Merge branch 'main' of github.com:microsoft/onnxruntime into abjindal…

…/update_deepspeed_version

cgmanifests/generated/cgmanifest.json

@@ -112,7 +112,7 @@
       "component": {
         "type": "git",
         "git": {
-          "commitHash": "5a5f8a5935762397aa68429b5493084ff970f774",
+          "commitHash": "9b7bca2a723ff94edcd007d93b5d0cf1838591dc",
           "repositoryUrl": "https://github.com/onnx/onnx.git"
         },
         "comments": "git submodule at cmake/external/onnx"
@@ -318,6 +318,16 @@
         "comments": "onnx_tensorrt"
       }
     },
+    {
+      "component": {
+        "type": "git",
+        "git": {
+          "commitHash": "a20c65f2cd549445fda907f7b83894c8eb7427d6",
+          "repositoryUrl": "https://github.com/protocolbuffers/protobuf.git"
+        },
+        "comments": "protobuf"
+      }
+    },
     {
       "component": {
         "type": "git",

cmake/CMakeLists.txt

@@ -1041,6 +1041,7 @@ endif()
 
 if (onnxruntime_USE_DNNL)
   include(dnnl)
+  add_compile_definitions(DNNL_OPENMP)
 endif()
 
 # TVM EP

cmake/deps.txt

@@ -23,10 +23,10 @@ microsoft_gsl;https://github.com/microsoft/GSL/archive/refs/tags/v4.0.0.zip;cf36
 microsoft_wil;https://github.com/microsoft/wil/archive/5f4caba4e7a9017816e47becdd918fcc872039ba.zip;fd119887d0d17c37adf1fc227b054befa28158ad
 mimalloc;https://github.com/microsoft/mimalloc/archive/refs/tags/v2.0.3.zip;e4f37b93b2da78a5816c2495603a4188d316214b
 mp11;https://github.com/boostorg/mp11/archive/refs/tags/boost-1.79.0.zip;c8f04e378535ededbe5af52c8f969d2dedbe73d5
-onnx;https://github.com/onnx/onnx/archive/1ba785612a79fe749aa1e478336e534743372639.zip;6c22c32fabafac81107186ea6bedb31741a3142d
+onnx;https://github.com/onnx/onnx/archive/refs/tags/v1.13.0.zip;8dda5079cdb5a134b08b0c73f4592a6404fc2dc6
 #use the commit where it's several commits after 8.5-GA branch (https://github.com/onnx/onnx-tensorrt/commit/369d6676423c2a6dbf4a5665c4b5010240d99d3c)
 onnx_tensorrt;https://github.com/onnx/onnx-tensorrt/archive/369d6676423c2a6dbf4a5665c4b5010240d99d3c.zip;62119892edfb78689061790140c439b111491275
-protobuf;https://github.com/protocolbuffers/protobuf/archive/refs/tags/v3.18.3.zip;b95bf7e9de9c2249b6c1f2ca556ace49999e90bd
+protobuf;https://github.com/protocolbuffers/protobuf/archive/refs/tags/v3.20.2.zip;9f71dad95fb83438e88822a9969fc93773fd8c48
 psimd;https://github.com/Maratyszcza/psimd/archive/072586a71b55b7f8c584153d223e95687148a900.zip;1f5454b01f06f9656b77e4a5e2e31d7422487013
 pthreadpool;https://github.com/Maratyszcza/pthreadpool/archive/1787867f6183f056420e532eec640cba25efafea.zip;e43e80781560c5ab404a4da20f34d846f5f5d101
 pybind11;https://github.com/pybind/pybind11/archive/refs/tags/v2.10.1.zip;769b6aa67a77f17a770960f604b727645b6f6a13

cmake/external/onnxruntime_external_deps.cmake

@@ -115,7 +115,7 @@ FetchContent_Declare(
   URL_HASH SHA1=${DEP_SHA1_protobuf}
   SOURCE_SUBDIR  cmake
   PATCH_COMMAND ${ONNXRUNTIME_PROTOBUF_PATCH_COMMAND}
-  FIND_PACKAGE_ARGS 3.18.0 NAMES Protobuf
+  FIND_PACKAGE_ARGS 3.20.2 NAMES Protobuf
 )
 set(protobuf_BUILD_TESTS OFF CACHE BOOL "Build protobuf tests" FORCE)
 if (CMAKE_SYSTEM_NAME STREQUAL "Android")

cmake/onnxruntime_providers.cmake

@@ -589,6 +589,11 @@ if (onnxruntime_USE_DNNL)
     target_include_directories(onnxruntime_providers_dnnl PRIVATE ${ORTTRAINING_ROOT})
   endif()
 
+  # Needed for threadpool handling
+  if(onnxruntime_BUILD_JAVA)
+    add_compile_definitions(DNNL_JAVA)
+  endif()
+
   if(APPLE)
     set_property(TARGET onnxruntime_providers_dnnl APPEND_STRING PROPERTY LINK_FLAGS "-Xlinker -exported_symbols_list ${ONNXRUNTIME_ROOT}/core/providers/dnnl/exported_symbols.lst")
     set_target_properties(onnxruntime_providers_dnnl PROPERTIES

cmake/onnxruntime_unittests.cmake

@@ -763,7 +763,7 @@ if (MSVC)
   # The warning means the type of two integral values around a binary operator is narrow than their result.
   # If we promote the two input values first, it could be more tolerant to integer overflow.
   # However, this is test code. We are less concerned.
-  target_compile_options(onnxruntime_test_all PRIVATE "/wd26451")
+  target_compile_options(onnxruntime_test_all PRIVATE "/wd26451" "/wd4244")
 else()
   target_compile_options(onnxruntime_test_all PRIVATE "-Wno-parentheses")
 endif()

cmake/patches/onnx/onnx.patch

@@ -1,8 +1,16 @@
 diff --git a/CMakeLists.txt b/CMakeLists.txt
-index 8e595855..789ec80a 100644
+index 4dd56b6e..018da488 100644
 --- a/CMakeLists.txt
 +++ b/CMakeLists.txt
-@@ -545,20 +545,9 @@ endif()
+@@ -397,6 +397,7 @@ if (MSVC)
+   endif()
+ else()
+   # On non-Windows, hide all symbols we don't need
++  set(EXTRA_FLAGS "-Wno-unused-parameter")
+   set(ONNX_API_DEFINE "-DONNX_API=__attribute__\(\(__visibility__\(\"default\"\)\)\)")
+   set_target_properties(onnx_proto PROPERTIES CXX_VISIBILITY_PRESET hidden)
+   set_target_properties(onnx_proto PROPERTIES VISIBILITY_INLINES_HIDDEN 1)
+@@ -548,20 +549,9 @@ endif()
  if(MSVC)
    target_compile_options(onnx_proto
                           PRIVATE /MP
@@ -23,3 +31,29 @@ index 8e595855..789ec80a 100644
                                   ${EXTRA_FLAGS})
    if(ONNX_USE_PROTOBUF_SHARED_LIBS)
        target_compile_options(onnx_proto
+diff --git a/onnx/onnx_pb.h b/onnx/onnx_pb.h
+index 0aab3e26..0f859267 100644
+--- a/onnx/onnx_pb.h
++++ b/onnx/onnx_pb.h
+@@ -47,10 +47,21 @@
+ #define ONNX_API ONNX_IMPORT
+ #endif
+
++//  onnx/onnx-operators-ml.pb.h:1178:25:   required from here
++//  protobuf/src/google/protobuf/repeated_ptr_field.h:752:66: error: unused parameter ‘prototype’ [-Werror=unused-parameter]
++#if defined(__GNUC__)
++#pragma GCC diagnostic push
++#pragma GCC diagnostic ignored "-Wunused-parameter"
++#endif
++
+ #ifdef ONNX_ML
+ #include "onnx/onnx-ml.pb.h"
+ #else
+ #include "onnx/onnx.pb.h"
+ #endif
+
++#if defined(__GNUC__)
++#pragma GCC diagnostic pop
++#endif
++
+ #endif // ! ONNX_ONNX_PB_H

cmake/patches/protobuf/protobuf_cmake.patch

@@ -1,25 +1,31 @@
 diff --git a/cmake/CMakeLists.txt b/cmake/CMakeLists.txt
-index 61a5c3ded..68d7c5598 100644
+index ac92442a1..e930cbd2e 100644
 --- a/cmake/CMakeLists.txt
 +++ b/cmake/CMakeLists.txt
-@@ -199,12 +199,8 @@ if (MSVC)
-   endif()
+@@ -240,9 +240,7 @@ if (MSVC)
    # MSVC warning suppressions
    add_definitions(
--    /wd4018 # 'expression' : signed/unsigned mismatch
      /wd4065 # switch statement contains 'default' but no 'case' labels
--    /wd4146 # unary minus operator applied to unsigned type, result still unsigned
 -    /wd4244 # 'conversion' conversion from 'type1' to 'type2', possible loss of data
      /wd4251 # 'identifier' : class 'type' needs to have dll-interface to be used by clients of class 'type2'
 -    /wd4267 # 'var' : conversion from 'size_t' to 'type', possible loss of data
      /wd4305 # 'identifier' : truncation from 'type1' to 'type2'
      /wd4307 # 'operator' : integral constant overflow
      /wd4309 # 'conversion' : truncation of constant value
-@@ -212,7 +208,6 @@ if (MSVC)
+@@ -250,7 +248,6 @@ if (MSVC)
      /wd4355 # 'this' : used in base member initializer list
      /wd4506 # no definition for inline function 'function'
      /wd4800 # 'type' : forcing value to bool 'true' or 'false' (performance warning)
 -    /wd4996 # The compiler encountered a deprecated declaration.
    )
    # Allow big object
    add_definitions(/bigobj)
+@@ -272,6 +269,8 @@ if (MSVC)
+   endif()
+
+   configure_file(version.rc.in ${CMAKE_CURRENT_BINARY_DIR}/version.rc @ONLY)
++else (MSVC)
++  set(CMAKE_CXX_FLAGS  "${CMAKE_CXX_FLAGS} -Wno-unused-parameter")
+ endif (MSVC)
+
+

include/onnxruntime/core/providers/dnnl/dnnl_provider_options.h

@@ -0,0 +1,8 @@
+// Copyright(C) 2022 Intel Corporation
+// Licensed under the MIT License
+#pragma once
+
+struct OrtDnnlProviderOptions {
+  int use_arena;          // If arena is used, use_arena 0 = not used, nonzero = used
+  void* threadpool_args;  // Used to enable configure the oneDNN threadpool
+};

include/onnxruntime/core/session/onnxruntime_c_api.h

@@ -268,6 +268,7 @@ ORT_RUNTIME_CLASS(PrepackedWeightsContainer);
 ORT_RUNTIME_CLASS(TensorRTProviderOptionsV2);
 ORT_RUNTIME_CLASS(CUDAProviderOptionsV2);
 ORT_RUNTIME_CLASS(CANNProviderOptions);
+ORT_RUNTIME_CLASS(DnnlProviderOptions);
 ORT_RUNTIME_CLASS(Op);
 ORT_RUNTIME_CLASS(OpAttr);
 
@@ -3879,6 +3880,73 @@ struct OrtApi {
 
   /// @}
 
+  /** \brief Append dnnl provider to session options
+   *
+   * If oneDNN is not available, this function will return failure.
+   *
+   * \param[in] options
+   * \param[in] cann_options
+   *
+   * \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.14.
+   */
+  ORT_API2_STATUS(SessionOptionsAppendExecutionProvider_Dnnl,
+                  _In_ OrtSessionOptions* options, _In_ const OrtDnnlProviderOptions* dnnl_options);
+
+   /** \brief Create an OrtDnnlProviderOptions
+   *
+   * \param[out] out Newly created ::OrtDnnlProviderOptions. Must be released with OrtApi::ReleaseDnnlProviderOptions
+   *
+   * \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.14.
+   */
+  ORT_API2_STATUS(CreateDnnlProviderOptions, _Outptr_ OrtDnnlProviderOptions** out);
+
+  /** \brief Set options in a oneDNN Execution Provider.
+   *
+   * Key should be in null terminated string format of the member of ::OrtDnnlProviderOptions
+   * and value should be its related range.
+   *
+   * For example, key="use_arena" and value="1"
+   *
+   * \param[in] dnnl_options
+   * \param[in] provider_options_keys Array of UTF-8 null-terminated string for provider options keys
+   * \param[in] provider_options_values Array of UTF-8 null-terminated string for provider options values
+   * \param[in] num_keys Number of elements in the `provider_option_keys` and `provider_options_values` arrays
+   *
+   * \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.14.
+   */
+  ORT_API2_STATUS(UpdateDnnlProviderOptions, _Inout_ OrtDnnlProviderOptions* dnnl_options,
+                  _In_reads_(num_keys) const char* const* provider_options_keys,
+                  _In_reads_(num_keys) const char* const* provider_options_values,
+                  _In_ size_t num_keys);
+
+  /**
+   * Get serialized oneDNN provider options string.
+   *
+   * For example, "use_arena=1;......"
+   *
+   * \param dnnl_options - OrtDnnlProviderOptions instance
+   * \param allocator - a ptr to an instance of OrtAllocator obtained with CreateAllocator() or GetAllocatorWithDefaultOptions()
+   *                      the specified allocator will be used to allocate continuous buffers for output strings and lengths.
+   * \param ptr - is a UTF-8 null terminated string allocated using 'allocator'. The caller is responsible for using the same allocator to free it.
+   *
+   * \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.14.
+   */
+  ORT_API2_STATUS(GetDnnlProviderOptionsAsString, _In_ const OrtDnnlProviderOptions* dnnl_options, _Inout_ OrtAllocator* allocator, _Outptr_ char** ptr);
+
+  /** \brief Release an ::OrtDnnlProviderOptions
+   *
+   * \since Version 1.14.
+   */
+  void(ORT_API_CALL* ReleaseDnnlProviderOptions)(_Frees_ptr_opt_ OrtDnnlProviderOptions* input);
+
 #ifdef __cplusplus
   OrtApi(const OrtApi&) = delete;  // Prevent users from accidentally copying the API structure, it should always be passed as a pointer
 #endif
@@ -3980,6 +4048,16 @@ ORT_API_STATUS(OrtSessionOptionsAppendExecutionProvider_CUDA, _In_ OrtSessionOpt
  */
 ORT_API_STATUS(OrtSessionOptionsAppendExecutionProvider_MIGraphX, _In_ OrtSessionOptions* options, int device_id);
 
+/*
+ * This is the old way to add the oneDNN provider to the session, please use
+ * SessionOptionsAppendExecutionProvider_oneDNN above to access the latest functionality
+ * This function always exists, but will only succeed if Onnxruntime was built with
+ * oneDNN support and the oneDNN provider shared library exists
+ *
+ * \param use_arena zero: false. non-zero: true.
+ */
+ORT_API_STATUS(OrtSessionOptionsAppendExecutionProvider_Dnnl, _In_ OrtSessionOptions* options, int use_arena);
+
 #ifdef __cplusplus
 }
 #endif

include/onnxruntime/core/session/onnxruntime_cxx_api.h

@@ -579,6 +579,8 @@ struct SessionOptionsImpl : ConstSessionOptionsImpl<T> {
   SessionOptionsImpl& AppendExecutionProvider_MIGraphX(const OrtMIGraphXProviderOptions& provider_options);       ///< Wraps OrtApi::SessionOptionsAppendExecutionProvider_MIGraphX
   ///< Wraps OrtApi::SessionOptionsAppendExecutionProvider_CANN
   SessionOptionsImpl& AppendExecutionProvider_CANN(const OrtCANNProviderOptions& provider_options);
+  ///< Wraps OrtApi::SessionOptionsAppendExecutionProvider_Dnnl
+  SessionOptionsImpl& AppendExecutionProvider_Dnnl(const OrtDnnlProviderOptions& provider_options); 
   /// Wraps OrtApi::SessionOptionsAppendExecutionProvider. Currently supports SNPE and XNNPACK.
   SessionOptionsImpl& AppendExecutionProvider(const std::string& provider_name,
                                               const std::unordered_map<std::string, std::string>& provider_options = {});

include/onnxruntime/core/session/onnxruntime_cxx_inline.h

@@ -732,6 +732,12 @@ inline SessionOptionsImpl<T>& SessionOptionsImpl<T>::AppendExecutionProvider_CAN
   return *this;
 }
 
+template <typename T>
+inline SessionOptionsImpl<T>& SessionOptionsImpl<T>::AppendExecutionProvider_Dnnl(const OrtDnnlProviderOptions& provider_options) {
+  ThrowOnError(GetApi().SessionOptionsAppendExecutionProvider_Dnnl(this->p_, &provider_options));
+  return *this;
+}
+
 template <typename T>
 inline SessionOptionsImpl<T>& SessionOptionsImpl<T>::AppendExecutionProvider(
     const std::string& provider_name,

java/src/main/native/ai_onnxruntime_OrtSession_SessionOptions.c

@@ -17,7 +17,6 @@
 // Providers
 #define ORT_JAVA_MAX_ARGUMENT_ARRAY_LENGTH 128
 #include "onnxruntime/core/providers/cpu/cpu_provider_factory.h"
-#include "onnxruntime/core/providers/dnnl/dnnl_provider_factory.h"
 #include "onnxruntime/core/providers/nnapi/nnapi_provider_factory.h"
 #include "onnxruntime/core/providers/tvm/tvm_provider_factory.h"
 #include "onnxruntime/core/providers/openvino/openvino_provider_factory.h"
@@ -29,6 +28,11 @@
 #include "onnxruntime/core/providers/dml/dml_provider_factory.h"
 #endif
 
+#ifdef USE_DNNL
+#include "core/providers/dnnl/dnnl_provider_options.h"
+#endif
+
+
 /*
  * Class:     ai_onnxruntime_OrtSession_SessionOptions
  * Method:    setExecutionMode
@@ -431,9 +435,12 @@ JNIEXPORT void JNICALL Java_ai_onnxruntime_OrtSession_00024SessionOptions_addDnn
   (JNIEnv * jniEnv, jobject jobj, jlong apiHandle, jlong handle, jint useArena) {
     (void)jobj;
   #ifdef USE_DNNL
-    checkOrtStatus(jniEnv,(const OrtApi*)apiHandle,OrtSessionOptionsAppendExecutionProvider_Dnnl((OrtSessionOptions*) handle,useArena));
-  #else
-    (void)apiHandle;(void)handle;(void)useArena; // Parameters used when DNNL is defined.
+    OrtDnnlProviderOptions dnnl_options;
+    dnnl_options.use_arena = useArena;  // Follow the user command
+    const OrtApi* api = (OrtApi*)apiHandle;
+    checkOrtStatus(jniEnv, api, api->SessionOptionsAppendExecutionProvider_Dnnl((OrtSessionOptions*)handle, &dnnl_options));
+#else
+    (void)apiHandle; (void)handle; (void)useArena; // Parameters used when DNNL is defined.
     throwOrtException(jniEnv,convertErrorCode(ORT_INVALID_ARGUMENT),"This binary was not compiled with DNNL support.");
   #endif
 }