Merge branch 'main' into gh/kimishpatel/195/orig

Author: kimishpatel

.github/workflows/apple.yml

@@ -39,7 +39,7 @@ jobs:
         id: set_version
         shell: bash
         run: |
-          VERSION="0.7.0.$(TZ='PST8PDT' date +%Y%m%d)"
+          VERSION="0.8.0.$(TZ='PST8PDT' date +%Y%m%d)"
           echo "version=$VERSION" >> "$GITHUB_OUTPUT"
 
   build-demo-ios:

.github/workflows/trunk.yml

@@ -271,6 +271,36 @@ jobs:
           exit 1
         fi
 
+  nxp-build-test:
+    name: nxp-build-test
+    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
+    permissions:
+      id-token: write
+      contents: read
+    with:
+      runner: linux.2xlarge
+      docker-image: executorch-ubuntu-22.04-arm-sdk
+      submodules: 'recursive'
+      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      timeout: 90
+      script: |
+        # The generic Linux job chooses to use base env, not the one setup by the image
+        CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
+        conda activate "${CONDA_ENV}"
+
+        # Build
+        cmake -DEXECUTORCH_BUILD_NXP_NEUTRON=ON -Bcmake-out .
+        cmake --build cmake-out --target executorch_delegate_neutron --config Release
+
+        # Build check for the neutron backend library
+        lib_neutron="cmake-out/backends/nxp/libexecutorch_delegate_neutron.a"
+        if [ -f $lib_neutron ]; then
+            echo "Neutron backend library built."
+        else
+            echo "Neutron backend library not found!"
+            exit 1
+        fi
+
   test-coreml-delegate:
     name: test-coreml-delegate
     uses: pytorch/test-infra/.github/workflows/macos_job.yml@main

CMakeLists.txt

@@ -496,6 +496,10 @@ if(EXECUTORCH_BUILD_CADENCE)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/backends/cadence)
 endif()
 
+if(EXECUTORCH_BUILD_NXP_NEUTRON)
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/backends/nxp)
+endif()
+
 if(EXECUTORCH_BUILD_COREML)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/backends/apple/coreml)
 endif()
@@ -706,11 +710,7 @@ if(EXECUTORCH_BUILD_EXECUTOR_RUNNER)
 
   add_executable(executor_runner ${_executor_runner__srcs})
   if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
-    if(APPLE)
-      target_link_options(executor_runner PRIVATE "LINKER:-dead_strip")
-    else()
-      target_link_options(executor_runner PRIVATE "LINKER:--gc-sections")
-    endif()
+    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_linear_pass.py

@@ -88,6 +88,7 @@ def call(self, graph_module):
                     op_target=exir_ops.edge.aten.view_copy.default,
                     args=(conv, list(output_shape)),
                     kwargs={},
+                    from_node=node,
                 )
 
             node.replace_all_uses_with(output)

backends/arm/_passes/insert_table_ops.py

@@ -55,6 +55,7 @@ class TableOps:
         exir_ops.edge.aten.hardsigmoid.default: torch.nn.functional.hardsigmoid,
         exir_ops.edge.aten.hardswish.default: torch.nn.functional.hardswish,
         exir_ops.edge.aten.sinh.default: torch.sinh,
+        exir_ops.edge.aten.acosh.default: torch.acosh,
     }
 
     # Targets that must be treated explicitly

backends/arm/_passes/match_arg_ranks_pass.py

@@ -51,6 +51,7 @@ def __init__(self, exported_program):
         exir_ops.edge.aten.gt.Tensor,
         exir_ops.edge.aten.ge.Tensor,
         exir_ops.edge.aten.lt.Tensor,
+        exir_ops.edge.aten.le.Tensor,
         exir_ops.edge.aten.pow.Tensor_Tensor,
         exir_ops.edge.aten.where.self,
     ]

backends/arm/_passes/replace_scalar_with_tensor_pass.py

@@ -32,6 +32,7 @@
     exir_ops.edge.aten.gt.Scalar: exir_ops.edge.aten.gt.Tensor,
     exir_ops.edge.aten.ge.Scalar: exir_ops.edge.aten.ge.Tensor,
     exir_ops.edge.aten.lt.Scalar: exir_ops.edge.aten.lt.Tensor,
+    exir_ops.edge.aten.le.Scalar: exir_ops.edge.aten.le.Tensor,
     exir_ops.edge.aten.ne.Scalar: exir_ops.edge.aten.ne.Tensor,
     torch.ops.aten.add.Scalar: torch.ops.aten.add.Tensor,
     torch.ops.aten.sub.Scalar: torch.ops.aten.sub.Tensor,
@@ -43,6 +44,7 @@
     torch.ops.aten.gt.Scalar: torch.ops.aten.gt.Tensor,
     torch.ops.aten.ge.Scalar: torch.ops.aten.ge.Tensor,
     torch.ops.aten.lt.Scalar: torch.ops.aten.lt.Tensor,
+    torch.ops.aten.le.Scalar: torch.ops.aten.le.Tensor,
     torch.ops.aten.ne.Scalar: torch.ops.aten.ne.Tensor,
 }