You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@harmony.apache.org by "Egor Pasko (JIRA)" <ji...@apache.org> on 2006/10/26 09:04:17 UTC
[jira] Created: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
[drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
---------------------------------------------------------------------------------------------------------------
Key: HARMONY-1969
URL: http://issues.apache.org/jira/browse/HARMONY-1969
Project: Harmony
Issue Type: Bug
Components: DRLVM
Environment: any
Reporter: Egor Pasko
Attachments: DoubleSync.java
this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
run DoubleSync (attached) with -Xem:opt
java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
SIGABRT in VM code.
Stack trace:
1: ?? (??:-1)
2: abort (??:-1)
3: __assert_fail (??:-1)
4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
<snip>
dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Commented: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Geir Magnusson Jr (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=comments#action_12447276 ]
Geir Magnusson Jr commented on HARMONY-1969:
--------------------------------------------
I've tried to get this bug to exhibit itself to no avail.
On Ubuntu 6 , gcc 3.4.6, I cannot get a debug build to break w/ -Xem:opt. I didn't comment out the asserts, but had a clean run w/ them uncommented
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Attachments: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt, DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Commented: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Alexei Fedotov (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=comments#action_12447314 ]
Alexei Fedotov commented on HARMONY-1969:
-----------------------------------------
Mikhail, thanks for a good comment. I was trying to unedrstnand how to get a debug version of jitrino. Shoulf I just change release to debug in the following line of build.bat or build.sh file?
drlvm/trunk/build/build.bat:
REM Note: vm.jitrino is always complied in release mode, otherwise it makes VM debug too slow
CALL "%ANT_COMMAND%" -f make/build.xml -Dvm.jitrino.cfg=release %*
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Attachments: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt, DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Commented: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Alexey Varlamov (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=comments#action_12449310 ]
Alexey Varlamov commented on HARMONY-1969:
------------------------------------------
Egor,
I applied the patch to fix HARMONY-1808. How should we resolve this JIRA, taking into account you statement about more general problem [1] - would you create new issue?
[1] http://issues.apache.org/jira/browse/HARMONY-1969#action_12445527
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Assigned To: Alexey Varlamov
> Attachments: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt, DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Updated: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Egor Pasko (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=all ]
Egor Pasko updated HARMONY-1969:
--------------------------------
Patch Info: [Patch Available]
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Attachments: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt, DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Commented: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Egor Pasko (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=comments#action_12449410 ]
Egor Pasko commented on HARMONY-1969:
-------------------------------------
Alexey,
nested synchronized blocks are fixed. So, this bug is to be closed. I planned to add a loop nesting check assertion to HARMONY-1908. Now since HARMONY-1908 is closed, I am going to create a special JIRA for the loop nesting check. This is going to be a low priority bug.
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Assigned To: Alexey Varlamov
> Attachments: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt, DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Updated: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Egor Pasko (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=all ]
Egor Pasko updated HARMONY-1969:
--------------------------------
Attachment: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt
0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt fixes the problem. But the general problem with unnested loops still persists (broken loop structure makes IA-32 CG (and statprof?) crash in several places). Need to write a detector of such situations, first of all, to be able to move them to JET.
c-unit, smoke, kernel, SUSE9
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Attachments: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt, DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Updated: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Egor Pasko (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=all ]
Egor Pasko updated HARMONY-1969:
--------------------------------
Attachment: DoubleSync.java
DoubleSync.java attached
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Attachments: DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Commented: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Alexei Fedotov (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=comments#action_12449506 ]
Alexei Fedotov commented on HARMONY-1969:
-----------------------------------------
I believe Alexey meant HARMONY-1908.
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Assigned To: Alexey Varlamov
> Attachments: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt, DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Assigned: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Alexey Varlamov (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=all ]
Alexey Varlamov reassigned HARMONY-1969:
----------------------------------------
Assignee: Alexey Varlamov
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Assigned To: Alexey Varlamov
> Attachments: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt, DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Commented: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Alexei Fedotov (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=comments#action_12447326 ]
Alexei Fedotov commented on HARMONY-1969:
-----------------------------------------
In addition to the change of build.sh file I disabled fatal warnings on Linux. Then I was able to reproduce an asserion.
Index: drlvm/trunk/build/make/components/vm/jitrino.xml
===================================================================
--- drlvm/trunk/build/make/components/vm/jitrino.xml (revision 470383)
+++ drlvm/trunk/build/make/components/vm/jitrino.xml (working copy)
@@ -234,7 +234,7 @@
<select cxx="gcc">
<compilerarg value="-fmessage-length=0" />
<compilerarg value="-Wall" />
- <compilerarg value="-Werror" />
+<!-- <compilerarg value="-Werror" />-->
</select>
<compilerarg value="-x" />
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Attachments: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt, DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Closed: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Alexey Varlamov (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=all ]
Alexey Varlamov closed HARMONY-1969.
------------------------------------
Resolution: Fixed
Yep, I meant 1908, thanks. Closing as requested.
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Assigned To: Alexey Varlamov
> Attachments: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt, DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Commented: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Egor Pasko (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=comments#action_12444831 ]
Egor Pasko commented on HARMONY-1969:
-------------------------------------
The above stack trace was taken with profiling asserts uncommented (by mistake). Here is the correct one:
1: ?? (??:-1)
2: abort (??:-1)
3: __assert_fail (??:-1)
4: Jitrino::Ia32::IRManager::calculateLivenessInfo() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/codegenerator/ia32/Ia32IRManager.cpp:1540)
5: Jitrino::Ia32::IRManager::updateLivenessInfo() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/codegenerator/ia32/Ia32IRManager.h:300)
6: Jitrino::Ia32::SessionAction::debugOutput(char const*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/codegenerator/ia32/Ia32IRManager.cpp:2195)
7: Jitrino::Ia32::SessionAction::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/codegenerator/ia32/Ia32IRManager.cpp:2149)
8: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
9: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
<snip>
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Attachments: DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Commented: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Alexei Fedotov (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=comments#action_12447327 ]
Alexei Fedotov commented on HARMONY-1969:
-----------------------------------------
With patch applied the test passes smoothly.
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Attachments: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt, DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Commented: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Alexei Fedotov (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=comments#action_12445644 ]
Alexei Fedotov commented on HARMONY-1969:
-----------------------------------------
This bug prevents Jitrino.OPT from running class library tests.
See http://wiki.apache.org/harmony/Unit_Tests_Pass_on_DRLVM
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Attachments: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt, DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
[jira] Commented: (HARMONY-1969) [drlvm][jit] nested synchronized
blocks make Jitrino.OPT crash meeting multiple loop entrances to a single
loop
Posted by "Mikhail Fursov (JIRA)" <ji...@apache.org>.
[ http://issues.apache.org/jira/browse/HARMONY-1969?page=comments#action_12447290 ]
Mikhail Fursov commented on HARMONY-1969:
-----------------------------------------
Geir, are you sure that libjitrino.so you tested was built in debug mode? When you build DRLVM in debug mode, jitrino builds in release.
?
> [drlvm][jit] nested synchronized blocks make Jitrino.OPT crash meeting multiple loop entrances to a single loop
> ---------------------------------------------------------------------------------------------------------------
>
> Key: HARMONY-1969
> URL: http://issues.apache.org/jira/browse/HARMONY-1969
> Project: Harmony
> Issue Type: Bug
> Components: DRLVM
> Environment: any
> Reporter: Egor Pasko
> Attachments: 0001-more-monexit-patterns-transformed-to-correct-loop-structure.txt, DoubleSync.java
>
>
> this is a sub-bug for HARMONY-1908. The reason is $subj. I suggest to restore a normal loop structure in this case (i. e. generalize the eliminateUnnestedLoopsOnDispatch() a little)
> of course, loop nesting may be broken in specially crafted bytecode samples (and we will not repair them with the proposed solution), but a reasonable behavoiur in this case would be to move bytecodes like this to JET. Here I am not for moving to JET because of the common nature of the code pattern (nested synches, javac) that we need to run heavily optimized.
> To reproduce, disable profiling-related asserts in ControlFlowGraph.cpp, rebuild Jitrino in debug mode.
> run DoubleSync (attached) with -Xem:opt
> java: /export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:384: bool Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool): Assertion `doAssert?false:true' failed.
> SIGABRT in VM code.
> Stack trace:
> 1: ?? (??:-1)
> 2: abort (??:-1)
> 3: __assert_fail (??:-1)
> 4: Jitrino::ControlFlowGraph::isEdgeProfileConsistent(bool, bool, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:385)
> 5: Jitrino::ControlFlowGraph::smoothEdgeProfile() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/shared/ControlFlowGraph.cpp:947)
> 6: Jitrino::StaticProfiler::estimateGraph(Jitrino::IRManager&, double, bool) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:253)
> 7: Jitrino::StaticProfilerPass::_run(Jitrino::IRManager&) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/dynopt/StaticProfiler.cpp:293)
> 8: Jitrino::OptPass::run() (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/optimizer/optpass.cpp:61)
> 9: Jitrino::runPipeline(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:226)
> 10: Jitrino::compileMethod(Jitrino::CompilationContext*) (/export/users/evpasko/svn/3/trunk/working_vm/vm/jitrino/src/main/Jitrino.cpp:261)
> <snip>
> dumping dot files helps significantly
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira