Remove unused functions for quantization handling

Remove functions not used for searching/finding quantization information.

Signed-off-by: Per Åstrand <[email protected]>
Change-Id: I5917e8b447dc21be78e79b2b0477c321fa09e262

Author: per

backends/arm/_passes/annotate_channels_last_dim_order_pass.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Arm Limited and/or its affiliates.
+# Copyright 2024-2025 Arm Limited and/or its affiliates.
 # All rights reserved.
 #
 # This source code is licensed under the BSD-style license found in the
@@ -15,7 +15,7 @@
     get_node_arg,
     insert_q_dq_pair,
 )
-from executorch.backends.arm.tosa_quant_utils import dq_op, q_op, register_passable_op
+from executorch.backends.arm.tosa_quant_utils import dq_op, q_op
 from executorch.backends.arm.tosa_utils import is_consumer_node_depthwise_conv2d
 from executorch.exir.dialects._ops import ops as exir_ops
 from executorch.exir.pass_base import ExportPass, PassResult
@@ -43,9 +43,6 @@ def _transpose_impl(*args, **kwargs):
     return args[0]
 
 
-register_passable_op(torch.ops.passthrough_to_tosa._transpose)
-
-
 class AnnotateChannelsLastDimOrder(ExportPass):
     """
     Annotates each node with a tosa_dim_order. tosa_dim_order can be seen as a channels-last dim-order

backends/arm/operators/__init__.py

@@ -13,7 +13,6 @@
     op_bmm,
     op_cat,
     op_conv2d,
-    op_dequant,
     op_exp,
     op_full,
     op_get_item,
@@ -24,7 +23,6 @@
     op_min,
     op_mul,
     op_permute,
-    op_quant,
     op_reciprocal,
     op_relu,
     op_repeat,

backends/arm/operators/op_dequant.py

@@ -1,4 +1,4 @@
-# Copyright 2023-2024 Arm Limited and/or its affiliates.
+# Copyright 2023-2025 Arm Limited and/or its affiliates.
 #
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
@@ -19,7 +19,6 @@
 )
 from executorch.backends.arm.tosa_mapping import TosaArg
 
-from executorch.backends.arm.tosa_quant_utils import quantize_value
 from serializer.tosa_serializer import TosaOp
 
 
@@ -44,8 +43,8 @@ def define_node(
             input_qparams = get_input_qparams(node)  # pyre-ignore[16]
             qargs = input_qparams[0]
             # Convert to quantized representation
-            clamp_min_qs = quantize_value(inputs[1].number, qargs)
-            clamp_max_qs = quantize_value(inputs[2].number, qargs)
+            clamp_min_qs = qargs.quantize_value(inputs[1].number).item()
+            clamp_max_qs = qargs.quantize_value(inputs[2].number).item()
             # Set fp values to 0.0 since they are not used
             clamp_min_fp = 0.0
             clamp_max_fp = 0.0

backends/arm/operators/op_hardtanh.py

@@ -1,11 +1,10 @@
-# Copyright 2024 Arm Limited and/or its affiliates.
+# Copyright 2024-2025 Arm Limited and/or its affiliates.
 #
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
 # pyre-unsafe
 
-import executorch.backends.arm.tosa_quant_utils as tqutils
 import serializer.tosa_serializer as ts
 import torch.fx
 
@@ -43,9 +42,8 @@ def define_node(
         clamp_max_qs = 0
         if inputs[0].dtype == ts.DType.INT8:
             out_qargs = get_output_qparams(node)  # pyre-ignore[16]
-            clamp_min_qs = tqutils.quantize_value(0, out_qargs[0])
-            clamp_max_qs = tqutils.quantize_value(float("inf"), out_qargs[0])
-
+            clamp_min_qs = out_qargs[0].quantize_value(0).item()
+            clamp_max_qs = out_qargs[0].quantize_value(float("inf")).item()
         else:
             clamp_min_fp = 0
             clamp_max_fp = float("inf")

backends/arm/operators/op_quant.py

@@ -12,12 +12,7 @@
 import torch
 import torch.fx
 from executorch.backends.arm.operators.node_visitor import NodeVisitor
-from executorch.backends.arm.tosa_mapping import map_dtype, TosaArg
-from executorch.backends.arm.tosa_quant_utils import (
-    dq_op,
-    get_quantized_node_output_dtype,
-    is_node_quantized,
-)
+from executorch.backends.arm.tosa_mapping import TosaArg
 from executorch.backends.arm.tosa_specification import TosaSpecification
 from executorch.backends.arm.tosa_utils import getNodeArgs, tosa_shape
 from torch.export.exported_program import ExportedProgram
@@ -35,15 +30,8 @@ def process_call_function(
     # Convert output (this node itself)
     output = TosaArg(node)
 
-    is_dq_node = node.target == dq_op
-    if is_dq_node:
-        output_dtype = ts.DType.INT8
-    else:
-        output_dtype = output.dtype
     tosa_graph.currRegion.currBasicBlock.addTensor(
-        output.name,
-        tosa_shape(output.shape, output.dim_order),
-        output_dtype,
+        output.name, tosa_shape(output.shape, output.dim_order), output.dtype
     )
 
     # Visiting each Node
@@ -79,11 +67,7 @@ def process_inputs(
     tensor = ts.TosaSerializerTensor(
         inputs[0].name,
         tosa_shape(input_shape, input_dim_order),
-        (
-            map_dtype(get_quantized_node_output_dtype(node))
-            if is_node_quantized(node)
-            else inputs[0].dtype
-        ),
+        inputs[0].dtype,
         data=None,
         placeholderFilename=inputs[0].name + ".npy",
     )