NXP Backend: Add padd to remove unnecessary Quantize/Dequantize nodes.#15148

Draft

MartinPavella wants to merge 15 commits intopytorch:mainfrom

nxp-upstream:upstream/main-nxp/EIEX-519-upstream-removeadditionalqdqclusters-pass

Commits on Oct 14, 2025

NXP backend: Store inferred node format in the `node.meta`.
MartinPavella
committed
NXP backend: Update pass which removes GetItem nodes, to preserve the node format.
MartinPavella
committed
NXP backend: Perform node format inference before partitioning.
MartinPavella
committed
NXP backend: Improve `cat` delegation by using inferred node formats.
MartinPavella
committed
NXP backend: Improve `constant_pad_nd` delegation by using inferred node formats.
MartinPavella
committed
NXP backend: Do not infer format for unknown nodes.
MartinPavella
committed
Fix rebase.
MartinPavella
committed
NXP backend: Propagate flag from `executorch_pipeline` to `model_builder`, to delegate format related transpositions of input/output tensors.
MartinPavella
committed
NXP backend: Move format related transpositions of partition inputs/outputs to Neutron when possible.
MartinPavella
committed
NXP backend: Add support for aten_permute_copy_default operator

roman-janik-nxp
authored and
MartinPavella
committed