Re: [Discuss-gnuradio] GrBlock Inconsistent IO Size

[Josh Blum]

Just FYI, I am integrating this work into gnuradio-core as we speak :-)

On 10/28/2011 10:05 AM, Jason Bonior wrote:
> I used grblock and numpy to create a block which would calculate the
> variance of an incoming vector then pass on the result:
> 
> class variance(grblock.sync_block):
> 
>     def __init__(self):
>         grblock.sync_block.__init__(
>             self,
>             name = "variance",
>             in_sig = [(numpy.float32, 2048)],
>             out_sig = [numpy.float32],
>         )
> 
>     def work(self, input_items, output_items):
>       output_items[0][:] = numpy.var(input_items[0])
>         print output_items[0][:].shape
>         print input_items[0].shape
>       return len(output_items[0])
> 
> I added the print .shape commands to make sure that the block IO was
> performing as expected. Most of the time I get (1, 2048) and (1, ) as
> I would expect but occasionally I get (2, 2048) (2, ) up to (5, 2048)
> (5,). Has anyone else seen anything like this or know how it might
> affect the performance of the block?
> 

Each element of the input vector is an array of items. In this case each
item is a vector of 2048 (I know... its a vector of vectors of vectors).
Its not guaranteed how many items the scheduler will give you. It looks
like usually you get 1 item, but sometimes several. This is consistent
with my observations when dealing with the output of the FFT block.

-Josh