It's not possible to set the size of a thread dim inside a PP code snippet, because of, well, the threading and the different-lengths-for-different-iterations problem -- but it *is* of course possible to do very tricky things if you duck out of the autothreading and use direct C manipulation of the code. Depending on your application, you can return the size of your active dim in a secondary output variable and truncate as needed in the wrapper (I think this is what you said you are doing now) or condition the input variable to make sure there are no thread dims, and then manipulate the pdl struct explicitly inside PP (I forget the names of them, but there are PDL methods that do explicit threadloops at the Perl level, which can help encapsulate your non-threadable operator and make it act like a regular PDL operator).
That second approach prevents you from using the autogenerated thread loops in the PP, but you still get access to its nice macros for accessing the data and looping over active dims.
Post by Diab JeriusDavid,
I'd convinced myself that this code couldn't be threadable, but now you've instilled a bit of doubt and I'll have to go and test it...
The idea of filling the unused elements with a BAD value is great. But, what about PDL installations without bad values? I could provide an option to specify a fill value, but that means all code using this module (which I do plan on releasing to CPAN) will need to first check if they're bad value enabled to determine which value to expect.
Another option would be to return an appropriately dimensioned piddle recording the valid extents. The caller could then slice the returned data as appropriate.
A combination of the two works best, I think.
Thanks,
Diab
Hey Diab,
I am not aware of any technique, and I don't believe it's possible. The problem is that any code written using PDL::PP will automatically thread to higher dimensions. Data-dependent truncation would naively try to truncate each "thread" to its own data-dependent length, which is obviously incommensurate with the uniform allocations assumed by PDL.
The closest thing to a PDL-endorsed method of doing this is to set all trailing values to BAD. It might be a good idea to have a slicing routine that trims off bad values from the end of the data, in a threadable way, to make this sort of operation easier to handle in the general case.
By the way, I have used an approach similar to yours to achieve the same ends, but that's only in code that I use for my own research and not for code distributed to CPAN.
David
Is there a PP accessible means of truncating piddles?
In code I'm writing the output piddle length will be less than the input length, but the exact value isn't known until the end of processing the input (so RedoDims doesn't apply).
Currently I pass back the number of valid elements and truncate the piddles in the Perl wrapper, but I'd prefer to do it in the PP code.
Thanks,
Diab
_______________________________________________
PDL-porters mailing list
http://mailman.jach.hawaii.edu/mailman/listinfo/pdl-porters
--
"Debugging is twice as hard as writing the code in the first place.
Therefore, if you write the code as cleverly as possible, you are,
by definition, not smart enough to debug it." -- Brian Kernighan
_______________________________________________
PDL-porters mailing list
http://mailman.jach.hawaii.edu/mailman/listinfo/pdl-porters