XNNPACK: Run recommended pre-partition transforms in AOT flow (pytorch#20446)

Wire get_transform_passes() into the XNNPACK example AOT compiler so
official exports run the same recommended pre-partition graph transforms
used by tests.

Update XNNPACK docs and README snippets to show the optional
transform_passes stage and clarify that it is recommended in general for
XNNPACK before partitioning.


cc @GregoryComer @digantdesai @cbilgin @freddan80 @per @zingo
@oscarandersson8218 @Sebastian-Larsson @robell @rascani

Signed-off-by: Måns Nilsson <mans.nilsson@arm.com>

Author: mansnils

backends/xnnpack/README.md

@@ -46,6 +46,7 @@ import torchvision.models as models
 from torch.export import export, ExportedProgram
 from torchvision.models.mobilenetv2 import MobileNet_V2_Weights
 from executorch.backends.xnnpack.partition.xnnpack_partitioner import XnnpackPartitioner
+from executorch.backends.xnnpack.utils.configs import get_transform_passes
 from executorch.exir import EdgeProgramManager, ExecutorchProgramManager, to_edge
 from executorch.exir.backend.backend_api import to_backend
 
@@ -56,6 +57,8 @@ sample_inputs = (torch.randn(1, 3, 224, 224), )
 exported_program: ExportedProgram = export(mobilenet_v2, sample_inputs)
 edge: EdgeProgramManager = to_edge(exported_program)
 
+# Optional in the generic API, but recommended for XNNPACK before partitioning.
+edge = edge.transform(get_transform_passes())
 edge = edge.to_backend(XnnpackPartitioner())
 ```
 

docs/source/backends/xnnpack/xnnpack-overview.md

@@ -32,6 +32,7 @@ import torch
 import torchvision.models as models
 from torchvision.models.mobilenetv2 import MobileNet_V2_Weights
 from executorch.backends.xnnpack.partition.xnnpack_partitioner import XnnpackPartitioner
+from executorch.backends.xnnpack.utils.configs import get_transform_passes
 from executorch.exir import to_edge_transform_and_lower
 
 mobilenet_v2 = models.mobilenetv2.mobilenet_v2(weights=MobileNet_V2_Weights.DEFAULT).eval()
@@ -40,12 +41,18 @@ sample_inputs = (torch.randn(1, 3, 224, 224), )
 et_program = to_edge_transform_and_lower(
     torch.export.export(mobilenet_v2, sample_inputs),
     partitioner=[XnnpackPartitioner()],
+    transform_passes=get_transform_passes(),
 ).to_executorch()
 
 with open("mv2_xnnpack.pte", "wb") as file:
     et_program.write_to_file(file)
 ```
 
+The `transform_passes` stage is optional in the generic lowering API. For
+XNNPACK, `get_transform_passes()` is recommended in general because it runs
+pre-partition graph transforms that expose supported patterns to the
+partitioner.
+
 See [Partitioner API](/backends/xnnpack/xnnpack-partitioner) for a reference on available partitioner options. <!-- @lint-ignore -->
 
 ----

docs/source/backends/xnnpack/xnnpack-quantization.md

@@ -31,6 +31,7 @@ import torchvision.models as models
 from torchvision.models.mobilenetv2 import MobileNet_V2_Weights
 from executorch.backends.xnnpack.quantizer.xnnpack_quantizer import XNNPACKQuantizer, get_symmetric_quantization_config
 from executorch.backends.xnnpack.partition.xnnpack_partitioner import XnnpackPartitioner
+from executorch.backends.xnnpack.utils.configs import get_transform_passes
 from executorch.exir import to_edge_transform_and_lower
 from torchao.quantization.pt2e.quantize_pt2e import convert_pt2e, prepare_pt2e
 
@@ -52,9 +53,15 @@ quantized_model = convert_pt2e(prepared_model) # (5)
 et_program = to_edge_transform_and_lower( # (6)
     torch.export.export(quantized_model, sample_inputs),
     partitioner=[XnnpackPartitioner()],
+    transform_passes=get_transform_passes(),
 ).to_executorch()
 ```
 
+The `transform_passes` stage is optional in the generic lowering API. For
+XNNPACK, `get_transform_passes()` is recommended in general because it runs
+pre-partition graph transforms that expose supported patterns to the
+partitioner.
+
 See [PyTorch 2 Export Post Training Quantization](https://docs.pytorch.org/ao/main/tutorials_source/pt2e_quant_ptq.html) for more information.
 
 ### LLM quantization with quantize_

examples/xnnpack/README.md

@@ -18,6 +18,9 @@ examples/xnnpack
 ## Delegating a Floating-point Model
 
 The following command will produce a floating-point XNNPACK delegated model `mv2_xnnpack_fp32.pte` that can be run using XNNPACK's operators. It will also print out the lowered graph, showing what parts of the models have been lowered to XNNPACK via `executorch_call_delegate`.
+The compiler uses the optional pre-partition transform stage. For XNNPACK this
+is recommended in general because it exposes supported patterns to the
+partitioner before delegation decisions are made.
 
 ```bash
 # For MobileNet V2

examples/xnnpack/aot_compiler.py

@@ -1,5 +1,6 @@
 # Copyright (c) Meta Platforms, Inc. and affiliates.
 # All rights reserved.
+# Copyright 2026 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.
@@ -13,6 +14,7 @@
 
 import torch
 from executorch.backends.xnnpack.partition.xnnpack_partitioner import XnnpackPartitioner
+from executorch.backends.xnnpack.utils.configs import get_transform_passes
 from executorch.exir import (
     EdgeCompileConfig,
     ExecutorchBackendConfig,
@@ -106,6 +108,7 @@
     edge = to_edge_transform_and_lower(
         ep,
         partitioner=[XnnpackPartitioner()],
+        transform_passes=get_transform_passes(),
         compile_config=EdgeCompileConfig(
             _check_ir_validity=False if args.quantize else True,
             _skip_dim_order=True,  # TODO(T182187531): enable dim order in xnnpack