980423 - Talos tcanvasmark regression on Android 2.2, March 5 2014

Status: RESOLVED WONTFIX

(Firefox for Android Graveyard :: General, defect)

Description

[Geoff Brown [:gbrown]]

Date: Thu, 06 Mar 2014 04:22:06 -0000
Subject: <Regression> Mozilla-Inbound - CanvasMark - Android 2.2
        (Native) -        19.2%

Regression: Mozilla-Inbound - CanvasMark - Android 2.2 (Native) - 19.2% decrease
--------------------------------------------------------------------------------
    Previous: avg 7225.583 stddev 167.720 of 12 runs up to revision 9f57c9fd8d52
    New     : avg 5839.625 stddev 132.772 of 12 runs since revision 6f287c42495b
    Change  : -1385.958 (19.2% / z=8.264)
    Graph   : http://mzl.la/1cbhMto

Changeset range: http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=9f57c9fd8d52&tochange=6f287c42495b

Changesets:
  * http://hg.mozilla.org/integration/mozilla-inbound/rev/1c5922e166f9
    : Milan Sreckovic <[e-mail]> - Bug 910860: Use gfxPrefs to make sure preferences are evaluated on the main thread. This still leaves the preference re-evaluated all the time, not just at startup.  Changing that is another bug. r=bgirard.
    : http://bugzilla.mozilla.org/show_bug.cgi?id=910860

  * http://hg.mozilla.org/integration/mozilla-inbound/rev/1db9c2e7c5a5
    : Robert Longson <[e-mail]> - Bug 979407 - Fix ASSERTION: viewport height must be nonnegative, with negative markerHeight or markerWidth attribute. r=dholbert
    : http://bugzilla.mozilla.org/show_bug.cgi?id=979407

  * http://hg.mozilla.org/integration/mozilla-inbound/rev/e91a61089c37
    : Yura Zenevich <[e-mail]> - Bug 976082 - added tests for Touch Adapter. r=eeejay

---
 accessible/tests/mochitest/jsat/a11y.ini           |   3 +
 accessible/tests/mochitest/jsat/dom_helper.js      | 157 +++++++++++++++++++++
 accessible/tests/mochitest/jsat/gestures.json      | 132 +++++++++++++++++
 accessible/tests/mochitest/jsat/jsatcommon.js      |  15 +-
 .../tests/mochitest/jsat/test_touch_adapter.html   |  54 +++++++
 5 files changed, 356 insertions(+), 5 deletions(-)
 create mode 100644 accessible/tests/mochitest/jsat/dom_helper.js
 create mode 100644 accessible/tests/mochitest/jsat/gestures.json
 create mode 100644 accessible/tests/mochitest/jsat/test_touch_adapter.html
    : http://bugzilla.mozilla.org/show_bug.cgi?id=976082

  * http://hg.mozilla.org/integration/mozilla-inbound/rev/ccf27e254ed8
    : James Willcox <[e-mail]> - Bug 956993 - Rely on OES_EGL_image_external for SharedSurface_EGLImage r=jgilbert
    : http://bugzilla.mozilla.org/show_bug.cgi?id=956993

  * http://hg.mozilla.org/integration/mozilla-inbound/rev/2dda16c0a398
    : James Willcox <[e-mail]> - Bug 939276 -  Use a single GLContext for all SkiaGL canvases r=jgilbert,vlad,gwright,bjacob
    : http://bugzilla.mozilla.org/show_bug.cgi?id=939276

  * http://hg.mozilla.org/integration/mozilla-inbound/rev/6f287c42495b
    : James Willcox <[e-mail]> - Bug 980048 - Don't use EGL fences on Pandaboard r=jgilbert
    : http://bugzilla.mozilla.org/show_bug.cgi?id=980048

Bugs:
  * http://bugzilla.mozilla.org/show_bug.cgi?id=956993 - Add SharedSurface backend that uses 'external' EGLImages
  * http://bugzilla.mozilla.org/show_bug.cgi?id=939276 - [SkiaGL] [Feature] Support multiple Skia DrawTargets backed by a single GLContext
  * http://bugzilla.mozilla.org/show_bug.cgi?id=910860 - PNG decoder gets preferences off the main thread
  * http://bugzilla.mozilla.org/show_bug.cgi?id=979407 - "ASSERTION: viewport height must be nonnegative", with negative 'markerHeight' attribute (and similar for negative markerWidth)
  * http://bugzilla.mozilla.org/show_bug.cgi?id=980048 - Pandaboard hangs in test_drawDiscardedImage.html
  * http://bugzilla.mozilla.org/show_bug.cgi?id=976082 - [AccessFu] Provide tests for touch adapter.

Comment 1

[Joel Maher ( :jmaher ) (UTC -8)]

I believe this is what snorp was talking about yesterday regarding a skia update, I will let him confirm!

Comment 2

[James Willcox (:snorp) ([e-mail]) (he/him)]

This is from disabling SkiaGL on 2.2. We removed the fastpath for GL on those devices because it was a giant PITA to make it work with single-GLContext Skia. Also, not sure we really care about pre-ICS phones running WebGL/SkiaGL stuff.