Cherry-picks for release 1.8.1 - Round3 (#8195)

* Revert the cuda algo finding change as this causes a significant memory bloat. (#8181)

* Revert the cuda algo finding change as this causes a significant memory bloat.

* Address PR comment

* Make pipelines to support torch1.8.1 and torch1.9.0 (#8084)

* Add post-install command to build PyTorch CPP extensions from within onnxruntime package (#8027)

ORTModule requires two PyTorch CPP extensions that are currently JIT compiled. The runtime compilation can cause issues in some environments without all build requirements or in environments with multiple instances of ORTModule running in parallel

This PR creates a custom command to compile such extensions that must be manually executed before ORTModule is executed for the first time. When users try to use ORTModule before the extensions are compiled, an error with instructions are raised

PyTorch CPP Extensions for ORTModule can be compiled by running:
python -m onnxruntime.training.ortmodule.torch_cpp_extensions.install

Full build environment is needed for this

* Patch orttraining-ortmodule pipeline with latest fix on master

* add cuda version to build config

* [rel-1.8.1][Objective-C API] Cherry-pick Objective-C API updates (#8197)

* [Objective-C API] Add ORTSession methods to get input, overridable initializer, and output names. (#7837)

* [Objective-C API] Fixes from package testing and clean up (#7866)

* [Objective-C API] Enable CoreML EP (#7914)

Enable CoreML EP in Objective-C API.

* [Objective-C API] Add script to assemble pod package files. (#7958)

Add a helper script for creating the Objective-C API pod package. It puts the necessary files and generates a podspec in a staging directory.

* [Objective-C API] Add support for documentation generation (#7999)

Adding support for generating API documentation with the Jazzy tool.
It's a manual process now, but we can eventually make it a part of the release pipeline.

Co-authored-by: Pranav Sharma <prs@microsoft.com>
Co-authored-by: liqunfu <liqfu@microsoft.com>
Co-authored-by: Thiago Crepaldi <thiago.crepaldi@microsoft.com>
Co-authored-by: Baiju Meswani <bmeswani@microsoft.com>
Co-authored-by: Edward Chen <18449977+edgchen1@users.noreply.github.com>

cmake/onnxruntime_objectivec.cmake

@@ -1,8 +1,18 @@
 # Copyright (c) Microsoft Corporation. All rights reserved.
 # Licensed under the MIT License.
 
-if(${CMAKE_VERSION} VERSION_LESS "3.18")
-    message(FATAL_ERROR "CMake 3.18+ is required when building the Objective-C API.")
+if(NOT APPLE)
+    message(FATAL_ERROR "The Objective-C API must be built on an Apple platform.")
+endif()
+
+set(ONNXRUNTIME_OBJC_MIN_CMAKE_VERSION "3.18")
+
+if(CMAKE_VERSION VERSION_LESS ONNXRUNTIME_OBJC_MIN_CMAKE_VERSION)
+    message(FATAL_ERROR "The Objective-C API requires CMake ${ONNXRUNTIME_OBJC_MIN_CMAKE_VERSION}+.")
+endif()
+
+if(NOT onnxruntime_BUILD_SHARED_LIB)
+    message(FATAL_ERROR "The Objective-C API requires onnxruntime_BUILD_SHARED_LIB to be enabled.")
 endif()
 
 check_language(OBJC)
@@ -34,41 +44,35 @@ set(OBJC_ARC_COMPILE_OPTIONS "-fobjc-arc" "-fobjc-arc-exceptions")
 # onnxruntime_objc target
 
 # these headers are the public interface
-# explicitly list them here so it is easy to see what is included
-set(onnxruntime_objc_headers
-    "${OBJC_ROOT}/include/onnxruntime.h"
-    "${OBJC_ROOT}/include/ort_enums.h"
-    "${OBJC_ROOT}/include/ort_env.h"
-    "${OBJC_ROOT}/include/ort_session.h"
-    "${OBJC_ROOT}/include/ort_value.h"
-    )
+file(GLOB onnxruntime_objc_headers CONFIGURE_DEPENDS
+    "${OBJC_ROOT}/include/*.h")
 
-file(GLOB onnxruntime_objc_srcs
+file(GLOB onnxruntime_objc_srcs CONFIGURE_DEPENDS
     "${OBJC_ROOT}/src/*.h"
     "${OBJC_ROOT}/src/*.m"
     "${OBJC_ROOT}/src/*.mm")
 
-# files common to implementation and test targets
-set(onnxruntime_objc_common_srcs
-    "${OBJC_ROOT}/common/assert_arc_enabled.mm")
-
 source_group(TREE "${OBJC_ROOT}" FILES
     ${onnxruntime_objc_headers}
-    ${onnxruntime_objc_srcs}
-    ${onnxruntime_objc_common_srcs})
+    ${onnxruntime_objc_srcs})
 
 add_library(onnxruntime_objc SHARED
     ${onnxruntime_objc_headers}
-    ${onnxruntime_objc_srcs}
-    ${onnxruntime_objc_common_srcs})
+    ${onnxruntime_objc_srcs})
 
 target_include_directories(onnxruntime_objc
     PUBLIC
         "${OBJC_ROOT}/include"
     PRIVATE
-        "${ONNXRUNTIME_ROOT}"
+        "${ONNXRUNTIME_INCLUDE_DIR}/core/session"
         "${OBJC_ROOT}")
 
+if(onnxruntime_USE_COREML)
+    target_include_directories(onnxruntime_objc
+        PRIVATE
+            "${ONNXRUNTIME_INCLUDE_DIR}/core/providers/coreml")
+endif()
+
 find_library(FOUNDATION_LIB Foundation REQUIRED)
 
 target_link_libraries(onnxruntime_objc
@@ -110,7 +114,7 @@ if(onnxruntime_BUILD_UNIT_TESTS)
 
     # onnxruntime_objc_test target
 
-    file(GLOB onnxruntime_objc_test_srcs
+    file(GLOB onnxruntime_objc_test_srcs CONFIGURE_DEPENDS
         "${OBJC_ROOT}/test/*.h"
         "${OBJC_ROOT}/test/*.m"
         "${OBJC_ROOT}/test/*.mm")
@@ -119,8 +123,7 @@ if(onnxruntime_BUILD_UNIT_TESTS)
 
     xctest_add_bundle(onnxruntime_objc_test onnxruntime_objc
         ${onnxruntime_objc_headers}
-        ${onnxruntime_objc_test_srcs}
-        ${onnxruntime_objc_common_srcs})
+        ${onnxruntime_objc_test_srcs})
 
     target_include_directories(onnxruntime_objc_test
         PRIVATE
@@ -129,12 +132,13 @@ if(onnxruntime_BUILD_UNIT_TESTS)
     target_compile_options(onnxruntime_objc_test PRIVATE ${OBJC_ARC_COMPILE_OPTIONS})
 
     set_target_properties(onnxruntime_objc_test PROPERTIES
-        FOLDER "ONNXRuntimeTest")
+        FOLDER "ONNXRuntimeTest"
+        XCODE_ATTRIBUTE_CODE_SIGNING_ALLOWED "NO")
 
     add_custom_command(TARGET onnxruntime_objc_test POST_BUILD
         COMMAND ${CMAKE_COMMAND} -E copy_directory
             "${OBJC_ROOT}/test/testdata"
-            "$<TARGET_BUNDLE_CONTENT_DIR:onnxruntime_objc_test>/Resources/testdata")
+            "$<TARGET_BUNDLE_CONTENT_DIR:onnxruntime_objc_test>/Resources")
 
     xctest_add_test(XCTest.onnxruntime_objc_test onnxruntime_objc_test)
 

cmake/onnxruntime_python.cmake

@@ -220,6 +220,15 @@ if (onnxruntime_ENABLE_TRAINING)
   file(GLOB onnxruntime_python_ortmodule_srcs CONFIGURE_DEPENDS
     "${ORTTRAINING_SOURCE_DIR}/python/training/ortmodule/*.py"
   )
+  file(GLOB onnxruntime_python_ortmodule_torch_cpp_ext_srcs CONFIGURE_DEPENDS
+    "${ORTTRAINING_SOURCE_DIR}/python/training/ortmodule/torch_cpp_extensions/*.py"
+  )
+  file(GLOB onnxruntime_python_ortmodule_torch_cpp_ext_aten_op_executor_srcs CONFIGURE_DEPENDS
+    "${ORTTRAINING_SOURCE_DIR}/python/training/ortmodule/torch_cpp_extensions/aten_op_executor/*"
+  )
+  file(GLOB onnxruntime_python_ortmodule_torch_cpp_ext_torch_gpu_allocator_srcs CONFIGURE_DEPENDS
+    "${ORTTRAINING_SOURCE_DIR}/python/training/ortmodule/torch_cpp_extensions/torch_gpu_allocator/*"
+  )
   file(GLOB onnxruntime_python_train_tools_srcs CONFIGURE_DEPENDS
     "${REPO_ROOT}/tools/python/register_custom_ops_pytorch_exporter.py"
   )
@@ -406,6 +415,9 @@ if (onnxruntime_ENABLE_TRAINING)
     COMMAND ${CMAKE_COMMAND} -E make_directory $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/amp
     COMMAND ${CMAKE_COMMAND} -E make_directory $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/optim
     COMMAND ${CMAKE_COMMAND} -E make_directory $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/ortmodule
+    COMMAND ${CMAKE_COMMAND} -E make_directory $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/ortmodule/torch_cpp_extensions
+    COMMAND ${CMAKE_COMMAND} -E make_directory $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/ortmodule/torch_cpp_extensions/aten_op_executor
+    COMMAND ${CMAKE_COMMAND} -E make_directory $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/ortmodule/torch_cpp_extensions/torch_gpu_allocator
     COMMAND ${CMAKE_COMMAND} -E copy
         ${onnxruntime_python_capi_training_srcs}
         $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/capi/training/
@@ -421,6 +433,15 @@ if (onnxruntime_ENABLE_TRAINING)
     COMMAND ${CMAKE_COMMAND} -E copy
         ${onnxruntime_python_ortmodule_srcs}
         $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/ortmodule/
+    COMMAND ${CMAKE_COMMAND} -E copy
+        ${onnxruntime_python_ortmodule_torch_cpp_ext_srcs}
+        $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/ortmodule/torch_cpp_extensions/
+    COMMAND ${CMAKE_COMMAND} -E copy
+        ${onnxruntime_python_ortmodule_torch_cpp_ext_aten_op_executor_srcs}
+        $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/ortmodule/torch_cpp_extensions/aten_op_executor/
+    COMMAND ${CMAKE_COMMAND} -E copy
+        ${onnxruntime_python_ortmodule_torch_cpp_ext_torch_gpu_allocator_srcs}
+        $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/ortmodule/torch_cpp_extensions/torch_gpu_allocator/
     COMMAND ${CMAKE_COMMAND} -E copy
         ${onnxruntime_python_train_tools_srcs}
         $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/

include/onnxruntime/core/providers/coreml/coreml_provider_factory.h

@@ -5,7 +5,7 @@
 #include "onnxruntime_c_api.h"
 
 // COREMLFlags are bool options we want to set for CoreML EP
-// This enum is defined as bit flats, and cannot have negative value
+// This enum is defined as bit flags, and cannot have negative value
 // To generate an uint32_t coreml_flags for using with OrtSessionOptionsAppendExecutionProvider_CoreML below,
 //   uint32_t coreml_flags = 0;
 //   coreml_flags |= COREML_FLAG_USE_CPU_ONLY;

include/onnxruntime/core/providers/nnapi/nnapi_provider_factory.h

@@ -5,7 +5,7 @@
 #include "onnxruntime_c_api.h"
 
 // NNAPIFlags are bool options we want to set for NNAPI EP
-// This enum is defined as bit flats, and cannot have negative value
+// This enum is defined as bit flags, and cannot have negative value
 // To generate an uint32_t nnapi_flags for using with OrtSessionOptionsAppendExecutionProvider_Nnapi below,
 //   uint32_t nnapi_flags = 0;
 //   nnapi_flags |= NNAPI_FLAG_USE_FP16;

include/onnxruntime/core/session/onnxruntime_c_api.h

@@ -507,7 +507,8 @@ struct OrtApi {
                   _Outptr_ OrtTypeInfo** type_info);
 
   /**
-   * \param value  is set to a null terminated string allocated using 'allocator'. The caller is responsible for freeing it.
+   * \param value is set to a null terminated UTF-8 encoded string allocated using 'allocator'.
+   *              The caller is responsible for freeing it.
    */
   ORT_API2_STATUS(SessionGetInputName, _In_ const OrtSession* sess, size_t index, _Inout_ OrtAllocator* allocator,
                   _Outptr_ char** value);

objectivec/docs/docs.readme.md

@@ -0,0 +1,5 @@
+# ONNX Runtime Objective-C API
+
+ONNX Runtime provides an Objective-C API.
+
+It can be used from Objective-C/C++ or Swift with a bridging header.

objectivec/docs/jazzy_config.yaml

@@ -0,0 +1,13 @@
+module: ONNX Runtime Objective-C API
+author: ONNX Runtime Authors
+author_url: https://www.onnxruntime.ai
+github_url: https://github.com/microsoft/onnxruntime
+
+objc: true
+umbrella_header: ../include/onnxruntime.h
+framework_root: ..
+
+readme: ./docs.readme.md
+
+hide_documentation_coverage: true
+undocumented_text: ""

objectivec/docs/readme.md

@@ -0,0 +1,17 @@
+# Objective-C API Documentation
+
+The API should be documented with comments in the [public header files](../include).
+
+## Documentation Generation
+
+The [Jazzy](https://github.com/realm/jazzy) tool is used to generate documentation from the code.
+
+For example, to generate documentation for a release version, from the repo root, run:
+
+```bash
+jazzy --config objectivec/docs/jazzy_config.yaml --output <output directory> --module-version $(cat VERSION_NUMBER)
+```
+
+The generated documentation website files will be in `<output directory>`.
+
+[docs.readme.md](./docs.readme.md) contains content for the main page of the generated documentation website.

objectivec/format_objc.sh

@@ -7,4 +7,3 @@ set -e
 SCRIPT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
 
 clang-format -i $(find ${SCRIPT_DIR} -name "*.h" -o -name "*.m" -o -name "*.mm")
-

objectivec/include/onnxruntime.h

@@ -4,6 +4,7 @@
 // this header contains the entire ONNX Runtime Objective-C API
 // the headers below can also be imported individually
 
+#import "ort_coreml_execution_provider.h"
 #import "ort_enums.h"
 #import "ort_env.h"
 #import "ort_session.h"

objectivec/include/ort_coreml_execution_provider.h

@@ -0,0 +1,54 @@
+// Copyright (c) Microsoft Corporation. All rights reserved.
+// Licensed under the MIT License.
+
+#import <Foundation/Foundation.h>
+
+#import "ort_session.h"
+
+NS_ASSUME_NONNULL_BEGIN
+
+/**
+ * Gets whether the CoreML execution provider is available.
+ */
+BOOL ORTIsCoreMLExecutionProviderAvailable(void);
+
+/**
+ * Options for configuring the CoreML execution provider.
+ */
+@interface ORTCoreMLExecutionProviderOptions : NSObject
+
+/**
+ * Whether the CoreML execution provider should run on CPU only.
+ */
+@property BOOL useCPUOnly;
+
+/**
+ * Whether the CoreML execution provider is enabled on subgraphs.
+ */
+@property BOOL enableOnSubgraphs;
+
+/**
+ * Whether the CoreML execution provider is only enabled for devices with Apple
+ * Neural Engine (ANE).
+ */
+@property BOOL onlyEnableForDevicesWithANE;
+
+@end
+
+@interface ORTSessionOptions (ORTSessionOptionsCoreMLEP)
+
+/**
+ * Enables the CoreML execution provider in the session configuration options.
+ * It is appended to the execution provider list which is ordered by
+ * decreasing priority.
+ *
+ * @param options The CoreML execution provider configuration options.
+ * @param error Optional error information set if an error occurs.
+ * @return Whether the provider was enabled successfully.
+ */
+- (BOOL)appendCoreMLExecutionProviderWithOptions:(ORTCoreMLExecutionProviderOptions*)options
+                                           error:(NSError**)error;
+
+@end
+
+NS_ASSUME_NONNULL_END

objectivec/include/ort_env.h

@@ -18,7 +18,7 @@ NS_ASSUME_NONNULL_BEGIN
  * Creates an ORT Environment.
  *
  * @param loggingLevel The environment logging level.
- * @param[out] error Optional error information set if an error occurs.
+ * @param error Optional error information set if an error occurs.
  * @return The instance, or nil if an error occurs.
  */
 - (nullable instancetype)initWithLoggingLevel:(ORTLoggingLevel)loggingLevel