93 lines
3.6 KiB
Python
93 lines
3.6 KiB
Python
|
from concurrent.futures import ThreadPoolExecutor
|
|||
|
|
import time
|
|||
|
|
import numpy as np
|
|||
|
|
import argparse
|
|||
|
|
|
|||
|
|
# Последовательное умножение
|
|||
|
|
def matrix_multiply_sequential(A, B):
|
|||
|
|
n = len(A)
|
|||
|
|
C = [[0] * n for _ in range(n)]
|
|||
|
|
|
|||
|
|
# Транспонируем матрицу B для оптимального доступа по строкам
|
|||
|
|
B_T = [[B[j][i] for j in range(n)] for i in range(n)]
|
|||
|
|
|
|||
|
|
for i in range(n):
|
|||
|
|
A_row = A[i]
|
|||
|
|
for j in range(n):
|
|||
|
|
B_col = B_T[j]
|
|||
|
|
sum_ij = 0
|
|||
|
|
for k in range(n):
|
|||
|
|
sum_ij += A_row[k] * B_col[k]
|
|||
|
|
C[i][j] = sum_ij
|
|||
|
|
|
|||
|
|
return C
|
|||
|
|
|
|||
|
|
# Вычисляет подматрицу C
|
|||
|
|
def worker(A, B, C, start_row, end_row):
|
|||
|
|
n = len(A)
|
|||
|
|
for i in range(start_row, end_row):
|
|||
|
|
for j in range(n):
|
|||
|
|
sum_ij = 0
|
|||
|
|
for k in range(n):
|
|||
|
|
sum_ij += A[i][k] * B[k][j]
|
|||
|
|
C[i][j] = sum_ij
|
|||
|
|
|
|||
|
|
# Параллельное умножение матриц
|
|||
|
|
def matrix_multiply_parallel(A, B, num_threads):
|
|||
|
|
n = len(A)
|
|||
|
|
C = [[0] * n for _ in range(n)]
|
|||
|
|
|
|||
|
|
# Разбиваем строки между потоками
|
|||
|
|
rows_per_thread = n // num_threads
|
|||
|
|
extra_rows = n % num_threads
|
|||
|
|
row_splits = [rows_per_thread + (1 if i < extra_rows else 0) for i in range(num_threads)]
|
|||
|
|
row_indices = [sum(row_splits[:i]) for i in range(num_threads + 1)]
|
|||
|
|
|
|||
|
|
# Параллельно выполняем умножение подматриц
|
|||
|
|
with ThreadPoolExecutor(max_workers=num_threads) as executor:
|
|||
|
|
futures = [
|
|||
|
|
executor.submit(worker, A, B, C, row_indices[i], row_indices[i+1])
|
|||
|
|
for i in range(num_threads)
|
|||
|
|
]
|
|||
|
|
for future in futures:
|
|||
|
|
future.result()
|
|||
|
|
|
|||
|
|
return C
|
|||
|
|
|
|||
|
|
def benchmark(matrix_sizes, num_threads_list):
|
|||
|
|
for size in matrix_sizes:
|
|||
|
|
A = [[1] * size for _ in range(size)] # Матрицы, заполненные единицами для упрощения теста
|
|||
|
|
B = [[1] * size for _ in range(size)]
|
|||
|
|
print(f"\nРазмер матриц: {size}x{size}")
|
|||
|
|
|
|||
|
|
# Бенчмарк для последовательного алгоритма
|
|||
|
|
start_time = time.time()
|
|||
|
|
C_seq = matrix_multiply_sequential(A, B)
|
|||
|
|
sequential_time = time.time() - start_time
|
|||
|
|
print(f"Последовательное умножение заняло: {sequential_time:.4f} секунд")
|
|||
|
|
|
|||
|
|
# Бенчмарк для параллельного алгоритма
|
|||
|
|
for num_threads in num_threads_list:
|
|||
|
|
start_time = time.time()
|
|||
|
|
C_par = matrix_multiply_parallel(A, B, num_threads)
|
|||
|
|
parallel_time = time.time() - start_time
|
|||
|
|
print(f"Параллельное умножение с {num_threads} потоками заняло: {parallel_time:.4f} секунд")
|
|||
|
|
|
|||
|
|
if __name__ == "__main__":
|
|||
|
|
parser = argparse.ArgumentParser(description="Запуск бенчмарков для умножения матриц.")
|
|||
|
|
parser.add_argument(
|
|||
|
|
"--threads",
|
|||
|
|
type=int,
|
|||
|
|
nargs="+",
|
|||
|
|
required=True,
|
|||
|
|
help="Список количества потоков для параллельного алгоритма (например, --threads 1 2 4 8)"
|
|||
|
|
)
|
|||
|
|
args = parser.parse_args()
|
|||
|
|
|
|||
|
|
# Задание размеров матриц для тестов
|
|||
|
|
matrix_sizes = [100, 300, 500]
|
|||
|
|
num_threads_list = args.threads # Получаем список потоков из аргументов командной строки
|
|||
|
|
|
|||
|
|
# Запуск бенчмарка
|
|||
|
|
benchmark(matrix_sizes, num_threads_list)
|