Bug 30293

Summary: [Qt] performance regressions on loading page after ImageDecoderQt changes
Product: WebKit Reporter: Holger Freyther <zecke>
Component: WebCore Misc.Assignee: QtWebKit Unassigned <webkit-qt-unassigned>
Status: RESOLVED FIXED    
Severity: Normal CC: benjamin, hausmann, jturcotte
Priority: P2 Keywords: Performance, Qt, Regression
Version: 528+ (Nightly build)   
Hardware: PC   
OS: OS X 10.5   
Bug Depends on:    
Bug Blocks: 34208    
Attachments:
Description Flags
Current loading speed regressions none

Description Holger Freyther 2009-10-11 19:33:41 PDT
The old and new code has been executed on the same data and 30 iterations each. The majority of cases are faster, some show a regression we should need to understand. The point of this bug is to understand the cause of this regression.

loading test
row   - slower in msecs on 30 iterations

row01 - 3msec
row02 - 6msec
row14 - 844msec
row16 - 1031msec
row19 - 538msec
row22 - 49msec
row24 - 73msec
row25 - 693msec
row26 - 705msec
row30 - 3msec
row32 - 1785msec
row37 - 35msec
row39 - 261msec


cycling: slower in msecs on 30 iterations

row09 -  9msec
row11 - 39msec
row17 - 1msec
row27 - 10msec
row30 - 8msec
row32 - 114msec
row38 - 1msec
row39 - 1051msec
Comment 1 Holger Freyther 2009-10-12 01:16:40 PDT
(In reply to comment #0)


> row32 - 1785msec

This does not happen anymore. I will have to redo tests and verify the /etc/hosts.
Comment 2 Holger Freyther 2009-10-12 01:46:33 PDT
Okay, new regression list:

 RESULT : tst_Loading::load():"row06":
-     576.83 msec per iteration (total: 17305, iterations: 30)
+     623.29 msec per iteration (total: 18699, iterations: 30)
 RESULT : tst_Loading::load():"row09":
-     3.33 msec per iteration (total: 100, iterations: 30)
+     5.23 msec per iteration (total: 157, iterations: 30)
 RESULT : tst_Loading::load():"row10":
-     872.39 msec per iteration (total: 26172, iterations: 30)
+     875.20 msec per iteration (total: 26256, iterations: 30)
 RESULT : tst_Loading::load():"row12":
-     411.80 msec per iteration (total: 12354, iterations: 30)
+     412.33 msec per iteration (total: 12370, iterations: 30)
 RESULT : tst_Loading::load():"row14":
-     215.7 msec per iteration (total: 6472, iterations: 30)
+     304.4 msec per iteration (total: 9134, iterations: 30)
 RESULT : tst_Loading::load():"row18":
-     1,000.6 msec per iteration (total: 30020, iterations: 30)
+     1,019.8 msec per iteration (total: 30597, iterations: 30)
 RESULT : tst_Loading::load():"row19":
-     23.6 msec per iteration (total: 710, iterations: 30)
+     30.1 msec per iteration (total: 905, iterations: 30)
 RESULT : tst_Loading::load():"row20":
-     80.29 msec per iteration (total: 2409, iterations: 30)
+     98.70 msec per iteration (total: 2961, iterations: 30)
 RESULT : tst_Loading::load():"row22":
-     2.7 msec per iteration (total: 83, iterations: 30)
+     3.2 msec per iteration (total: 96, iterations: 30)
 RESULT : tst_Loading::load():"row25":
-     49.03 msec per iteration (total: 1471, iterations: 30)
+     49.23 msec per iteration (total: 1477, iterations: 30)
 RESULT : tst_Loading::load():"row26":
-     138.1 msec per iteration (total: 4144, iterations: 30)
+     152.2 msec per iteration (total: 4567, iterations: 30)
 RESULT : tst_Loading::load():"row30":
-     5.36 msec per iteration (total: 161, iterations: 30)
+     7.36 msec per iteration (total: 221, iterations: 30)
 RESULT : tst_Loading::load():"row31":
-     40.66 msec per iteration (total: 1220, iterations: 30)
+     44.93 msec per iteration (total: 1348, iterations: 30)
 RESULT : tst_Loading::load():"row32":
-     327.3 msec per iteration (total: 9820, iterations: 30)
+     354.60 msec per iteration (total: 10638, iterations: 30)
 RESULT : tst_Loading::load():"row37":
-     44.33 msec per iteration (total: 1330, iterations: 30)
+     62.53 msec per iteration (total: 1876, iterations: 30)
 RESULT : tst_Loading::load():"row38":
-     238.1 msec per iteration (total: 7145, iterations: 30)
+     259.9 msec per iteration (total: 7798, iterations: 30)
Comment 3 Holger Freyther 2009-10-12 03:12:01 PDT
Created attachment 41029 [details]
Current loading speed regressions

Current update. This is coming where everything is locally served. We are seeing 12 cases that seem to be slower than before. The problem is QBENCHMARK is not really good for statistics. If one loading took extremely long our whole result is screwed..
Comment 4 Benjamin Poulain 2010-05-04 13:31:31 PDT
Do you remember what you are comparing against?
I don't have regressions compared to 4.6.2.
Comment 5 Benjamin Poulain 2010-09-23 11:24:40 PDT
The benchmarks are now run regularily. Image loading in with 4.7 if faster than with 4.6.