Deeploy.CommonExtensions.OptimizationPasses.TopologyOptimizationPasses.LoweringOptimizationPasses.isPULPPointwise

Deeploy.CommonExtensions.OptimizationPasses.TopologyOptimizationPasses.LoweringOptimizationPasses.isPULPPointwise(node: Node) → bool

1x1 / stride 1 / no-pad convolutions the PULP kernel can write channels-first.

The lowering pass, the parser and the template must agree on this exactly, or a node ends up producing one layout and being read as the other.