Merge branch 'main' into gh/lucylq/79/orig

Author: Gasoonjia

.ci/scripts/tests/test_gather_benchmark_configs.py

@@ -53,11 +53,11 @@ def test_skip_disabled_configs(self):
                 "mv3": [
                     self.gather_benchmark_configs.DisabledConfig(
                         config_name="disabled_config1",
-                        github_issue="https://github.com/org/repo/issues/123",
+                        github_issue="https://github.com/org/repo/issues/123",  # @lint-ignore
                     ),
                     self.gather_benchmark_configs.DisabledConfig(
                         config_name="disabled_config2",
-                        github_issue="https://github.com/org/repo/issues/124",
+                        github_issue="https://github.com/org/repo/issues/124",  # @lint-ignore
                     ),
                 ]
             },
@@ -84,7 +84,9 @@ def test_skip_disabled_configs(self):
             self.assertIn("enabled_config2", result)
 
     def test_disabled_configs_have_github_links(self):
-        github_issue_regex = re.compile(r"https://github\.com/.+/.+/issues/\d+")
+        github_issue_regex = re.compile(
+            r"https://github\.com/.+/.+/issues/\d+"  # @lint-ignore
+        )
 
         for (
             model_name,

.ci/scripts/unittest-linux.sh

@@ -20,8 +20,9 @@ if [[ "$BUILD_TOOL" == "cmake" ]]; then
     # Setup swiftshader and Vulkan SDK which are required to build the Vulkan delegate
     source .ci/scripts/setup-vulkan-linux-deps.sh
 
+    # We need the runner to test the built library.
     PYTHON_EXECUTABLE=python \
-    CMAKE_ARGS="-DEXECUTORCH_BUILD_PYBIND=ON -DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
+    CMAKE_ARGS="-DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
     .ci/scripts/setup-linux.sh "$@"
 
     .ci/scripts/unittest-linux-cmake.sh

.ci/scripts/unittest-macos.sh

@@ -20,8 +20,9 @@ export PATH="${TMP_DIR}:$PATH"
 trap 'rm -rfv ${TMP_DIR}' EXIT
 
 # Setup MacOS dependencies as there is no Docker support on MacOS atm
+# We need the runner to test the built library.
 PYTHON_EXECUTABLE=python \
-CMAKE_ARGS="-DEXECUTORCH_BUILD_PYBIND=ON -DEXECUTORCH_BUILD_COREML=ON -DEXECUTORCH_BUILD_MPS=ON -DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
+CMAKE_ARGS="-DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
 ${CONDA_RUN} --no-capture-output \
 .ci/scripts/setup-macos.sh "$@"
 

.ci/scripts/wheel/envvar_base.sh

@@ -11,7 +11,3 @@
 # Ensure that CMAKE_ARGS is defined before referencing it. Defaults to empty
 # if not defined.
 export CMAKE_ARGS="${CMAKE_ARGS:-}"
-
-# Link the XNNPACK backend into the pybindings runtime so that users can execute
-# ExecuTorch programs that delegate to it.
-CMAKE_ARGS="${CMAKE_ARGS} -DEXECUTORCH_BUILD_PYBIND=ON -DEXECUTORCH_BUILD_XNNPACK=ON"

.ci/scripts/wheel/envvar_macos.sh

@@ -9,7 +9,3 @@
 # any variables so that subprocesses will see them.
 
 source "${GITHUB_WORKSPACE}/${REPOSITORY}/.ci/scripts/wheel/envvar_base.sh"
-
-# When building for macOS, link additional backends into the pybindings runtime.
-CMAKE_ARGS="${CMAKE_ARGS} -DEXECUTORCH_BUILD_COREML=ON"
-CMAKE_ARGS="${CMAKE_ARGS} -DEXECUTORCH_BUILD_MPS=ON"

.github/scripts/label_utils.py

@@ -51,7 +51,6 @@ def get_last_page_num_from_header(header: Any) -> int:
     # rel="next", <https://api.github.com/repositories/65600975/labels?per_page=100&page=3>; rel="last"
     link_info = header["link"]
     # Docs does not specify that it should be present for projects with just few labels
-    # And https://github.com/malfet/deleteme/actions/runs/7334565243/job/19971396887 it's not the case
     if link_info is None:
         return 1
     prefix = "&page="

.github/scripts/test_extract_benchmark_results.py

@@ -29,8 +29,8 @@ def get_mock_happy_flow_content(app_type: str = "IOS_APP"):
                 "name": "Syslog",
                 "type": "DEVICE_LOG",
                 "extension": "syslog",
-                "url": "https://job_arn_1_device_log",
-                "s3_url": "https://job_arn_1/test-workflow1/1/syslog.syslog",
+                "url": "https://job_arn_1_device_log",  # @lint-ignore
+                "s3_url": "https://job_arn_1/test-workflow1/1/syslog.syslog",  # @lint-ignore
                 "app_type": app_type,
                 "job_name": "job_arn_1_name",
                 "os": "14",
@@ -55,8 +55,8 @@ def get_mock_happy_flow_content(app_type: str = "IOS_APP"):
                 "name": "Customer Artifacts",
                 "type": "CUSTOMER_ARTIFACT",
                 "extension": "zip",
-                "url": "https://job_arn_1_customer_artifact",
-                "s3_url": "https://job_arn_1_customer_artifact1",
+                "url": "https://job_arn_1_customer_artifact",  # @lint-ignore
+                "s3_url": "https://job_arn_1_customer_artifact1",  # @lint-ignore
                 "app_type": app_type,
                 "job_name": "job_arn_1_device_name",
                 "os": "14",
@@ -68,8 +68,8 @@ def get_mock_happy_flow_content(app_type: str = "IOS_APP"):
                 "name": "Syslog",
                 "type": "DEVICE_LOG",
                 "extension": "syslog",
-                "url": "https://job_arn_1_device_log",
-                "s3_url": "https://job_arn_1/test-workflow1/1/syslog.syslog",
+                "url": "https://job_arn_1_device_log",  # @lint-ignore
+                "s3_url": "https://job_arn_1/test-workflow1/1/syslog.syslog",  # @lint-ignore
                 "app_type": app_type,
                 "job_name": "job_arn_2_name",
                 "os": "14",
@@ -94,8 +94,8 @@ def get_mock_happy_flow_content(app_type: str = "IOS_APP"):
                 "name": "Customer Artifacts",
                 "type": "CUSTOMER_ARTIFACT",
                 "extension": "zip",
-                "url": "https://job_arn_1_customer_artifact",
-                "s3_url": "https://job_arn_1_customer_artifact1",
+                "url": "https://job_arn_1_customer_artifact",  # @lint-ignore
+                "s3_url": "https://job_arn_1_customer_artifact1",  # @lint-ignore
                 "app_type": app_type,
                 "job_name": "job_arn_2_name",
                 "os": "14",

.github/scripts/trymerge.py

@@ -1964,7 +1964,7 @@ def check_for_sev(org: str, project: str, skip_mandatory_checks: bool) -> None:
     response = cast(
         Dict[str, Any],
         gh_fetch_json_list(
-            "https://api.github.com/search/issues",
+            "https://api.github.com/search/issues",  # @lint-ignore
             params={"q": f'repo:{org}/{project} is:open is:issue label:"ci: sev"'},
         ),
     )

CMakePresets.json

@@ -1,17 +1,10 @@
 {
-  "version": 10,
-  "cmakeMinimumRequired": {
-    "major": 3,
-    "minor": 31,
-    "patch": 0
-  },
-  "$comment": "On-device AI across mobile, embedded and edge for PyTorch.",
+  "version": 6,
   "configurePresets": [
     {
       "name": "common",
       "hidden": true,
-      "binaryDir": "${sourceDir}/cmake-out",
-      "generator": "Unix Makefiles"
+      "binaryDir": "${sourceDir}/cmake-out"
     },
     {
       "name": "macos-arm64",

backends/cadence/aot/TARGETS

@@ -211,6 +211,7 @@ python_library(
     typing = True,
     deps = [
         ":pass_utils",
+        ":utils",
         "//executorch/backends/cadence/aot:pass_utils",
         "//executorch/exir:pass_base",
         "//executorch/exir/dialects:lib",

backends/cadence/aot/simplify_ops.py

@@ -16,10 +16,10 @@
     CadencePassAttribute,
     register_cadence_pass,
 )
+from executorch.backends.cadence.aot.utils import rebind
 from executorch.exir.dialects._ops import ops as exir_ops
 from executorch.exir.dialects.edge._ops import EdgeOpOverload
 from executorch.exir.pass_base import ExportPass, ProxyValue
-from torch.fx.operator_schemas import get_signature_for_torch_op
 
 
 @register_cadence_pass(CadencePassAttribute(opt_level=0))
@@ -117,32 +117,11 @@ class BindOptionalArgsPass(ExportPass):
     def call_operator(self, op, args, kwargs, meta):
         if not isinstance(op, EdgeOpOverload):
             return super().call_operator(op, args, kwargs, meta)
-        assert callable(op)
 
-        torch_op_schemas = get_signature_for_torch_op(op._op)
-        if len(torch_op_schemas) == 0:
-            return super().call_operator(op, args, kwargs, meta)
-
-        matched_schemas = []
-        # Iterate through all of the schema until we find one that matches
-        # If one matches, populate `new_args_and_kwargs` with the new args/kwargs
-        # values. If none matches, `new_args_and_kwargs` will be None
-        for candidate_signature in torch_op_schemas:
-            try:
-                candidate_signature.bind(*args, **kwargs)
-                matched_schemas.append(candidate_signature)
-            except TypeError:
-                continue
-
-        if len(matched_schemas) != 1:
-            # Did not match any schema. Cannot normalize
-            return super().call_operator(op, args, kwargs, meta)
-
-        sig = matched_schemas[0]
-        bound_args = sig.bind(*args, **kwargs)
-        bound_args.apply_defaults()
+        if (updated_args := rebind(op, args, kwargs)) is not None:
+            args, kwargs = updated_args
 
-        return super().call_operator(op, bound_args.args, bound_args.kwargs, meta)
+        return super().call_operator(op, args, kwargs, meta)
 
 
 # This class encapsulates all the functions that simplify the op's args

backends/cadence/aot/tests/test_fusion_ops_passes.py

@@ -248,32 +248,28 @@ def forward(self, x):
         )
 
     def test_force_quant_dequant_fusion(self):
-        class M(torch.nn.Module):
-            def __init__(self):
-                super().__init__()
-
-            def forward(self, x):
-                x = torch.ops.quantized_decomposed.quantize_per_tensor(
-                    x, 1.2, 3, 0, 127, torch.int8
-                )
-                x = torch.permute(x, [2, 0, 1, 3])
-                x = torch.ops.quantized_decomposed.dequantize_per_tensor(
-                    x, 4.5, 6, 0, 127, torch.int8
-                )
-                return x
-
-        inputs = torch.randn(2, 12, 1, 6)
-        model = M()
-        graph_module = export_to_edge(model, (inputs,)).exported_program().graph_module
-
-        graph_module = FuseQuantDequantToRequantizePass(
+        builder = GraphBuilder()
+        x = builder.placeholder("x", torch.randn(2, 12, 1, 6, dtype=torch.float32))
+        quant = builder.call_operator(
+            op=exir_ops.edge.quantized_decomposed.quantize_per_tensor.default,
+            args=(x, 1.2, 3, 0, 127, torch.int8),
+        )
+        permute = builder.call_operator(
+            op=exir_ops.edge.aten.permute_copy.default, args=(quant, [2, 0, 1, 3])
+        )
+        dequant = builder.call_operator(
+            op=exir_ops.edge.quantized_decomposed.dequantize_per_tensor.default,
+            args=(permute, 4.5, 6, 0, 127, torch.int8),
+        )
+        builder.output(dequant)
+        original_graph = builder.get_graph_module()
+        converted_graph = FuseQuantDequantToRequantizePass(
             force_quant_dequant_fusion=True
-        )(graph_module).graph_module
+        )(original_graph).graph_module
         self.check_op_counts(
-            graph_module,
+            converted_graph,
             expected_op_counts={
-                # Verify that no dequant/quant pair was replaced with requantize.
-                # quantize -> permute -> dequantize should not be replaced with requantize.
+                # Verify that dequant/quant pair was replaced with requantize.
                 exir_ops.edge.quantized_decomposed.quantize_per_tensor.default: 0,
                 exir_ops.edge.quantized_decomposed.dequantize_per_tensor.default: 0,
                 exir_ops.edge.cadence.requantize.default: 1,

backends/cadence/aot/utils.py

@@ -18,7 +18,9 @@
 from executorch.exir import ExecutorchProgramManager, memory
 from executorch.exir.dialects._ops import ops as exir_ops
 from executorch.exir.dialects.edge._ops import EdgeOpOverload, EdgeOpOverloadPacket
+from executorch.exir.pass_base import Argument
 from tabulate import tabulate
+from torch.fx.operator_schemas import get_signature_for_torch_op
 
 from torch.utils._pytree import tree_flatten
 
@@ -308,3 +310,30 @@ def get_size(self, exir_id: int) -> int:
 # Return default memory config for the backend
 def get_default_memory_config() -> MemoryConfig:
     return MemoryConfig(memory_sizes=[0x1000000000])
+
+
+def rebind(
+    op: EdgeOpOverload, args: tuple[Argument, ...], kwargs: dict[str, Argument]
+) -> Optional[tuple[tuple[Argument, ...], dict[str, Argument]]]:
+    """Populates optional args and binds args/kwargs based on schema."""
+    torch_op_schemas = get_signature_for_torch_op(op._op)
+
+    matched_schemas = []
+    # Iterate through all of the schema until we find one that matches
+    # If one matches, populate `new_args_and_kwargs` with the new args/kwargs
+    # values. If none matches, `new_args_and_kwargs` will be None
+    for candidate_signature in torch_op_schemas:
+        try:
+            candidate_signature.bind(*args, **kwargs)
+            matched_schemas.append(candidate_signature)
+        except TypeError:
+            continue
+
+    if len(matched_schemas) != 1:
+        # Did not match any schema. Cannot normalize
+        return None
+
+    bound_args = matched_schemas[0].bind(*args, **kwargs)
+    bound_args.apply_defaults()
+
+    return bound_args.args, bound_args.kwargs

backends/openvino/scripts/openvino_build.sh

@@ -48,16 +48,11 @@ main() {
         # Set parameters to configure the project with CMake
         # Note: Add any additional configuration options you need here
         export CMAKE_ARGS="-DEXECUTORCH_BUILD_OPENVINO=ON \
-                           -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
-                           -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
-                           -DEXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL=ON \
-                           -DEXECUTORCH_ENABLE_LOGGING=ON \
-                           -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
-                           -DEXECUTORCH_BUILD_PYBIND=ON"
+                           -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON"
         export CMAKE_BUILD_ARGS="--target openvino_backend"
 
         # Build the package
-        pip install . --no-build-isolation
+        ./install_executorch.sh
 
     else
         echo "Error: Argument is not valid: $build_type"

devtools/inspector/_inspector.py

@@ -1240,7 +1240,7 @@ def find_total_for_module(self, module_name: str) -> float:
         for block in self.event_blocks:
             for event in block.events:
                 # Skip OPERATOR_CALL events to avoid double-counting and exclude framework tax
-                if event.event_name == "OPERATOR_CALL":
+                if event.name == "OPERATOR_CALL":
                     continue
 
                 module_hierarchy = event.module_hierarchy.values()