Julien Schueller

do you have some benchmarking results ?

then it also depends on the parallel flag of your weight function

it starts to look good, does the tests still fail ?

I dont know, some kind of deadlock, maybe something else is not thread-safe here ?

so not a deadlock but only a timeout ? it might be related to the way TBBContext disables openblas threading in a nested tbb parallel region since the new loops...

this is likely not macos-specific but rather due to the openblas code (we'll see this on linux too), see my previous comment

I'd try to disable openblas threading at the global level in the cxx test main function and comment the tbbcontext instanciation in the tbb for loop to see if the...

`OPENBLAS_NUMTHREADS=1 OMP_NUM_THREADS=1 ctest ...`

Maybe it uses some numerical estimation to compute de correlation this way instead of yielding the exact value, but you should use BlockIndependentDistribution to assemble mutivariate distributions anyway: ``` dist_V...

are you saying UseGenericCovarianceAlgorithm should be flipped on ?