Open telegraphic opened 5 years ago
Out of curiosity, what is the use case you are thinking about for this block?
The plan is to recombine F-engine data that arrives on two different UDP ports. So input tensors are ['time', 'ant', 'channel', 'pol']
, and concatenating on channel axis.
Our capture code is actually PSRDADA-based; I'm using the bifrost bindings to attach to existing ring buffer capture code. It's working! (I'll add some details on how to do this to the bifrost documentation at some stage)
That's cool. Were you also able to get this running on the GPU with bf.map()
or are you still on the CPU with numpy.concatenate()
?
It's working with numpy.concatenate; haven't made a gpu version (yet)
Hi all, I'm experimenting with a pipeline that joins two input streams (blocks) together, and was hoping for some clarification on how to implement it. Firstly, what should happens on the
on_sequence
event if there are N input streams, and M output streams? Is it as simple as parsing the N input headers, and returning a list of M output headers? Similarly, theon_data
method should return a list with M entries, each specifying the number of frames per stream?Here is my attempt, using numpy to concantenate two inputs. Is there a suggested way to implement this on the GPU, perhaps using
bf.map
?Cheers, Danny