inline void FELighting::platformApplyGeneric(LightingData& data, LightSource::PaintingData& paintingData)
{
int optimalThreadNumber = ((data.widthDecreasedByOne - 1) * (data.heightDecreasedByOne - 1)) / s_minimalRectDimension;
if (optimalThreadNumber > 1) {
// Initialize parallel jobs
ParallelJobs<PlatformApplyGenericParameters> parallelJobs(&platformApplyGenericWorker, optimalThreadNumber);
// Fill the parameter array
int job = parallelJobs.numberOfJobs();
if (job > 1) {
// Split the job into "yStep"-sized jobs but there a few jobs that need to be slightly larger since
// yStep * jobs < total size. These extras are handled by the remainder "jobsWithExtra".
const int yStep = (data.heightDecreasedByOne - 1) / job;
const int jobsWithExtra = (data.heightDecreasedByOne - 1) % job;
int yStart = 1;
for (--job; job >= 0; --job) {
PlatformApplyGenericParameters& params = parallelJobs.parameter(job);
params.filter = this;
params.data = data;
params.paintingData = paintingData;
params.yStart = yStart;
yStart += job < jobsWithExtra ? yStep + 1 : yStep;
params.yEnd = yStart;
}
parallelJobs.execute();
return;
}
// Fallback to single threaded mode.
}
platformApplyGenericPaint(data, paintingData, 1, data.heightDecreasedByOne);
}
void FEMorphology::platformApply(PaintingData* paintingData)
{
int optimalThreadNumber = (paintingData->width * paintingData->height) / s_minimalArea;
if (optimalThreadNumber > 1) {
ParallelJobs<PlatformApplyParameters> parallelJobs(&WebCore::FEMorphology::platformApplyWorker, optimalThreadNumber);
int numOfThreads = parallelJobs.numberOfJobs();
if (numOfThreads > 1) {
// Split the job into "jobSize"-sized jobs but there a few jobs that need to be slightly larger since
// jobSize * jobs < total size. These extras are handled by the remainder "jobsWithExtra".
const int jobSize = paintingData->height / numOfThreads;
const int jobsWithExtra = paintingData->height % numOfThreads;
int currentY = 0;
for (int job = numOfThreads - 1; job >= 0; --job) {
PlatformApplyParameters& param = parallelJobs.parameter(job);
param.filter = this;
param.startY = currentY;
currentY += job < jobsWithExtra ? jobSize + 1 : jobSize;
param.endY = currentY;
param.paintingData = paintingData;
}
parallelJobs.execute();
return;
}
// Fallback to single thread model
}
platformApplyGeneric(paintingData, 0, paintingData->height);
}
void FEMorphology::platformApply(PaintingData* paintingData)
{
#if ENABLE(PARALLEL_JOBS)
int optimalThreadNumber = (paintingData->width * paintingData->height) / s_minimalArea;
if (optimalThreadNumber > 1) {
ParallelJobs<PlatformApplyParameters> parallelJobs(&WebCore::FEMorphology::platformApplyWorker, optimalThreadNumber);
int numOfThreads = parallelJobs.numberOfJobs();
if (numOfThreads > 1) {
const int deltaY = 1 + paintingData->height / numOfThreads;
int currentY = 0;
for (int job = numOfThreads - 1; job >= 0; --job) {
PlatformApplyParameters& param = parallelJobs.parameter(job);
param.filter = this;
param.startY = currentY;
currentY += deltaY;
param.endY = job ? currentY : paintingData->height;
param.paintingData = paintingData;
}
parallelJobs.execute();
return;
}
// Fallback to single thread model
}
#endif
platformApplyGeneric(paintingData, 0, paintingData->height);
}
inline void FELighting::platformApplyGeneric(LightingData& data, LightSource::PaintingData& paintingData)
{
int optimalThreadNumber = ((data.widthDecreasedByOne - 1) * (data.heightDecreasedByOne - 1)) / s_minimalRectDimension;
if (optimalThreadNumber > 1) {
// Initialize parallel jobs
WTF::ParallelJobs<PlatformApplyGenericParameters> parallelJobs(&platformApplyGenericWorker, optimalThreadNumber);
// Fill the parameter array
int job = parallelJobs.numberOfJobs();
if (job > 1) {
int yStart = 1;
int yStep = (data.heightDecreasedByOne - 1) / job;
for (--job; job >= 0; --job) {
PlatformApplyGenericParameters& params = parallelJobs.parameter(job);
params.filter = this;
params.data = data;
params.paintingData = paintingData;
params.yStart = yStart;
if (job > 0) {
params.yEnd = yStart + yStep;
yStart += yStep;
} else
params.yEnd = data.heightDecreasedByOne;
}
parallelJobs.execute();
return;
}
// Fallback to single threaded mode.
}
platformApplyGenericPaint(data, paintingData, 1, data.heightDecreasedByOne);
}
void FETurbulence::apply()
{
if (hasResult())
return;
ByteArray* pixelArray = createUnmultipliedImageResult();
if (!pixelArray)
return;
if (absolutePaintRect().isEmpty())
return;
PaintingData paintingData(m_seed, roundedIntSize(filterPrimitiveSubregion().size()));
initPaint(paintingData);
#if ENABLE(PARALLEL_JOBS)
int optimalThreadNumber = (absolutePaintRect().width() * absolutePaintRect().height()) / s_minimalRectDimension;
if (optimalThreadNumber > 1) {
// Initialize parallel jobs
ParallelJobs<FillRegionParameters> parallelJobs(&WebCore::FETurbulence::fillRegionWorker, optimalThreadNumber);
// Fill the parameter array
int i = parallelJobs.numberOfJobs();
if (i > 1) {
int startY = 0;
int stepY = absolutePaintRect().height() / i;
for (; i > 0; --i) {
FillRegionParameters& params = parallelJobs.parameter(i-1);
params.filter = this;
params.pixelArray = pixelArray;
params.paintingData = &paintingData;
params.startY = startY;
if (i != 1) {
params.endY = startY + stepY;
startY = startY + stepY;
} else
params.endY = absolutePaintRect().height();
}
// Execute parallel jobs
parallelJobs.execute();
return;
}
}
// Fallback to sequential mode if there is no room for a new thread or the paint area is too small
#endif // ENABLE(PARALLEL_JOBS)
fillRegion(pixelArray, paintingData, 0, absolutePaintRect().height());
}
void FETurbulence::applySoftware()
{
Uint8ClampedArray* pixelArray = createUnmultipliedImageResult();
if (!pixelArray)
return;
if (absolutePaintRect().isEmpty()) {
pixelArray->zeroFill();
return;
}
PaintingData paintingData(m_seed, roundedIntSize(filterPrimitiveSubregion().size()));
initPaint(paintingData);
int optimalThreadNumber = (absolutePaintRect().width() * absolutePaintRect().height()) / s_minimalRectDimension;
if (optimalThreadNumber > 1) {
// Initialize parallel jobs
ParallelJobs<FillRegionParameters> parallelJobs(&WebCore::FETurbulence::fillRegionWorker, optimalThreadNumber);
// Fill the parameter array
int i = parallelJobs.numberOfJobs();
if (i > 1) {
// Split the job into "stepY"-sized jobs but there a few jobs that need to be slightly larger since
// stepY * jobs < total size. These extras are handled by the remainder "jobsWithExtra".
const int stepY = absolutePaintRect().height() / i;
const int jobsWithExtra = absolutePaintRect().height() % i;
int startY = 0;
for (; i > 0; --i) {
FillRegionParameters& params = parallelJobs.parameter(i-1);
params.filter = this;
params.pixelArray = pixelArray;
params.paintingData = &paintingData;
params.startY = startY;
startY += i < jobsWithExtra ? stepY + 1 : stepY;
params.endY = startY;
params.baseFrequencyX = m_baseFrequencyX;
params.baseFrequencyY = m_baseFrequencyY;
}
// Execute parallel jobs
parallelJobs.execute();
return;
}
}
// Fallback to single threaded mode if there is no room for a new thread or the paint area is too small.
fillRegion(pixelArray, paintingData, 0, absolutePaintRect().height(), m_baseFrequencyX, m_baseFrequencyY);
}
void FEConvolveMatrix::platformApplySoftware()
{
FilterEffect* in = inputEffect(0);
Uint8ClampedArray* resultImage;
if (m_preserveAlpha)
resultImage = createUnmultipliedImageResult();
else
resultImage = createPremultipliedImageResult();
if (!resultImage)
return;
IntRect effectDrawingRect = requestedRegionOfInputImageData(in->absolutePaintRect());
RefPtr<Uint8ClampedArray> srcPixelArray;
if (m_preserveAlpha)
srcPixelArray = in->asUnmultipliedImage(effectDrawingRect);
else
srcPixelArray = in->asPremultipliedImage(effectDrawingRect);
IntSize paintSize = absolutePaintRect().size();
PaintingData paintingData;
paintingData.srcPixelArray = srcPixelArray.get();
paintingData.dstPixelArray = resultImage;
paintingData.width = paintSize.width();
paintingData.height = paintSize.height();
paintingData.bias = m_bias * 255;
// Drawing fully covered pixels
int clipRight = paintSize.width() - m_kernelSize.width();
int clipBottom = paintSize.height() - m_kernelSize.height();
if (clipRight >= 0 && clipBottom >= 0) {
int optimalThreadNumber = (absolutePaintRect().width() * absolutePaintRect().height()) / s_minimalRectDimension;
if (optimalThreadNumber > 1) {
WTF::ParallelJobs<InteriorPixelParameters> parallelJobs(&WebCore::FEConvolveMatrix::setInteriorPixelsWorker, optimalThreadNumber);
const int numOfThreads = parallelJobs.numberOfJobs();
// Split the job into "heightPerThread" jobs but there a few jobs that need to be slightly larger since
// heightPerThread * jobs < total size. These extras are handled by the remainder "jobsWithExtra".
const int heightPerThread = clipBottom / numOfThreads;
const int jobsWithExtra = clipBottom % numOfThreads;
int startY = 0;
for (int job = 0; job < numOfThreads; ++job) {
InteriorPixelParameters& param = parallelJobs.parameter(job);
param.filter = this;
param.paintingData = &paintingData;
param.clipRight = clipRight;
param.clipBottom = clipBottom;
param.yStart = startY;
startY += job < jobsWithExtra ? heightPerThread + 1 : heightPerThread;
param.yEnd = startY;
}
parallelJobs.execute();
} else {
// Fallback to single threaded mode.
setInteriorPixels(paintingData, clipRight, clipBottom, 0, clipBottom);
}
clipRight += m_targetOffset.x() + 1;
clipBottom += m_targetOffset.y() + 1;
if (m_targetOffset.y() > 0)
setOuterPixels(paintingData, 0, 0, paintSize.width(), m_targetOffset.y());
if (clipBottom < paintSize.height())
setOuterPixels(paintingData, 0, clipBottom, paintSize.width(), paintSize.height());
if (m_targetOffset.x() > 0)
setOuterPixels(paintingData, 0, m_targetOffset.y(), m_targetOffset.x(), clipBottom);
if (clipRight < paintSize.width())
setOuterPixels(paintingData, clipRight, m_targetOffset.y(), paintSize.width(), clipBottom);
} else {
// Rare situation, not optimizied for speed
setOuterPixels(paintingData, 0, 0, paintSize.width(), paintSize.height());
}
}
inline void FEGaussianBlur::platformApply(Uint8ClampedArray* srcPixelArray, Uint8ClampedArray* tmpPixelArray, unsigned kernelSizeX, unsigned kernelSizeY, IntSize& paintSize)
{
int scanline = 4 * paintSize.width();
int extraHeight = 3 * kernelSizeY * 0.5f;
int optimalThreadNumber = (paintSize.width() * paintSize.height()) / (s_minimalRectDimension + extraHeight * paintSize.width());
if (optimalThreadNumber > 1) {
WTF::ParallelJobs<PlatformApplyParameters> parallelJobs(&platformApplyWorker, optimalThreadNumber);
int jobs = parallelJobs.numberOfJobs();
if (jobs > 1) {
int blockHeight = paintSize.height() / jobs;
--jobs;
for (int job = jobs; job >= 0; --job) {
PlatformApplyParameters& params = parallelJobs.parameter(job);
params.filter = this;
int startY;
int endY;
if (!job) {
startY = 0;
endY = blockHeight + extraHeight;
params.srcPixelArray = srcPixelArray;
params.dstPixelArray = tmpPixelArray;
} else {
if (job == jobs) {
startY = job * blockHeight - extraHeight;
endY = paintSize.height();
} else {
startY = job * blockHeight - extraHeight;
endY = (job + 1) * blockHeight + extraHeight;
}
int blockSize = (endY - startY) * scanline;
params.srcPixelArray = Uint8ClampedArray::createUninitialized(blockSize);
params.dstPixelArray = Uint8ClampedArray::createUninitialized(blockSize);
memcpy(params.srcPixelArray->data(), srcPixelArray->data() + startY * scanline, blockSize);
}
params.width = paintSize.width();
params.height = endY - startY;
params.kernelSizeX = kernelSizeX;
params.kernelSizeY = kernelSizeY;
}
parallelJobs.execute();
// Copy together the parts of the image.
for (int job = jobs; job >= 1; --job) {
PlatformApplyParameters& params = parallelJobs.parameter(job);
int sourceOffset;
int destinationOffset;
int size;
if (job == jobs) {
sourceOffset = extraHeight * scanline;
destinationOffset = job * blockHeight * scanline;
size = (paintSize.height() - job * blockHeight) * scanline;
} else {
sourceOffset = extraHeight * scanline;
destinationOffset = job * blockHeight * scanline;
size = blockHeight * scanline;
}
memcpy(srcPixelArray->data() + destinationOffset, params.srcPixelArray->data() + sourceOffset, size);
}
return;
}
// Fallback to single threaded mode.
}
// The selection here eventually should happen dynamically on some platforms.
#if CPU(ARM_NEON) && COMPILER(GCC)
platformApplyNeon(srcPixelArray, tmpPixelArray, kernelSizeX, kernelSizeY, paintSize);
#else
platformApplyGeneric(srcPixelArray, tmpPixelArray, kernelSizeX, kernelSizeY, paintSize);
#endif
}
void FEConvolveMatrix::apply()
{
if (hasResult())
return;
FilterEffect* in = inputEffect(0);
in->apply();
if (!in->hasResult())
return;
ByteArray* resultImage;
if (m_preserveAlpha)
resultImage = createUnmultipliedImageResult();
else
resultImage = createPremultipliedImageResult();
if (!resultImage)
return;
IntRect effectDrawingRect = requestedRegionOfInputImageData(in->absolutePaintRect());
RefPtr<ByteArray> srcPixelArray;
if (m_preserveAlpha)
srcPixelArray = in->asUnmultipliedImage(effectDrawingRect);
else
srcPixelArray = in->asPremultipliedImage(effectDrawingRect);
IntSize paintSize = absolutePaintRect().size();
PaintingData paintingData;
paintingData.srcPixelArray = srcPixelArray.get();
paintingData.dstPixelArray = resultImage;
paintingData.width = paintSize.width();
paintingData.height = paintSize.height();
paintingData.bias = m_bias * 255;
// Drawing fully covered pixels
int clipRight = paintSize.width() - m_kernelSize.width();
int clipBottom = paintSize.height() - m_kernelSize.height();
if (clipRight >= 0 && clipBottom >= 0) {
#if ENABLE(PARALLEL_JOBS)
int optimalThreadNumber = (absolutePaintRect().width() * absolutePaintRect().height()) / s_minimalRectDimension;
if (optimalThreadNumber > 1) {
ParallelJobs<InteriorPixelParameters> parallelJobs(&WebCore::FEConvolveMatrix::setInteriorPixelsWorker, optimalThreadNumber);
const int numOfThreads = parallelJobs.numberOfJobs();
const int heightPerThread = clipBottom / numOfThreads;
int startY = 0;
for (int job = 0; job < numOfThreads; ++job) {
InteriorPixelParameters& param = parallelJobs.parameter(job);
param.filter = this;
param.paintingData = &paintingData;
param.clipRight = clipRight;
param.clipBottom = clipBottom;
param.yStart = startY;
if (job < numOfThreads - 1) {
startY += heightPerThread;
param.yEnd = startY - 1;
} else
param.yEnd = clipBottom;
}
parallelJobs.execute();
} else
// Fallback to the default setInteriorPixels call.
#endif
setInteriorPixels(paintingData, clipRight, clipBottom, 0, clipBottom);
clipRight += m_targetOffset.x() + 1;
clipBottom += m_targetOffset.y() + 1;
if (m_targetOffset.y() > 0)
setOuterPixels(paintingData, 0, 0, paintSize.width(), m_targetOffset.y());
if (clipBottom < paintSize.height())
setOuterPixels(paintingData, 0, clipBottom, paintSize.width(), paintSize.height());
if (m_targetOffset.x() > 0)
setOuterPixels(paintingData, 0, m_targetOffset.y(), m_targetOffset.x(), clipBottom);
if (clipRight < paintSize.width())
setOuterPixels(paintingData, clipRight, m_targetOffset.y(), paintSize.width(), clipBottom);
} else {
// Rare situation, not optimizied for speed
setOuterPixels(paintingData, 0, 0, paintSize.width(), paintSize.height());
}
}
inline void FEGaussianBlur::platformApply(Uint8ClampedArray* srcPixelArray, Uint8ClampedArray* tmpPixelArray, unsigned kernelSizeX, unsigned kernelSizeY, IntSize& paintSize)
{
int scanline = 4 * paintSize.width();
int extraHeight = 3 * kernelSizeY * 0.5f;
int optimalThreadNumber = (paintSize.width() * paintSize.height()) / (s_minimalRectDimension + extraHeight * paintSize.width());
if (optimalThreadNumber > 1) {
WTF::ParallelJobs<PlatformApplyParameters> parallelJobs(&platformApplyWorker, optimalThreadNumber);
int jobs = parallelJobs.numberOfJobs();
if (jobs > 1) {
// Split the job into "blockHeight"-sized jobs but there a few jobs that need to be slightly larger since
// blockHeight * jobs < total size. These extras are handled by the remainder "jobsWithExtra".
const int blockHeight = paintSize.height() / jobs;
const int jobsWithExtra = paintSize.height() % jobs;
int currentY = 0;
for (int job = 0; job < jobs; job++) {
PlatformApplyParameters& params = parallelJobs.parameter(job);
params.filter = this;
int startY = !job ? 0 : currentY - extraHeight;
currentY += job < jobsWithExtra ? blockHeight + 1 : blockHeight;
int endY = job == jobs - 1 ? currentY : currentY + extraHeight;
int blockSize = (endY - startY) * scanline;
if (!job) {
params.srcPixelArray = srcPixelArray;
params.dstPixelArray = tmpPixelArray;
} else {
params.srcPixelArray = Uint8ClampedArray::createUninitialized(blockSize);
params.dstPixelArray = Uint8ClampedArray::createUninitialized(blockSize);
memcpy(params.srcPixelArray->data(), srcPixelArray->data() + startY * scanline, blockSize);
}
params.width = paintSize.width();
params.height = endY - startY;
params.kernelSizeX = kernelSizeX;
params.kernelSizeY = kernelSizeY;
}
parallelJobs.execute();
// Copy together the parts of the image.
currentY = 0;
for (int job = 1; job < jobs; job++) {
PlatformApplyParameters& params = parallelJobs.parameter(job);
int sourceOffset;
int destinationOffset;
int size;
int adjustedBlockHeight = job < jobsWithExtra ? blockHeight + 1 : blockHeight;
currentY += adjustedBlockHeight;
sourceOffset = extraHeight * scanline;
destinationOffset = currentY * scanline;
size = adjustedBlockHeight * scanline;
memcpy(srcPixelArray->data() + destinationOffset, params.srcPixelArray->data() + sourceOffset, size);
}
return;
}
// Fallback to single threaded mode.
}
// The selection here eventually should happen dynamically on some platforms.
platformApplyGeneric(srcPixelArray, tmpPixelArray, kernelSizeX, kernelSizeY, paintSize);
}