944429 - Talos regressions tp4 7% on Android 4.0.4, SVG-ASAP 400% on Android 4.0.4, 500% on Android 2.2, on Nov 27 2013

Status: RESOLVED WONTFIX

(Firefox for Android Graveyard :: General, defect)

Description

[Geoff Brown [:gbrown]]

From dev-tree-management Digest, Vol 59, Issue 205:

Regression: Mozilla-Inbound - Tp4 Mobile - Android 4.0.4 - 7.66% increase
-------------------------------------------------------------------------
    Previous: avg 917.654 stddev 27.905 of 12 runs up to revision 8fbc23103cbc
    New     : avg 987.942 stddev 25.824 of 12 runs since revision 2402eb839597
    Change  : +70.288 (7.66% / z=2.519)
    Graph   : http://mzl.la/1erV3Ge

Regression: Mozilla-Inbound - SVG-ASAP - Android 4.0.4 - 429% increase
----------------------------------------------------------------------
    Previous: avg 1171.181 stddev 18.715 of 12 runs up to revision 8fbc23103cbc
    New     : avg 6197.546 stddev 53.229 of 12 runs since revision 2402eb839597
    Change  : +5026.365 (429% / z=268.568)
    Graph   : http://mzl.la/1erXN6k

Regression: Mozilla-Inbound - SVG-ASAP - Android 2.2 (Native) - 504% increase
-----------------------------------------------------------------------------
    Previous: avg 1256.183 stddev 13.218 of 12 runs up to revision 8fbc23103cbc
    New     : avg 7581.538 stddev 68.847 of 12 runs since revision e7e1771dd294
    Change  : +6325.355 (504% / z=478.543)
    Graph   : http://mzl.la/1aZHRrd

All blame bug 931823.

Comment 1

[Chris Lord [:cwiiis]]

A note about these regressions, the tests that regress terribly before regressed because they weren't rendering correctly for - if you grep the logs for 'aborting' prior to bug 931823, you'll notice that these tests abort drawing constantly. This means that either they weren't drawing anything to the screen, or more likely they were drawing in low resolution (which explains the roughly 4x speed decrease).

So I wouldn't really call this a performance regression (although it is a slight one, the real magnitude of which you can see in the panning and checkerboarding tests), so much as these numbers were just incorrect previously.

I'd also like to add, that we *really* need this patch, it's not an option to back it out - without it, several sites flicker terribly while scrolling.

Comment 2

[Geoff Brown [:gbrown]]

On Aurora now:

Date: Thu, 12 Dec 2013 17:40:50 -0000
From: [e-mail]
To: [e-mail]
Subject: <Regression> Mozilla-Aurora - SVG-ASAP - Android 2.2 (Native)
        - 567%
Message-ID:
        <[e-mail]>
Content-Type: text/plain; charset="us-ascii"

Regression: Mozilla-Aurora - SVG-ASAP - Android 2.2 (Native) - 567% increase
----------------------------------------------------------------------------
    Previous: avg 1122.613 stddev 35.786 of 12 runs up to revision eec4eefa594c
    New     : avg 7492.537 stddev 166.124 of 12 runs since revision ecb386453084
    Change  : +6369.924 (567% / z=178.001)
    Graph   : http://mzl.la/1gsFQJ0

Comment 3

[Geoff Brown [:gbrown]]

Comment 1 seems like a good argument to resolve WONTFIX or INVALID. Any reason to keep this open?