Path: csiph.com!usenet.pasdenom.info!weretis.net!feeder4.news.weretis.net!feeds.phibee-telecom.net!newsfeed.xs4all.nl!newsfeed4a.news.xs4all.nl!xs4all!newsgate.cistron.nl!newsgate.news.xs4all.nl!post.news.xs4all.nl!not-for-mail
To: python-list@python.org
From: Terry Reedy <tjreedy@udel.edu>
Subject: Re: Does CPython already has Peephole optimizations?
Date: Mon, 17 Feb 2014 08:51:16 -0500
References: <mailman.7086.1392625473.18130.python-list@python.org> <5301cf82$0$29985$c3e8da3$5496439d@news.astraweb.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:24.0) Gecko/20100101 Thunderbird/24.3.0
In-Reply-To: <5301cf82$0$29985$c3e8da3$5496439d@news.astraweb.com>
Precedence: list
Newsgroups: comp.lang.python
Message-ID: <mailman.7096.1392645125.18130.python-list@python.org>
Lines: 61
NNTP-Posting-Host: 2001:888:2000:d::a6
Xref: csiph.com comp.lang.python:66606

On 2/17/2014 3:59 AM, Steven D'Aprano wrote:
> On Mon, 17 Feb 2014 13:54:25 +0530, Laxmikant Chitare wrote:
>
>> I read about this article:
>> http://www.python.org/workshops/1998-11/proceedings/papers/montanaro/
> montanaro.html
>>
>> Just wanted to clarify whether CPython already includes these kind of
>> byte code optimizations?

Most of the easily seen and obviously safe low-hanging fruits for the 
compile step have been plucked. Note that the effect of the peephole 
process would only save a few percent, if any, for real apps*. Improving 
the C code invoked by bytecode has resulted in much larger gains.

* We now have a much better benchmark suite with some real apps. This is 
thanks in part to the pypy project.

>> Are all the temporary variables removed when byte code is generated?
>
> You can check these things for yourself:
>
> import dis
> dis.dis(function)
>
> will show you the byte code.
>
> But in general, I would expect not. CPython (that's the Python you
> probably use) doesn't do a lot of optimization apart from some simple
> constant folding. If you're interested in optimizing Python, you should
> look at the JIT optimizing Python compiler, PyPy.

For CPython, new optimization has mostly moved to AST tranformations 
prior to compilation. (Python ASTs are new since Skip started the 
peephole work.)  I believe there are some open issues on the tracker.

Once optimization constraint Skip did not mention is the correspondence 
between source lines and blocks of bytecode, which is used by profiling, 
tracing, and tracebacks. Effectively transforming

if type(a) == types.ComplexType:
     x = cmath.sin(a)
     foo(x)
else:
     x = math.sin(a)
     foo(x)

into

if type(a) == types.ComplexType:
     x = cmath.sin(a)
else:
     x = math.sin(a)
foo(x)

breaks the correspondence. If foo(x) raises, which original line should 
be reported as the source of the exception?

-- 
Terry Jan Reedy