Update on "[et] generate debug handle before opeartor decomposition"

This diff update the debug handle generation, from each node in the edge program having a individual debug handle, to all nodes having a same ancestor in export graph sharing a same debug handle, which update the start point of tracing our node transformation from edge graph to exported graph.

Differential Revision: [D76860368](https://our.internmc.facebook.com/intern/diff/D76860368/)

[ghstack-poisoned]

Author: Gasoonjia

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-5616fa4a68718ead203314a3467f7dd9547153ae
+9b498d3bb28b8e3411ce464dd2755c5b96d92c8f

.ci/docker/common/install_conda.sh

@@ -13,7 +13,7 @@ source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
 install_miniconda() {
   BASE_URL="https://repo.anaconda.com/miniconda"
   CONDA_FILE="Miniconda3-py${PYTHON_VERSION//./}_${MINICONDA_VERSION}-Linux-x86_64.sh"
-  if [[ $(uname -m) == "aarch64" ]]; then 
+  if [[ $(uname -m) == "aarch64" ]]; then
     CONDA_FILE="Miniconda3-py${PYTHON_VERSION//./}_${MINICONDA_VERSION}-Linux-aarch64.sh"
   fi
 
@@ -71,4 +71,8 @@ fix_conda_ubuntu_libstdcxx() {
 install_miniconda
 install_python
 install_pip_dependencies
-fix_conda_ubuntu_libstdcxx
+# Hack breaks the job on aarch64 but is still necessary everywhere
+# else.
+if [ "$(uname -m)" != "aarch64" ]; then
+    fix_conda_ubuntu_libstdcxx
+fi

.ci/docker/conda-env-ci.txt

@@ -1,4 +1,4 @@
-cmake=3.26.4
+cmake=3.31.2
 ninja=1.10.2
 libuv
 llvm-openmp

.ci/scripts/check_c10_sync.sh

@@ -12,3 +12,4 @@ pushd pytorch
 git checkout "$pytorch_pin"
 popd
 "$(dirname "${BASH_SOURCE[0]}")"/compare_dirs.sh runtime/core/portable_type/c10/c10 pytorch/c10
+"$(dirname "${BASH_SOURCE[0]}")"/compare_dirs.sh runtime/core/portable_type/c10/torch/standalone pytorch/torch/standalone

.lintrunner.toml

@@ -76,6 +76,7 @@ exclude_patterns = [
     'examples/demo-apps/apple_ios/**',
     'examples/demo-apps/react-native/rnllama/ios/**',
     'extension/apple/**',
+    'extension/llm/apple/**',
     # File contains @generated
     'extension/llm/custom_ops/spinquant/fast_hadamard_transform_special.h',
     'extension/llm/custom_ops/spinquant/test/fast_hadamard_transform_special_unstrided_cpu.h',

CMakeLists.txt

@@ -705,12 +705,8 @@ if(EXECUTORCH_BUILD_EXECUTOR_RUNNER)
   endif()
 
   add_executable(executor_runner ${_executor_runner__srcs})
-  if(CMAKE_BUILD_TYPE STREQUAL "Release")
-    if(APPLE)
-      target_link_options(executor_runner PRIVATE "LINKER:-dead_strip")
-    else()
-      target_link_options(executor_runner PRIVATE "LINKER:--gc-sections")
-    endif()
+  if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
+    target_link_options_gc_sections(executor_runner)
   endif()
   target_link_libraries(executor_runner ${_executor_runner_libs})
   target_compile_options(executor_runner PUBLIC ${_common_compile_options})

backends/arm/_passes/__init__.py

@@ -22,6 +22,7 @@
 from .convert_split_to_slice import ConvertSplitToSlicePass  # noqa
 from .convert_squeezes_to_view import ConvertSqueezesToViewPass  # noqa
 from .convert_to_clamp import ConvertToClampPass  # noqa
+from .decompose_acosh_pass import DecomposeAcoshPass  # noqa
 from .decompose_atan_pass import DecomposeAtanPass  # noqa
 from .decompose_avg_pool2d import DecomposeAvgPool2d  # noqa
 from .decompose_batch_norm_no_stats import DecomposeBatchNormNoStatsPass  # noqa

backends/arm/_passes/arm_pass_manager.py

@@ -25,6 +25,7 @@
     ConvertSplitToSlicePass,
     ConvertSqueezesToViewPass,
     ConvertToClampPass,
+    DecomposeAcoshPass,
     DecomposeAtanPass,
     DecomposeAvgPool2d,
     DecomposeBatchNormNoStatsPass,
@@ -151,6 +152,7 @@ def _tosa_080_BI_pipeline(self, exported_program: ExportedProgram) -> GraphModul
 
     def _tosa_080_MI_pipeline(self, exported_program: ExportedProgram) -> GraphModule:
         self.add_pass(DecomposeRoundPass())
+        self.add_pass(DecomposeAcoshPass())
         self.add_pass(DecomposeSqrtPass())
         self.add_pass(DecomposeAtanPass())
         self.add_pass(ConvertIntPowToMuls())

backends/arm/_passes/decompose_acosh_pass.py

@@ -0,0 +1,52 @@
+# 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.
+
+# pyre-unsafe
+
+from executorch.backends.arm._passes import ArmPass
+from executorch.exir.dialects._ops import ops as exir_ops
+
+# For MI case
+edge_acosh_op = exir_ops.edge.aten.acosh.default
+
+
+class DecomposeAcoshPass(ArmPass):
+    """
+    Decomposes acosh to supported TOSA-operations.
+    This decomposition is based on the mathematical identity:
+        acosh(x) = log(x + sqrt((x-1)(x+1))
+    """
+
+    def call_operator(self, op, args, kwargs, meta, updated=False):
+
+        if op is not edge_acosh_op:
+            return super().call_operator(op, args, kwargs, meta, updated)
+
+        log_op, sqrt_op, mul_op, sub_op, add_op, add_op_scalar = (
+            exir_ops.edge.aten.log.default,
+            exir_ops.edge.aten.sqrt.default,
+            exir_ops.edge.aten.mul.Tensor,
+            exir_ops.edge.aten.sub.Scalar,
+            exir_ops.edge.aten.add.Tensor,
+            exir_ops.edge.aten.add.Scalar,
+        )
+
+        x = args[0]
+
+        # (x-1)(x+1)
+        sub = super().call_operator(sub_op, (x, 1.0), {}, meta, True)
+        add = super().call_operator(add_op_scalar, (x, 1.0), {}, meta, True)
+        mul = super().call_operator(mul_op, (sub, add), {}, meta, True)
+
+        # sqrt((x-1)(x+1))
+        sqrt = super().call_operator(sqrt_op, (mul,), {}, meta, True)
+
+        # x + sqrt((x-1)(x+1))
+        add = super().call_operator(add_op, (x, sqrt), {}, meta, True)
+
+        # out = ln(x + sqrt((x-1)(x+1))
+        out = super().call_operator(log_op, (add,), {}, meta, True)
+
+        return out

backends/arm/_passes/decompose_sqrt_pass.py

@@ -4,6 +4,8 @@
 # LICENSE file in the root directory of this source tree.
 
 # pyre-unsafe
+from typing import Tuple, Union
+
 import torch
 from executorch.exir.dialects._ops import ops as exir_ops
 from executorch.exir.pass_base import ExportPass
@@ -15,7 +17,7 @@
 )
 
 
-def get_sqrt_decomposition(op) -> tuple:
+def get_sqrt_decomposition(op) -> Union[Tuple, torch._ops.OpOverload]:
     # TODO : "MLETORCH-863 : Replace current sqrt -> pow.Tensor_Scalar workaround with pow.Tensor_Tensor"
     if op in edge_sqrt_ops:
         return exir_ops.edge.aten.pow.Tensor_Scalar