Custom kernels unused bits

Option 2:

@cp.fuse()
def primeFactorizationSum(n):
    num = n
    total = 0
    if num % 2 == 0:
        count = 0
        while num % 2 == 0:
            num //= 2
            count += 1
        total += count * 2
    i = 3
    while i * i <= num:
        if num % i == 0:
            count = 0
            while num % i == 0:
                num //= i
                count += 1
            total += count * i
        i += 2
    if num > 1:
        total += num
    return total


n = 1_000_000
A = cp.empty(n - 1, dtype=cp.int64)


exec_time = timeit.timeit(
    "for i in range(0, n - 1): A[i] = primeFactorizationSum(i + 2)",
    number=20,
    globals=globals()
)

print(benchmark(primeFactorizationSum, (n,), n_repeat=20))

print(f"Execution in {round(exec_time, 3)} s")
print(f"\nPrimes sum array: {A}")

General concepts

Inputs and output format: type + name.

Examples:

float32 a   # NumPy data types can be used
T x         # T = generic type

Elementwise kernels

<kernel name> = cp.ElementwiseKernel(
    '<list of inputs>',
    '<list of outputs>',
    '<operation to perform>',
    '<kernel name>'
)

Example:

squared_diff = cp.ElementwiseKernel(
   'float32 x, float32 y',
   'float32 z',
   'z = (x - y) * (x - y)',
   'squared_diff'
)