This is an example of accessing sliding windows of data with channel operations. This example is rudimentary and not ideal because all data must be read in at once. A better example would allow more of a data streaming mechanism, but I'm not sure how best to do that with the current channel API. I think to do this properly you would need some sort of peek() mechanism.
If I don't have the peek() mechanism, if the sliding window is in an scf.for, I believe the only options are to put nested if/else (which I haven't yet ever used, but should be in mlir-python-extras) statements or an scf.index_switch (which is not in mlir-python-extras) with one statement per unique window in the loop... which is not a scalable way to write this sort of code.
I made this example along the way towards working on a large example, and I thought it was a useful building block, so that is why I am adding it to the programming examples (even with the sub-optimal implementation). I'm very open to suggestions for how to make it better!
This is an example of accessing sliding windows of data with channel operations. This example is rudimentary and not ideal because all data must be read in at once. A better example would allow more of a data streaming mechanism, but I'm not sure how best to do that with the current channel API. I think to do this properly you would need some sort of
peek()mechanism.If I don't have the
peek()mechanism, if the sliding window is in anscf.for, I believe the only options are to put nestedif/else(which I haven't yet ever used, but should be inmlir-python-extras) statements or anscf.index_switch(which is not inmlir-python-extras) with one statement per unique window in the loop... which is not a scalable way to write this sort of code.I made this example along the way towards working on a large example, and I thought it was a useful building block, so that is why I am adding it to the programming examples (even with the sub-optimal implementation). I'm very open to suggestions for how to make it better!