGB9_E5;
break;
case gpu::DEPTH:
texel.format = GL_DEPTH_COMPONENT; // It's depth component to load it
texel.internalFormat = GL_DEPTH_COMPONENT32;
@ -694,6 +712,9 @@ GLTexelFormat GLTexelFormat::evalGLTexelFormat(const Element& dstFormat, const E
case gpu::COMPRESSED_BC5_XY:
texel.internalFormat = GL_COMPRESSED_RG_RGTC2;
break;
case gpu::COMPRESSED_BC6_RGB:
texel.internalFormat = GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT;
break;
case gpu::COMPRESSED_BC7_SRGBA:
texel.internalFormat = GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM;
break;

2
libraries/gpu-gl/src/gpu/gl41/GL41BackendTexture.cpp
View file

@ -114,6 +114,7 @@ Size GL41Texture::copyMipFaceLinesFromTexture(uint16_t mip, uint8_t face, const
case GL_COMPRESSED_RED_RGTC1:
case GL_COMPRESSED_RG_RGTC2:
case GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM:
case GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT:
glCompressedTexSubImage2D(_target, mip, 0, yOffset, size.x, size.y, internalFormat,
static_cast<GLsizei>(sourceSize), sourcePointer);
break;
@ -131,6 +132,7 @@ Size GL41Texture::copyMipFaceLinesFromTexture(uint16_t mip, uint8_t face, const
case GL_COMPRESSED_RED_RGTC1:
case GL_COMPRESSED_RG_RGTC2:
case GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM:
case GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT:
glCompressedTexSubImage2D(target, mip, 0, yOffset, size.x, size.y, internalFormat,
static_cast<GLsizei>(sourceSize), sourcePointer);
break;

2
libraries/gpu-gl/src/gpu/gl45/GL45BackendTexture.cpp
View file

@ -143,6 +143,7 @@ Size GL45Texture::copyMipFaceLinesFromTexture(uint16_t mip, uint8_t face, const
case GL_COMPRESSED_RED_RGTC1:
case GL_COMPRESSED_RG_RGTC2:
case GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM:
case GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT:
glCompressedTextureSubImage2D(_id, mip, 0, yOffset, size.x, size.y, internalFormat,
static_cast<GLsizei>(sourceSize), sourcePointer);
break;
@ -158,6 +159,7 @@ Size GL45Texture::copyMipFaceLinesFromTexture(uint16_t mip, uint8_t face, const
case GL_COMPRESSED_RED_RGTC1:
case GL_COMPRESSED_RG_RGTC2:
case GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM:
case GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT:
if (glCompressedTextureSubImage2DEXT) {
auto target = GLTexture::CUBE_FACE_LAYOUT[face];
glCompressedTextureSubImage2DEXT(_id, target, mip, 0, yOffset, size.x, size.y, internalFormat,

4
libraries/gpu/src/gpu/Format.cpp
View file

@ -24,11 +24,13 @@ const Element Element::COLOR_COMPRESSED_SRGB { TILE4x4, COMPRESSED, COMPRESSED_B
const Element Element::COLOR_COMPRESSED_SRGBA_MASK { TILE4x4, COMPRESSED, COMPRESSED_BC1_SRGBA };
const Element Element::COLOR_COMPRESSED_SRGBA { TILE4x4, COMPRESSED, COMPRESSED_BC3_SRGBA };
const Element Element::COLOR_COMPRESSED_XY { TILE4x4, COMPRESSED, COMPRESSED_BC5_XY };
const Element Element::COLOR_COMPRESSED_SRGBA_HIGH { TILE4x4, COMPRESSED, COMPRESSED_BC7_SRGBA };
const Element Element::COLOR_COMPRESSED_SRGBA_HIGH{ TILE4x4, COMPRESSED, COMPRESSED_BC7_SRGBA };
const Element Element::COLOR_COMPRESSED_HDR_RGB{ TILE4x4, COMPRESSED, COMPRESSED_BC6_RGB };
const Element Element::VEC2NU8_XY{ VEC2, NUINT8, XY };
const Element Element::COLOR_R11G11B10{ SCALAR, FLOAT, R11G11B10 };
const Element Element::COLOR_RGB9E5{ SCALAR, FLOAT, RGB9E5 };
const Element Element::VEC4F_COLOR_RGBA{ VEC4, FLOAT, RGBA };
const Element Element::VEC2F_UV{ VEC2, FLOAT, UV };
const Element Element::VEC2F_XY{ VEC2, FLOAT, XY };

6
libraries/gpu/src/gpu/Format.h
View file

@ -187,11 +187,13 @@ enum Semantic : uint8_t {
COMPRESSED_BC3_SRGBA,
COMPRESSED_BC4_RED,
COMPRESSED_BC5_XY,
COMPRESSED_BC6_RGB,
COMPRESSED_BC7_SRGBA,
_LAST_COMPRESSED,
R11G11B10,
RGB9E5,
UNIFORM,
UNIFORM_BUFFER,
@ -240,11 +242,13 @@ static const int SEMANTIC_SIZE_FACTOR[NUM_SEMANTICS] = {
16, //COMPRESSED_BC3_SRGBA, 1 byte/pixel * 4x4 pixels = 16 bytes
8, //COMPRESSED_BC4_RED, 1/2 byte/pixel * 4x4 pixels = 8 bytes
16, //COMPRESSED_BC5_XY, 1 byte/pixel * 4x4 pixels = 16 bytes
16, //COMPRESSED_BC6_RGB, 1 byte/pixel * 4x4 pixels = 16 bytes
16, //COMPRESSED_BC7_SRGBA, 1 byte/pixel * 4x4 pixels = 16 bytes
1, //_LAST_COMPRESSED,
1, //R11G11B10,
1, //RGB9E5
1, //UNIFORM,
1, //UNIFORM_BUFFER,
@ -306,12 +310,14 @@ public:
static const Element COLOR_BGRA_32;
static const Element COLOR_SBGRA_32;
static const Element COLOR_R11G11B10;
static const Element COLOR_RGB9E5;
static const Element COLOR_COMPRESSED_RED;
static const Element COLOR_COMPRESSED_SRGB;
static const Element COLOR_COMPRESSED_SRGBA_MASK;
static const Element COLOR_COMPRESSED_SRGBA;
static const Element COLOR_COMPRESSED_XY;
static const Element COLOR_COMPRESSED_SRGBA_HIGH;
static const Element COLOR_COMPRESSED_HDR_RGB;
static const Element VEC2NU8_XY;
static const Element VEC4F_COLOR_RGBA;
static const Element VEC2F_UV;

47
libraries/gpu/src/gpu/Texture.cpp
View file

@ -14,6 +14,7 @@
#include <glm/gtc/constants.hpp>
#include <glm/gtx/component_wise.hpp>
#include <glm/gtc/packing.hpp>
#include <QtCore/QDebug>
#include <QtCore/QThread>
@ -683,6 +684,21 @@ bool sphericalHarmonicsFromTexture(const gpu::Texture& cubeTexture, std::vector<
PROFILE_RANGE(render_gpu, "sphericalHarmonicsFromTexture");
auto mipFormat = cubeTexture.getStoredMipFormat();
std::function<glm::vec3(uint32)> unpackFunc;
switch (mipFormat.getSemantic()) {
case gpu::R11G11B10:
unpackFunc = glm::unpackF2x11_1x10;
break;
case gpu::RGB9E5:
unpackFunc = glm::unpackF3x9_E1x5;
break;
default:
assert(false);
break;
}
const uint sqOrder = order*order;
// allocate memory for calculations
@ -716,17 +732,7 @@ bool sphericalHarmonicsFromTexture(const gpu::Texture& cubeTexture, std::vector<
for(int face=0; face < gpu::Texture::NUM_CUBE_FACES; face++) {
PROFILE_RANGE(render_gpu, "ProcessFace");
auto mipFormat = cubeTexture.getStoredMipFormat();
auto numComponents = mipFormat.getScalarCount();
int roffset { 0 };
int goffset { 1 };
int boffset { 2 };
if ((mipFormat.getSemantic() == gpu::BGRA) || (mipFormat.getSemantic() == gpu::SBGRA)) {
roffset = 2;
boffset = 0;
}
auto data = cubeTexture.accessStoredMipFace(0, face)->readData();
auto data = reinterpret_cast<const uint32*>( cubeTexture.accessStoredMipFace(0, face)->readData() );
if (data == nullptr) {
continue;
}
@ -806,29 +812,24 @@ bool sphericalHarmonicsFromTexture(const gpu::Texture& cubeTexture, std::vector<
// index of texel in texture
// get color from texture and map to range [0, 1]
float red { 0.0f };
float green { 0.0f };
float blue { 0.0f };
// get color from texture
glm::vec3 color{ 0.0f, 0.0f, 0.0f };
for (int i = 0; i < stride; ++i) {
for (int j = 0; j < stride; ++j) {
int k = (int)(x + i - halfStride + (y + j - halfStride) * width) * numComponents;
red += ColorUtils::sRGB8ToLinearFloat(data[k + roffset]);
green += ColorUtils::sRGB8ToLinearFloat(data[k + goffset]);
blue += ColorUtils::sRGB8ToLinearFloat(data[k + boffset]);
int k = (int)(x + i - halfStride + (y + j - halfStride) * width);
color += unpackFunc(data[k]);
}
}
glm::vec3 clr(red, green, blue);
// scale color and add to previously accumulated coefficients
// red
sphericalHarmonicsScale(shBuffB.data(), order, shBuff.data(), clr.r * fDiffSolid);
sphericalHarmonicsScale(shBuffB.data(), order, shBuff.data(), color.r * fDiffSolid);
sphericalHarmonicsAdd(resultR.data(), order, resultR.data(), shBuffB.data());
// green
sphericalHarmonicsScale(shBuffB.data(), order, shBuff.data(), clr.g * fDiffSolid);
sphericalHarmonicsScale(shBuffB.data(), order, shBuff.data(), color.g * fDiffSolid);
sphericalHarmonicsAdd(resultG.data(), order, resultG.data(), shBuffB.data());
// blue
sphericalHarmonicsScale(shBuffB.data(), order, shBuff.data(), clr.b * fDiffSolid);
sphericalHarmonicsScale(shBuffB.data(), order, shBuff.data(), color.b * fDiffSolid);
sphericalHarmonicsAdd(resultB.data(), order, resultB.data(), shBuffB.data());
}
}

19
libraries/gpu/src/gpu/Texture_ktx.cpp
View file

@ -515,8 +515,6 @@ TexturePointer Texture::build(const ktx::KTXDescriptor& descriptor) {
return texture;
}
TexturePointer Texture::unserialize(const cache::FilePointer& cacheEntry) {
std::unique_ptr<ktx::KTX> ktxPointer = ktx::KTX::create(std::make_shared<storage::FileStorage>(cacheEntry->getFilepath().c_str()));
if (!ktxPointer) {
@ -536,7 +534,7 @@ TexturePointer Texture::unserialize(const std::string& ktxfile) {
if (!ktxPointer) {
return nullptr;
}
auto texture = build(ktxPointer->toDescriptor());
if (texture) {
texture->setKtxBacking(ktxfile);
@ -570,6 +568,12 @@ bool Texture::evalKTXFormat(const Element& mipFormat, const Element& texelFormat
header.setCompressed(ktx::GLInternalFormat::COMPRESSED_RG_RGTC2, ktx::GLBaseInternalFormat::RG);
} else if (texelFormat == Format::COLOR_COMPRESSED_SRGBA_HIGH && mipFormat == Format::COLOR_COMPRESSED_SRGBA_HIGH) {
header.setCompressed(ktx::GLInternalFormat::COMPRESSED_SRGB_ALPHA_BPTC_UNORM, ktx::GLBaseInternalFormat::RGBA);
} else if (texelFormat == Format::COLOR_COMPRESSED_HDR_RGB && mipFormat == Format::COLOR_COMPRESSED_HDR_RGB) {
header.setCompressed(ktx::GLInternalFormat::COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT, ktx::GLBaseInternalFormat::RGB);
} else if (texelFormat == Format::COLOR_RGB9E5 && mipFormat == Format::COLOR_RGB9E5) {
header.setUncompressed(ktx::GLType::UNSIGNED_INT_5_9_9_9_REV, 1, ktx::GLFormat::RGB, ktx::GLInternalFormat::RGB9_E5, ktx::GLBaseInternalFormat::RGB);
} else if (texelFormat == Format::COLOR_R11G11B10 && mipFormat == Format::COLOR_R11G11B10) {
header.setUncompressed(ktx::GLType::UNSIGNED_INT_10F_11F_11F_REV, 1, ktx::GLFormat::RGB, ktx::GLInternalFormat::R11F_G11F_B10F, ktx::GLBaseInternalFormat::RGB);
} else {
return false;
}
@ -612,6 +616,12 @@ bool Texture::evalTextureFormat(const ktx::Header& header, Element& mipFormat, E
} else {
return false;
}
} else if (header.getGLFormat() == ktx::GLFormat::RGB && header.getGLType() == ktx::GLType::UNSIGNED_INT_10F_11F_11F_REV) {
mipFormat = Format::COLOR_R11G11B10;
texelFormat = Format::COLOR_R11G11B10;
} else if (header.getGLFormat() == ktx::GLFormat::RGB && header.getGLType() == ktx::GLType::UNSIGNED_INT_5_9_9_9_REV) {
mipFormat = Format::COLOR_RGB9E5;
texelFormat = Format::COLOR_RGB9E5;
} else if (header.isCompressed()) {
if (header.getGLInternaFormat() == ktx::GLInternalFormat::COMPRESSED_SRGB_S3TC_DXT1_EXT) {
mipFormat = Format::COLOR_COMPRESSED_SRGB;
@ -631,6 +641,9 @@ bool Texture::evalTextureFormat(const ktx::Header& header, Element& mipFormat, E
} else if (header.getGLInternaFormat() == ktx::GLInternalFormat::COMPRESSED_SRGB_ALPHA_BPTC_UNORM) {
mipFormat = Format::COLOR_COMPRESSED_SRGBA_HIGH;
texelFormat = Format::COLOR_COMPRESSED_SRGBA_HIGH;
} else if (header.getGLInternaFormat() == ktx::GLInternalFormat::COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT) {
mipFormat = Format::COLOR_COMPRESSED_HDR_RGB;
texelFormat = Format::COLOR_COMPRESSED_HDR_RGB;
} else {
return false;
}

331
libraries/image/src/image/Image.cpp
View file

@ -11,9 +11,10 @@
#include "Image.h"
#include <nvtt/nvtt.h>
#include <glm/gtc/packing.hpp>
#include <QtCore/QtGlobal>
#include <QUrl>
#include <QImage>
#include <QBuffer>
@ -42,6 +43,8 @@ bool DEV_DECIMATE_TEXTURES = false;
std::atomic<size_t> DECIMATED_TEXTURE_COUNT{ 0 };
std::atomic<size_t> RECTIFIED_TEXTURE_COUNT{ 0 };
static const auto HDR_FORMAT = gpu::Element::COLOR_R11G11B10;
static std::atomic<bool> compressColorTextures { false };
static std::atomic<bool> compressNormalTextures { false };
static std::atomic<bool> compressGrayscaleTextures { false };
@ -71,6 +74,10 @@ glm::uvec2 rectifyToSparseSize(const glm::uvec2& size) {
namespace image {
enum {
QIMAGE_HDR_FORMAT = QImage::Format_RGB30
};
TextureUsage::TextureLoader TextureUsage::getTextureLoaderForType(Type type, const QVariantMap& options) {
switch (type) {
case ALBEDO_TEXTURE:
@ -213,6 +220,25 @@ void setCubeTexturesCompressionEnabled(bool enabled) {
compressCubeTextures.store(enabled);
}
static float denormalize(float value, const float minValue) {
return value < minValue ? 0.0f : value;
}
uint32 packR11G11B10F(const glm::vec3& color) {
// Denormalize else unpacking gives high and incorrect values
// See https://www.khronos.org/opengl/wiki/Small_Float_Formats for this min value
static const auto minValue = 6.10e-5f;
static const auto maxValue = 6.50e4f;
glm::vec3 ucolor;
ucolor.r = denormalize(color.r, minValue);
ucolor.g = denormalize(color.g, minValue);
ucolor.b = denormalize(color.b, minValue);
ucolor.r = std::min(ucolor.r, maxValue);
ucolor.g = std::min(ucolor.g, maxValue);
ucolor.b = std::min(ucolor.b, maxValue);
return glm::packF2x11_1x10(ucolor);
}
gpu::TexturePointer processImage(const QByteArray& content, const std::string& filename,
int maxNumPixels, TextureUsage::Type textureType,
const std::atomic<bool>& abortProcessing) {
@ -270,8 +296,6 @@ gpu::TexturePointer processImage(const QByteArray& content, const std::string& f
return texture;
}
QImage processSourceImage(const QImage& srcImage, bool cubemap) {
PROFILE_RANGE(resource_parse, "processSourceImage");
const glm::uvec2 srcImageSize = toGlm(srcImage.size());
@ -303,8 +327,8 @@ QImage processSourceImage(const QImage& srcImage, bool cubemap) {
}
#if defined(NVTT_API)
struct MyOutputHandler : public nvtt::OutputHandler {
MyOutputHandler(gpu::Texture* texture, int face) : _texture(texture), _face(face) {}
struct OutputHandler : public nvtt::OutputHandler {
OutputHandler(gpu::Texture* texture, int face) : _texture(texture), _face(face) {}
virtual void beginImage(int size, int width, int height, int depth, int face, int miplevel) override {
_size = size;
@ -313,12 +337,14 @@ struct MyOutputHandler : public nvtt::OutputHandler {
_data = static_cast<gpu::Byte*>(malloc(size));
_current = _data;
}
virtual bool writeData(const void* data, int size) override {
assert(_current + size <= _data + _size);
memcpy(_current, data, size);
_current += size;
return true;
}
virtual void endImage() override {
if (_face >= 0) {
_texture->assignStoredMipFace(_miplevel, _face, _size, static_cast<const gpu::Byte*>(_data));
@ -336,6 +362,51 @@ struct MyOutputHandler : public nvtt::OutputHandler {
int _size = 0;
int _face = -1;
};
struct PackedFloatOutputHandler : public OutputHandler {
PackedFloatOutputHandler(gpu::Texture* texture, int face, gpu::Element format) : OutputHandler(texture, face) {
if (format == gpu::Element::COLOR_RGB9E5) {
_packFunc = glm::packF3x9_E1x5;
} else if (format == gpu::Element::COLOR_R11G11B10) {
_packFunc = packR11G11B10F;
} else {
qCWarning(imagelogging) << "Unknown handler format";
Q_UNREACHABLE();
}
}
virtual void beginImage(int size, int width, int height, int depth, int face, int miplevel) override {
// Divide by 3 because we will compress from 3*floats to 1 uint32
OutputHandler::beginImage(size / 3, width, height, depth, face, miplevel);
}
virtual bool writeData(const void* data, int size) override {
// Expecting to write multiple of floats
if (_packFunc) {
assert((size % sizeof(float)) == 0);
auto floatCount = size / sizeof(float);
const float* floatBegin = (const float*)data;
const float* floatEnd = floatBegin + floatCount;
while (floatBegin < floatEnd) {
_pixel[_coordIndex] = *floatBegin;
floatBegin++;
_coordIndex++;
if (_coordIndex == 3) {
uint32 packedRGB = _packFunc(_pixel);
_coordIndex = 0;
OutputHandler::writeData(&packedRGB, sizeof(packedRGB));
}
}
return true;
}
return false;
}
std::function<uint32(const glm::vec3&)> _packFunc;
glm::vec3 _pixel;
int _coordIndex{ 0 };
};
struct MyErrorHandler : public nvtt::ErrorHandler {
virtual void error(nvtt::Error e) override {
qCWarning(imagelogging) << "Texture compression error:" << nvtt::errorString(e);
@ -343,10 +414,115 @@ struct MyErrorHandler : public nvtt::ErrorHandler {
};
#endif
void generateMips(gpu::Texture* texture, QImage& image, const std::atomic<bool>& abortProcessing = false, int face = -1) {
#if CPU_MIPMAPS
PROFILE_RANGE(resource_parse, "generateMips");
class SequentialTaskDispatcher : public nvtt::TaskDispatcher {
public:
SequentialTaskDispatcher(const std::atomic<bool>& abortProcessing) : _abortProcessing(abortProcessing) {};
const std::atomic<bool>& _abortProcessing;
virtual void dispatch(nvtt::Task* task, void* context, int count) override {
for (int i = 0; i < count; i++) {
if (!_abortProcessing.load()) {
task(context, i);
} else {
break;
}
}
}
};
void generateHDRMips(gpu::Texture* texture, const QImage& image, const std::atomic<bool>& abortProcessing, int face) {
assert(image.format() == QIMAGE_HDR_FORMAT);
const int width = image.width(), height = image.height();
std::vector<glm::vec4> data;
std::vector<glm::vec4>::iterator dataIt;
auto mipFormat = texture->getStoredMipFormat();
std::function<glm::vec3(uint32)> unpackFunc;
nvtt::TextureType textureType = nvtt::TextureType_2D;
nvtt::InputFormat inputFormat = nvtt::InputFormat_RGBA_32F;
nvtt::WrapMode wrapMode = nvtt::WrapMode_Mirror;
nvtt::RoundMode roundMode = nvtt::RoundMode_None;
nvtt::AlphaMode alphaMode = nvtt::AlphaMode_None;
nvtt::CompressionOptions compressionOptions;
compressionOptions.setQuality(nvtt::Quality_Production);
if (mipFormat == gpu::Element::COLOR_COMPRESSED_HDR_RGB) {
compressionOptions.setFormat(nvtt::Format_BC6);
} else if (mipFormat == gpu::Element::COLOR_RGB9E5) {
compressionOptions.setFormat(nvtt::Format_RGB);
compressionOptions.setPixelType(nvtt::PixelType_Float);
compressionOptions.setPixelFormat(32, 32, 32, 0);
} else if (mipFormat == gpu::Element::COLOR_R11G11B10) {
compressionOptions.setFormat(nvtt::Format_RGB);
compressionOptions.setPixelType(nvtt::PixelType_Float);
compressionOptions.setPixelFormat(32, 32, 32, 0);
} else {
qCWarning(imagelogging) << "Unknown mip format";
Q_UNREACHABLE();
return;
}
if (HDR_FORMAT == gpu::Element::COLOR_RGB9E5) {
unpackFunc = glm::unpackF3x9_E1x5;
} else if (HDR_FORMAT == gpu::Element::COLOR_R11G11B10) {
unpackFunc = glm::unpackF2x11_1x10;
} else {
qCWarning(imagelogging) << "Unknown HDR encoding format in QImage";
Q_UNREACHABLE();
return;
}
data.resize(width*height);
dataIt = data.begin();
for (auto lineNb = 0; lineNb < height; lineNb++) {
const uint32* srcPixelIt = reinterpret_cast<const uint32*>( image.constScanLine(lineNb) );
const uint32* srcPixelEnd = srcPixelIt + width;
while (srcPixelIt < srcPixelEnd) {
*dataIt = glm::vec4(unpackFunc(*srcPixelIt), 1.0f);
++srcPixelIt;
++dataIt;
}
}
assert(dataIt == data.end());
nvtt::OutputOptions outputOptions;
outputOptions.setOutputHeader(false);
std::unique_ptr<nvtt::OutputHandler> outputHandler;
MyErrorHandler errorHandler;
outputOptions.setErrorHandler(&errorHandler);
nvtt::Context context;
int mipLevel = 0;
if (mipFormat == gpu::Element::COLOR_RGB9E5 || mipFormat == gpu::Element::COLOR_R11G11B10) {
// Don't use NVTT (at least version 2.1) as it outputs wrong RGB9E5 and R11G11B10F values from floats
outputHandler.reset(new PackedFloatOutputHandler(texture, face, mipFormat));
} else {
outputHandler.reset( new OutputHandler(texture, face) );
}
outputOptions.setOutputHandler(outputHandler.get());
nvtt::Surface surface;
surface.setImage(inputFormat, width, height, 1, &(*data.begin()));
surface.setAlphaMode(alphaMode);
surface.setWrapMode(wrapMode);
SequentialTaskDispatcher dispatcher(abortProcessing);
nvtt::Compressor compressor;
context.setTaskDispatcher(&dispatcher);
context.compress(surface, face, mipLevel++, compressionOptions, outputOptions);
while (surface.canMakeNextMipmap() && !abortProcessing.load()) {
surface.buildNextMipmap(nvtt::MipmapFilter_Box);
context.compress(surface, face, mipLevel++, compressionOptions, outputOptions);
}
}
void generateLDRMips(gpu::Texture* texture, QImage& image, const std::atomic<bool>& abortProcessing, int face) {
if (image.format() != QImage::Format_ARGB32) {
image = image.convertToFormat(QImage::Format_ARGB32);
}
@ -400,10 +576,10 @@ void generateMips(gpu::Texture* texture, QImage& image, const std::atomic<bool>&
compressionOptions.setPixelType(nvtt::PixelType_UnsignedNorm);
compressionOptions.setPitchAlignment(4);
compressionOptions.setPixelFormat(32,
0x000000FF,
0x0000FF00,
0x00FF0000,
0xFF000000);
0x000000FF,
0x0000FF00,
0x00FF0000,
0xFF000000);
inputGamma = 1.0f;
outputGamma = 1.0f;
} else if (mipFormat == gpu::Element::COLOR_BGRA_32) {
@ -411,10 +587,10 @@ void generateMips(gpu::Texture* texture, QImage& image, const std::atomic<bool>&
compressionOptions.setPixelType(nvtt::PixelType_UnsignedNorm);
compressionOptions.setPitchAlignment(4);
compressionOptions.setPixelFormat(32,
0x00FF0000,
0x0000FF00,
0x000000FF,
0xFF000000);
0x00FF0000,
0x0000FF00,
0x000000FF,
0xFF000000);
inputGamma = 1.0f;
outputGamma = 1.0f;
} else if (mipFormat == gpu::Element::COLOR_SRGBA_32) {
@ -422,19 +598,19 @@ void generateMips(gpu::Texture* texture, QImage& image, const std::atomic<bool>&
compressionOptions.setPixelType(nvtt::PixelType_UnsignedNorm);
compressionOptions.setPitchAlignment(4);
compressionOptions.setPixelFormat(32,
0x000000FF,
0x0000FF00,
0x00FF0000,
0xFF000000);
0x000000FF,
0x0000FF00,
0x00FF0000,
0xFF000000);
} else if (mipFormat == gpu::Element::COLOR_SBGRA_32) {
compressionOptions.setFormat(nvtt::Format_RGBA);
compressionOptions.setPixelType(nvtt::PixelType_UnsignedNorm);
compressionOptions.setPitchAlignment(4);
compressionOptions.setPixelFormat(32,
0x00FF0000,
0x0000FF00,
0x000000FF,
0xFF000000);
0x00FF0000,
0x0000FF00,
0x000000FF,
0xFF000000);
} else if (mipFormat == gpu::Element::COLOR_R_8) {
compressionOptions.setFormat(nvtt::Format_RGB);
compressionOptions.setPixelType(nvtt::PixelType_UnsignedNorm);
@ -454,32 +630,26 @@ void generateMips(gpu::Texture* texture, QImage& image, const std::atomic<bool>&
nvtt::OutputOptions outputOptions;
outputOptions.setOutputHeader(false);
MyOutputHandler outputHandler(texture, face);
OutputHandler outputHandler(texture, face);
outputOptions.setOutputHandler(&outputHandler);
MyErrorHandler errorHandler;
outputOptions.setErrorHandler(&errorHandler);
class SequentialTaskDispatcher : public nvtt::TaskDispatcher {
public:
SequentialTaskDispatcher(const std::atomic<bool>& abortProcessing) : _abortProcessing(abortProcessing) {};
const std::atomic<bool>& _abortProcessing;
virtual void dispatch(nvtt::Task* task, void* context, int count) override {
for (int i = 0; i < count; i++) {
if (!_abortProcessing.load()) {
task(context, i);
} else {
break;
}
}
}
};
SequentialTaskDispatcher dispatcher(abortProcessing);
nvtt::Compressor compressor;
compressor.setTaskDispatcher(&dispatcher);
compressor.process(inputOptions, compressionOptions, outputOptions);
}
void generateMips(gpu::Texture* texture, QImage& image, const std::atomic<bool>& abortProcessing = false, int face = -1) {
#if CPU_MIPMAPS
PROFILE_RANGE(resource_parse, "generateMips");
if (image.format() == QIMAGE_HDR_FORMAT) {
generateHDRMips(texture, image, abortProcessing, face);
} else {
generateLDRMips(texture, image, abortProcessing, face);
}
#else
texture->setAutoGenerateMips(true);
#endif
@ -961,6 +1131,62 @@ const CubeLayout CubeLayout::CUBEMAP_LAYOUTS[] = {
};
const int CubeLayout::NUM_CUBEMAP_LAYOUTS = sizeof(CubeLayout::CUBEMAP_LAYOUTS) / sizeof(CubeLayout);
//#define DEBUG_COLOR_PACKING
QImage convertToHDRFormat(QImage srcImage, gpu::Element format) {
QImage hdrImage(srcImage.width(), srcImage.height(), (QImage::Format)QIMAGE_HDR_FORMAT);
std::function<uint32(const glm::vec3&)> packFunc;
#ifdef DEBUG_COLOR_PACKING
std::function<glm::vec3(uint32)> unpackFunc;
#endif
switch (format.getSemantic()) {
case gpu::R11G11B10:
packFunc = packR11G11B10F;
#ifdef DEBUG_COLOR_PACKING
unpackFunc = glm::unpackF2x11_1x10;
#endif
break;
case gpu::RGB9E5:
packFunc = glm::packF3x9_E1x5;
#ifdef DEBUG_COLOR_PACKING
unpackFunc = glm::unpackF3x9_E1x5;
#endif
break;
default:
qCWarning(imagelogging) << "Unsupported HDR format";
Q_UNREACHABLE();
return srcImage;
}
srcImage = srcImage.convertToFormat(QImage::Format_ARGB32);
for (auto y = 0; y < srcImage.height(); y++) {
const QRgb* srcLineIt = reinterpret_cast<const QRgb*>( srcImage.constScanLine(y) );
const QRgb* srcLineEnd = srcLineIt + srcImage.width();
uint32* hdrLineIt = reinterpret_cast<uint32*>( hdrImage.scanLine(y) );
glm::vec3 color;
while (srcLineIt < srcLineEnd) {
color.r = qRed(*srcLineIt);
color.g = qGreen(*srcLineIt);
color.b = qBlue(*srcLineIt);
// Normalize and apply gamma
color /= 255.0f;
color.r = powf(color.r, 2.2f);
color.g = powf(color.g, 2.2f);
color.b = powf(color.b, 2.2f);
*hdrLineIt = packFunc(color);
#ifdef DEBUG_COLOR_PACKING
glm::vec3 ucolor = unpackFunc(*hdrLineIt);
assert(glm::distance(color, ucolor) <= 5e-2);
#endif
++srcLineIt;
++hdrLineIt;
}
}
return hdrImage;
}
gpu::TexturePointer TextureUsage::processCubeTextureColorFromImage(const QImage& srcImage, const std::string& srcImageName,
bool generateIrradiance,
const std::atomic<bool>& abortProcessing) {
@ -969,18 +1195,19 @@ gpu::TexturePointer TextureUsage::processCubeTextureColorFromImage(const QImage&
gpu::TexturePointer theTexture = nullptr;
if ((srcImage.width() > 0) && (srcImage.height() > 0)) {
QImage image = processSourceImage(srcImage, true);
if (image.format() != QImage::Format_ARGB32) {
image = image.convertToFormat(QImage::Format_ARGB32);
}
gpu::Element formatMip;
gpu::Element formatGPU;
if (isCubeTexturesCompressionEnabled()) {
formatMip = gpu::Element::COLOR_COMPRESSED_SRGBA_HIGH;
formatGPU = gpu::Element::COLOR_COMPRESSED_SRGBA_HIGH;
formatMip = gpu::Element::COLOR_COMPRESSED_HDR_RGB;
formatGPU = gpu::Element::COLOR_COMPRESSED_HDR_RGB;
} else {
formatMip = gpu::Element::COLOR_SRGBA_32;
formatGPU = gpu::Element::COLOR_SRGBA_32;
formatMip = HDR_FORMAT;
formatGPU = HDR_FORMAT;
}
if (image.format() != QIMAGE_HDR_FORMAT) {
image = convertToHDRFormat(image, HDR_FORMAT);
}
// Find the layout of the cubemap in the 2D image
@ -1028,9 +1255,9 @@ gpu::TexturePointer TextureUsage::processCubeTextureColorFromImage(const QImage&
// Generate irradiance while we are at it
if (generateIrradiance) {
PROFILE_RANGE(resource_parse, "generateIrradiance");
auto irradianceTexture = gpu::Texture::createCube(gpu::Element::COLOR_SRGBA_32, faces[0].width(), gpu::Texture::MAX_NUM_MIPS, gpu::Sampler(gpu::Sampler::FILTER_MIN_MAG_MIP_LINEAR, gpu::Sampler::WRAP_CLAMP));
auto irradianceTexture = gpu::Texture::createCube(HDR_FORMAT, faces[0].width(), gpu::Texture::MAX_NUM_MIPS, gpu::Sampler(gpu::Sampler::FILTER_MIN_MAG_MIP_LINEAR, gpu::Sampler::WRAP_CLAMP));
irradianceTexture->setSource(srcImageName);
irradianceTexture->setStoredMipFormat(gpu::Element::COLOR_SBGRA_32);
irradianceTexture->setStoredMipFormat(HDR_FORMAT);
for (uint8 face = 0; face < faces.size(); ++face) {
irradianceTexture->assignStoredMipFace(0, face, faces[face].byteCount(), faces[face].constBits());
}

137
libraries/ktx/src/khronos/KHR.h
View file

@ -209,6 +209,137 @@ namespace khronos {
COMPRESSED_SIGNED_RG11_EAC = 0x9273,
};
inline uint8_t evalUncompressedBlockBitSize(InternalFormat format) {
switch (format) {
case InternalFormat::R8:
case InternalFormat::R8_SNORM:
return 8;
case InternalFormat::R16:
case InternalFormat::R16_SNORM:
case InternalFormat::RG8:
case InternalFormat::RG8_SNORM:
return 16;
case InternalFormat::RG16:
case InternalFormat::RG16_SNORM:
return 16;
case InternalFormat::R3_G3_B2:
return 8;
case InternalFormat::RGB4:
return 12;
case InternalFormat::RGB5:
case InternalFormat::RGB565:
return 16;
case InternalFormat::RGB8:
case InternalFormat::RGB8_SNORM:
return 24;
case InternalFormat::RGB10:
// TODO: check if this is really the case
return 32;
case InternalFormat::RGB12:
// TODO: check if this is really the case
return 48;
case InternalFormat::RGB16:
case InternalFormat::RGB16_SNORM:
return 48;
case InternalFormat::RGBA2:
return 8;
case InternalFormat::RGBA4:
case InternalFormat::RGB5_A1:
return 16;
case InternalFormat::RGBA8:
case InternalFormat::RGBA8_SNORM:
case InternalFormat::RGB10_A2:
case InternalFormat::RGB10_A2UI:
return 32;
case InternalFormat::RGBA12:
return 48;
case InternalFormat::RGBA16:
case InternalFormat::RGBA16_SNORM:
return 64;
case InternalFormat::SRGB8:
return 24;
case InternalFormat::SRGB8_ALPHA8:
return 32;
case InternalFormat::R16F:
return 16;
case InternalFormat::RG16F:
return 32;
case InternalFormat::RGB16F:
return 48;
case InternalFormat::RGBA16F:
return 64;
case InternalFormat::R32F:
return 32;
case InternalFormat::RG32F:
return 64;
case InternalFormat::RGB32F:
return 96;
case InternalFormat::RGBA32F:
return 128;
case InternalFormat::R11F_G11F_B10F:
case InternalFormat::RGB9_E5:
return 32;
case InternalFormat::R8I:
case InternalFormat::R8UI:
return 8;
case InternalFormat::R16I:
case InternalFormat::R16UI:
return 16;
case InternalFormat::R32I:
case InternalFormat::R32UI:
return 32;
case InternalFormat::RG8I:
case InternalFormat::RG8UI:
return 16;
case InternalFormat::RG16I:
case InternalFormat::RG16UI:
return 32;
case InternalFormat::RG32I:
case InternalFormat::RG32UI:
return 64;
case InternalFormat::RGB8I:
case InternalFormat::RGB8UI:
return 24;
case InternalFormat::RGB16I:
case InternalFormat::RGB16UI:
return 48;
case InternalFormat::RGB32I:
case InternalFormat::RGB32UI:
return 96;
case InternalFormat::RGBA8I:
case InternalFormat::RGBA8UI:
return 32;
case InternalFormat::RGBA16I:
case InternalFormat::RGBA16UI:
return 64;
case InternalFormat::RGBA32I:
case InternalFormat::RGBA32UI:
return 128;
case InternalFormat::DEPTH_COMPONENT16:
return 16;
case InternalFormat::DEPTH_COMPONENT24:
return 24;
case InternalFormat::DEPTH_COMPONENT32:
case InternalFormat::DEPTH_COMPONENT32F:
case InternalFormat::DEPTH24_STENCIL8:
return 32;
case InternalFormat::DEPTH32F_STENCIL8:
// TODO : check if this true
return 40;
case InternalFormat::STENCIL_INDEX1:
return 1;
case InternalFormat::STENCIL_INDEX4:
return 4;
case InternalFormat::STENCIL_INDEX8:
return 8;
case InternalFormat::STENCIL_INDEX16:
return 16;
default:
return 0;
}
}
template <uint32_t ALIGNMENT>
inline uint32_t evalAlignedCompressedBlockCount(uint32_t value) {
enum { val = ALIGNMENT && !(ALIGNMENT & (ALIGNMENT - 1)) };
@ -225,6 +356,7 @@ namespace khronos {
case InternalFormat::COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT: // BC3
case InternalFormat::COMPRESSED_RED_RGTC1: // BC4
case InternalFormat::COMPRESSED_RG_RGTC2: // BC5
case InternalFormat::COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT: // BC6
case InternalFormat::COMPRESSED_SRGB_ALPHA_BPTC_UNORM: // BC7
return evalAlignedCompressedBlockCount<4>(value);
@ -241,6 +373,7 @@ namespace khronos {
return 8;
case InternalFormat::COMPRESSED_SRGB_ALPHA_BPTC_UNORM:
case InternalFormat::COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT:
case InternalFormat::COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT:
case InternalFormat::COMPRESSED_RG_RGTC2:
return 16;
@ -250,6 +383,10 @@ namespace khronos {
}
}
inline uint8_t evalCompressedBlockBitSize(InternalFormat format) {
return evalCompressedBlockSize(format) * 8;
}
enum class BaseInternalFormat : uint32_t {
// GL 4.4 Table 8.11
DEPTH_COMPONENT = 0x1902,

17
libraries/ktx/src/ktx/KTX.cpp
View file

@ -54,15 +54,13 @@ uint32_t Header::evalPixelOrBlockDepth(uint32_t level) const {
return evalMipDimension(level, getPixelDepth());
}
size_t Header::evalPixelOrBlockSize() const {
size_t Header::evalPixelOrBlockBitSize() const {
size_t result = 0;
auto format = getGLInternaFormat();
if (isCompressed()) {
auto format = getGLInternaFormat();
result = khronos::gl::texture::evalCompressedBlockSize(format);
result = khronos::gl::texture::evalCompressedBlockBitSize(format);
} else {
// FIXME should really be using the internal format, not the base internal format
auto baseFormat = getGLBaseInternalFormat();
result = khronos::gl::texture::evalComponentCount(baseFormat);
result = khronos::gl::texture::evalUncompressedBlockBitSize(format);
}
if (0 == result) {
@ -73,11 +71,14 @@ size_t Header::evalPixelOrBlockSize() const {
size_t Header::evalRowSize(uint32_t level) const {
auto pixWidth = evalPixelOrBlockWidth(level);
auto pixSize = evalPixelOrBlockSize();
auto pixSize = evalPixelOrBlockBitSize();
if (pixSize == 0) {
return 0;
}
return evalPaddedSize(pixWidth * pixSize);
auto totalByteSize = pixWidth * pixSize;
// Round to the nearest upper byte size
totalByteSize = (totalByteSize / 8) + (((totalByteSize % 8) != 0) & 1);
return evalPaddedSize(totalByteSize);
}
size_t Header::evalFaceSize(uint32_t level) const {

2
libraries/ktx/src/ktx/KTX.h
View file

@ -170,7 +170,7 @@ namespace ktx {
uint32_t evalPixelOrBlockHeight(uint32_t level) const;
uint32_t evalPixelOrBlockDepth(uint32_t level) const;
size_t evalPixelOrBlockSize() const;
size_t evalPixelOrBlockBitSize() const;
size_t evalRowSize(uint32_t level) const;
size_t evalFaceSize(uint32_t level) const;
size_t evalImageSize(uint32_t level) const;