JDK-6891435 : Improve D3D preloading
  • Type: Enhancement
  • Component: client-libs
  • Sub-Component: java.awt
  • Affected Version: 6u18
  • Priority: P2
  • Status: Closed
  • Resolution: Fixed
  • OS: windows
  • CPU: generic
  • Submitted: 2009-10-14
  • Updated: 2013-11-01
  • Resolved: 2010-01-13
The Version table provides details related to the release that this issue/RFE will be addressed.

Unresolved : Release in which this issue/RFE will be addressed.
Resolved: Release in which this issue/RFE has been resolved.
Fixed : Release in which this issue/RFE has been fixed. The release containing this fix may be available for download as an Early Access Release or a General Availability Release.

To download the current JDK release, click here.
JDK 6 JDK 7
6u18 b04Fixed 7Fixed
Related Reports
Relates :  
JDK-6946559 - AWTToolKit thread crashes in JNU_GetEnv
Relates :  
JDK-6943693 - with 6u18 a considerably smaller initial/max-heap-size is possible compared to 6u17
Relates :  
JDK-6867515 - Reduce impact of D3D initializion on startup time
Relates :  
JDK-2200223 - Reduce impact of D3D initializion on startup time
Relates :  
JDK-6987967 - D3D preloading thread should initialize COM
Description
Fix for 6867515 added "D3D preload" feature.
Currently it includes creation of "main" Direct3D object (IDirect3D9)
Additional research shown that 2nd delay in D3D initialization occurs on creation/initialization of D3D Adapter objects (implemented in D3DContext class).
This delay becomes longer when D3D Adapter are created after some period of time after IDirect3D9 creation and this fact causes decreasing of performance improvement due D3D preload (for example plugins require more time for initializetion and in this case time period between IDirect3D9 creation & D3DAdapter creation often is too big).
D3DPipelineManager supports caching of D3DContext (when D3DContext for some adapter is requested, D3DPipelineManager returns existing object if it was created before), so we can do creation/initialization of D3DContext objects during preloading.
Comments
EVALUATION http://hg.openjdk.java.net/jdk7/build/jdk/rev/1f45c4c1f3a7
04-12-2010
EVALUATION The fix changes enabling/disabling algorithm for AWT warmup (vs fix for 6867515): - for applications (java/javaw) "-Dsun.awt.warmup=true" JVM parameter turns AWT warmup on; - javaws/plugins always add the parameter launching JVM (so warmup is turned on for javaws/plugin by default). J2D_D3D_PRELOAD env. variable (values: true/false) TAKES PRECEDENCE of "-Dsun.awt.warmup" JMV parameter, so setting J2D_D3D_PRELOAD to "false" disables AWT warmup for javaws/plugin too. The following settings can disable warmup (regardless of "-Dsun.awt.warmup" and "J2D_D3D_PRELOAD" values): - "-Dsun.java2d.noddraw=true" JVM parameter; - "-Dsun.java2d.d3d=false" JVM parameter; - "-Dsun.java2d.opengl=true" JVM parameter; if none of the "-Dsun.java2d.noddraw" "-Dsun.java2d.d3d", "-Dsun.java2d.opengl" parameters are specified, "J2D_D3D" env. variable is tested; if it is set to "false", the warmup is disabled.
19-10-2009
EVALUATION applet_awt & applet_swing tests use appletviewer to run applets. appletviewer doesn't use preload mechanism so there is no improvement in the tests. To evaluate plugin improvement applet_swing_jpi (which uses browser) should be considered.
16-10-2009
EVALUATION Improvement measurement results: Hi-end system, (6867515 fix made unsignificant improvement on the system): ============================================================================== results.6u18b04/no_preload: Benchmark Samples Mean Stdev Geomean Weight app_awt 20 268,65 40,88 0,25 app_swing 20 291,45 10,46 0,25 app_swing_jws 20 526,45 20,33 0,25 applet_swing_jpi 20 588,40 19,32 0,25 ============================================================================== results.6u18b04/preload: Benchmark Samples Mean Stdev %Diff P Significant app_awt 20 170,40 8,75 36,57 0,000 Yes app_swing 20 191,45 6,98 34,31 0,000 Yes app_swing_jws 20 468,70 35,72 10,97 0,000 Yes applet_swing_jpi 20 504,00 13,65 14,34 0,000 Yes ============================================================================== Middle-end system: ============================================================================== results.6u18b04/no_preload/: Benchmark Samples Mean Stdev Geomean Weight startup2 20 544,35 7,25 app_awt 20 362,40 18,80 0,17 app_swing 20 399,20 38,41 0,17 app_swing_jws 20 753,85 18,89 0,17 applet_awt 20 492,25 12,76 0,17 applet_swing 20 536,80 10,32 0,17 applet_swing_jpi 20 904,65 36,74 0,17 ============================================================================== results.6u18b04/preload/: Benchmark Samples Mean Stdev %Diff P Significant startup2 20 468,53 10,22 13,93 0,000 Yes app_awt 20 250,80 15,61 30,79 0,000 Yes app_swing 20 284,60 17,13 28,71 0,000 Yes app_swing_jws 20 685,95 28,58 9,01 0,000 Yes applet_awt 20 489,75 7,72 0,51 0,459 * applet_swing 20 539,15 10,75 -0,44 0,485 * applet_swing_jpi 20 815,60 20,57 9,84 0,000 Yes ============================================================================== results for middle-end system without the fix (unfortunately compare doesn't work due test sets are different): ============================================================================== workload.results/6u18_b03/preload: Benchmark Samples Mean Stdev Geomean Weight startup2 20 495,73 8,79 app_awt 20 349,30 15,56 0,20 app_swing 20 387,45 18,18 0,20 applet_awt 20 497,65 37,77 0,20 applet_swing 20 529,60 14,10 0,20 applet_swing_jpi 20 842,95 19,23 0,20 ==============================================================================
16-10-2009
EVALUATION should be fixed to get additional performance increase.
14-10-2009