[Jikesrvm-issues] [ jikesrvm-Bugs-1723869 ] VM_OptMachineCodeMap.generateMCInformation bad gcmap

[SourceForge.net <no...@so...>]

Bugs item #1723869, was opened at 2007-05-23 10:28
Message generated for change (Comment added) made by peter_donald
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=712768&aid=1723869&group_id=128805

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: optimizing compiler
Group: None
Status: Open
Resolution: None
Priority: 7
Private: No
Submitted By: Peter Donald (peter_donald)
Assigned to: Nobody/Anonymous (nobody)
Summary: VM_OptMachineCodeMap.generateMCInformation bad gcmap

Initial Comment:
The method VM_OptMachineCodeMap.generateMCInformation seems to produce and invalid gc map at O2 level optimization. Note: This is after r12223 where a few of explicit optimizations where moved from O1 to O2

To reproduce you can change the value of "config.bootimage.compiler.args" in "build/configs/gcstress.properties" from "-X:bc:O1" to "-X:bc:O2" and then run the gcmap-sanity test run via something like

ant -f test.xml -Dtest-run.name=gcmap-sanity

Tests will start failing with output like 


validRef: TIB outside heap, ref = 0x9d800a24 tib = 0x00000000

Invalid ref reported while scanning stack

--- METHOD (OPT) Lorg/jikesrvm/compilers/opt/VM_OptMachineCodeMap;.generateMCInformation (Lorg/jikesrvm/compilers/opt/ir/OPT_GCIRMap;)Lorg/jikesrvm/compilers/opt/VM_OptMachineCodeMap;
--- fp = 0x6b29fcc4     code base = 0x5b8c1ef4     code offset = 0x00003243
0x6dc2fbf4:REF=0x9d800a24 TIB=0x00000000 STATUS=0x00000000 (INVALID TIB: CLASS NOT ACCESSIBLE)
...
Dumping stack starting at frame with bad ref:
-- Stack --
   Lorg/jikesrvm/compilers/opt/VM_OptMachineCodeMap; generateMCInformation(Lorg/jikesrvm/compilers/opt/ir/OPT_GCIRMap;)Lorg/jikesrvm/compilers/opt/VM_OptMachineCodeMap; at line 398
   Lorg/jikesrvm/compilers/opt/VM_OptMachineCodeMap; create(Lorg/jikesrvm/compilers/opt/ir/OPT_IR;I)Lorg/jikesrvm/compilers/opt/VM_OptMachineCodeMap; at line 99
   Lorg/jikesrvm/compilers/opt/VM_OptCompiledMethod; createFinalMCMap(Lorg/jikesrvm/compilers/opt/ir/OPT_IR;I)V at line 438
...

There are a number of sample failures attached.

----------------------------------------------------------------------

>Comment By: Peter Donald (peter_donald)
Date: 2007-05-23 11:20

Message:
Logged In: YES 
user_id=1642927
Originator: YES

"[ 1488798 ] Leave SSA broken - lack of loop unrolling breaks MMTk" may be
a duplicate of this bug but it is difficult to determine without further
analysis

----------------------------------------------------------------------

Comment By: Peter Donald (peter_donald)
Date: 2007-05-23 11:18

Message:
Logged In: YES 
user_id=1642927
Originator: YES

Unfortunately I made an error and ran tests against r12222 but r12223
seems to have fixed it which means the bug was caused by one of the
optimizations that was moved from O2 to O3 in r12223

----------------------------------------------------------------------

Comment By: Peter Donald (peter_donald)
Date: 2007-05-23 10:43

Message:
Logged In: YES 
user_id=1642927
Originator: YES

Note that 398 seems to be the earliest line number where the bug is
expressed.

----------------------------------------------------------------------

Comment By: Peter Donald (peter_donald)
Date: 2007-05-23 10:38

Message:
Logged In: YES 
user_id=1642927
Originator: YES

File Added: jBYTEmark-output.txt

----------------------------------------------------------------------

Comment By: Peter Donald (peter_donald)
Date: 2007-05-23 10:35

Message:
Logged In: YES 
user_id=1642927
Originator: YES

File Added: JLex.Main-trimmed.txt.gz

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=712768&aid=1723869&group_id=128805