RVM

Failing DaCapo benchmarks on PPC 32 OS X

Details

  • Number of attachments :
    4

Description

I see the following failures in a pre-commit test-run:

test:
[echo] Test Result for [development|dacapo] antlr : SUCCESS
[texec] Result: 124
[echo] Test Result for [development|dacapo] bloat : FAILURE Unexpected exit code.
[touch] Creating /Users/ian/jikesrvm/results/tests/pre-commit/development/dacapo/chart-output.txt
[echo] Test Result for [development|dacapo] chart : EXCLUDED Test excluded.
[texec] Timeout: killed the sub-process
[texec] Result: 143
[echo] Test Result for [development|dacapo] eclipse : OVERTIME Test exceeded time limit.
[echo] Test Result for [development|dacapo] fop : SUCCESS
[echo] Test Result for [development|dacapo] hsqldb : SUCCESS
[texec] Result: 255
[echo] Test Result for [development|dacapo] jython : FAILURE Unexpected exit code.
[echo] Test Result for [development|dacapo] luindex : SUCCESS
[touch] Creating /Users/ian/jikesrvm/results/tests/pre-commit/development/dacapo/lusearch-output.txt
[echo] Test Result for [development|dacapo] lusearch : EXCLUDED Test excluded.
[echo] Test Result for [development|dacapo] pmd : SUCCESS
[texec] Result: 124
[echo] Test Result for [development|dacapo] xalan : FAILURE Unexpected exit code.

I've attached the failing test results. I also see eclipse fail when run independently.

Attachments

  1. Text File
    bloat-output.txt
    03/Aug/07 11:35 AM
    18 kB
    Ian Rogers
  2. Text File
    eclipse-output.txt
    03/Aug/07 11:42 AM
    34 kB
    Ian Rogers
  3. Text File
    jython-output.txt
    03/Aug/07 11:35 AM
    5 kB
    Ian Rogers
  4. Text File
    xalan-output.txt
    03/Aug/07 11:35 AM
    32 kB
    Ian Rogers

Activity

Ascending order - Click to sort in descending order
Hide
Permalink
Ian Rogers added a comment -

GC error

Show
Ian Rogers added a comment - GC error
Hide
Permalink
Peter Donald added a comment -

I think the failure in bloat is a reference map problem that is present in other ppc runs. I have not seen xalan or eclipse failures before and they may or may not be related to recent timer changes but regardless I would add them as a separate issue.

Show
Peter Donald added a comment - I think the failure in bloat is a reference map problem that is present in other ppc runs. I have not seen xalan or eclipse failures before and they may or may not be related to recent timer changes but regardless I would add them as a separate issue.
Hide
Permalink
Ian Rogers added a comment -

I can't find the other bloat error. The other errors look new and timer related, it could be the mach implementation of sysNanoTime should be better, it's hard to find this code by googling and by finding libraries actually installed in a normal system. Maybe I should revise the name of the issue, but I think it captures the problem pretty well at the moment.

Show
Ian Rogers added a comment - I can't find the other bloat error. The other errors look new and timer related, it could be the mach implementation of sysNanoTime should be better, it's hard to find this code by googling and by finding libraries actually installed in a normal system. Maybe I should revise the name of the issue, but I think it captures the problem pretty well at the moment.
Hide
Permalink
Ian Rogers added a comment -

one option is to use gettimeofday to return nano second timing on OS/X.. comments on the net suggest that this is derived from mach_absolute_time. Using it may cause a performance hit, but it's arguably better than occasionally seeing what may be the clock going backwards.

Show
Ian Rogers added a comment - one option is to use gettimeofday to return nano second timing on OS/X.. comments on the net suggest that this is derived from mach_absolute_time. Using it may cause a performance hit, but it's arguably better than occasionally seeing what may be the clock going backwards.
Hide
Permalink
Ian Rogers added a comment -

so the eclipse and xalan failures seem to be because the compare on line 205 of org.mmtk.utuility.heap.HeapGrowthManager assumes the gcLoad should be not < 0.0, whereas the reported gcLoad is actually -0.0. Looks like timer resolution is missing and we should do something with this compare.

Show
Ian Rogers added a comment - so the eclipse and xalan failures seem to be because the compare on line 205 of org.mmtk.utuility.heap.HeapGrowthManager assumes the gcLoad should be not < 0.0, whereas the reported gcLoad is actually -0.0. Looks like timer resolution is missing and we should do something with this compare.
Hide
Permalink
Ian Rogers added a comment -

As of r13260 I the HeapGrowthManager problems are gone.

Show
Ian Rogers added a comment - As of r13260 I the HeapGrowthManager problems are gone.
Hide
Permalink
Ian Rogers added a comment -

I don't see this problem any more.

Show
Ian Rogers added a comment - I don't see this problem any more.

People

Vote (0)
Watch (0)

Dates

  • Created:
    Updated:
    Resolved:
Atlassian JIRA (v4.4.3#663-r165197) | Report a problem
Powered by a free Atlassian JIRA open source license for Codehaus. Try JIRA - bug tracking software for your team.