OpenAnalysis

log

find changesets by author, revision, files, or words in the commit message
rev 39: (0) tip
age author description
3 months ago utke on macbook better debug messagesdefault tip
7 months ago azamat Produce debug output
7 months ago utke on macbook try to handle pointers
14 months ago utke looks ok to me - no explanation what exactly is incomplete or wrong...
14 months ago utke additional debug information to verify alias settings
15 months ago utke typos
15 months ago utke This is a fix for a case where a formal parameter does not get activated when it should as described in bug report http://trac.mcs.anl.gov/projects/openAD/ticket/268
16 months ago utke updated the year
16 months ago utke third attempt at the passivation on dummy paramters. The problem is that in both cases (for varied and usefull) it is plausible to propagate the respective property through the dependency graph because the fact that the parameter is not referenced doesn't imply that the dependency is killed. Instead it is a special case where certain variables despite being both usefull and varied are nothing but "unused" references and therefore do not need to become active. This is now done be separate logic that tests that there be no dependency edges other than call/return and no selfdependencies in which case we then decide to not make the variable active. Extra debug information has been added to make following the analysis more comprehensive.
16 months ago utke it appears that the previous the previous changeset impacts the algorithm in other ways that lead to deactivation of more variables than intended because the varied property has additonal implications; instead we change to the
16 months ago utke we want to unset varied also for formal parameters of stub routines that are (intentionally) not referenced in stub routine. That means the may have multiple outedges (nonParentSuccessors) but they are all return edges. Consequently we introcude a variable that establishes if there are any out edges that are not RETURN edges and amend the test.
16 months ago utke for debugging
17 months ago utke initialized with const string literal
17 months ago utke we already are inside of these namespaces and gcc 4.5.1 now yields en error on these
2009-10-22 m_utkej fix supplied by Jaewook Shin to get the correct activity for testcases OpenADFortTk/Regression/TestSources/PostProcess_ComBlock[1|3].f90
2009-10-07 m_utkej fix for changeset 117154ac1fde to address problem with OpenADFortTk/Regression/TestSources/PostProcess_optParm1.f90 that failed after introducing that changeset
2009-10-05 m_utkej Jaewooks fix for the performance problem with long parameter lists (see the updated README in CSFIActivity)
2009-07-23 m_utkej add entry
2008-11-06 m_utkej should use proper C++ headers
2008-11-06 m_utkej branch merge
2008-11-06 m_utkej include fix needed for gcc v 4.3.2
2008-11-05 utke more ignores
2008-10-21 m_utkej branch merge
2008-10-21 m_utkej Jaewook's fix
2008-08-06 m_utkej introduce flag to make all varied variables active
2008-04-08 m_utkej sync logic fix
2008-04-08 m_utkej extra output
2008-04-08 m_utkej Jaewook's fix
2008-02-07 m_utkej sync
2008-02-04 m_utkej resynch
2007-10-23 m_utkej branch merge
2007-10-23 m_utkej resynch
2007-10-08 m_utkej added log file
2007-10-05 m_utkej resynched
2007-10-05 m_utkej Added tag preOA for changeset 07efa9cabd8f
2007-08-10 m_utkej ignorepreOA
2007-08-01 m_utkej fix up uddu logic (also done in svn trunk)
2006-11-27 m_utkej add a flag to turn restriction to basic block level chain information on and off (emacs reformatting)
2006-11-13 m_utkej baseline Jaewooks changes
2006-11-13 utke baseline identical to CVS at Rice
rev 39: (0) tip