[NE16] integrate Pu DENG's NE16 Linear PR with NE16-w platform

- TargetLibraries/GAP9/CMakeLists.txt: rename CNN_Libraries_NE16 →
  CNN_Libraries_HWPE (the actual gap9-sdk path); skip SDK
  CNN_BasicKernels_NE16.c source for GAP9_w_NE16 platform (it uses the
  pulp-nnx ne16 stack, so the SDK NE16 kernels are not needed).
- Deeploy/Targets/NE16/Platform.py: instantiate the GAP9ClusterEngine
  with a trimmed includeList (no CNN_BasicKernels_NE16.h /
  ne16_utils.h / CNN_Copy.h) so the generated Network.c does not pull
  in the SDK NE16 header alongside pulp-nnx ne16_task_defs.h — the
  NE16_REG_* macros are defined in both and trigger -Werror redefs.

Author: runwangdl

Deeploy/Targets/GAP9/Platform.py

@@ -46,8 +46,8 @@
     IntegerDivRequantMergePass, MatMulAddMergePass, MergeConstAddAndRequantPass, MergeTrueIntegerDivRequantShiftPass, \
     QuantPatternPass, RQSSplitPass, SkipEmptyConcatPass, SkipUnityRequantPass, iGELURequantMergePass, \
     iHardswishRequantMergePass
-from Deeploy.Targets.PULPOpen.Bindings import BasicDequantBindings, BasicQuantBindings, PULPConv1DBinding, \
-    PULPDMASliceBindings, PULPDWConv1DBinding, PULPReduceMeanBindings, PULPRQSConv1DBindings, PULPSliceBindings
+from Deeploy.Targets.PULPOpen.Bindings import BasicDequantBindings, BasicQuantBindings, PULPDMASliceBindings, \
+    PULPDWConv1DBinding, PULPReduceMeanBindings, PULPRQSConv1DBindings, PULPSliceBindings
 from Deeploy.Targets.PULPOpen.Layers import PULPRQSConvLayer, PULPRQSGEMMLayer
 from Deeploy.Targets.PULPOpen.Parsers import PULPConv1DParser, PULPConv2DParser, PULPDWConv1DParser, \
     PULPDWConv2DParser, PULPFPConv2DParser, PULPFPDWConv2DParser, PULPGEMMParser, PULPMatrixVecParser, \

Deeploy/Targets/NE16/Platform.py

@@ -28,7 +28,17 @@ def __init__(self,
                  structBuffer = GAP9StructBuffer,
                  transientBuffer = GAP9TransientBuffer) -> None:
         if engines is None:
-            engines = [NE16Engine("NE16"), GAP9ClusterEngine("GAP9Cluster")]
+            # Drop SDK NE16 headers from the cluster engine include list so the
+            # generated Network.c does not pull in CNN_BasicKernels_NE16.h /
+            # ne16_utils.h alongside pulp-nnx's ne16_task_defs.h
+            # (NE16_REG_* macros are defined in both, causing -Werror redefs).
+            cluster = GAP9ClusterEngine(
+                "GAP9Cluster",
+                includeList = [
+                    "pmsis.h", "DeeployGAP9Math.h", "pulp_nn_kernels.h", "DeeployMchan.h", "CNN_BasicKernels_fp32.h"
+                ],
+            )
+            engines = [NE16Engine("NE16"), cluster]
         super().__init__(engines, variableBuffer, constantBuffer, structBuffer, transientBuffer)
 
 
@@ -44,7 +54,17 @@ def __init__(self,
                  structBuffer = GAP9StructBuffer,
                  transientBuffer = GAP9TransientBuffer) -> None:
         if engines is None:
-            engines = [NE16Engine("NE16"), GAP9ClusterEngine("GAP9Cluster")]
+            # Drop SDK NE16 headers from the cluster engine include list so the
+            # generated Network.c does not pull in CNN_BasicKernels_NE16.h /
+            # ne16_utils.h alongside pulp-nnx's ne16_task_defs.h
+            # (NE16_REG_* macros are defined in both, causing -Werror redefs).
+            cluster = GAP9ClusterEngine(
+                "GAP9Cluster",
+                includeList = [
+                    "pmsis.h", "DeeployGAP9Math.h", "pulp_nn_kernels.h", "DeeployMchan.h", "CNN_BasicKernels_fp32.h"
+                ],
+            )
+            engines = [NE16Engine("NE16"), cluster]
         super().__init__(memoryHierarchy, defaultTargetMemoryLevel, engines, variableBuffer, constantBuffer,
                          structBuffer, transientBuffer)
         self.weightMemoryLevel = weightMemoryLevel

TargetLibraries/GAP9/CMakeLists.txt

@@ -5,10 +5,18 @@
 file(GLOB_RECURSE SOURCES
   "src/**"
   "$ENV{GAP_SDK_HOME}/tools/autotiler_v3/CNN_Libraries_fp32/CNN_Bias_Linear_Activation_fp32.c"
-  "$ENV{GAP_SDK_HOME}/tools/autotiler_v3/CNN_Libraries_NE16/CNN_BasicKernels_NE16.c"
   "$ENV{GAP_SDK_HOME}/tools/autotiler_v3/CNN_Libraries/CNN_Copy.c"
 )
 
+# CNN_BasicKernels_NE16 from gap9-sdk redefines NE16_REG_* macros that
+# pulp-nnx's ne16 hal also defines. For GAP9_w_NE16 we use the pulp-nnx
+# NE16 stack; for plain GAP9 (Pu DENG's NE16-Linear path) we use the SDK's.
+if(NOT platform STREQUAL "GAP9_w_NE16")
+  list(APPEND SOURCES
+    "$ENV{GAP_SDK_HOME}/tools/autotiler_v3/CNN_Libraries_HWPE/CNN_BasicKernels_NE16.c"
+  )
+endif()
+
 
 # Exclude dory_mem and dory_dma from SOURCES (they need different optimization)
 list(FILTER SOURCES EXCLUDE REGEX ".*dory_(mem|dma).*")
@@ -48,7 +56,6 @@ target_include_directories(deeploygap9
   ${TILER_EMU_INC}
   ${TILER_CNN_KERNEL_PATH_FP32}
   ${TILER_CNN_KERNEL_PATH_FP16}
-  $ENV{GAP_SDK_HOME}/tools/autotiler_v3/CNN_Libraries_NE16
   $ENV{GAP_SDK_HOME}/tools/autotiler_v3/CNN_Libraries_SQ8
   $ENV{GAP_SDK_HOME}/tools/autotiler_v3/CNN_Libraries
   ${TILER_DSP_KERNEL_V2_PATH}