RVM

Pure on mathMagic breaks dacapo sunflow

Details

  • Type: Bug Bug
  • Status: Closed Closed
  • Priority: Major Major
  • Resolution: Fixed
  • Affects Version/s: None
  • Fix Version/s: 3.0
  • Component/s: Compiler: Optimizing
  • Labels:
    None
  • Number of attachments :
    0

Description

The change in commit 14036 (specifically to the single parameter version of mathMagic) causes dacapo sunflow (from the dacapo head) to reliably fail with an image validation check (I assume some floating point error has crept in).

This may be related to the intermittent lusearch problem.

Activity

Ascending order - Click to sort in descending order
Hide
Permalink
Steve Blackburn added a comment -

Further info:

Sunflow worked on March 13 [1], r14029 [2] but has been broken since March 14 [3] , r14037 [4]. The stack trace is here [5].

I understand that Sunflow works in our current head if you simply disable the pure annotation (Daniel?).

[1] http://jikesrvm.anu.edu.au/~dacapo/regression/results-20080313/
[2] http://jikesrvm.anu.edu.au/~dacapo/regression/results-20080313/jikesrvm.build.log
[3] http://jikesrvm.anu.edu.au/~dacapo/regression/results-20080314/
[4] http://jikesrvm.anu.edu.au/~dacapo/regression/results-20080314/jikesrvm.build.log
[5] http://jikesrvm.anu.edu.au/~dacapo/regression/results-20080314/JIKESRVM/sunflow.default.tmp.log

Show
Steve Blackburn added a comment - Further info: Sunflow worked on March 13 [1] , r14029 [2] but has been broken since March 14 [3] , r14037 [4] . The stack trace is here [5] . I understand that Sunflow works in our current head if you simply disable the pure annotation (Daniel?). [1] http://jikesrvm.anu.edu.au/~dacapo/regression/results-20080313/ [2] http://jikesrvm.anu.edu.au/~dacapo/regression/results-20080313/jikesrvm.build.log [3] http://jikesrvm.anu.edu.au/~dacapo/regression/results-20080314/ [4] http://jikesrvm.anu.edu.au/~dacapo/regression/results-20080314/jikesrvm.build.log [5] http://jikesrvm.anu.edu.au/~dacapo/regression/results-20080314/JIKESRVM/sunflow.default.tmp.log
Hide
Permalink
Steve Blackburn added a comment -

The meta point here is that I am trying to move forward with the next DaCapo release, and as a background task, trying to see if we can fix Jikes RVM bugs exposed by the dacapo head.

Show
Steve Blackburn added a comment - The meta point here is that I am trying to move forward with the next DaCapo release, and as a background task, trying to see if we can fix Jikes RVM bugs exposed by the dacapo head.
Hide
Permalink
Daniel Frampton added a comment -

As Steve mentioned I investigated the change in more detail and narrowed it down to just un-@Pure-ing that mathMagic method.

Show
Daniel Frampton added a comment - As Steve mentioned I investigated the change in more detail and narrowed it down to just un-@Pure-ing that mathMagic method.
Hide
Permalink
Ian Rogers added a comment -

With Daniel's help I've confirmed there's a bug in LocalCSE, we transform:

y = tan x
z = sin x

into

y = tan x
z = y

this is because the local CSE is only comparing arguments and not the method or address parameters.

Show
Ian Rogers added a comment - With Daniel's help I've confirmed there's a bug in LocalCSE, we transform: y = tan x z = sin x into y = tan x z = y this is because the local CSE is only comparing arguments and not the method or address parameters.
Hide
Permalink
Ian Rogers added a comment -

Patch committed in r14323.

Show
Ian Rogers added a comment - Patch committed in r14323.
Hide
Permalink
David Grove added a comment -

reopening so I can modify fix target to 3.0

Show
David Grove added a comment - reopening so I can modify fix target to 3.0

People

  • Assignee:
    Ian Rogers
    Reporter:
    Daniel Frampton
Vote (0)
Watch (0)

Dates

  • Created:
    Updated:
    Resolved: