Open rcclay opened 7 years ago
All CSVMC cases or just PbyP no drift?
Don't know yet. I'll check.
I checked the all-electron driver with drift. Same issue. On the plus side, all observables seem correct to within error bars, which is good.
https://cdash.qmcpack.org/CDash/queryTests.php?project=QMCPACK&date=2017-09-07&filtercount=1&showfilters=1&field1=testname&compare1=63&value1=short-bccH_1x1x1_ae-csvmc The statistics from last night shows a higher failure rate than expected.
Looking into it. I suspect the error bars might have been incorrectly calculated in the reference (bit too small). I notice that on my workstation, the failures are statistical... rerunning will often set it right. And it passes more often than it fails.
The code is currently guarded against OMP_NUM_THREADS>1. However, attempts to run CSVMC (CSVMCUpdatePbyP driver specifically) with OMP_NUM_THREADS>1 seem to increase BlockCPU by a factor of OMP_NUM_THREADS. Currently running with OMP_NUM_THREADS=32 to see if observables are incorrect as well.