Impressive but
> Is PyPy a drop in replacement for CPython?
> Almost!
> The most likely stumbling block for any given project is support for extension modules. PyPy supports ... mostly only those found in the standard library
| [reply] |
Talking about JIT, maybe try benchmarking a JS version too. :)
| [reply] |
That's intriguing. I admit I've never used Perl's bignum or bigint features for anything serious. Are you able to post short benchmark programs for this? It would be interesting to profile and figure out where the bottleneck is. Which version of Perl, by the way?
| [reply] |
Well, I hate to post a complete solution to one of their problems, so I've changed it around. The question is, given the first 10 outputs of this pseudo-random number generator:
for (1..10) {
$s = ($s*0x5deece66d + 0xb) % 2**48;
say(($s>>17)%1000);
}
Find the initial value of the seed $s, which is a randomly-chosen number less than 2**17. Here are the comparisons of a brute-force search using several different modules:
s/iter BigInt BigIntGMP GMP Perl
BigInt 25.0 -- -42% -94% -100%
BigIntGMP 14.4 73% -- -90% -99%
GMP 1.47 1605% 885% -- -93%
Perl 0.110 22701% 13070% 1238% --
I'm running perl 5.24.1 on an aging MacBook. I could make the Perl faster, but it would require a 64-bit perl build. You could probably get good performance with Math::Int64, but I'll leave that as an exercise for the reader. Python is running in time comparable to the pure-Perl, and PyPy smokes them all at under a millisecond. Full code is below. You don't need to tell me that my benchmarking methodology is bad.
| [reply]
[d/l]
[select] |
