numpy, for comparison: In [8]: vec = np.random.randint(-200, 200, (100_000_000,)) In [9]: %timeit vec.sum() 63 ms ± 4.81 ms per loop (mean ± std. dev

numpy, for comparison: In [8]: vec = np.random.randint(-200, 200, (100_000_000,)) In [9]: %timeit vec.sum() 63 ms ± 4.81 ms per loop (mean ± std. dev. of 7 runs, 10 loops each) The branchless C++ version took 125ms, and the AVX 512 version took ~9ms.