This code demonstrates parallel matrix multiplication in Python using the C API, bypassing the Global Interpreter Lock (GIL). The data structure employed for matrix representation is a 2D list on the Python side, and the resulting type is also a standard 2D list, with double-precision values. The computation is performed in C, while the variables and Python object interaction remain in Python.
-
Data Structure: The data structure is defined on the Python side as a 2D list, and the returned type is also a standard 2D list. The results are in double precision.
-
Parallelization: The computations are parallelized across the CPUs.
To compile the C code, follow replace python3.10 with your specific Python version if necessary.
gcc -I/usr/include/python3.10 -shared -fPIC -O3 -o matrix_multiplication.so matrix_multiplication.c
The compiled library can be included and used in Python as usual. Number of Threads:
You can specify the number of threads to be used for computation in the multiply_matrices function.
import random as rd
import matrix_multiplication
# Define matrix dimensions
d1 = 50
d2 = 800
# Generate random matrices
matrix1 = [[rd.random() for _ in range(d2)] for _ in range(d1)]
matrix2 = [[rd.random() for _ in range(d1)] for _ in range(d2)]
# Specify the number of threads
num_threads = 6
# Call the parallel_matrix_multiply function
# The result is a standart 2D Python list
result = matrix_multiplication.multiply_matrices(matrix1, matrix2, num_threads)