Platform/GFX/SWWRmeetingnotes

Purpose of this document is to collect meeting notes for our work on Software WebRender

July 10th 2020

Purpose of meeting:

• Set next milestone for SWGL
• Discuss potential rollout plans

Attendees:

• LS, JB, JM, JCB

Notes:

• What do we consider the MVP for SWGL?
• So far, there are some cases were we seem better than basic, some worse
• If everything is in the picture cache, works well. If not, it chugs rather than nice scroll

Next major milestones to look at:

• Video performance
• Image performance
• What other areas of incorrectness turn up when we run Gecko reftests?

• Ship (in Nightly) to a slice Linux first as that population is on basic layers

Goals by end of Aug:

• Have video fast path implemented (LS)
• After JB is done clip API work, determine if working on the specialized path for images makes sense or start getting an idea of how much work Gecko reftests will be
• After August LS will likely have to make time for a Skia update

May 7th 2020

Core purpose of meeting:

• Set May milestone for sw-wr

Agenda:

• Review status/how far we came in April
• Discuss core priorities for May
• Performance measurement idea from Jim
• Any other SWGL discussions/concerns we should surface (quality, security)

Notes:

• In terms of correctness, majority of places where we are still failing is png reference.
• To get better than that would impact performance
• Add fuzz bonus to Wrench harness, ideally SWGL specific
• Whatever solution is easiest so we can get SWGL running in wrench CI is suitable
• When do we want to start attacking Gecko reftests?
• Right now performance and getting software composition for Linux and Windows are likely top priorities

Goals for May: Set up infrastructure we need to get good perf measurements

• Compositing for Windows and Linux
• Wrench and TP5scroll in CI - need to make sure it supports the options we need and is green in try before we get more help to hook that up
• Do a run of TP5scroll locally to get early results
• Back burner task - get a handle on how bad we are with Gecko reftests

March 31st 2020

Core purpose of meeting:

• Set April milestone for sw-wr

Notes:

• SWGL is in tree, work underway to fix key breakages/features
• Lee has been working on implementing support for perspective correct rasterization
• Need to investigate failing Wrench reftests, looks like many failures potentially caused by common problems (anti-aliasing and stuff offset to the left a bit)

Goals for April:

• Look at figuring out the two main problems mentioned above that seem to be a common cause of many failures - Lee
• We are also crashing while running reftests right now - Jim to take an initial look at that, see if we can figure out why (https://bugzilla.mozilla.org/show_bug.cgi?id=1626702)
• After that, we can start looking at and categorizing failing reftets - Jim (with support from JeffM and Lee)
  • Then we can get a rough idea of how many we might aim to fix throughout April
• Reducing shaders exposed to perspective? Background list of tasks for Glenn (not super urgent but chipping away next couple of weeks)
• Aim to have something in place for perspective (even if not yet final) - Lee
• Performance is a little on the backburner for now, Jeff notes gmail is slow so some profiling could happen there if we have time/decide it is a prority