Mobile/BostonBrainstorm/PanZoom

Panning &Zooming

• Potential Causes
  • Root cause: Texture upload (we do it synchronously, and too much of it)
• Causes of too much texture upload:
  • Pixel-aligned scrolling - we need to verify that we are doing the correct thing here. (if not, it means we will constantly redraw the whole page when scrolling)
  • Display list issues where we're not able to use retained layers, because of the frames in the retained layer changing. (unsure, but same consequence as above - whole page redraws while scrolling)
• Short Term:
  • Figure out why we're repainting all the time
  • Determine why desktop is better (zoom?)
  • We currently upload the entire displayport. Why can't we be more intelligent about what we do here? (tile prioritization?)
  • Eideticker (check panning jank tests, write more/better tests)
• Long Term (8 months):
  • Fix the issues identified
  • Special display items (images, gradients)
  • SurfaceTexture?
• Things that can be done tomorrow:
  • Check pixel alignment test skip line
  • Check out what the status of Benwa's tile priotization patches are.
  • Check with Matt Woodrow to see if dlbi will help us with repainting all the time
• Things that could be done this week:
  • Invalidation tests
  • Is this still important?
  • Yes. We are currently janky on panning and this may be causing us to lose/not retain users.
• How does this relate to ADIs?
  • More people will use the browser if it's better. If we do well here, we can vastly surpass performance on the gingerbread stock browser, the ICS/Jellybean stock browser, Chrome, and perhaps even iOS Safari.
• Relevant bugs:

bug 751629 - Split tile upload into multiple transactions (upload progressively/asynchronously)
bug 681192 - Give scrolling APIs a flexible "allowed scroll destination range" and use it inside Gecko
bug 539356 - Replace Invalidate() calls in reflow with display list analysis
bug 771154 - Large amount of over painting on some pages on fennec
bug 781516 - Fennec invalidation regression on Nightly