Merge branch 'main' into addGeluSupport

Author: leafs1

.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

@@ -1,11 +1,5 @@
 {
-  "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",

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/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"

backends/vulkan/runtime/graph/ComputeGraph.cpp

@@ -644,6 +644,11 @@ void ComputeGraph::prepare() {
   if (config_.enable_querypool) {
     context_->initialize_querypool();
   }
+
+  for (SharedObject& shared_object : shared_objects_) {
+    shared_object.allocate(this);
+    shared_object.bind_users(this);
+  }
 }
 
 void ComputeGraph::encode_prepack() {
@@ -668,11 +673,6 @@ void ComputeGraph::encode_execute() {
 
   context_->cmd_reset_querypool();
 
-  for (SharedObject& shared_object : shared_objects_) {
-    shared_object.allocate(this);
-    shared_object.bind_users(this);
-  }
-
   for (std::unique_ptr<ExecuteNode>& node : execute_nodes_) {
     node->encode(this);
   }

backends/vulkan/runtime/graph/ComputeGraph.h

@@ -21,6 +21,7 @@
 #include <executorch/backends/vulkan/runtime/graph/containers/Value.h>
 
 #include <executorch/backends/vulkan/runtime/graph/ops/DispatchNode.h>
+#include <executorch/backends/vulkan/runtime/graph/ops/DynamicDispatchNode.h>
 #include <executorch/backends/vulkan/runtime/graph/ops/ExecuteNode.h>
 #include <executorch/backends/vulkan/runtime/graph/ops/PrepackNode.h>
 

backends/vulkan/runtime/graph/ops/DispatchNode.h

@@ -22,7 +22,7 @@ class ComputeGraph;
 /*
  * Represents a single shader execution op in a ML model.
  */
-class DispatchNode final : public ExecuteNode {
+class DispatchNode : public ExecuteNode {
   friend class ComputeGraph;
 
  public:
@@ -43,9 +43,9 @@ class DispatchNode final : public ExecuteNode {
   void encode(ComputeGraph* graph) override;
 
  protected:
-  const vkapi::ShaderInfo shader_;
-  const utils::uvec3 global_workgroup_size_;
-  const utils::WorkgroupSize local_workgroup_size_;
+  vkapi::ShaderInfo shader_;
+  utils::uvec3 global_workgroup_size_;
+  utils::WorkgroupSize local_workgroup_size_;
   const vkapi::ParamsBindList params_;
   const vkapi::SpecVarList spec_vars_;
   const std::vector<PushConstantDataInfo> push_constants_;

backends/vulkan/runtime/graph/ops/DynamicDispatchNode.cpp

@@ -0,0 +1,49 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD-style license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+#include <executorch/backends/vulkan/runtime/graph/ops/DynamicDispatchNode.h>
+
+#include <executorch/backends/vulkan/runtime/graph/ComputeGraph.h>
+
+namespace vkcompute {
+
+DynamicDispatchNode::DynamicDispatchNode(
+    ComputeGraph& graph,
+    const PickShaderFn& pick_shader_fn,
+    const PickGlobalFn& pick_global_wg_fn,
+    const PickLocalFn& pick_local_wg_fn,
+    const std::vector<ArgGroup>& args,
+    const vkapi::ParamsBindList& params,
+    const std::vector<PushConstantDataInfo>& push_constants,
+    const vkapi::SpecVarList& spec_vars,
+    const std::vector<ValueRef>& resize_args,
+    const ResizeFunction& resize_fn)
+    : DispatchNode(
+          graph,
+          pick_shader_fn(&graph, args, resize_args),
+          pick_global_wg_fn(&graph, args, resize_args),
+          pick_local_wg_fn(&graph, args, resize_args),
+          args,
+          params,
+          push_constants,
+          spec_vars,
+          resize_args,
+          resize_fn),
+      pick_shader_fn_(pick_shader_fn),
+      pick_global_wg_fn_(pick_global_wg_fn),
+      pick_local_wg_fn_(pick_local_wg_fn) {}
+
+void DynamicDispatchNode::encode(ComputeGraph* graph) {
+  shader_ = pick_shader_fn_(graph, args_, resize_args_);
+  global_workgroup_size_ = pick_global_wg_fn_(graph, args_, resize_args_);
+  local_workgroup_size_ =
+      utils::WorkgroupSize(pick_local_wg_fn_(graph, args_, resize_args_));
+  DispatchNode::encode(graph);
+}
+
+} // namespace vkcompute

backends/vulkan/runtime/graph/ops/DynamicDispatchNode.h

@@ -0,0 +1,69 @@
+/*
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD-style license found in the
+ * LICENSE file in the root directory of this source tree.
+ */
+
+#pragma once
+
+#include <executorch/backends/vulkan/runtime/api/api.h>
+
+#include <executorch/backends/vulkan/runtime/graph/containers/PushConstantData.h>
+#include <executorch/backends/vulkan/runtime/graph/containers/Value.h>
+
+#include <executorch/backends/vulkan/runtime/graph/ops/DispatchNode.h>
+
+namespace vkcompute {
+
+class ComputeGraph;
+
+/*
+ * Represents a single shader execution op in a ML model.
+ */
+class DynamicDispatchNode final : public DispatchNode {
+  friend class ComputeGraph;
+
+ public:
+  using PickShaderFn = const std::function<vkapi::ShaderInfo(
+      ComputeGraph*,
+      const std::vector<ArgGroup>&,
+      const std::vector<ValueRef>&)>;
+  using PickGlobalFn = const std::function<utils::uvec3(
+      ComputeGraph*,
+      const std::vector<ArgGroup>&,
+      const std::vector<ValueRef>&)>;
+  using PickLocalFn = const std::function<utils::uvec3(
+      ComputeGraph*,
+      const std::vector<ArgGroup>&,
+      const std::vector<ValueRef>&)>;
+
+  explicit DynamicDispatchNode(
+      ComputeGraph& graph,
+      const PickShaderFn& pick_shader_fn,
+      const PickGlobalFn& pick_global_wg_fn,
+      const PickLocalFn& pick_local_wg_fn,
+      const std::vector<ArgGroup>& args,
+      const vkapi::ParamsBindList& params,
+      const std::vector<PushConstantDataInfo>& push_constants,
+      const vkapi::SpecVarList& spec_vars,
+      const std::vector<ValueRef>& resize_args,
+      const ResizeFunction& resize_fn = nullptr);
+
+  ~DynamicDispatchNode() override = default;
+
+  void encode(ComputeGraph* graph) override;
+
+ protected:
+  const PickShaderFn pick_shader_fn_;
+  const PickGlobalFn pick_global_wg_fn_;
+  const PickLocalFn pick_local_wg_fn_;
+
+ public:
+  operator bool() const {
+    return shader_;
+  }
+};
+
+} // namespace vkcompute