Applied fixes suggested in the PR review

Author: diaconuccalin

Deeploy/CommonExtensions/CodeTransformationPasses/Closure.py

@@ -156,7 +156,7 @@ def apply(self,
               name: str,
               verbose: CodeGenVerbosity = _NoVerbosity) -> Tuple[NetworkContext, ExecutionBlock]:
         # Add underscore to avoid name issues when beginning with problematic characters (like numbers)
-        self.closureName = "_" + name + self.closureSuffix
+        self.closureName = "op_" + name + self.closureSuffix
         self.functionCall = executionBlock.generate(ctxt)
         self._generateClosureStruct(ctxt, executionBlock)
         ctxt = self._generateClosureCtxt(ctxt, name)

Deeploy/CommonExtensions/OptimizationPasses/TopologyOptimizationPasses/LoweringOptimizationPasses.py

@@ -283,6 +283,26 @@ def __init__(self, default_channels_first: bool = True):
         super().__init__(graph, partial(_NCHWtoNHWC_fun, default_channels_first = default_channels_first), name)
 
 
+@contextagnostic
+class NCHWtoNHWCPass(SequentialPass):
+
+    def __init__(self, default_channels_first: bool = True):
+        passes = [
+            NCHWtoNHWCPadPass(default_channels_first),
+            NCHWtoNHWCMaxPoolPass(default_channels_first),
+            NCHWtoNHWCConvPass(default_channels_first),
+            NCHWtoNHWCRequantizedConvPass(default_channels_first),
+        ]
+        super().__init__(*passes)
+
+
+def _PULPDWNCHWtoNHWC_fun(graph: gs.Graph, match: Match, name: str, default_channels_first: bool = True):
+
+    matched_nodes = [m for k, m in match.nodes_map.items()]
+    opNode = matched_nodes[0]
+    node_op = opNode.op
+
+    if 'group' in opNode.attrs and opNode.attrs['group'] == 1:
 def _NCWHtoNHWC_dw_fun(graph: gs.Graph, match: Match, name: str, default_channels_first: bool) -> gs.Graph:
     node = next(iter((match.nodes_map.values())))
 

Deeploy/DeeployTypes.py

@@ -359,7 +359,7 @@ def get_aliases_of(self):
         """
 
         if hasattr(self, "alias_of"):
-            return self.alias_of
+            return list(self.alias_of)
         else:
             return list()
 
@@ -407,6 +407,11 @@ class TransientBuffer(VariableBuffer):
 
     def __init__(self, name: str = '', size = 0):
         super().__init__(name, shape = (size,))
+        self.name = name
+        self.size = size  # int or str: Total BYTE size, or regular size if _type manually set for this TransientBuffer
+
+        # Do not override - Should be written in the parsing passes
+        self._users = []
 
         # Do not override - Should be written in the parsing passes
         self._type: Type[Pointer] = PointerClass(VoidType)
@@ -417,7 +422,7 @@ def __eq__(self, other):
         return ret
 
     def _bufferRepresentation(self) -> Dict:
-        return {"type": self._type, "name": self.name, "size": int(self.size)}
+        return {"type": self._type, "name": self.name, "size": self.size}
 
     def __str__(self) -> str:
         return f'TransientBuffer: name: {self.name}, size: {self.size}'
@@ -434,8 +439,10 @@ class ConstantBuffer(VariableBuffer):
 
     """
 
-    def __init__(self, name: str = '', shape = [1], values = [0], alias_of: Optional[List[str]] = []):
-        super().__init__(name, shape, alias_of)
+    def __init__(self, name: str = '', shape = [1], values = [0], alias_of: Optional[List[str]] = None):
+        # Pass a copy of alias_of to avoid shared references
+        super().__init__(name, shape, list(alias_of) if alias_of is not None else None)
+
         values = np.asarray(values)
         # intArray = values.astype(int)
         # assert (np.abs(values - intArray)).max() < 0.001, "Constant value {name} is NOT an integer!"
@@ -868,7 +875,7 @@ def is_buffer(self, value: Any) -> bool:
         obj = self.lookup(value)
         return isinstance(obj, VariableBuffer)
 
-    def hoistTransientBuffer(self, name: str, size: int) -> str:
+    def hoistTransientBuffer(self, name: str, size: Union[int, str]) -> str:
         """Registers a new TransientBuffer in the local context
 
         Parameters

Deeploy/Targets/Generic/Parsers.py

@@ -1089,7 +1089,7 @@ def parseNodeCtxt(self,
         # Add new aliases to output node
         output_node.add_aliases(aliases_to_add = new_output_node_aliases)
 
-        # Add output node as alias to its aliases (alias relationship is reflexive)
+        # Add output node as alias to its aliases (alias relationship is symmetric)
         for alias in new_output_node_aliases:
             alias_node = ctxt.lookup(alias)
             alias_node.add_aliases(aliases_to_add = [

Deeploy/Targets/PULPOpen/Parsers.py

@@ -116,7 +116,7 @@ def parseNodeCtxt(self,
 
             return newCtxt, True
 
-        return newCtxt, False
+        return ctxt, False
 
 
 class PULPFPDWConv2DParser(Conv2DParser):
@@ -185,7 +185,7 @@ def parseNodeCtxt(self,
             if self.operatorRepresentation['group'] == self.operatorRepresentation['ch_im_in']:
                 return newCtxt, True
 
-        return newCtxt, False
+        return ctxt, False
 
 
 class PULPDWConv1DParser(RQSConv1DParser):

Deeploy/Targets/PULPOpen/Templates/FloatAddTemplate.py

@@ -6,13 +6,13 @@
 
 referenceTemplate = NodeTemplate("""
 // Add Parallel with 1x6 unrolling (Name: ${nodeName}, Op: ${nodeOp})
-int8_t ${nodeName}_core_id = (int8_t) pi_core_id();
-int8_t ${nodeName}_log2Core = (int8_t) log2(NUM_CORES);
-int16_t ${nodeName}_chunk = (${size} >> ${nodeName}_log2Core) + ((${size} & (NUM_CORES-1))!=0);
-int16_t ${nodeName}_chunk_start = (int16_t) MIN(${nodeName}_chunk*${nodeName}_core_id, ${size});
-int16_t ${nodeName}_chunk_stop = (int16_t) MIN(${nodeName}_chunk_start + ${nodeName}_chunk, ${size});
+uint8_t ${nodeName}_core_id = (uint8_t) pi_core_id();
+uint8_t ${nodeName}_log2Core = (uint8_t) log2(NUM_CORES);
+uint32_t ${nodeName}_chunk = (${size} >> ${nodeName}_log2Core) + ((${size} & (NUM_CORES-1))!=0);
+uint32_t ${nodeName}_chunk_start = (uint32_t) MIN(${nodeName}_chunk*${nodeName}_core_id, (uint32_t) ${size});
+uint32_t ${nodeName}_chunk_stop = (uint32_t) MIN(${nodeName}_chunk_start + ${nodeName}_chunk, (uint32_t) ${size});
 
-int32_t i = ${nodeName}_chunk_start;
+uint32_t i = ${nodeName}_chunk_start;
 for (; i + 5 < ${nodeName}_chunk_stop; i += 6) {
     ${data_out}[i] = ${data_in_1}[i] + ${data_in_2}[i];
     ${data_out}[i+1] = ${data_in_1}[i+1] + ${data_in_2}[i+1];

TargetLibraries/PULPOpen/inc/DeeployPULPMath.h

@@ -32,7 +32,6 @@
 #include "kernel/RequantShift.h"
 #include "kernel/Softmax.h"
 #include "kernel/UniformRequantShift.h"
-#include "kernel/gemm.h"
 #include "kernel/gemv.h"
 #include "kernel/iRMSnorm.h"