Mercurial > hg > svgui
view layer/Colour3DPlotRenderer.cpp @ 1090:c8c747783110 spectrogram-minor-refactor
Cut over to using the renderer (though it's very incomplete) and fix some cache problems and pointer lifetime misunderstandings
author | Chris Cannam |
---|---|
date | Tue, 05 Jul 2016 17:48:26 +0100 |
parents | db976e9f385a |
children | 8a815776151c |
line wrap: on
line source
/* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */ /* Sonic Visualiser An audio file viewer and annotation editor. Centre for Digital Music, Queen Mary, University of London. This file copyright 2006-2016 Chris Cannam and QMUL. This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. See the file COPYING included with this distribution for more information. */ #include "Colour3DPlotRenderer.h" #include "RenderTimer.h" #include "data/model/DenseThreeDimensionalModel.h" #include "data/model/Dense3DModelPeakCache.h" #include "data/model/FFTModel.h" #include "LayerGeometryProvider.h" #include "VerticalBinLayer.h" #include <vector> using namespace std; Colour3DPlotRenderer::RenderResult Colour3DPlotRenderer::render(LayerGeometryProvider *v, QPainter &paint, QRect rect) { return render(v, paint, rect, false); } Colour3DPlotRenderer::RenderResult Colour3DPlotRenderer::renderTimeConstrained(LayerGeometryProvider *v, QPainter &paint, QRect rect) { return render(v, paint, rect, true); } Colour3DPlotRenderer::RenderResult Colour3DPlotRenderer::render(LayerGeometryProvider *v, QPainter &paint, QRect rect, bool timeConstrained) { sv_frame_t startFrame = v->getStartFrame(); int x0 = v->getXForViewX(rect.x()); int x1 = v->getXForViewX(rect.x() + rect.width()); if (x0 < 0) x0 = 0; if (x1 > v->getPaintWidth()) x1 = v->getPaintWidth(); m_cache.resize(v->getPaintSize()); m_cache.setZoomLevel(v->getZoomLevel()); cerr << "cache start " << m_cache.getStartFrame() << " view start " << startFrame << " valid left " << m_cache.getValidLeft() << " valid right " << m_cache.getValidRight() << " x0 " << x0 << " x1 " << x1 << endl; if (m_cache.isValid()) { // some part of the cache is valid if (v->getXForFrame(m_cache.getStartFrame()) == v->getXForFrame(startFrame) && m_cache.getValidLeft() <= x0 && m_cache.getValidRight() >= x1) { cerr << "cache hit" << endl; // cache is valid for the complete requested area paint.drawImage(rect, m_cache.getImage(), rect); return { rect, {} }; } else { cerr << "cache partial hit" << endl; // cache doesn't begin at the right frame or doesn't // contain the complete view, but might be scrollable or // partially usable m_cache.scrollTo(v, startFrame); // if we are not time-constrained, then we want to paint // the whole area in one go; we don't return a partial // paint. To avoid providing the more complex logic to // handle painting discontiguous areas, if the only valid // part of cache is in the middle, just make the whole // thing invalid and start again. if (!timeConstrained) { if (m_cache.getValidLeft() > x0 && m_cache.getValidRight() < x1) { m_cache.invalidate(); } } } } else { // cache completely invalid m_cache.setStartFrame(startFrame); } bool rightToLeft = false; if (!m_cache.isValid() && timeConstrained) { // When rendering the whole area, in a context where we might // not be able to complete the work, start from somewhere near // the middle so that the region of interest appears first //!!! (perhaps we should avoid doing this if past repaints //!!! have been fast enough to do the whole in one shot) if (x0 == 0 && x1 == v->getPaintWidth()) { x0 = int(x1 * 0.3); } } if (m_cache.isValid()) { cerr << "cache somewhat valid" << endl; // When rendering only a part of the cache, we need to make // sure that the part we're rendering is adjacent to (or // overlapping) a valid area of cache, if we have one. The // alternative is to ditch the valid area of cache and render // only the requested area, but that's risky because this can // happen when just waving the pointer over a small part of // the view -- if we lose the partly-built cache every time // the user does that, we'll never finish building it. int left = x0; int width = x1 - x0; bool isLeftOfValidArea = false; m_cache.adjustToTouchValidArea(left, width, isLeftOfValidArea); x0 = left; x1 = x0 + width; // That call also told us whether we should be painting // sub-regions of our target region in right-to-left order in // order to ensure contiguity rightToLeft = isLeftOfValidArea; } renderToCache(v, x0, x1 - x0, rightToLeft, timeConstrained); QRect pr = rect & m_cache.getValidArea(); paint.drawImage(pr.x(), pr.y(), m_cache.getImage(), pr.x(), pr.y(), pr.width(), pr.height()); if (!timeConstrained && (pr != rect)) { //!!! on a first cut, there is a risk that this will happen //!!! when we are at start/end of model -- trap, report, and //!!! then fix throw std::logic_error("internal error: failed to render entire requested rect even when not time-constrained"); } return { pr, {} }; //!!! todo: timing/incomplete paint //!!! todo: peak frequency style //!!! todo: transparent style from Colour3DPlot //!!! todo: bin boundary alignment when in BinResolution //!!! todo: view magnitudes / normalise visible area //!!! todo: alter documentation for view mag stuff (cached paints //!!! do not update MagnitudeRange) //!!! todo, here or in caller: illuminateLocalFeatures //!!! fft model scaling? //!!! should we own the Dense3DModelPeakCache here? or should it persist } void Colour3DPlotRenderer::renderToCache(LayerGeometryProvider *v, int x0, int repaintWidth, bool rightToLeft, bool timeConstrained) { // Draw to the draw buffer, and then scale-copy from there. DenseThreeDimensionalModel *model = m_sources.source; if (!model || !model->isOK() || !model->isReady()) { throw std::logic_error("no source model provided, or model not ready"); } // The draw buffer contains a fragment at either our pixel // resolution (if there is more than one time-bin per pixel) or // time-bin resolution (if a time-bin spans more than one pixel). // We need to ensure that it starts and ends at points where a // time-bin boundary occurs at an exact pixel boundary, and with a // certain amount of overlap across existing pixels so that we can // scale and draw from it without smoothing errors at the edges. // If (getFrameForX(x) / increment) * increment == // getFrameForX(x), then x is a time-bin boundary. We want two // such boundaries at either side of the draw buffer -- one which // we draw up to, and one which we subsequently crop at. bool bufferIsBinResolution = false; int binResolution = model->getResolution(); int zoomLevel = v->getZoomLevel(); if (binResolution > zoomLevel) bufferIsBinResolution = true; sv_frame_t leftBoundaryFrame = -1, leftCropFrame = -1; sv_frame_t rightBoundaryFrame = -1, rightCropFrame = -1; int drawWidth; if (bufferIsBinResolution) { for (int x = x0; ; --x) { sv_frame_t f = v->getFrameForX(x); if ((f / binResolution) * binResolution == f) { if (leftCropFrame == -1) leftCropFrame = f; else if (x < x0 - 2) { leftBoundaryFrame = f; break; } } } for (int x = x0 + repaintWidth; ; ++x) { sv_frame_t f = v->getFrameForX(x); if ((f / binResolution) * binResolution == f) { if (rightCropFrame == -1) rightCropFrame = f; else if (x > x0 + repaintWidth + 2) { rightBoundaryFrame = f; break; } } } drawWidth = int((rightBoundaryFrame - leftBoundaryFrame) / binResolution); } else { drawWidth = repaintWidth; } // We always paint the full height. Smaller heights can be used // when painting direct from cache (outside this function), but we // want to ensure the cache is coherent without having to worry // about vertical matching of required and valid areas as well as // horizontal. That's why this function didn't take any y/height // parameters. int h = v->getPaintHeight(); clearDrawBuffer(drawWidth, h); vector<int> binforx(drawWidth); vector<double> binfory(h); bool usePeaksCache = false; int binsPerPeak = 1; if (bufferIsBinResolution) { for (int x = 0; x < drawWidth; ++x) { binforx[x] = int(leftBoundaryFrame / binResolution) + x; } // calculating boundaries later will be too fiddly for partial // paints, and painting should be fast anyway when this is the // case because it means we're well zoomed in timeConstrained = false; } else { for (int x = 0; x < drawWidth; ++x) { sv_frame_t f0 = v->getFrameForX(x0 + x); double s0 = double(f0 - model->getStartFrame()) / binResolution; binforx[x] = int(s0 + 0.0001); } if (m_sources.peaks) { // peaks cache exists binsPerPeak = m_sources.peaks->getColumnsPerPeak(); usePeaksCache = (binResolution * binsPerPeak) < zoomLevel; if (m_params.colourScale.getScale() == ColourScale::PhaseColourScale) { usePeaksCache = false; } } } for (int y = 0; y < h; ++y) { binfory[y] = m_sources.verticalBinLayer->getBinForY(v, h - y - 1); } int attainedWidth = renderDrawBuffer(repaintWidth, h, binforx, binfory, usePeaksCache, rightToLeft, timeConstrained); //!!! now scale-copy to cache if (attainedWidth == 0) return; int paintedLeft = x0; if (rightToLeft) { paintedLeft += (repaintWidth - attainedWidth); } if (bufferIsBinResolution) { int scaledLeft = v->getXForFrame(leftBoundaryFrame); int scaledRight = v->getXForFrame(rightBoundaryFrame); QImage scaled = m_drawBuffer.scaled (scaledRight - scaledLeft, h, Qt::IgnoreAspectRatio, (m_params.interpolate ? Qt::SmoothTransformation : Qt::FastTransformation)); int scaledLeftCrop = v->getXForFrame(leftCropFrame); int scaledRightCrop = v->getXForFrame(rightCropFrame); int targetLeft = scaledLeftCrop; if (targetLeft < 0) { targetLeft = 0; } int targetWidth = scaledRightCrop - targetLeft; if (targetLeft + targetWidth > m_cache.getSize().width()) { targetWidth = m_cache.getSize().width() - targetLeft; } int sourceLeft = targetLeft - scaledLeft; if (sourceLeft < 0) { sourceLeft = 0; } int sourceWidth = targetWidth; if (targetWidth > 0) { m_cache.drawImage(targetLeft, targetWidth, scaled, sourceLeft, sourceWidth); } } else { m_cache.drawImage(paintedLeft, attainedWidth, m_drawBuffer, paintedLeft - x0, attainedWidth); } } int Colour3DPlotRenderer::renderDrawBuffer(int w, int h, const vector<int> &binforx, const vector<double> &binfory, bool usePeaksCache, bool rightToLeft, bool timeConstrained) { // Callers must have checked that the appropriate subset of // Sources data members are set for the supplied flags (e.g. that // peaks model exists if usePeaksCache) RenderTimer timer(timeConstrained ? RenderTimer::FastRender : RenderTimer::NoTimeout); int minbin = int(binfory[0] + 0.0001); int maxbin = int(binfory[h-1]); if (minbin < 0) minbin = 0; if (maxbin < 0) maxbin = minbin+1; int divisor = 1; DenseThreeDimensionalModel *sourceModel = m_sources.source; if (usePeaksCache) { divisor = m_sources.peaks->getColumnsPerPeak(); sourceModel = m_sources.peaks; } int psx = -1; int start = 0; int finish = w; int step = 1; if (rightToLeft) { start = w-1; finish = -1; step = -1; } int columnCount = 0; vector<float> preparedColumn; for (int x = start; x != finish; x += step) { // x is the on-canvas pixel coord; sx (later) will be the // source column index ++columnCount; if (binforx[x] < 0) continue; int sx0 = binforx[x] / divisor; int sx1 = sx0; if (x+1 < w) sx1 = binforx[x+1] / divisor; if (sx0 < 0) sx0 = sx1 - 1; if (sx0 < 0) continue; if (sx1 <= sx0) sx1 = sx0 + 1; vector<float> pixelPeakColumn; for (int sx = sx0; sx < sx1; ++sx) { #ifdef DEBUG_SPECTROGRAM_REPAINT // cerr << "sx = " << sx << endl; #endif if (sx < 0 || sx >= sourceModel->getWidth()) { continue; } if (sx != psx) { // order: // get column -> scale -> record extents -> // normalise -> peak pick -> apply display gain -> // distribute/interpolate ColumnOp::Column fullColumn = sourceModel->getColumn(sx); cerr << "x " << x << ", sx " << sx << ", col height " << fullColumn.size() << ", minbin " << minbin << ", maxbin " << maxbin << endl; ColumnOp::Column column = vector<float>(fullColumn.data() + minbin, fullColumn.data() + maxbin + 1); //!!! fft scale if (m_colourScale != PhaseColourScale) { // column = ColumnOp::fftScale(column, m_fftSize); // } //!!! extents recordColumnExtents(column, // sx, // overallMag, // overallMagChanged); // if (m_colourScale != PhaseColourScale) { column = ColumnOp::normalize(column, m_params.normalization); // } if (m_params.binDisplay == PeakBins) { column = ColumnOp::peakPick(column); } preparedColumn = ColumnOp::distribute(column, //!!! gain? ColumnOp::applyGain(column, m_gain), h, binfory, minbin, m_params.interpolate); psx = sx; } if (sx == sx0) { pixelPeakColumn = preparedColumn; } else { for (int i = 0; in_range_for(pixelPeakColumn, i); ++i) { pixelPeakColumn[i] = std::max(pixelPeakColumn[i], preparedColumn[i]); } } } if (!pixelPeakColumn.empty()) { for (int y = 0; y < h; ++y) { m_drawBuffer.setPixel (x, h-y-1, m_params.colourScale.getPixel(pixelPeakColumn[y])); } } double fractionComplete = double(columnCount) / double(w); if (timer.outOfTime(fractionComplete)) { return columnCount; } } return columnCount; } void Colour3DPlotRenderer::clearDrawBuffer(int w, int h) { if (m_drawBuffer.width() < w || m_drawBuffer.height() != h) { m_drawBuffer = QImage(w, h, QImage::Format_Indexed8); for (int pixel = 0; pixel < 256; ++pixel) { //!!! todo: colour rotation (here 0) m_drawBuffer.setColor ((unsigned char)pixel, m_params.colourScale.getColourForPixel(pixel, 0).rgb()); } } m_drawBuffer.fill(0); }