<azonenberg>
the scopehal side of the refactor is done
<azonenberg>
still working on the glscopeclient side, so the submodule in scopehal-apps will continue to point to the pre-refactor HEAD for another day or so
<_whitenotifier-7>
[scopehal] electroniceel e6b1201 - install the compiled .spv filter files into share/glscopeclient/shaders glscopeclient won't start if it can't load these shader files
<_whitenotifier-7>
[scopehal] azonenberg 2906b8f - Merge pull request #671 from electroniceel/spv-install install the compiled .spv filter files into share/glscopeclient/shaders
<_whitenotifier-7>
[scopehal-apps] azonenberg 18304c3 - Updated submodules and copyright date in header
<_whitenotifier-7>
[scopehal-apps] azonenberg 460503b - Removed half-baked OpenCL renderer so there's less stuff to undo when we do the Vulkan transition
<_whitenotifier-7>
[scopehal-apps] azonenberg 06dc07f - Fixed a bunch of stuff around test cases
<_whitenotifier-7>
[scopehal-apps] ... and 6 more commits.
<azonenberg>
well, it compiles and runs. i'm seeing a crash in the pico driver, unclear if it's a regression or was always there
<azonenberg>
investigating now
<azonenberg>
Performance for my initial benchmark (just streaming a single 10M point channel from the picoscope) is about 49% faster just from the reduced memory bandwidth consumption
<_whitenotifier-7>
[scopehal] azonenberg 388fd78 - Bug fix: AcceleratorBuffer sometimes did not grow large enough if resizing from a very small capacity
<azonenberg>
and there's still a lot more performance i can squeeze out
<azonenberg>
in my one minute benchmark we spend 7 seconds in the Vulkan memory allocate routine
<azonenberg>
about 14ms per waveform
<azonenberg>
I want to add a memory pool model so when a new trigger occurs, if we're also deleting an old waveform from history with a buffer that's the same size
<azonenberg>
we reuse that buffer vs making a new one
massi has quit [Remote host closed the connection]
<_whitenotifier-7>
[scopehal] azonenberg 71f700a - Set up memory pool for Vulkan memory objects, currently only used by Pico and LeCroy drivers as those are the highest WFM/s and benefit most from it. Fixes #670.
<_whitenotifier-7>
[scopehal-apps] azonenberg 2cf39f3 - HistoryWindow: push waveforms to memory pool instead of deleting them
mikolajw has joined #scopehal
whitequark has joined #scopehal
sajattack[m] has joined #scopehal
<azonenberg>
Oook so now that the new data model is in place and the vulkan renderer work is planned, i guess the next thing to try is actually implementing more filter blocks on the GPU
<azonenberg>
and porting more of the legacy opencl filters to use vulkan so we can ditch opencl
<azonenberg>
And then of course getting back to all of the probes sitting on my bench that i've been too busy coding to assemble