Update base for Update on "[cortex-m] Add scalar c++ op for dequantize_per_tensor"

Only buck build for now, CMake is next.
No MVE, scalar only.
Strictly the dtypes we care about update arg_meta to reflect that.

Differential Revision: [D73164576](https://our.internmc.facebook.com/intern/diff/D73164576/)

[ghstack-poisoned]

Author: digantdesai

.github/workflows/android-release-artifacts.yml

@@ -80,6 +80,12 @@ jobs:
 
         echo -n "$SECRET_EXECUTORCH_MAVEN_SIGNING_GPG_KEY_CONTENTS" | base64 -d > /tmp/secring.gpg
 
+        # Update the version name in build.gradle in case of maven publish
+        VERSION="${{ inputs.version }}"
+        if [ ! -z "$VERSION" ]; then
+          sed -i "s/\(coordinates(\"org.pytorch\", \"executorch-android\", \"\)\([0-9]\+.[0-9]\+.[0-9]\+\)\(\")\)/\1$VERSION\3/" extension/android/executorch_android/build.gradle
+        fi
+
         # Build AAR Package
         mkdir aar-out
         export BUILD_AAR_DIR=aar-out
@@ -92,7 +98,7 @@ jobs:
         # Publish to maven staging
         UPLOAD_TO_MAVEN="${{ inputs.upload_to_maven }}"
         if [[ "$UPLOAD_TO_MAVEN" == "true" ]]; then
-          (cd aar-out; ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:publishToMavenCentral)
+          (cd extension/android; ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:publishToMavenCentral)
         fi
 
   upload-release-aar:

CMakeLists.txt

@@ -761,12 +761,16 @@ if(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/flat_tensor)
 endif()
 
+if(EXECUTORCH_BUILD_EXTENSION_MODULE)
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/module)
+endif()
+
 if(EXECUTORCH_BUILD_EXTENSION_LLM)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/llm/tokenizers)
 endif()
 
-if(EXECUTORCH_BUILD_EXTENSION_MODULE)
-  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/module)
+if(EXECUTORCH_BUILD_EXTENSION_LLM_RUNNER)
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/llm/runner)
 endif()
 
 if(EXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL)

CONTRIBUTING.md

@@ -45,7 +45,7 @@ executorch
 │   └── <a href="devtools/visualization">visualization</a> - Visualization tools for representing model structure and performance metrics.
 ├── <a href="docs">docs</a> - Static docs tooling and documentation source files.
 ├── <a href="examples">examples</a> - Examples of various user flows, such as model export, delegates, and runtime execution.
-├── <a href="exir">exir</a> - Ahead-of-time library: model capture and lowering APIs. EXport Intermediate Representation (EXIR) is a format for representing the result of <a href="https://pytorch.org/docs/main/export.ir_spec.html">torch.export</a>. This directory contains utilities and passes for lowering the EXIR graphs into different <a href="/docs/source/ir-exir.md">dialects</a> and eventually suitable to run on target hardware.
+├── <a href="exir">exir</a> - Ahead-of-time library: model capture and lowering APIs. EXport Intermediate Representation (EXIR) is a format for representing the result of <a href="https://pytorch.org/docs/stable/export.html">torch.export</a>. This directory contains utilities and passes for lowering the EXIR graphs into different <a href="/docs/source/ir-exir.md">dialects</a> and eventually suitable to run on target hardware.
 │   ├── <a href="exir/_serialize">_serialize</a> - Serialize final export artifact.
 │   ├── <a href="exir/backend">backend</a> - Backend delegate ahead of time APIs.
 │   ├── <a href="exir/capture">capture</a> - Program capture.

backends/arm/test/ops/test_tanh.py

@@ -9,9 +9,11 @@
 
 from typing import Tuple
 
+import pytest
+
 import torch
 
-from executorch.backends.arm.test import common
+from executorch.backends.arm.test import common, conftest
 from executorch.backends.arm.test.tester.arm_tester import ArmTester
 from executorch.exir.backend.compile_spec_schema import CompileSpec
 from parameterized import parameterized
@@ -40,7 +42,7 @@ def forward(self, x):
     def _test_tanh_tosa_MI_pipeline(
         self, module: torch.nn.Module, test_data: Tuple[torch.tensor]
     ):
-        (
+        tester = (
             ArmTester(
                 module,
                 example_inputs=test_data,
@@ -54,11 +56,13 @@ def _test_tanh_tosa_MI_pipeline(
             .check_not(["executorch_exir_dialects_edge__ops_aten_tanh_default"])
             .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
             .to_executorch()
-            .run_method_and_compare_outputs(inputs=test_data)
         )
 
+        if conftest.is_option_enabled("tosa_ref_model"):
+            tester.run_method_and_compare_outputs(inputs=test_data)
+
     def _test_tanh_tosa_BI_pipeline(self, module: torch.nn.Module, test_data: Tuple):
-        (
+        tester = (
             ArmTester(
                 module,
                 example_inputs=test_data,
@@ -73,9 +77,11 @@ def _test_tanh_tosa_BI_pipeline(self, module: torch.nn.Module, test_data: Tuple)
             .check_not(["executorch_exir_dialects_edge__ops_aten_tanh_default"])
             .check_count({"torch.ops.higher_order.executorch_call_delegate": 1})
             .to_executorch()
-            .run_method_and_compare_outputs(inputs=test_data)
         )
 
+        if conftest.is_option_enabled("tosa_ref_model"):
+            tester.run_method_and_compare_outputs(inputs=test_data)
+
     def _test_tanh_tosa_ethos_BI_pipeline(
         self,
         compile_spec: list[CompileSpec],
@@ -114,6 +120,7 @@ def _test_tanh_tosa_u85_BI_pipeline(
         )
 
     @parameterized.expand(test_data_suite)
+    @pytest.mark.tosa_ref_model
     def test_tanh_tosa_MI(
         self,
         test_name: str,
@@ -122,6 +129,7 @@ def test_tanh_tosa_MI(
         self._test_tanh_tosa_MI_pipeline(self.Tanh(), (test_data,))
 
     @parameterized.expand(test_data_suite)
+    @pytest.mark.tosa_ref_model
     def test_tanh_tosa_BI(self, test_name: str, test_data: torch.Tensor):
         self._test_tanh_tosa_BI_pipeline(self.Tanh(), (test_data,))
 

backends/arm/test/targets.bzl

@@ -16,6 +16,7 @@ def define_arm_tests():
         "ops/test_linear.py", 
         "ops/test_slice.py",
         "ops/test_sigmoid.py",
+        "ops/test_tanh.py",
     ]
 
     TESTS = {}

backends/cadence/aot/replace_ops.py

@@ -1806,30 +1806,6 @@ def call_operator(self, op, args, kwargs, meta):
         return super().call_operator(op, tuple(new_args), kwargs, meta)
 
 
-@register_cadence_pass(CadencePassAttribute(opt_level=0))
-class ReplaceAtenLinalgVectorNormWithCadenceLinalgVectorNormPass(ExportPass):
-    """
-    Replace the aten.linalg_vector_norm op with a custom op.
-    aten.linalg_vector_norm is not supported by Jarvis, so we
-    need to replace it with native_batch_norm at all optimization levels.
-    """
-
-    def call_operator(self, op, args, kwargs, meta):
-        if op != exir_ops.edge.aten.linalg_vector_norm.default:
-            return super().call_operator(op, args, kwargs, meta)
-
-        assert (
-            len(args) == 1
-        ), "aten.linalg_vector_norm should have 1 argument (a tensor), we do not support any custom variants"
-
-        return super().call_operator(
-            exir_ops.edge.cadence.linalg_vector_norm.default,
-            args,
-            kwargs,
-            meta,
-        )
-
-
 @register_cadence_pass(CadencePassAttribute(opt_level=1))
 class ReplaceSingleElementTensorArgumentsFromFullOpWithScalarPass(ExportPass):
     """
@@ -2243,7 +2219,6 @@ class CadenceReplaceOpsInGraph:
         ReplacePT2DequantWithCadenceDequantPass,
         ReplaceSingleElementTensorArgumentsFromFullOpWithScalarPass,
         ReplaceAtenAvgPoolWithJarvisAvgPoolPass,
-        ReplaceAtenLinalgVectorNormWithCadenceLinalgVectorNormPass,
         ReplaceWhereWithFullArgsWithWhereScalar,
         # ReplaceGeluWithApproximateGeluPass,
     ]

backends/cadence/aot/tests/test_replace_ops_passes.py

@@ -23,7 +23,6 @@
     MakeSliceAndCatDimOutermostPass,
     ReplaceAddMMWithLinearPass,
     ReplaceAtenConvolutionWithJarvisConvolutionPass,
-    ReplaceAtenLinalgVectorNormWithCadenceLinalgVectorNormPass,
     ReplaceConstantPadNdWithSlicePass,
     ReplaceConvolutionOptionalArgsWithConcreteArgsPass,
     ReplaceConvWithIm2RowAndLinear,
@@ -1189,36 +1188,6 @@ def forward(self, x):
             count_node(graph_after_passes, exir_ops.edge.aten.transpose_copy.int), 0
         )
 
-    def test_replace_aten_linalg_vector_norm_with_cadence_linalg_vector_norm(self):
-        class LinalgVectorNorm(torch.nn.Module):
-            def forward(self, x: torch.Tensor):
-                return torch.linalg.vector_norm(x)
-
-        x = torch.randn(32)
-
-        graph_module = (
-            export_to_edge(LinalgVectorNorm(), (x,)).exported_program().graph_module
-        )
-
-        p = ReplaceAtenLinalgVectorNormWithCadenceLinalgVectorNormPass()
-        graph_after_passes = cast(PassResult, p(graph_module)).graph_module
-
-        # Assert that aten.linalg_vector_norm op was replaced by a
-        # cadence.linalg_vector_norm op
-        self.assertEqual(
-            count_node(
-                graph_after_passes,
-                exir_ops.edge.aten.linalg_vector_norm.default,
-            ),
-            0,
-        )
-        self.assertEqual(
-            count_node(
-                graph_after_passes, exir_ops.edge.cadence.linalg_vector_norm.default
-            ),
-            1,
-        )
-
     def test_replace_aten_where_with_cadence_where_Scalar(self):
         class WhereScalarModel(torch.nn.Module):
             def forward(self, cond: torch.Tensor):

backends/cadence/runtime/et_pal.cpp

@@ -6,7 +6,7 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-#if defined(XTENSA)
+#if defined(__XTENSA__)
 
 #include <stdio.h>
 #include <sys/times.h>

backends/cortex_m/test/targets.bzl

@@ -21,6 +21,7 @@ def define_operator_test_target(op):
             "//executorch/kernels/test:test_util",
             "//executorch/backends/cortex_m/ops:op_{}".format(op),
             "//executorch/backends/cortex_m/ops:cortex_m_generated_lib",
+            "//executorch/backends/cortex_m/ops:cortex_m_generated_lib_headers",
         ]
     )
 

backends/qualcomm/aot/ir/targets.bzl

@@ -4,7 +4,7 @@ load(
 )
 load("@fbsource//xplat/executorch/build:runtime_wrapper.bzl", "runtime")
 load("@fbsource//xplat/executorch/backends/qualcomm:targets.bzl", "generate_schema_header")
-load("@fbsource//xplat/executorch/backends/qualcomm/qnn_version.bzl", "get_qnn_library_verision")
+load("@fbsource//xplat/executorch/backends/qualcomm/qnn_version.bzl", "get_qnn_library_version")
 
 QCIR_NAME = "qcir"
 INPUT_QCIR = QCIR_NAME + ".fbs"
@@ -56,7 +56,7 @@ def define_common_targets():
         platforms = [ANDROID],
         visibility = ["@EXECUTORCH_CLIENTS"],
         deps = [
-            "fbsource//third-party/qualcomm/qnn/qnn-{0}:api".format(get_qnn_library_verision()),
+            "fbsource//third-party/qualcomm/qnn/qnn-{0}:api".format(get_qnn_library_version()),
             "//executorch/runtime/backend:interface",
             "//executorch/runtime/core:core",
             "//executorch/backends/qualcomm/aot/wrappers:wrappers",

backends/qualcomm/aot/python/targets.bzl

@@ -3,7 +3,7 @@ load(
     "ANDROID",
 )
 load("@fbsource//xplat/executorch/build:runtime_wrapper.bzl", "runtime")
-load("@fbsource//xplat/executorch/backends/qualcomm/qnn_version.bzl", "get_qnn_library_verision")
+load("@fbsource//xplat/executorch/backends/qualcomm/qnn_version.bzl", "get_qnn_library_version")
 
 PYTHON_MODULE_NAME = "PyQnnManagerAdaptor"
 
@@ -34,7 +34,7 @@ def define_common_targets():
             "//executorch/backends/qualcomm/aot/ir:qcir_utils",
             "//executorch/backends/qualcomm/runtime:runtime",
             "fbsource//third-party/pybind11:pybind11",
-            "fbsource//third-party/qualcomm/qnn/qnn-{0}:api".format(get_qnn_library_verision()),
+            "fbsource//third-party/qualcomm/qnn/qnn-{0}:api".format(get_qnn_library_version()),
         ],
         external_deps = [
             "libtorch_python",
@@ -67,7 +67,7 @@ def define_common_targets():
             "//executorch/backends/qualcomm/aot/ir:qcir_utils",
             "//executorch/backends/qualcomm/runtime:runtime",
             "fbsource//third-party/pybind11:pybind11",
-            "fbsource//third-party/qualcomm/qnn/qnn-{0}:api".format(get_qnn_library_verision()),
+            "fbsource//third-party/qualcomm/qnn/qnn-{0}:api".format(get_qnn_library_version()),
         ],
         external_deps = [
             "libtorch_python",
@@ -94,6 +94,6 @@ def define_common_targets():
             "//executorch/backends/qualcomm/aot/ir:qcir_utils",
             "//executorch/backends/qualcomm/runtime:runtime",
             "fbsource//third-party/pybind11:pybind11",
-            "fbsource//third-party/qualcomm/qnn/qnn-{0}:api".format(get_qnn_library_verision()),
+            "fbsource//third-party/qualcomm/qnn/qnn-{0}:api".format(get_qnn_library_version()),
         ],
     )

backends/qualcomm/aot/wrappers/targets.bzl

@@ -3,7 +3,7 @@ load(
     "ANDROID",
 )
 load("@fbsource//xplat/executorch/build:runtime_wrapper.bzl", "runtime")
-load("@fbsource//xplat/executorch/backends/qualcomm/qnn_version.bzl", "get_qnn_library_verision")
+load("@fbsource//xplat/executorch/backends/qualcomm/qnn_version.bzl", "get_qnn_library_version")
 
 def define_common_targets():
     """Defines targets that should be shared between fbcode and xplat.
@@ -23,7 +23,7 @@ def define_common_targets():
         platforms = [ANDROID],
         visibility = ["@EXECUTORCH_CLIENTS"],
         deps = [
-            "fbsource//third-party/qualcomm/qnn/qnn-{0}:api".format(get_qnn_library_verision()),
+            "fbsource//third-party/qualcomm/qnn/qnn-{0}:api".format(get_qnn_library_version()),
             "//executorch/runtime/backend:interface",
             "//executorch/runtime/core:core",
         ],