Downsampling average/peak mode slow with large datasets

[oyvindlr]

The idea of downsampling is to speed up the painting by having fewer data points. But when there are sufficiently many points in view, the downsampling-processing itself slows down the updating significantly (except when using the subsampling method).

A possible fix may be to keep a cached version of the downsampled data. This would work well for fixed downsampling, but for auto-mode it becomes a bit more complicated. One could possibly create a cached version at a new "maximum" downsampling rate, which is always used as long as the amount of data in view is greater than some threshold, where the threshold should take into consideration the tradeoff between painting many points and downsampling processing. Below this threshold, one would keep the current downsampling method, which is fast enough for smaller amounts of data.

By the way, this is an issue that becomes way more pronounced when there are multiple plotDataItems in the plot, even with the same total amount of points. Enhancements as discussed in #2777 would be highly welcome.

[pijyoi]

For the situations with large data sets that warrant enabling "autoDownSampling", I would usually also enable "clipToView".
Of course, your definition of what constitutes "large" may be different from mine.

import pyqtgraph as pg
import numpy as np

rng = np.random.default_rng(0)
y = rng.uniform(-1, 1, size=100_000_000)
x = np.arange(y.size, dtype=np.float64)

pg.mkQApp()

pw = pg.PlotWidget()
pw.setMouseEnabled(True, False)
pw.setClipToView(True)
pw.setDownsampling(auto=True)
pdi = pw.plot()
pdi.setData(x, y)

pw.show()
pg.exec()