there is a bug in end of run analysis. The processor who starts CkExit(), and use...
authorGengbin Zheng <gzheng@illinois.edu>
Sun, 14 Mar 2010 00:09:47 +0000 (18:09 -0600)
committerGengbin Zheng <gzheng@illinois.edu>
Sun, 14 Mar 2010 00:09:47 +0000 (18:09 -0600)
commit13fcc1c1cf012cb86d614ac39b4cdfd1bb23c4e8
treebb42624773907c9d39e0874288cc947f0647e556
parent45f72eb784fbedc03fa00bb69c24e4c3800c67e9
there is a bug in end of run analysis. The processor who starts CkExit(), and use exithandler, must pass control back to the original processor to continue calling CkExit(). The old code always call secondary CkExit() from PE0, which is wrong when the CkExit() is called from a processor other than 0.
src/ck-perf/trace-projections.C
src/ck-perf/trace-projections.ci
src/ck-perf/trace-projectionsBOC.h