diff --git a/aidge_export_cpp/operators.py b/aidge_export_cpp/operators.py
index 24c2d63c0cc33f5eda845ab7c6c8ec0427016701..6c16f9ec45d0a1bc008ca0302f163f4a734d9dca 100644
--- a/aidge_export_cpp/operators.py
+++ b/aidge_export_cpp/operators.py
@@ -98,56 +98,28 @@ class ReLUCPP(ExportNodeCpp):
@operator_register("Conv")
-class ConvCPP(ExportNode):
+class ConvCPP(ExportNodeCpp):
def __init__(self, node):
super().__init__(node)
-
- # self.kernel = node.get_operator().attr.kernel_dims
- # self.stride = node.get_operator().attr.stride_dims
- # self.dilation = node.get_operator().attr.dilation_dims
-
# No padding with Conv
# Use PaddedConv to add padding attribute
self.attributes["padding"] = [0, 0]
-
- # self.nb_channels = node.get_operator().in_channels()
- # self.nb_outputs = node.get_operator().out_channels()
-
-
- def export(self, export_folder:Path, list_configs:list):
-
- copyfile(str(ROOT / "kernels" / "convolution.hpp"),
- str(export_folder / "include" / "kernels"))
- copyfile(str(ROOT / "kernels" / "macs.hpp"),
- str(export_folder / "include" / "kernels"))
- copyfile(str(ROOT / "kernels" / "activation.hpp"),
- str(export_folder / "include" / "kernels"))
-
- list_configs.append("kernels/convolution.hpp")
- list_configs.append(f"layers/{self.attributes['name']}.h")
- generate_file(
- str(export_folder / "layers" / f"{self.attributes['name']}.h"),
- str(ROOT / "templates" / "configuration" / "convolution_config.jinja"),
- activation="Linear",
- rescaling="NoScaling",
- **self.attributes,
- )
-
- return list_configs
-
- def forward(self, list_actions:list):
- list_actions.append(generate_str(
- str(ROOT / "templates" / "kernel_forward" / "convolution_forward.jinja"),
- **self.attributes
- ))
- return list_actions
+ self.attributes["activation"] = "Linear"
+ self.attributes["rescaling"] = "NoScaling"
+ self.config_template = str(ROOT / "templates" / "configuration" / "convolution_config.jinja")
+ self.forward_template = str(ROOT / "templates" / "kernel_forward" / "convolution_forward.jinja")
+ self.include_list = []
+ self.kernels_to_copy = [
+ str(ROOT / "kernels" / "convolution.hpp"),
+ str(ROOT / "kernels" / "macs.hpp"),
+ str(ROOT / "kernels" / "activation.hpp"),
+ ]
@operator_register("PaddedConv")
-class PaddedConvCPP(ConvCPP):
+class PaddedConvCPP(ExportNodeCpp):
def __init__(self, node):
- ExportNode.__init__(self, node)
-
+ super().__init__(node)
# TODO find a way to retrive attr for meta op
for n in self.operator.get_micro_graph().get_nodes():
if n.type() == "Pad":
@@ -156,105 +128,83 @@ class PaddedConvCPP(ConvCPP):
self.attributes["kernel_dims"] = n.get_operator().attr.kernel_dims
self.attributes["stride_dims"] = n.get_operator().attr.stride_dims
self.attributes["dilation_dims"] = n.get_operator().attr.dilation_dims
-
+ self.attributes["activation"] = "Linear"
+ self.attributes["rescaling"] = "NoScaling"
+ self.config_template = str(ROOT / "templates" / "configuration" / "convolution_config.jinja")
+ self.forward_template = str(ROOT / "templates" / "kernel_forward" / "convolution_forward.jinja")
+ self.include_list = []
+ self.kernels_to_copy = [
+ str(ROOT / "kernels" / "convolution.hpp"),
+ str(ROOT / "kernels" / "macs.hpp"),
+ str(ROOT / "kernels" / "activation.hpp"),
+ ]
@operator_register("Add")
-class AddCPP(ExportNode):
+class AddCPP(ExportNodeCpp):
def __init__(self, node):
super().__init__(node)
-
- def export(self, export_folder:str, list_configs:list):
- list_configs.append(f"layers/{self.attributes['name']}.h")
- list_configs.append("kernels/elemwise.hpp")
-
- copyfile(str(ROOT / "kernels" / "elemwise.hpp"),
- str(export_folder / "include" / "kernels"))
- copyfile(str(ROOT / "kernels" / "activation.hpp"),
- str(export_folder / "include" / "kernels"))
-
- generate_file(
- str(export_folder / "layers" / f"{self.attributes['name']}.h"),
- str(ROOT / "templates" / "configuration" / "elemwise_config.jinja"),
- activation="Linear",
- elemwise_op="Add",
- rescaling="NoScaling",
- **self.attributes)
-
- return list_configs
-
- def forward(self, list_actions:list):
- list_actions.append(generate_str(
- str(ROOT / "templates" / "kernel_forward" / "elemwise_forward.jinja"),
- **self.attributes
- ))
- return list_actions
+ self.attributes["elemwise_op"] = "Add"
+ self.attributes["activation"] = "Linear"
+ self.attributes["rescaling"] = "NoScaling"
+ self.config_template = str(ROOT / "templates" / "configuration" / "elemwise_config.jinja")
+ self.forward_template = str(ROOT / "templates" / "kernel_forward" / "elemwise_forward.jinja")
+ self.include_list = []
+ self.kernels_to_copy = [
+ str(ROOT / "kernels" / "elemwise.hpp"),
+ str(ROOT / "kernels" / "activation.hpp"),
+ ]
@operator_register("Sub")
-class SubCPP(ExportNode):
+class SubCPP(ExportNodeCpp):
def __init__(self, node):
super().__init__(node)
-
- def export(self, export_folder:str, list_configs:list):
- list_configs.append(f"layers/{self.attributes['name']}.h")
- list_configs.append("kernels/elemwise.hpp")
- copyfile(str(ROOT / "kernels" / "elemwise.hpp"),
- str(export_folder / "include" / "kernels"))
- copyfile(str(ROOT / "kernels" / "activation.hpp"),
- str(export_folder / "include" / "kernels"))
-
- generate_file(
- str(export_folder / "layers" / f"{self.attributes['name']}.h"),
- str(ROOT / "templates" / "configuration" / "elemwise_config.jinja"),
- activation="Linear",
- elemwise_op="Sub",
- rescaling="NoScaling",
- **self.attributes)
-
- return list_configs
-
- def forward(self, list_actions:list):
- list_actions.append(generate_str(
- str(ROOT / "templates" / "kernel_forward" / "elemwise_forward.jinja"),
- **self.attributes
- ))
- return list_actions
+ self.attributes["elemwise_op"] = "Sub"
+ self.attributes["activation"] = "Linear"
+ self.attributes["rescaling"] = "NoScaling"
+ self.config_template = str(ROOT / "templates" / "configuration" / "elemwise_config.jinja")
+ self.forward_template = str(ROOT / "templates" / "kernel_forward" / "elemwise_forward.jinja")
+ self.include_list = []
+ self.kernels_to_copy = [
+ str(ROOT / "kernels" / "elemwise.hpp"),
+ str(ROOT / "kernels" / "activation.hpp"),
+ ]
@operator_register("Mul")
-class MulCPP(ExportNode):
+class MulCPP(ExportNodeCpp):
def __init__(self, node):
super().__init__(node)
+ self.attributes["elemwise_op"] = "Mul"
+ self.attributes["activation"] = "Linear"
+ self.attributes["rescaling"] = "NoScaling"
+ self.config_template = str(ROOT / "templates" / "configuration" / "elemwise_config.jinja")
+ self.forward_template = str(ROOT / "templates" / "kernel_forward" / "elemwise_forward.jinja")
+ self.include_list = []
+ self.kernels_to_copy = [
+ str(ROOT / "kernels" / "elemwise.hpp"),
+ str(ROOT / "kernels" / "activation.hpp"),
+ ]
- def export(self, export_folder:str, list_configs:list):
- list_configs.append(f"layers/{self.name}.h")
- list_configs.append("kernels/elemwise.hpp")
- copyfile(str(ROOT / "kernels" / "elemwise.hpp"),
- str(export_folder / "include" / "kernels"))
- copyfile(str(ROOT / "kernels" / "activation.hpp"),
- str(export_folder / "include" / "kernels"))
-
- generate_file(
- str(export_folder / "layers" / f"{self.name}.h"),
- str(ROOT / "templates" / "configuration" / "elemwise_config.jinja"),
- name=self.name,
- nb_elts=np.prod(self.inputs_dims[0]),
- activation="Linear",
- elemwise_op="Mul",
- rescaling="NoScaling")
-
- return list_configs
+@operator_register("MaxPooling")
+class MaxPoolCPP(ExportNodeCpp):
+ def __init__(self, node):
+ super().__init__(node)
- def forward(self, list_actions:list):
- if not self.is_last:
- list_actions.append(set_up_output(self.name, "float"))
+ # No padding with MaxPooling
+ # Use PaddedMaxPooling to add padding attribute
+ self.attributes["padding"] = [0, 0]
+ self.attributes["pool_type"] = "Max"
+ self.attributes["activation"] = "Linear"
- list_actions.append(generate_str(
- str(ROOT / "templates" / "kernel_forward" / "elemwise_forward.jinja"),
- **self.attributes
- ))
- return list_actions
+ self.config_template = str(ROOT / "templates" / "configuration" / "pooling_config.jinja")
+ self.forward_template = str(ROOT / "templates" / "kernel_forward" / "pooling_forward.jinja")
+ self.include_list = []
+ self.kernels_to_copy = [
+ str(ROOT / "kernels" / "pooling.hpp"),
+ str(ROOT / "kernels" / "activation.hpp"),
+ ]
@operator_register("PaddedMaxPooling")
-class PaddedMaxPoolCPP(ExportNode):
+class PaddedMaxPoolCPP(ExportNodeCpp):
def __init__(self, node):
super().__init__(node)
for n in self.operator.get_micro_graph().get_nodes():
@@ -263,66 +213,19 @@ class PaddedMaxPoolCPP(ExportNode):
if n.type() == "MaxPooling":
self.attributes["kernel_dims"] = n.get_operator().attr.kernel_dims
self.attributes["stride_dims"] = n.get_operator().attr.stride_dims
+ self.attributes["pool_type"] = "Max"
+ self.attributes["activation"] = "Linear"
- def export(self, export_folder:Path, list_configs:list):
-
- copyfile(str(ROOT / "kernels" / "pooling.hpp"),
- str(export_folder / "include" / "kernels"))
-
- list_configs.append("kernels/pooling.hpp")
- list_configs.append(f"layers/{self.attributes['name']}.h")
-
- generate_file(
- str(export_folder / "layers" / f"{self.attributes['name']}.h"),
- str(ROOT / "templates" / "configuration" / "pooling_config.jinja"),
- pool_type="Max",
- activation="Linear",
- **self.attributes)
-
- return list_configs
-
- def forward(self, list_actions:list):
- list_actions.append(generate_str(
- str(ROOT / "templates" / "kernel_forward" / "pooling_forward.jinja"),
- **self.attributes
- ))
- return list_actions
-
-@operator_register("MaxPooling")
-class MaxPoolCPP(ExportNode):
- def __init__(self, node):
- super().__init__(node)
-
- # No padding with MaxPooling
- # Use PaddedMaxPooling to add padding attribute
- self.attributes["padding"] = [0, 0]
-
- def export(self, export_folder:Path, list_configs:list):
-
- copyfile(str(ROOT / "kernels" / "pooling.hpp"),
- str(export_folder / "include" / "kernels"))
-
- list_configs.append("kernels/pooling.hpp")
- list_configs.append(f"layers/{self.attributes['name']}.h")
-
- generate_file(
- str(export_folder / "layers" / f"{self.attributes['name']}.h"),
- str(ROOT / "templates" / "configuration" / "pooling_config.jinja"),
- pool_type="Max",
- activation="Linear",
- **self.attributes)
-
- return list_configs
-
- def forward(self, list_actions:list):
- list_actions.append(generate_str(
- str(ROOT / "templates" / "kernel_forward" / "pooling_forward.jinja"),
- **self.attributes
- ))
- return list_actions
+ self.config_template = str(ROOT / "templates" / "configuration" / "pooling_config.jinja")
+ self.forward_template = str(ROOT / "templates" / "kernel_forward" / "pooling_forward.jinja")
+ self.include_list = []
+ self.kernels_to_copy = [
+ str(ROOT / "kernels" / "pooling.hpp"),
+ str(ROOT / "kernels" / "activation.hpp"),
+ ]
@operator_register("GlobalAveragePooling")
-class GlobalAveragePoolCPP(ExportNode):
+class GlobalAveragePoolCPP(ExportNodeCpp):
def __init__(self, node):
super().__init__(node)
@@ -334,64 +237,29 @@ class GlobalAveragePoolCPP(ExportNode):
self.attributes["in_height"][0],
self.attributes["in_width"][0],
]
+ self.attributes["pool_type"] = "Average"
+ self.attributes["activation"] = "Linear"
- def export(self, export_folder:Path, list_configs:list):
-
- copyfile(str(ROOT / "kernels" / "pooling.hpp"),
- str(export_folder / "include" / "kernels"))
-
- list_configs.append("kernels/pooling.hpp")
- list_configs.append(f"layers/{self.attributes['name']}.h")
-
- generate_file(
- str(export_folder / "layers" / f"{self.attributes['name']}.h"),
- str(ROOT / "templates" / "configuration" / "pooling_config.jinja"),
- pool_type="Average",
- activation="Linear",
- **self.attributes)
-
- return list_configs
-
- def forward(self, list_actions:list):
- list_actions.append(generate_str(
- str(ROOT / "templates" / "kernel_forward" / "pooling_forward.jinja"),
- **self.attributes
- ))
- return list_actions
+ self.config_template = str(ROOT / "templates" / "configuration" / "pooling_config.jinja")
+ self.forward_template = str(ROOT / "templates" / "kernel_forward" / "pooling_forward.jinja")
+ self.include_list = []
+ self.kernels_to_copy = [
+ str(ROOT / "kernels" / "pooling.hpp"),
+ str(ROOT / "kernels" / "activation.hpp"),
+ ]
@operator_register("FC")
-class FcCPP(ExportNode):
+class FcCPP(ExportNodeCpp):
def __init__(self, node):
super().__init__(node)
-
- def export(self, export_folder:Path, list_configs:list):
-
- copyfile(str(ROOT / "kernels" / "fullyconnected.hpp"),
- str(export_folder / "include" / "kernels"))
- copyfile(str(ROOT / "kernels" / "macs.hpp"),
- str(export_folder / "include" / "kernels"))
- copyfile(str(ROOT / "kernels" / "activation.hpp"),
- str(export_folder / "include" / "kernels"))
-
- # Add to config list the include of configurations
- list_configs.append("kernels/fullyconnected.hpp")
- list_configs.append(f"layers/{self.attributes['name']}.h")
-
- # Export configuration file
- generate_file(
- str(export_folder / "layers" / f"{self.attributes['name']}.h"),
- str(ROOT / "templates" / "configuration" / "fullyconnected_config.jinja"),
- activation="Linear",
- rescaling="NoScaling",
- **self.attributes)
-
- return list_configs
-
- def forward(self, list_actions:list):
- list_actions.append(generate_str(
- str(ROOT / "templates" / "kernel_forward" / "fullyconnected_forward.jinja"),
- **self.attributes
- ))
- return list_actions
-
+ self.attributes["activation"] = "Linear"
+ self.attributes["rescaling"] = "NoScaling"
+ self.config_template = str(ROOT / "templates" / "configuration" / "fullyconnected_config.jinja")
+ self.forward_template = str(ROOT / "templates" / "kernel_forward" / "fullyconnected_forward.jinja")
+ self.include_list = []
+ self.kernels_to_copy = [
+ str(ROOT / "kernels" / "fullyconnected.hpp"),
+ str(ROOT / "kernels" / "macs.hpp"),
+ str(ROOT / "kernels" / "activation.hpp"),
+ ]