BLAS : program is terminated. because you tried to allocate too many memory regions.

[zhuqunyan]

I try to set
os.environ['OPENBLAS_NUM_THREADS'] = '1'
os.environ['GOTO_NUM_THREADS'] = '1'
os.environ['OMP_NUM_THREADS'] = '1'
but, the error info still occur

then This is how I fix the error:

from threadpoolctl import threadpool_info
from threadpoolctl import threadpool_limits
pprint(threadpool_info())
threadpool_limits(limits=1, user_api='blas')
threadpool_limits(limits=1, user_api='openmp')

[zhuqunyan]

os.environ['OPENBLAS_NUM_THREADS'] = '1'
os.environ['GOTO_NUM_THREADS'] = '1'
os.environ['OMP_NUM_THREADS'] = '1'

print('before threadpool_limits:')
pprint(threadpool_info())

threadpool_limits(limits=1, user_api='blas')
threadpool_limits(limits=1, user_api='openmp')
print('after threadpool_limits:')
pprint(threadpool_info())

The following are The stdout:

before threadpool_limits:
[{'architecture': 'Haswell',
'filepath': '/home/pai/lib/python3.6/site-packages/numpy/.libs/libopenblasp-r0-34a18dc3.3.7.so',
'internal_api': 'openblas',
'num_threads': 64,
'prefix': 'libopenblas',
'threading_layer': 'pthreads',
'user_api': 'blas',
'version': '0.3.7'},
{'filepath': '/home/pai/lib/libgomp.so.1.0.0',
'internal_api': 'openmp',
'num_threads': 104,
'prefix': 'libgomp',
'user_api': 'openmp',
'version': None},
{'architecture': 'Haswell',
'filepath': '/home/pai/lib/python3.6/site-packages/scipy.libs/libopenblasp-r0-085ca80a.3.9.so',
'internal_api': 'openblas',
'num_threads': 64,
'prefix': 'libopenblas',
'threading_layer': 'pthreads',
'user_api': 'blas',
'version': '0.3.9'}]
after threadpool_limits:
[{'architecture': 'Haswell',
'filepath': '/home/pai/lib/python3.6/site-packages/numpy/.libs/libopenblasp-r0-34a18dc3.3.7.so',
'internal_api': 'openblas',
'num_threads': 1,
'prefix': 'libopenblas',
'threading_layer': 'pthreads',
'user_api': 'blas',
'version': '0.3.7'},
{'filepath': '/home/pai/lib/libgomp.so.1.0.0',
'internal_api': 'openmp',
'num_threads': 1,
'prefix': 'libgomp',
'user_api': 'openmp',
'version': None},
{'architecture': 'Haswell',
'filepath': '/home/pai/lib/python3.6/site-packages/scipy.libs/libopenblasp-r0-085ca80a.3.9.so',
'internal_api': 'openblas',
'num_threads': 1,
'prefix': 'libopenblas',
'threading_layer': 'pthreads',
'user_api': 'blas',
'version': '0.3.9'}]

[benfred]

Going os.environ['OPENBLAS_NUM_THREADS'] = '1' won't work if you've imported any code that uses numpy before setting this (the environment variable needs to be set when importing the blas library, not before trying to run). Can you try setting as an environment variable before running to see if this also fixes ?

The threadpoolctl library looks neat! thanks for sharing.

However, implicit uses openmp for multithreading - and going threadpool_limits(limits=1, user_api='openmp') might limit implicit to running on a single thread. Setting the blas limits with threadpoolctl should also work though,

[benfred]

I've switched over to using threadpoolctl (instead of relying on the environment variables in #692) to detect threadpool usage in blas.

let me know if you're still having any problems here,