<_whitenotifier-f>
[scopehal-apps] azonenberg a10f8c5 - Initial work on new waterfall rendering shader. Works but doesn't account for pan/zoom of viewport yet
<_whitenotifier-f>
[scopehal-apps] azonenberg 7452429 - Waterfall display in ngscopeclient now works. Fixes #552.
<d1b2>
<johnsel> I see some commits refering to a new rendering shader. Still compute/2d or something more spicy?
<azonenberg>
@johnsel waterfall
<azonenberg>
Still optimizing and tweaking
<d1b2>
<johnsel> Oh cool
<d1b2>
<johnsel> I still have want to experiment with trying to generate mesh based graphs (i.e. measurement, those graphs, not filter graph) at some point
<azonenberg>
you mean drawing waveforms as triangle geometry?
<azonenberg>
i dont think it's possible for that to be faster, you get memory bound and the compute shader renderer on uniform waveforms only reads a single fp32 per sample
<d1b2>
<johnsel> Yep, it's something I want to experiment with regardless. But yeah such in implementation would not be trivial at all
<d1b2>
<johnsel> it's definitely not something you could do with the raw data
<azonenberg>
Anywy this is what i'm working on right now. the waterfall display is significantly faster (~4x) than it was before as a result of moving the waterfall image to GPU local memory vs doing pcie accesses every pixel
<azonenberg>
but i still am slow because of moving the waveform GPU->CPU then integrating the persistence map on CPU then back to GPU to run the shader
<azonenberg>
so next step will be moving the integration path to GPU and cutting those big blue PCIE transfers out
<d1b2>
<johnsel> feeding something into their microscope view that is, not electronics I think