Re: Changesets: Re: Parallel access to data created with Octave

[Jaroslav Hajek]

On Thu, May 27, 2010 at 9:08 PM, Jarno Rajahalme
<address@hidden> wrote:
>
> On May 26, 2010, at 10:54 PM, ext Jaroslav Hajek wrote:
>>>>>
>>>>> Btw. I also noticed that cellfun() itself has some overhead. My own 
>>>>> "celltest.cc", which copied only the relevant parts of cellfun() was 
>>>>> faster.
>>>>
>>>> Hmm. Maybe you omitted something that cellfun needs to do? If you
>>>> don't think so, please show me the code. Or if you can figure out how
>>>> cellfun can be further sped up (preferably without invasive changes),
>>>> that would be even better :)
>>>>
>>>
>>> I just cut&pasted the cellfun itself, but only upto the point where the 
>>> arg0 is a string recognized by cellfun (e.g., "numel") and without any of 
>>> the preceding C++ helper code.
>>>
>>
>> That still sounds like it should be equally fast, so I'd like to see it.
>>
>
> Please find the attached, stripped down cellfun2.cc. Testing with the script 
> you sent (ticcellfun etc.), but with N = 1e3, cellfun2 is about 10-20% faster 
> than cellfun.
>

It seems a bit like so, but the numbers are so varying for me that I
can't say this for sure.


-- 
RNDr. Jaroslav Hajek, PhD
computing expert & GNU Octave developer
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz