Overlapping data

In the end there would be two labels created for one selection ID...probably there's some issue with default value in the GPU implementation

1D version: all shader changes?


Milestone: working GPU version of max distance

key was simplifying to 1D access and directly getting UVs from ID buffer
Delete debug memory copying

already a significant speedup
small code clean up


Use devicePixelRatio in determining final label position


unify, remove code duplicity

It looks like there's not really a way to optimize this implementation and it will just stay as a reference

WIP: continued work on GPU version of max distance part


WIP: no webgpu errors


WIP: trying to get to correct results


WIP: cleaning up trying to figure out stuff


MaxDT: attempt using atomics

speed looks pretty fine, but there seem to be still racing conditions: getting very jittery results

I believe this anyway will be reverted because adding "fake" data points is a stupid thing to do...

semi-transparent colored bg boxes


Making labeling optional (and disabled by default)


Don't show debug viewport by default on startup

In order to be able to output it to the ID attachment in G buffer and use it for label placement
WIP: interfacing selections from ChromatinViewport to LabelLayoutGener.


WIP: im just confused


Redo interfacing layout generator with current selections

Prepared for final version, for now it doesn't work only because I'm not getting the right IDs in the ID attachment of g buffer
Computing selection ID information for bins and sending to gpu write

Doesn't work at this point (still getting zeroed out buffer values in shader)
Guess I figured out the problem

WIP: refactoring big ass code file


Delete removed code


nother small refactor


WIP: figuring out coordinate conversions


comment out some debug outputting

There's still a bug where the behaviour randomly switches for some reason

starting DT cpu side


WIP: DT step compiles


Global Blit function (WIP)

Ideally I'd like to unify it so that you don't have to have different versions for regular/float textures and different formats...

Distance Transform seems to be somewhat working...

The issues was in initial contour seed value.
It also seems that there is some scaling problem
Found the bug in DT


Small cleanup


WIP: getting label placement using maximum distance from contour seed


WIP: whole labeling process somehow works

but is extremely unstable plus i had to do some magic constants to get the position to match

Another canvas overlaid for labeling debug views

Was having trouble with events not going through the top canvas, pointerEvents: none seems to work
WIP: setting up the debug viewport/canvas for debug blitting


WIP: sets up and clear the second canvas


WIP: Bliting texture seems to work


WIP: adding reference to labeling generator (to access debug textures)


dispatch -> dispatchWorkgroups


WIP: contours compute pass not working yet


switch back to showing contours buffer


Thanks for Matus, contour pass finally computes


Uncommenting shader code and stuff


WIP: adding ID texture to G buffer

thought I'd only do it for spheres and rounded cones but it produces invalid pipelines right now
Set up selection ID attachment in G buffer rendering


Matus saved the day once  :-)

WIP: refactoring labeling layout / debug viewport setup


WIP: restructured, outlines showing in debug viewport again


Moved webgpu objects to common graphics library places


Contours: output UVs for distance transform seed

still not working but I have all the webgpu structures in place

WIP: interfacing labeling algs with component/graphics lib

Testing with SVG overlaid over the canvas for now
Playing around with labels in SVG over canvas