Basic Python Questions - Oct. 31, 2013

[Robert Kern]

On 2013-10-31 14:49, Chris Angelico wrote:
> On Fri, Nov 1, 2013 at 1:41 AM, Robert Kern <robert.kern at gmail.com> wrote:
>> On 2013-10-31 14:05, Chris Angelico wrote:
>>>
>>> On Fri, Nov 1, 2013 at 12:17 AM, Alain Ketterlin
>>> <alain at dpt-info.u-strasbg.fr> wrote:
>>>>
>>>> "E.D.G." <edgrsprj at ix.netcom.com> writes:
>>>>
>>>>>         The calculation speed question just involves relatively simple
>>>>> math such as multiplications and divisions and trig calculations such
>>>>> as sin and tan etc.
>>>>
>>>>
>>>> These are not "simple" computations.
>>>>
>>>> Any compiled language (Fortran, C, C++, typically) will probably go much
>>>> faster than any interpreted/bytecode-based language (like python or
>>>> perl, anything that does not use a jit).
>>>
>>>
>>> Well, they may not be simple to do, but chances are you can push the
>>> work down to the CPU/FPU on most modern hardware - that is, if you're
>>> working with IEEE floating point, which I'm pretty sure CPython always
>>> does; not sure about other Pythons. No need to actually calculate trig
>>> functions unless you need arbitrary precision (and even then, I'd bet
>>> the GMP libraries have that all sewn up for you). So the language
>>> doesn't make a lot of difference.
>>
>>
>> Sure it does. Python boxes floats into a PyObject structure. Both Python and
>> C will ultimately implement the arithmetic of "a + b" with an FADD
>> instruction, but Python will do a bunch of pointer dereferencing, hash
>> lookups, and function calls before it gets down to that. All of that
>> overhead typically outweighs the floating point computations down at the
>> bottom, even for the more expensive trig functions.
>
> Of course that's true, but that difference is just as much whether
> you're working with addition or trig functions. That overhead is the
> same. So if, as I said in the other post, you're doing some heavy
> crypto work or something, then yes, all that boxing and unboxing is
> expensive.

Yes, Alain was wrong to suggest that these are not "simple calculations" and 
thus will benefit from a lower-level language. In fact, the relationship is 
reversed. These are *such* simple operations that they do benefit from the 
immediate surrounding code being done in C. The amount of time spent on overhead 
doesn't change based on the operation itself but the immediate surrounding code, 
the inner loops. That's where the language (implementation) matters.

 > Most programs aren't doing that, so the advantage is far
 > less (by proportion).

But we're not talking about most programs. We are talking about the OP's 
programs, which apparently *do* involve lots of iterated floating point 
calculations.

-- 
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless enigma
  that is made terrible by our own mad attempt to interpret it as though it had
  an underlying truth."
   -- Umberto Eco