antocuni changed the topic of #pypy to: PyPy, the flexible snake (IRC logs: https://botbot.me/freenode/pypy/ ) | use cffi for calling C | "PyPy: the Gradual Reduction of Magic (tm)"
<kenaan>
fijal canraise-assertionerror 904915e5425e /rpython/translator/backendopt/: branch to experiment with canraise not raising for assertion error
<fijal>
ok, that was a rabbit hole....
<cfbolz>
fijal: getting somewhere?
<fijal>
cfbolz: giving up
<fijal>
I dont
<fijal>
understand the flow model enough to spend half a day
<cfbolz>
:-(
<fijal>
*less than
cstratak has quit [Quit: Leaving]
cstratak has joined #pypy
realitix has quit [Read error: Connection reset by peer]
realitix has joined #pypy
<kenaan>
antocuni default 84a682f2e00a /pypy/tool/release/: the .debug files don't need to be executable, remove the 'x' bit
<kenaan>
antocuni default 060286c2177e /pypy/goal/getnightly.py: make sure to extract also the .debug files from the nightly
<magniff>
fijal should have get the audience prepared for tech details
<magniff>
and at the end it turned out that fijals talk was too softcore and mine too hardcore with gdb, jit bytecode and whatnot)
<antocuni>
hardcore talks are the best :)
<magniff>
agreed :)
aboudreault has quit [Excess Flood]
aboudreault has joined #pypy
aboudreault has quit [Excess Flood]
aboudreault has joined #pypy
aboudreault has quit [Excess Flood]
aboudreault has joined #pypy
aboudreault has quit [Excess Flood]
tos9 has quit [Ping timeout: 248 seconds]
aboudreault has joined #pypy
Thinh has quit [Quit: Bye!]
Thinh has joined #pypy
aboudreault has quit [Excess Flood]
marky1991 has quit [Ping timeout: 260 seconds]
aboudreault has joined #pypy
aboudreault has quit [Excess Flood]
aboudreault has joined #pypy
marky1991 has joined #pypy
tos9 has joined #pypy
mattip has joined #pypy
<magniff>
Is there any objective comparations of performance for Numba and PyPy?
raynold has quit [Quit: Connection closed for inactivity]
<fijal>
magniff: I haven't seen one
<fijal>
it's a bit hard to do
aboudreault has quit [Excess Flood]
<magniff>
yes, it`s hard in general
<magniff>
yet you could try some narrow algorithmic examples
yuyichao has quit [Ping timeout: 264 seconds]
<ionelmc>
so if i have a out-of-line binding and i have a wheel for it do i still need to have cffi installed?
exarkun has quit [Read error: Connection reset by peer]
aboudreault has joined #pypy
aboudreault has quit [Excess Flood]
<fijal>
magniff: so the key is - how do you optimize?
<fijal>
I'm sure full-optimized numba is better than full-optimized pypy
<fijal>
because they sacrifice some semantics and they have more money
<fijal>
how easy it is to go there? that's a very loaded question
<fijal>
what's "natural"? does "natural" even make sense?
aboudreault has joined #pypy
aboudreault has quit [Excess Flood]
forgottenone has quit [Ping timeout: 248 seconds]
forgottenone has joined #pypy
aboudreault has joined #pypy
<ionelmc>
so actually it;s not needed (it just had some `No module named _cffi_backend`) i suspect i had some unclean build issue but what is that module anyway?
aboudreault has quit [Excess Flood]
yuyichao has joined #pypy
aboudreault has joined #pypy
<magniff>
@fijal i suspect the difference in performance would depend very much on the actual task, yet even though numba is more tailored for numerical calculations I just got better result on pypy for naiive fibonacci function.
<fijal>
naive fib is the worst benchamrk ever :-P
<antocuni>
and it's not even particularly good on pypy (the recursive version, at least)
<mattip>
in rpython/translator/platform there is one called distutils_platform, it is tested (a little) but never actually used in practice,
<mattip>
may I remove it (otherwise it fails on win32 with the dreaded "vcvars not found" error)
<mattip>
fails = its own tests fail
forgottenone has quit [Ping timeout: 248 seconds]
exarkun has joined #pypy
forgottenone has joined #pypy
<magniff>
okok, on iterative factorial numba seems to be way faster
<ionelmc>
fijal: how are the cffi manylinux wheels built?
magniff has quit [Quit: Page closed]
<fijal>
ionelmc: no idea
<ionelmc>
fijal: aren't you the owner of the pypi entry?
<ionelmc>
actually what i wonder is why there's a customized --lib-sdir in the wheel ( i assume someone run auditwheel repair --lib-sdir=.libs_cffi_backend)
<ionelmc>
actually nevermind, it makes sense, as it goes in the root of site-packages
marky1991 has quit [Remote host closed the connection]