Hello. We're working on a tracing JIT for PHP using PyPy
(https://bitbucket.org/asuhan/happy). It's still in its infancy, but it
has a bunch of features already in place and beats Zend & HipHop by a
large margin in some microbenchmarks -- sieve (and other arithmetical
stuff) and lots of tests based on PHPbench.
However, we’re struggling with poor performance (2x slowdown compared to
Zend) for dictionaries -- especially set operations. The implementation
in pypy/rpython/lltypesystem/rdict.py seems perfectly fine at a first
glance and we’re relying on it for implementing PHP arrays.
pypy-c (built from 43932 changeset) is also about 2x slower than CPython
for something like:
a = {}
t = 0
while t != 1000000:
a[t] = t
t += 1
While we’re not expecting dictionaries to benefit from JIT magic, it’s a
little strange that the “baseline” performance is not up to par. Is
there a way of circumventing this issue?
Many thanks for this piece of software and keep up the good work.
--
asuhan
_______________________________________________
pypy-dev mailing list
pypy-dev@python.org
http://mail.python.org/mailman/listinfo/pypy-dev
