[Numpy-discussion] Problems with Numexpr and discontiguous arrays

Tim Hochberg tim.hochberg at ieee.org
Thu Oct 5 08:24:42 CDT 2006


Tim Hochberg wrote:
> David M. Cooke wrote:
>> On Wed, 04 Oct 2006 10:19:08 -0700
>> Tim Hochberg <tim.hochberg at ieee.org> wrote:
>>
>>  
>>> Ivan Vilata i Balaguer wrote:
>>>    
>>>> It seemed that discontiguous arrays worked OK in Numexpr since 
>>>> r1977 or
>>>> so, but I have come across some alignment or striding problems 
>>>> which can
>>>> be seen with the following code::
>>>>       
>>> I looked at this just a little bit and clearly this bit from 
>>> interp_body cannot work in the presence of recor arrays:
>>>
>>> //....
>>>         intp sf1 = sb1 / sizeof(double);        \
>>> //...
>>>         #define f1    ((double *)x1)[j*sf1]
>>>
>>>
>>> There are clearly some assumptions that sb1 is evenly divisible by 
>>> sizeof(double). Blech!. This is likely my fault, and I expect it 
>>> won't be too horrible to fix, but I don't know that I'll have time 
>>> immediately.
>>>     
>>
>> My thinking is that this should be handled by a copy, so that the 
>> opcodes
>> always work on contiguous data. The copy can be another opcode. One 
>> advantage
>> of operating on contiguous data is that it's easier to use the 
>> processor's
>> vector instructions, if applicable.
>>   
>
> That would be easy to do. Right now the opcodes should work correctly 
> on data that is spaced in multiples of the itemsize on the last axis. 
> Other arrays are copied (no opcode required, it's embedded at the top 
> of interp_body lines 64-80). The record array case apparently slips 
> through the cracks when we're checking whether an array is suitable to 
> be used correctly (interpreter.c 1086-1103). It would certainly not be 
> any harder to only allow contiguous arrays than to correctly deal with 
> record arrays. Only question I have is whether the extra copy will 
> overwhelm the savings of that operating on contiguous data gives.  The 
> thing to do is probably try it and see what happens.

OK, I've checked in a fix for this that makes a copy when the array is 
not strided in an even multiple of the itemsize. I first tried copying 
for all discontiguous array, but this resulted in a large speed hit for 
vanilla strided arrays (a=arange(10)[::2], etc.), so I was more frugal 
with my copying. I'm not entirely certain that I caught all of the 
problematic cases, so let me know if you run into any more issues like this.

-tim





More information about the Numpy-discussion mailing list