Update base for Update on "Reuse GELU implementation from PyTorch core"

kernels/optimized doesn't need to support embedded systems, so it can just take a header-only dep on PyTorch.

Note that, because we will pick up Sleef internally and ignore it
externally thanks to ATen vec, this PR gets to enable optimized GELU in OSS.

Testing: CI to make sure this doesn't break mobile build modes; happy to take advice on anything not currently covered that might break.

Differential Revision: [D66335522](https://our.internmc.facebook.com/intern/diff/D66335522/)

[ghstack-poisoned]

Author: Github Executorch

.buckconfig

@@ -33,3 +33,6 @@
       **/.git, \
       cmake-out, \
       pip-out
+
+[buck2]
+restarter=true

.ci/scripts/gather_benchmark_configs.py

@@ -238,7 +238,7 @@ def set_output(name: str, val: Any) -> None:
     try:
         with open(github_output, "a") as env:
             env.write(f"{name}={val}\n")
-    except PermissionError:
+    except (PermissionError, FileNotFoundError):
         # Fall back to printing in case of permission error in unit tests
         print(f"::set-output name={name}::{val}")
 

.ci/scripts/setup-linux.sh

@@ -23,4 +23,7 @@ fi
 # of nightly. This allows CI to test against latest commits from PyTorch
 install_executorch "use-pt-pinned-commit"
 build_executorch_runner "${BUILD_TOOL}"
-do_not_use_nightly_on_ci
+
+if [[ "${GITHUB_BASE_REF:-}" == *main* || "${GITHUB_BASE_REF:-}" == *gh* ]]; then
+  do_not_use_nightly_on_ci
+fi

.ci/scripts/setup-macos.sh

@@ -137,4 +137,7 @@ install_pytorch_and_domains
 # the pinned commit from PyTorch
 install_executorch "use-pt-pinned-commit"
 build_executorch_runner "${BUILD_TOOL}"
-do_not_use_nightly_on_ci
+
+if [[ "${GITHUB_BASE_REF:-}" == *main* || "${GITHUB_BASE_REF:-}" == *gh* ]]; then
+  do_not_use_nightly_on_ci
+fi

.gitmodules

@@ -64,3 +64,6 @@
 [submodule "third-party/ao"]
 	path = third-party/ao
 	url = https://github.com/pytorch/ao.git
+[submodule "backends/cadence/utils/FACTO"]
+	path = backends/cadence/utils/FACTO
+	url = https://github.com/pytorch-labs/FACTO.git

backends/arm/_passes/_debug_passes.py

@@ -0,0 +1,23 @@
+# Copyright 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.
+
+import torch
+from executorch.devtools.visualization.visualization_utils import visualize_graph
+from executorch.exir import ExportedProgram
+from executorch.exir.pass_base import ExportPass, PassResult
+
+
+class VisualizePass(ExportPass):
+    """
+    This pass visualizes the graph at the point of insertion in the pass manager
+    """
+
+    def __init__(self, exported_program: ExportedProgram) -> None:
+        super().__init__()
+        self.exported_program = exported_program
+
+    def call(self, graph_module: torch.fx.GraphModule) -> PassResult:
+        visualize_graph(graph_module, self.exported_program)
+        return PassResult(graph_module, False)

backends/arm/_passes/annotate_decomposed_matmul.py

@@ -4,6 +4,8 @@
 # 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 itertools
 
 from typing import List

backends/arm/_passes/arm_pass_manager.py

@@ -123,6 +123,7 @@ def _tosa_080_MI_pipeline(self, exported_program: ExportedProgram) -> GraphModul
         self.add_pass(FuseQuantizedActivationPass())
         self.add_pass(RemoveGetItemPass())
         self.add_pass(ConvertSplitToSlicePass())
+        self.add_pass(FuseBatchnorm2DPass(exported_program))
         self.add_pass(ConvertMmToBmmPass())
         self.add_pass(DecomposeLinearPass())
         self.add_pass(DecomposeBatchNormPass())
@@ -132,7 +133,6 @@ def _tosa_080_MI_pipeline(self, exported_program: ExportedProgram) -> GraphModul
         self.add_pass(ConvertMeanDimToAveragePoolPass())
         self.add_pass(DecomposeDivPass())
         self.add_pass(DecomposeSoftmaxesPass())
-        self.add_pass(FuseBatchnorm2DPass(exported_program))
 
         self.add_pass(AnnotateDecomposedMatmulPass())
         self.add_pass(QuantizeOperatorArguments())

backends/arm/_passes/fold_qdq_with_annotated_qparams_pass.py

@@ -4,6 +4,8 @@
 # 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 copy
 
 from typing import cast, Dict, Set, Tuple

backends/arm/_passes/fuse_batchnorm2d_pass.py

@@ -3,6 +3,8 @@
 # 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 torch
 from executorch.exir import ExportedProgram
 from executorch.exir.dialects._ops import ops as exir_ops