Description

This PR adds LayerNorm operator support across:

• Core C++ IR/Op (CPU reference + x86 oneDNN backend)
• Python bindings and functional API
• Unit tests

PR Type

Please check the type of change your PR introduces:

• Bugfix
• Feature
• Refactoring
• Documentation
• Other (please specify):

Related Issues

Resolves #278 (LayerNorm implementation item)

Changes Made

1. Core C++ IR & Op

   • Added kOpTypeLayerNormalization in op_param.h/.cc and string mappings.
   • Added LayerNormalizationParam with epsilon_.
   • Implemented CPU reference operator:
     • framework/include/nndeploy/op/op_layernorm.h
     • framework/source/nndeploy/op/op_layernorm.cc
   • Implemented x86 oneDNN operator:
     • framework/source/nndeploy/op/x86/op_layernorm.cc

2. Python API Exposure

   • Added IR enum/param exposure in python/src/ir/op_param.cc.
   • Added op bridge and pybind exposure:
     • python/src/op/op_func.h
     • python/src/op/op_func.cc
     • python/src/op/op.cc
   • Added user-facing functional API:
     • python/nndeploy/op/functional.py (layer_norm)
   • Updated Python OpType mapping:
     • python/nndeploy/ir/op_param.py

3. Testing

   • Added test/op/test_layer_norm.py with:
     • last-dim LayerNorm with affine (weight + bias)
     • multi-dim normalization without bias
   • Numpy reference is used for correctness checking.

Checklist

• Follows the project's coding style
• Adds unit tests for the new feature
• Code compiles successfully
• Clear commit history

Testing

cmake -S . -B build && cmake --build build -j4
# passed

PYTHONPATH=python LD_LIBRARY_PATH=build:$LD_LIBRARY_PATH python test/op/test_layer_norm.py
..
----------------------------------------------------------------------
Ran 2 tests in 0.012s