Two similar codes differ in speed in PyPy3

Question

Here is a simplified code of nested loop and some if-conditions regarding index-access to a list.
I noticed that the second code (# In separate loops) works twice faster than the first one (# In the same loops) in PyPy3 (ver. 7.3.1).
Both run in the almost same speed in CPython (3.8.4rc1).

What is the factor making the second one faster?

# In the same loop (slower)
import random

LIM = 10 ** 3


def main():
    random.seed(0)

    arr = [random.randint(0, 1) for _ in range(LIM * LIM)]
    cnt = 0
    for i in range(LIM):
        for j in range(LIM):
            if arr[i]:
                cnt += 1
            if arr[i * 2]:
                cnt += 1
            if arr[j]:
                cnt += 1
            if arr[j * 2]:
                cnt += 1
            if arr[i + LIM]:
                cnt += 1
            if arr[j + LIM]:
                cnt += 1
            if arr[i * j]:
                cnt += 1
    print(cnt)


if __name__ == "__main__":
    main()

# In separate loops (faster)
import random

LIM = 10 ** 3


def main():
    random.seed(0)

    arr = [random.randint(0, 1) for _ in range(LIM * LIM)]
    cnt = 0
    for i in range(LIM):
        for j in range(LIM):
            if arr[i]:
                cnt += 1
    for i in range(LIM):
        for j in range(LIM):
            if arr[i * 2]:
                cnt += 1
    for i in range(LIM):
        for j in range(LIM):
            if arr[j]:
                cnt += 1
    for i in range(LIM):
        for j in range(LIM):
            if arr[j * 2]:
                cnt += 1
    for i in range(LIM):
        for j in range(LIM):
            if arr[i + LIM]:
                cnt += 1
    for i in range(LIM):
        for j in range(LIM):
            if arr[j + LIM]:
                cnt += 1
    for i in range(LIM):
        for j in range(LIM):
            if arr[i * j]:
                cnt += 1
    print(cnt)


if __name__ == "__main__":
    main()

I don't know if that's the cause, but the two programs do different things. For example, the second program doesn't do `i // 2`. — Armin Rigo, Jul 16 '20 at 18:32
Likely that the second one can be optimized a lot because of the single condition in the loop — Mad Physicist, Jul 17 '20 at 00:29
@MadPhysicist What else can be optimized in PyPy? I want to know the rules/tips to write a fast PyPy code. — solzard, Jul 17 '20 at 01:08
@MadPhysicist I skimmed the official document (https://doc.pypy.org/), but couldn't find the rule which conditions PyPy optimizes loop statements in. — solzard, Jul 18 '20 at 22:38

Two similar codes differ in speed in PyPy3

0 Answers0