Closed
Bug 1004437
Opened 12 years ago
Closed 12 years ago
11% SVGX regression on OSX for fx 31,32 from bug 997551
Categories
(Core :: Graphics: Layers, defect)
Tracking
()
People
(Reporter: jmaher, Assigned: mattwoodrow)
References
Details
(Keywords: perf, regression, Whiteboard: [talos_regression])
when http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=d9a6a378349a&tochange=7eba5b89f831 landed on inbound it caused a tsvgx regression
here is a graph to show the difference:
http://graphs.mozilla.org/graph.html#tests=[[281,63,21]]&sel=1398575014000,1398747814000&displayrange=7&datatype=running
I did some retriggers, there is no doubt:
https://tbpl.mozilla.org/?tree=Mozilla-Inbound&fromchange=61b91c85a7aa&tochange=f3a57f0d89b0&jobname=Rev4%20MacOSX%20Snow%20Leopard%2010.6%20mozilla-inbound%20talos%20svgr
| Reporter | ||
Updated•12 years ago
|
tracking-firefox31:
--- → ?
| Assignee | ||
Comment 1•12 years ago
|
||
Is this 10.6 only (again)?
Do we have any idea why 10.6 continually shows regressions that don't affect 10.8?
Comment 2•12 years ago
|
||
(In reply to Matt Woodrow (:mattwoodrow) from comment #1)
> Do we have any idea why 10.6 continually shows regressions that don't affect
> 10.8?
Bug 990490 comment 1.
| Reporter | ||
Comment 3•12 years ago
|
||
this is 10.6 and 10.8, I see it on aurora now:
http://graphs.mozilla.org/graph.html#tests=[[281,63,21],[281,52,24],[281,52,21],[281,63,24]]&sel=1398575014000,1398747814000&displayrange=7&datatype=running
Luckily we have a few weeks to figure it out!
| Assignee | ||
Comment 4•12 years ago
|
||
Ok, we should fix this then.
Detailed breakdown of results:
Pre regression:
O - |i|pagename|runs|
18:59:02 INFO - |0;gearflowers.svg;381;164;161;161;164;161;162;164;160;163;165;161;165;165;167;162;163;167;174;168;164;165;169;166;169
18:59:02 INFO - |1;composite-scale.svg;84;46;44;47;47;55;45;46;45;54;47;45;47;50;45;45;47;56;47;49;46;53;46;43;45
18:59:08 INFO - |2;composite-scale-opacity.svg;226;213;211;222;210;211;223;211;213;223;210;214;224;212;210;221;210;211;221;212;211;226;211;212;223
18:59:08 INFO - |3;composite-scale-rotate.svg;147;141;140;140;142;141;141;142;143;141;141;141;141;141;142;141;142;140;142;141;142;140;143;142;140
18:59:08 INFO - |4;composite-scale-rotate-opacity.svg;375;373;371;370;371;373;371;370;370;372;369;369;369;371;370;369;370;368;371;368;371;368;374;370;371
18:59:08 INFO - |5;hixie-001.xml;525;493;489;490;493;495;492;483;492;490;485;487;488;494;487;484;489;484;493;493;490;487;507;490;500
18:59:08 INFO - |6;hixie-002.xml;505;494;495;492;490;493;491;487;493;491;494;491;487;497;486;490;490;484;486;489;485;488;490;488;492
18:59:08 INFO - |7;hixie-003.xml;323;193;186;186;181;186;188;186;189;187;186;189;187;185;189;187;189;187;188;186;192;189;189;187;187
18:59:08 INFO - |8;hixie-004.xml;1112;1118;1116;1118;1123;1117;1119;1104;1120;1121;1119;1103;1098;1093;1095;1097;1095;1095;1095;1094;1093;1097;1108;1095;1102
18:59:08 INFO - |9;hixie-005.xml;1022;1019;1015;1018;1018;1021;1018;1016;1014;1017;1017;1013;1019;1018;1017;1012;1018;1016;1017;1014;1015;1016;1015;1015;1016
18:59:08 INFO - |10;hixie-006.xml;1925;1923;1915;1923;1917;1918;1920;1926;1916;1922;1918;1920;1924;1927;1917;1930;1923;1916;1917;1925;1917;1919;1919;1915;1917
18:59:08 INFO - |11;hixie-007.xml;363;346;322;348;322;346;323;346;323;343;321;350;324;345;321;347;325;348;324;346;322;347;322;346;323
Post regression:
22:08:56 INFO - |0;gearflowers.svg;221;165;163;164;162;161;164;164;161;164;162;160;164;164;161;168;164;161;163;163;162;163;164;160;164
22:08:56 INFO - |1;composite-scale.svg;183;153;154;153;154;154;154;153;154;154;153;153;155;154;154;156;157;159;155;156;155;156;156;155;157
22:09:01 INFO - |2;composite-scale-opacity.svg;216;212;211;219;207;210;219;212;213;220;212;211;224;210;212;220;210;214;222;209;211;219;211;210;221
22:09:01 INFO - |3;composite-scale-rotate.svg;262;258;255;256;256;259;257;259;258;257;257;256;258;258;258;258;258;257;258;259;259;257;256;258;257
22:09:01 INFO - |4;composite-scale-rotate-opacity.svg;376;370;368;370;368;366;369;370;368;371;369;368;371;370;372;368;371;368;371;369;370;368;377;367;368
22:09:01 INFO - |5;hixie-001.xml;533;504;493;483;496;501;493;494;504;494;487;490;496;498;492;500;494;495;496;495;497;488;491;492;498
22:09:01 INFO - |6;hixie-002.xml;501;498;495;493;495;499;503;494;493;502;499;493;494;494;493;494;488;501;489;504;496;492;493;493;493
22:09:01 INFO - |7;hixie-003.xml;324;181;184;183;185;184;181;186;192;189;188;190;190;190;189;188;190;190;192;189;190;188;185;189;189
22:09:01 INFO - |8;hixie-004.xml;1313;1324;1323;1326;1319;1317;1316;1312;1317;1317;1303;1303;1301;1306;1303;1302;1370;1299;1313;1309;1306;1305;1301;1304;1303
22:09:01 INFO - |9;hixie-005.xml;1022;1014;1016;1016;1014;1019;1017;1022;1021;1020;1026;1027;1022;1019;1017;1023;1026;1021;1018;1021;1025;1020;1017;1022;1020
22:09:01 INFO - |10;hixie-006.xml;1930;1924;1930;1927;1922;1924;1920;1920;1929;1922;1922;1934;1923;1931;1927;1925;1934;1943;1929;1923;1955;1919;1930;1922;1957
22:09:01 INFO - |11;hixie-007.xml;368;350;325;349;325;352;327;348;325;351;325;350;324;351;324;349;323;348;326;349;324;346;324;349;327
Looks like composite-scale.svg, composite-scale-rotate.svg and hixie-004.xml all regressed significantly and the others didn't.
| Assignee | ||
Comment 5•12 years ago
|
||
This is because Quartz caches scaled copies of images opaquely within a CGImage.
My patch made it so we re-create the CGImage each draw, losing this benefit.
Since we didn't actually win on any benchmarks, maybe we should backout.
Updated•12 years ago
|
| Reporter | ||
Comment 6•12 years ago
|
||
do you know if this helped out on other non talos benchmarks?
Comment 7•12 years ago
|
||
Untracking. Beta 6 should be released today and we won't block the release on this bug.
| Reporter | ||
Comment 8•12 years ago
|
||
this regression is partially fixed by the 32 uplift to aurora:
http://graphs.mozilla.org/graph.html#tests=[[281,63,21],[281,52,21]]&sel=none&displayrange=90&datatype=running
Comment 9•12 years ago
|
||
Partially fixed enough to stop tracking this?
| Reporter | ||
Comment 10•12 years ago
|
||
we are still around 7% regressed since May 1:
http://graphs.mozilla.org/graph.html#tests=[[281,52,21],[281,63,21]]&sel=1375644425473,1407180425473&displayrange=365&datatype=running
Better than 11% though- I will provide the data, you guys can make any decisions :)
Comment 11•12 years ago
|
||
(In reply to Matt Woodrow (:mattwoodrow) from comment #5)
> This is because Quartz caches scaled copies of images opaquely within a
> CGImage.
>
> My patch made it so we re-create the CGImage each draw, losing this benefit.
>
> Since we didn't actually win on any benchmarks, maybe we should backout.
Matt - Is there any other benefit of your change in bug 997551? Should we just backout? What is the practical implication of the 7% regression that remains? Note that I have assigned this bug to you to help determine next steps, which may be do nothing, backout, fix.
Assignee: nobody → matt.woodrow
Flags: needinfo?(matt.woodrow)
Updated•12 years ago
|
| Assignee | ||
Comment 12•12 years ago
|
||
I don't think we should bother backing this out. imagelib has been converted to moz2d now, so I suspect we've actually fixed the underlying issue.
It was only ever likely to show up on testcases where we repeatedly draw the same image over and over with a significant downscale. It's hard to imagine that mattering much outside of synthetic benchmarks.
Flags: needinfo?(matt.woodrow)
Comment 13•12 years ago
|
||
From comment 12 it sounds like the current state is as 'fixed' as this bug is going to get. I'm resolving as fixed. Feel free to reopen if you disagree.
Status: NEW → RESOLVED
Closed: 12 years ago
Resolution: --- → FIXED
You need to log in
before you can comment on or make changes to this bug.
Description
•