Arm backend: Refactor Ops Tests for AvgPool, Clamp, Clone, Conv1d, Sub (pytorch#9039)

Refactored multiple operator test files:
- Replaced `@parameterized.expand` with `@common.parametrize`.
- Removed `_test_*_pipeline()` functions and replaced them with
`TosaPipelineMI` and `TosaPipelineBI`.
- Fixed padding handling for max pooling and avg pooling.
- Updated pipeline infrastructure to simplify replacing the default
quantize stage.

Signed-off-by: Fang-Ching <[email protected]>

Author: fumchin

backends/arm/operators/op_avg_pool2d.py

@@ -43,12 +43,19 @@ def _build_generic_avgpool2d(
         output_zp: int,
         accumulator_type: ts.DType,
     ) -> None:
-        input_tensor = inputs[0]
 
+        input_tensor = inputs[0]
         kernel_size_list = inputs[1].special
         stride_size_list = inputs[2].special
+
         try:
             pad_size_list = inputs[3].special
+            pad_size_list = [
+                pad_size_list[0],
+                pad_size_list[0],
+                pad_size_list[1],
+                pad_size_list[1],
+            ]
         except IndexError:
             pad_size_list = [0, 0, 0, 0]
 

backends/arm/operators/op_max_pool2d.py

@@ -42,9 +42,15 @@ def define_node(
         stride = inputs[2].special
 
         try:
-            padding = [*inputs[3].special, *inputs[3].special]
+            pad_size_list = inputs[3].special
+            pad_size_list = [
+                pad_size_list[0],
+                pad_size_list[0],
+                pad_size_list[1],
+                pad_size_list[1],
+            ]
         except IndexError:
-            padding = [0, 0, 0, 0]
+            pad_size_list = [0, 0, 0, 0]
 
         accumulator_type = output.dtype
 
@@ -63,7 +69,7 @@ def define_node(
         attr.PoolAttribute(
             kernel=kernel_size,
             stride=stride,
-            pad=padding,
+            pad=pad_size_list,
             input_zp=input_zp,
             output_zp=output_zp,
             accum_dtype=accumulator_type,