Cadence - refactor quantizer and passes (#3539)

Summary:
Pull Request resolved: #3539

Update the quantizer to the latest internal version, and heavily refactor it for readability and maintenance.

Reviewed By: cccclai

Differential Revision: D57082021

fbshipit-source-id: 947753859f6ffc5ea9ea0e803d2c4a0fd30bf608

backends/cadence/aot/__init__.py

@@ -0,0 +1,5 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.

backends/cadence/aot/export_example.py

@@ -14,12 +14,12 @@
 from typing import Any, Tuple
 
 from executorch.backends.cadence.aot.compiler import export_to_edge
-from executorch.backends.cadence.aot.quantizer import (
-    CadenceBaseQuantizer,
-    QuantFusion,
+from executorch.backends.cadence.aot.passes import (
     ReplacePT2DequantWithCadenceDequant,
     ReplacePT2QuantWithCadenceQuant,
 )
+from executorch.backends.cadence.aot.quantizer.fusion_pass import QuantFusion
+from executorch.backends.cadence.aot.quantizer.quantizer import CadenceQuantizer
 from executorch.exir import ExecutorchProgramManager
 from torch import nn
 from torch._export import capture_pre_autograd_graph
@@ -52,7 +52,7 @@ def export_model(
     model: nn.Module, example_inputs: Tuple[Any], file_name: str = "CadenceDemoModel"
 ):
     # Quantizer
-    quantizer = CadenceBaseQuantizer()
+    quantizer = CadenceQuantizer()
 
     # Export
     model_exp = capture_pre_autograd_graph(model, example_inputs)

backends/cadence/aot/passes.py

@@ -0,0 +1,42 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+from executorch.exir.dialects._ops import ops as exir_ops
+from executorch.exir.pass_base import ExportPass
+
+
+class ReplacePT2QuantWithCadenceQuant(ExportPass):
+    """
+    Replace the pt2 quantization ops with custom cadence quantization ops.
+    """
+
+    def call_operator(self, op, args, kwargs, meta):
+        if op not in {exir_ops.edge.quantized_decomposed.quantize_per_tensor.default}:
+            return super().call_operator(op, args, kwargs, meta)
+
+        return super().call_operator(
+            exir_ops.edge.cadence.quantize_per_tensor.default,
+            args,
+            kwargs,
+            meta,
+        )
+
+
+class ReplacePT2DequantWithCadenceDequant(ExportPass):
+    """
+    Replace the pt2 dequantization ops with custom cadence dequantization ops.
+    """
+
+    def call_operator(self, op, args, kwargs, meta):
+        if op not in {exir_ops.edge.quantized_decomposed.dequantize_per_tensor.default}:
+            return super().call_operator(op, args, kwargs, meta)
+
+        return super().call_operator(
+            exir_ops.edge.cadence.dequantize_per_tensor.default,
+            args,
+            kwargs,
+            meta,
+        )