Segfaults with Gaussian backend involving calculating probabilities on MacOS x86_64

[sansiro77]

Environment: Conda 23.1.0

• Create a Python environment, then install all related packages via pip

Example

cir = dq.QumodeCircuit(nmode=2, init_state='vac', cutoff=5, backend='gaussian')
cir.s(0)
cir.s(1)
cir.d(0)
cir.d(1)
cir.bs([0,1])

cir()
cir.measure()
# or
cir(is_prob=True)

Error information

• For cir.measure()

  OMP: Info #276: omp_set_nested routine deprecated, please use omp_set_max_active_levels instead. [1] 17214 segmentation fault python /usr/local/anaconda3/envs/test/lib/python3.10/multiprocessing/resource_tracker.py:224: UserWarning: resource_tracker: There appear to be 1 leaked semaphore objects to clean up at shutdown warnings.warn('resource_tracker: There appear to be %d '

• For cir(is_prob=True)

  OMP: Info #276: omp_set_nested routine deprecated, please use omp_set_max_active_levels instead.

Reasons

• PyTorch is based on Intel OpenMP ('libiomp') and MKL by default, but other packages (scipy, scikit-learn, threadpoolctl, etc.) are based on LLVM OpenMP ('libomp') and OpenBLAS
• There is a known conflict between libiomp and libomp on MacOS x86_64
• At present, we use Hafnian and Torontonian functions from thewalrus

Possible solutions

• Build PyTorch based on OpenBLAS from source
• Implement Hafnian and Torontonian functions by PyTorch

Reference

• Found Intel OpenMP ('libiomp') and LLVM OpenMP ('libomp') loaded at the same time. Both libraries are known to be incompatible and this can cause random crashes or deadlocks on Linux when loaded in the same Python program. Using threadpoolctl may cause crashes or deadlocks. For more information and possible workarounds, please see https://github.com/joblib/threadpoolctl/blob/master/multiple_openmp.md

• https://github.com/ContinuumIO/anaconda-issues/issues/13221

• https://github.com/XanaduAI/thewalrus/issues/367

[sansiro77]

Solution

• use Numba==0.55.2