Add some comments.

dicp/dicp/vendor/TopsGraph/codegen/enflame.py

@@ -944,7 +944,12 @@ def ComplexMul(op_var, out_shape, out_dtype, x, y):
     def Concatenate(op_var, out_shape, out_dtype, tensors, dim):
         return f"builder::Op {op_var} = builder::Concatenate({'{' + ', '.join(tensors) + '}'}, {dim});"
 
-    # Add an additional true flag for accuration in tops softmax.
+    """
+    Add an additional true flag for accuration in hlir_builder Softmax.
+    The third parameter, half_to_float, in aten._softmax represents whether cast
+    inputs from float16 to float32 or not, while the third parameter ,accurate,
+    in hlir_builder represents whether precision calculation is performed.
+    """
     @staticmethod
     def Softmax(op_var, out_shape, out_dtype, x, y):
         return f"builder::Op {op_var} = builder::Softmax({x}, {y}, true);"

dicp/dicp/vendor/TopsGraph/conversion.py

@@ -400,6 +400,12 @@ def BatchNorm(self, *args, **kwargs):
     def BatchNormBackward(*args, **kwargs):
         return tops_op.BatchNormBackward(*args, **kwargs)
 
+    """
+    Add an additional true flag for accuration in hlir_builder Softmax.
+    The third parameter, half_to_float, in aten._softmax represents whether cast
+    inputs from float16 to float32 or not, while the third parameter ,accurate,
+    in hlir_builder represents whether precision calculation is performed.
+    """
     @register_conversion(aten._softmax)
     def Softmax(self, a, dim, half_to_float):
         out_shape = fx_traceback.get_current_meta()["val"].shape

dicp/dicp/vendor/TopsGraph/tops_op.py

@@ -551,6 +551,7 @@ def __init__(self, *args, **kwargs):
         self.kwargs = kwargs
         self.torch_op = aten.expand.default
 
+    # The third parameter broadcast_dims is only required in hlir_builder Expand.
     def __call__(self, *args, **kwargs):
         new_args = args[:2]
         return super().__call__(*new_args, **kwargs)