Mailman 3 November 2014 - pypy-dev

use as benchmark pypy vs python if you please
by Dima Tisnek Sept. 13, 2015

Sept. 13, 2015

Attached is n-queens solver (pardon my naive algorithm), it runs: python 2.7.6: 17s pypy 2.4.0 alpha: 23s same nojit: 32s I've tried similar-looking algorithm for another problem before, and has similar results -- somehow pypy was slower. feel free to investigate / tweak or even use on speed.pypy.org

4 4

Poor performance for Krakatau
by Robert Grosse Aug. 10, 2015

Aug. 10, 2015

In some cases, Pypy performs worse than CPython when running Krakatau, and it has gotten a lot worse from 2.3 to 2.5 My benchmark is https://github.com/Storyyeller/Krakatau/tree/pypy_benchmark (for future reference, commit d889c7f44723e6d66a3630681f0385f173317dc9) You can run it with pypy Krakatau\benchmark.py -path <path to Java rt.jar> I get the following timings CPython: 17.43s Pypy 2.3: 38.48s (build pypy-c-jit-71056-c8e3b8cbc843-win32) Pypy 2.4: 39.92s (build pypy-c-jit-72200-375133966c12-win32) Pypy 2.5: 49.94s (build pypy-c-jit-74404-9fd586fe0fe5-win32) Not only was it slower than CPython to begin with, but it got much worse recently.

3 4

GSoC 2015: cpyext project?
by Toby St Clere Smithe Dec. 3, 2014

Dec. 3, 2014

Hi all, I've posted a couple of times on here before: I maintain a Python extension for GPGPU linear algebra[1], but it uses boost.python. I do most of my scientific computing in Python, but often am forced to use CPython where I would prefer to use PyPy, largely because of the availability of extensions. I'm looking for an interesting Google Summer of Code project for next year, and would like to continue working on things that help make high-performance computing in Python straight-forward. In particular, I've had my eye on the 'optimising cpyext'[2] project for a while: might work in that area be available? I notice that it is described with difficulty 'hard', and so I'm keen to enquire early so that I can get up to speed before making a potential application in the spring. I would love to work on getting cpyext into a good enough shape that both Cython and Boost.Python extensions are functional with minimal effort on behalf of the user. Does anyone have any advice? Are there particular things I should familiarise myself with? I know there is the module/cpyext tree, but it is quite formidable for someone uninitiated! Of course, I recognise that cpyext is a much trickier proposition in comparison with things like cffi and cppyy. In particular, I'm very excited by cppyy and PyCling, but they seem quite bound up in CERN's ROOT infrastructure, which is a shame. But it's also clear that very many useful extensions currently use the CPython API, and so -- as I have often found -- the apparent relative immaturity of cpyext keeps people away from PyPy, which is also a shame! [1] https://pypi.python.org/pypi/pyviennacl [2] https://bitbucket.org/pypy/pypy/wiki/GSOC%202014 Best, Toby -- Toby St Clere Smithe http://tsmithe.net

4 7

Was dict subclass discrepancy "fixed" (issue 708)?
by Luciano Ramalho Nov. 27, 2014

Nov. 27, 2014

Hello, I am writing a book about Python 3 [0] and while researching the caveats of subclassing built-in types I discovered the page "Differences between PyPy and CPython" [1] and issue #708 "Discrepancy in dict subclass __getitem__ calls between CPython 2.7 and PyPy 1.5" [2]. [0] http://shop.oreilly.com/product/0636920032519.do [1] http://pypy.readthedocs.org/en/latest/cpython_differences.html#subclasses-o… [2] https://bitbucket.org/pypy/pypy/issue/708/discrepancy-in-dict-subclass-__ge… However, when testing with pypy3-2.4.0 and pypy-2.4.0 my results were the same as with CPython, and not as documented in [1]. So was issue 708 "fixed" and now PyPy misbehaves in the same way as CPython? Thanks! Best, Luciano -- Luciano Ramalho Twitter: @ramalhoorg Professor em: http://python.pro.br Twitter: @pythonprobr

3 3

Re: [pypy-dev] An idea about automatic parallelization in PyPy/RPython
by haael＠interia.pl Nov. 26, 2014

Nov. 26, 2014

Hi 黄若尘, Armin, > > I thought about that too, but the granularity is very wrong for STM: > > the overhead of running tiny transactions will completely dwarf any > > potential speed gains. Then maybe we should go with a slightly different version of STM, specific to assembler loops in particular. A loop in assembly language operates on a linear memory model and usually modifies only a small number of memory cells. In fact, most changes made by a loop iteration get overwritten in the next iteration. Assuming that a single assembler instruction may modify only one memory cell, the number of cells changed will be no more than the count of loop iterations. We could replace (para-virtualize) any instruction that changes a memory cell with two stack pushes: save the memory address to the stack and the actual value that is written. Any memory read will also be replaced with the stack search before falling back to reading the actual memory contents. This is a big penalty of course but it's worth checking whether it pays in the future. This is a simple, assembler-specific flavor of STM. Then we could employ loop scheduling and run the modified code on all cores. Then we could check whether all the memory modifications agree, that means whether any two cores did not try to write different value to the same memory address. If not, then we could commit the transaction and exit the loop. The kind of loops that would benefit most from such optimization would be memset, memcpy and all map-like constructs: dst = map(fun, src) ---> for(int i = 0; i < len(src); i++) dst[i] = fun(src[i]); > can we just make a hybrid-system that firstly slightly screen > some loops that is not suitable for parallelization and then run others with STM? Following the general "try and fail" philosophy of Python, I would suggest the following: Just run the unmodified loop on one core and use the other cores to optimize/execute the modified version. If the optimization turns out unsuitable or the serial execution ends first, just abort the optimized run. If the loop turns out to be parralelizable, return the results instead. Thanks haael Od: "黄若塵" <hrc706(a)gmail.com> Do: "Armin Rigo" <arigo(a)tunes.org>; haael(a)interia.pl; Wysłane: 9:07 Środa 2014-11-26 Temat: Re: [pypy-dev] An idea about automatic parallelization in PyPy/RPython > Hi Haael, Rigo, > > 2014/11/21 19:21、Armin Rigo のメール： > > > > Hi Haael, hi 黄若尘, > > > > On 21 November 2014 10:55, wrote: > >> I would suggest a different approach, more similar to Armin's idea of parallelization. > >> > >> You could just optimistically assume that the loop is parallelizable. Just execute few steps at once (each in its own memory sandbox) and check for conflicts later. This also plays nice with STM. > > > > I thought about that too, but the granularity is very wrong for STM: > > the overhead of running tiny transactions will completely dwarf any > > potential speed gains. If we're talking about tiny transactions then > > maybe HTM would be more suitable. I have no idea if HTM will ever > > start appearing on GPU, though. Moreover, you still have the general > > hard problems of automatic parallelization, like communicating between > > threads the progress made; unless it is carefully done on a > > case-by-case basis by a human, this often adds (again) considerable > > overheads. > > Well, recently I have read some papers about TLS, and also realized the heavy > performance penalty of STM. What am I considering is that, is it possible to > simplify a STM for the trace generated by RPython using some features of it (for > example there is no control flow but only guard; there are some jit.elidable functions > in the interpreter), or, can we just make a hybrid-system that firstly slightly screen > some loops that is not suitable for parallelization and then run others with STM? > > > > > To 黄若尘: here's a quick answer to your question. It's not very clean, > > but I would patch rpython/jit/backend/x86/regalloc.py, prepare_loop(), > > just after it calls _prepare(). It gets a list of rewritten > > operations ready to be turned into assembler. I guess you'd need to > > check at this point if the loop contains only operations you support, > > and if so, produce some different code (possibly GPU). Then either > > abort the job here by raising some exception, or if it makes sense, > > change the 'operations' list so that it becomes just a few assembler > > instructions that will start and stop the GPU code. > > > > My own two cents about this project, however, is that it's relatively > > easy to support a few special cases, but it quickly becomes very, very > > hard to support more general code. You are likely to end up with a > > system that only compiles to GPU some very specific templates and > > nothing else. The end result for a user is obscure, because he won't > > get to use the GPU unless he writes loops that follow exactly some > > very strict rules. I certainly see why the end user might prefer to > > use a DSL instead: i.e. he knows he wants to use the GPU at specific > > places, and he is ready to use a separate very restricted "language" > > to express what he wants to do, as long as it is guaranteed to use the > > GPU. (The needs in this case are very different from the general PyPy > > JIT, which tries to accelerate any Python code.) > > > > > > A bientôt, > > > > Armin. > >

1 0

An idea about automatic parallelization in PyPy/RPython
by 黄若尘 Nov. 21, 2014

Nov. 21, 2014

Hi everyone, I’m a master student in Japan and I want to do some research in PyPy/RPython. I have read some papers about PyPy and I also had some ideas about it. I have communicated with Mr. Bloz and been advised to send my question here. Actually, I wonder if it is possible to make an automatic parallelization for the trace generated by JIT, that is, check if the hot loop is a parallel loop, if so, then try to run the trace parallel in multi-core CPU or GPU, make it faster. I think it maybe suitable because: 1. The traced-base JIT is targeting on loops, which is straight to parallel computation. 2. There is no control-flow in trace, which is suitable to the fragment program in GPU. 3. We may use the hint of @elidable in interpreter codes, since the elidable functions are nonsensitive in the execution ordering so can be executed parallel. What do you think about it? Best Regards, Huang Ruochen

4 7

Python Users list
by Adam Savey Nov. 20, 2014

Nov. 20, 2014

Hi, I was reviewing your website and thoughts you would be interested in reaching out to Python Users from USA. A few other technologies include: Java Users, J2EE Users, Linux Users, Dot net Users, Oracle RDBMS Users, Spring Users, Puppet Users. Ø 100% permission based contacts. Ø We guarantee 95% accuracy on data fields and 85% on email deliverability. Ø Once you purchase the list you can use it for multiple times, no restrictions. Ø The list can be used for Email Marketing, Direct Mail Marketing, Fax Marketing and Tele Marketing. We also have: Oracle Users, IBM Users, AWS Users, Salesforce Users, Google Apps Users, Google Analytics Users, Zoho Users, Sage Users, Sugar CRM Users, Infor Users, Epicor Users, Kronos Users, Netsuite Users, VMWare Users, IBM Cognos Users, SAP Users, SaaS Users, Middleware Users & many more.... Hit reply & send in your target criteria, I'll get back to you with counts, cost and other details for your review. If you've different target, feel free to get back to me with it. Regards, Adam Savey Demand Generation Executive We respect your privacy, if you do not wish to receive any further emails from our end, please reply with a subject Cancel

1 0

How to get your iPhone 6
by Shelly Greer Nov. 18, 2014

Nov. 18, 2014

Get a brand new iPhone 6*! Enter your ZIP for availability here http://www.desocom.com/Chance+to+Get+an++iPhone+6/ed6V86hN2uf2CkakadNOXFkaV… This advertisement was sent to you by a third party. If you are not interested in receiving future RewardZoneUsa advertisement, please go here http://www.desocom.com/enterprises/See+where+you+can+find+your+iPhone+6/d8d… *Participation required. See offer for details Alternatively, you can opt out by sending a letter to: 128 Court Street, 3rd FL White Plains, NY 10601

1 0

trying to install aubio
by Stuart Axon Nov. 17, 2014

Nov. 17, 2014

Hi, I've been trying to install aubio .. + it looks like it gets stuck because of numpy, and possibly python c bindings .. any idea whats going on ? S++

3 4

Does Numpypy works on Py3k?
by Alessandro Stamatto Nov. 16, 2014

Nov. 16, 2014

Hi, I'm very happy with the new Py3k branch of PyPy. Does Numpypy works on it? (If not, will it work in the future?) I tried installing it (on windows) with the recommended way (pip install git+ https://bitbucket.org/pypy/numpy.git) but it failed with a lot of errors. On the same vein, will Py3k have support for the other PyPy projects, like STM? Kudos for the awesome work on PyPy, it's a fantastic project! Thanks in advance, Alessandro Stamatto.

2 2