Prerequisites
None
Details
New behavior
It can happen that a specified target fps could not be met because ConvInputGen doesn't get unfolded further as long as the corresponding Pool layer meets the target fps. So maybe force the Pool layer to be unfolded even further to enable ConvInputGen to be further parallelized as well.
Motivation
Desired target fps not met because bottleneck layer ConvInputGen is not further parallelized as long as the corresponding Pool layer still meets the target fps.
Parts of FINN affected
Transformation, FPGA Dataflow, Set Folding