With --jit off, PyPy's interpreter is much slower than CPython

[gitlab-importer]

In Heptapod by @mattip on Feb 13, 2022, 07:40

Thinking about issue #3667, I was wondering how much slower is our base interpreter than CPython's interpreter. When comparing the code below to CPython, I get these results. Can we do better? The first is startup and import time, the difference between the second and third is that in the inner loop both an __and__ and a __add__ are used. Why should that slow down PyPy even more?

command line	time	relative to CPython
python3.8 /tmp/simple.py	14ms	1x
pypy3.8-HEAD/bin/pypy --jit off /tmp/simple.py	41ms	2.9x
---	---	---
python3.8 /tmp/simple.py sumup	264ms	1x
python3.8 /tmp/simple.py sumup	1082ms	4.1x
---	---	---
python3.8 /tmp/simple.py suminv	325ms	1x
python3.8 /tmp/simple.py suminv	1654ms	5.1x

/tmp/simple.py

bits = [1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384,
32768, 65536, 131072, 262144, 524288, 1048576, 2097152, 4194304, 8388608,
16777216, 33554432, 67108864, 134217728, 268435456, 536870912, 1073741824,
2147483648, 4294967296, 8589934592, 17179869184, 34359738368, 68719476736,
137438953472, 274877906944, 549755813888, 1099511627776, 2199023255552,
4398046511104, 8796093022208, 17592186044416, 35184372088832, 70368744177664,
140737488355328, 281474976710656, 562949953421312, 1125899906842624,
2251799813685248, 4503599627370496, 9007199254740992, 18014398509481984,
36028797018963968, 72057594037927936, 144115188075855872, 288230376151711744,
576460752303423488, 1152921504606846976, 2305843009213693952,
4611686018427387904, 9223372036854775808]

total = 0

def sumup(bits):
    total = 0
    for b, binv in zip(bits, bits[::-1]):
        total += b
    return total

def suminv(bits):
    total = 0
    for b, binv in zip(bits, bits[::-1]):
        total += b & binv
    return total

if __name__ == '__main__':
    import sys
    if len(sys.argv) < 2:
        pass
    elif sys.argv[1] == 'sumup':
        for i in range(100_000):
            total += sumup(bits)
    elif sys.argv[1] == 'suminv':
        for i in range(100_000):
            total += suminv(bits)

[gitlab-importer]

In Heptapod by @arigo on Feb 13, 2022, 07:45

Why should that slow down PyPy even more?

Unless I'm missing something, in the third case all the computations are involving longs. In the second case almost all iterations of the loop are done with machine integers. Maybe try to remove the last entry in bits?

[gitlab-importer]

In Heptapod by @mattip on Feb 13, 2022, 08:21

The last entries do seem to influence the relative timings. Without the last three values in `bits', the ratio is more consistent

command line	time	relative to CPython
python3.8 /tmp/simple.py	14ms	1x
pypy3.8-HEAD/bin/pypy --jit off /tmp/simple.py	41ms	2.9x
---	---	---
python3.8 /tmp/simple.py sumup	256ms	1x
python3.8 /tmp/simple.py sumup	956ms	3.7x
---	---	---
python3.8 /tmp/simple.py suminv	340ms	1x
python3.8 /tmp/simple.py suminv	1250ms	3.7x