01:31 <_whitenotifier-3> [scopehal] azonenberg pushed 3 commits to master [+0/-0/±3] https://git.io/JvRGR

01:31 <_whitenotifier-3> [scopehal] azonenberg 08ed8c2 - ClockRecoveryDecoder: fixed array bounds check

01:31 <_whitenotifier-3> [scopehal] azonenberg 8684233 - TMDSRenderer: changed control character rendering to CTL instead of C, to avoid confusion with hex characters

01:31 <_whitenotifier-3> [scopehal] azonenberg bbb1e23 - CaptureChannel: added GetSampleEnd convenience function

01:35 <_whitenotifier-3> [scopehal-apps] azonenberg pushed 1 commit to master [+2/-0/±4] https://git.io/JvRGu

01:35 <_whitenotifier-3> [scopehal-apps] azonenberg b9fcfeb - Initial library support for OpenGL compute shaders

07:30 <_whitenotifier-3> [scopehal-apps] azonenberg opened issue #56: Persistence doesn't work with new rendering engine - https://git.io/JvRWv

07:31 <_whitenotifier-3> [scopehal-apps] azonenberg labeled issue #56: Persistence doesn't work with new rendering engine - https://git.io/JvRWv

07:31 <_whitenotifier-3> [scopehal-apps] azonenberg labeled issue #56: Persistence doesn't work with new rendering engine - https://git.io/JvRWv

07:31 <_whitenotifier-3> [scopehal-apps] azonenberg pushed 2 commits to master [+1/-0/±10] https://git.io/JvRWJ

07:31 <_whitenotifier-3> [scopehal-apps] azonenberg d3cfabf - Experimental new rendering engine in compute shaders

07:31 <_whitenotifier-3> [scopehal-apps] azonenberg 87a4524 - Significant performance improvements to new compute-shader rendering engine

07:49 <_whitenotifier-3> [scopehal-apps] azonenberg pushed 2 commits to master [+0/-0/±4] https://git.io/JvRWG

07:49 <_whitenotifier-3> [scopehal-apps] azonenberg 20f956a - Removed dead code from old rendering engine

07:49 <_whitenotifier-3> [scopehal-apps] azonenberg b19e713 - Continued initial performance work on new renderer

08:19 <_whitenotifier-3> [scopehal-apps] azonenberg pushed 1 commit to master [+0/-0/±2] https://git.io/JvRWp

08:19 <_whitenotifier-3> [scopehal-apps] azonenberg 9464c9e - Fixed incorrect index calculation on very zoomed-out waveforms causing extremely poor performance

08:24 <azonenberg> well, the new renderer is officially significantly faster than the old one

08:24 <azonenberg> and i don't seem to be anywhere near 100% load on any GPU component so i likely have some bottlenecks i can optimize

08:24 <azonenberg> (testing on 1M point waveforms for now)

08:26 <azonenberg> On my Quadro K5200 (high end 2014 era workstation card, significantly slower than the 2080 Ti that I just ordered) the compute shader call for rendering a single 1M point waveform takes 6.82 ms

08:27 <azonenberg> So a theoretical 146 FPS if you only have one waveform on screen and don't do any compositing or UI or anything else

08:27 <azonenberg> I'm still tweaking, but i think at this point the waveform renderer itself is no longer the bottleneck

08:28 <azonenberg> for reference, the glBufferData() call to push the waveform and index data to the GPU took 15.5 ms

08:29 <azonenberg> I spend a fair bit of time in software rendering still, especially if you have a lot of protocol decodes. So i think the top priority for performance optimization at this point is to make digital waveforms use the same hardware accelerated path that analog ones do

08:32 <azonenberg> Next step will be making the outline boxes for complex decodes go GPU

08:33 <azonenberg> Text will stay software, since it's very hard to do good text on the GPU. And, more importantly, because with proper culling and such the amount of text on screen at any time should not be too high. It's not like we're going to have 1M text blocks on screen at a time

08:33 <azonenberg> At some point I want to add better (well, ANY) antialiasing to the new renderer and generally tweak appearances. But less of a priority

08:34 <azonenberg> Another performance boost to consider at some point would be decimating the waveform if you zoom out. When you go out super far and all you see is intensity graded outlines, rendering every 4th or 16th sample is probably fine, especially if instrument sample rate is a lot higher than your signal bandwidth

08:34 <azonenberg> (This will of course be an end-user-configurable setting)

08:41 <_whitenotifier-3> [scopehal-cmake] azonenberg pushed 1 commit to master [+0/-0/±4] https://git.io/JvR8J

08:41 <_whitenotifier-3> [scopehal-cmake] azonenberg f6b11d7 - Updated to latest submodules