In both of these examples, I attempt to broadcast an input a to 3 workers. In single_herd, there is a single herd of size=[1, 3] whereas in multi_herd there are 3 herds each of size=[1, 1]. The workers then add a unique value to each element in the input image and output the new image to a unique per-worker output channel. Both of these examples now work!
To make this work cleanly, I created a Channel class to wrap ChannelOp and set the broadcast_size attribute.
This is part of the effort to write channel examples (https://github.com/Xilinx/mlir-air/issues/648).
In both of these examples, I attempt to broadcast an input
ato 3 workers. Insingle_herd, there is a single herd ofsize=[1, 3]whereas inmulti_herdthere are 3 herds each ofsize=[1, 1]. The workers then add a unique value to each element in the input image and output the new image to a unique per-worker output channel. Both of these examples now work!To make this work cleanly, I created a
Channelclass to wrapChannelOpand set thebroadcast_sizeattribute.