Present: Peter, Frank, Christian, Roberto, Ian, Bhavesh, Vassili, Steve
ET release:
* ML issues: ** -O0 and -O1 produce results as in correct values ** printf does make it work as well ** -O2 shows the issue but cannot be debugged due to most variables being optimized out ** has a bit of a feel of a compiler bug, Roland mentions that since Intel defaults to -ffast-math then we cannot really expect that results of -O0 and -O2 are identical and for the GRHydro code this resulted in NaN vs no NaN due to divisions by zero / infinity. ** want to test which machines are affected (Roland will test bluewaters, wheeler, bethe, comet, cori, edison, zwicky, qb, minerva) to check which machines are affected ** have test similar to what is in bbox.cc to check for compiler versions
GRHydro issues: * seems gone, need to verify on different machines to be certain
Intel compiler workaround for bbox.cc: * current patch to reduce optimization setting for Inel 17 to -O1 also needs to be applied for the (beta) of Intel 18
Jenkins: * machine is up and running * want to check with NCSA security persons to make get their ok * the flaw that was exploited should be fixed but want to be sure * web-interface turned off and only accessible for persons with login credent
Next meeting: * Steve will host
Yours, Roland