[pypy-dev] Speeding up zlib in standard library

Tue Mar 20 15:52:42 CET 2012

On Tue, Mar 20, 2012 at 4:40 PM, Peter Cock <p.j.a.cock at googlemail.com> wrote:
> Hi again,
>
> Based on the structure of pypy/translator/goal/richards.py and
> pypy/translator/goal/targetrichards.py I have tried to make a
> simple ZLIB based benchmark based on what I have been using.
>
> $ python2.6 green_bottles.py
> Green bottles ZLIB benchmark starting... [<function entry_point at 0x1004c7500>]
> Trying ZLIB on 176683 bytes of text 10000 times
> finished.
> Total time for 10000 iterations: 34.25 secs
> Average time per iteration: 3.42 ms
>
> $ pypy green_bottles.py
> Green bottles ZLIB benchmark starting... [<function entry_point at
> 0x0000000102f3b3d0>]
> Trying ZLIB on 176683 bytes of text 10000 times
> finished.
> Total time for 10000 iterations: 45.12 secs
> Average time per iteration: 4.51 ms
>
> Here PyPy v1.8 is about 1.3 times as slow as C Python 2.6, not
> as bad as the factor of almost two I was seeing in some cases,
> but enough to tackle as a measurable target?
>
> Files here: https://gist.github.com/2136181
>
> However, I have a problem with using translate.py on this
> (trying the PyPy release-1.8 tag or the latest from bitbucket):
>
> $ python2.6 translate.py --run targetzlib.py
> [platform:msg] Setting platform to 'host' cc=None
> [translation:info] Translating target as defined by targetzlib
> [platform:execute] gcc ...
> [translation] translate.py configuration:
> [translation] [translate]
> [translation]     [goal_options]
> [translation]         run = True
> [translation]     targetspec = targetzlib
> [translation] translation configuration:
> [translation] [translation]
> [translation]     gc = minimark
> [translation]     gctransformer = framework
> [translation]     list_comprehension_operations = True
> [translation]     withsmallfuncsets = 5
> [translation:info] Annotating&simplifying...
> [translation:info] with policy: pypy.annotation.policy.AnnotatorPolicy
> [platform:execute] gcc ...
> [Timer] Timings:
> [Timer] annotate                       --- 1.6 s
> [Timer] ========================================
> [Timer] Total:                         --- 1.6 s
> [translation:ERROR] Error:
> [translation:ERROR]  Traceback (most recent call last):
> [translation:ERROR]    File "translate.py", line 309, in main
> [translation:ERROR]     drv.proceed(goals)
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/translator/driver.py", line 810, in
> proceed
> [translation:ERROR]     return self._execute(goals, task_skip =
> self._maybe_skip())
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/translator/tool/taskengine.py", line 116,
> in _execute
> [translation:ERROR]     res = self._do(goal, taskcallable, *args, **kwds)
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/translator/driver.py", line 287, in _do
> [translation:ERROR]     res = func()
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/translator/driver.py", line 324, in
> task_annotate
> [translation:ERROR]     s = annotator.build_types(self.entry_point,
> self.inputtypes)
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/annotation/annrpython.py", line 107, in
> build_types
> [translation:ERROR]     return self.build_graph_types(flowgraph,
> inputcells, complete_now=complete_now)
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/annotation/annrpython.py", line 198, in
> build_graph_types
> [translation:ERROR]     self.complete()
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/annotation/annrpython.py", line 254, in
> complete
> [translation:ERROR]     self.processblock(graph, block)
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/annotation/annrpython.py", line 452, in
> processblock
> [translation:ERROR]     self.flowin(graph, block)
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/annotation/annrpython.py", line 512, in
> flowin
> [translation:ERROR]     self.consider_op(block.operations[i])
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/annotation/annrpython.py", line 695, in
> consider_op
> [translation:ERROR]     argcells = [self.binding(a) for a in op.args]
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/annotation/annrpython.py", line 297, in
> binding
> [translation:ERROR]     return self.bookkeeper.immutableconstant(arg)
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/annotation/bookkeeper.py", line 318, in
> immutableconstant
> [translation:ERROR]     return self.immutablevalue(const.value)
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/annotation/bookkeeper.py", line 469, in
> immutablevalue
> [translation:ERROR]     result = SomePBC([self.getdesc(x)])
> [translation:ERROR]    File
> "/Users/pjcock/pypy-hg/pypy/annotation/bookkeeper.py", line 535, in
> getdesc
> [translation:ERROR]     pyobj,))
> [translation:ERROR]  Exception: unexpected prebuilt constant:
> <built-in function compressobj>
> [translation:ERROR] Processing block:
> [translation:ERROR]  block at 27 is a <class
> 'pypy.objspace.flow.flowcontext.SpamBlock'>
> [translation:ERROR]  in (pypy.translator.goal.green_bottles:10)compress
> [translation:ERROR]  containing the following operations:
> [translation:ERROR]        v0 =
> simple_call((builtin_function_or_method compressobj), (6), (8), (-15),
> (8), (0))
> [translation:ERROR]        v1 = getattr(v0, ('compress'))
> [translation:ERROR]        v2 = simple_call(v1, orig_data_0)
> [translation:ERROR]        v3 = getattr(v0, ('flush'))
> [translation:ERROR]        v4 = simple_call(v3)
> [translation:ERROR]        v5 = add(v2, v4)
> [translation:ERROR]        v6 =
> simple_call((builtin_function_or_method crc32), orig_data_0)
> [translation:ERROR]        v7 = newtuple(v5, v6)
> [translation:ERROR]  --end--
> [translation] start debugger...
>> /Users/pjcock/pypy-hg/pypy/annotation/bookkeeper.py(535)getdesc()
> -> pyobj,))
> (Pdb+) quit
>
> Calling it without the --run gives the same:
>
> $ python2.6 translate.py targetzlib.py
> ...
>
> Should my benchmark Python code be RPython?
>
> Thanks,
>
> Peter
> _______________________________________________
> pypy-dev mailing list
> pypy-dev at python.org
> http://mail.python.org/mailman/listinfo/pypy-dev

Hey

You're confusing levels. Your benchmark code cannot call zlib. I
suggested you benchmark not zlib, but your changes to StringBuilder
that you proposed. That does not (and cannot) require zlib. You'll be
testing something smaller that you're trying to change instead. Then
obviously a comparison with CPython is meaningless.

Cheers,
fijal