try to get some parallelism / openmp running on TACO's benchmarks

[rohany]

numpy is most likely parallelizing operations (we need to look into how to control the number of threads here), so we should as well.

[rohany]

Based on some googling, we need to set all of

OMP_NUM_THREADS=1
OPENBLAS_NUM_THREADS=1 
MKL_NUM_THREADS=1 
VECLIB_MACIMUM_THREADS=1 
NUMEXPR_NUM_THREADS=1

OPENBLAS_NUM_THREADS seems to control the number of threads for matrix multiply.

[rohany]

playing around with these different settings didn't appear to change the runtime of pydata's benchmarks

[stephenchouca]

For reference, you can build TACO with OpenMP support by running make OPENMP=ON. The TACO mxv benchmark in taco/graphblas.cpp shows how you can configure TACO to run with arbitrary numbers of OpenMP threads.