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());
}
}
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());
}
}
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) {
// Rare situation, not optimizied for speed
setOuterPixels(paintingData, 0, 0, paintSize.width(), paintSize.height());
return;
}
int iterations = (absolutePaintRect().width() * absolutePaintRect().height()) / s_minimalRectDimension;
if (!iterations) {
setInteriorPixels(paintingData, clipRight, clipBottom, 0, clipBottom);
return;
}
int stride = clipBottom / iterations;
int chunkCount = (clipBottom + stride - 1) / stride;
WorkQueue::concurrentApply(chunkCount, [&](size_t index) {
int yStart = (stride * index);
int yEnd = std::min<int>(stride * (index + 1), clipBottom);
setInteriorPixels(paintingData, clipRight, clipBottom, yStart, yEnd);
});
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);
}
