Optimize BinaryElementWise and BiasGeluGrad kernels for ROCm

.gitmodules

@@ -56,9 +56,6 @@
 [submodule "cmake/external/mp11"]
 	path = cmake/external/mp11
 	url = https://github.com/boostorg/mp11.git
-[submodule "cmake/external/coremltools"]
-	path = cmake/external/coremltools
-	url = https://github.com/apple/coremltools.git
 [submodule "cmake/external/dlpack"]
 	path = cmake/external/dlpack
 	url = https://github.com/dmlc/dlpack.git

CODEOWNERS

@@ -12,3 +12,8 @@ samples/python/training/** @thiagocrepaldi @tlh20 @liqunfu @baijumeswani @Sherlo
 
 # Mobile
 /onnxruntime/test/testdata/kernel_def_hashes/ @skottmckay @gwang-msft @YUNQIUGUO @edgchen1
+
+# Contrib Ops
+onnxruntime/core/graph/contrib_ops/nhwc_schema_defs.cc @zhanghuanrong @chenfucn @yufenglee @yihonglyu @snnn
+onnxruntime/core/graph/contrib_ops/nchwc_schema_defs.cc @zhanghuanrong @chenfucn @yufenglee @yihonglyu @snnn
+onnxruntime/core/graph/contrib_ops/quantization_defs.* @zhanghuanrong @chenfucn @yufenglee @yihonglyu @snnn

cgmanifests/generated/cgmanifest.json

@@ -110,16 +110,6 @@
         "comments": "git submodule at cmake/external/SafeInt/safeint"
       }
     },
-    {
-      "component": {
-        "type": "git",
-        "git": {
-          "commitHash": "523d5e03d86c26267ee6bdf17dd20f6ce6bdadd7",
-          "repositoryUrl": "https://github.com/apple/coremltools.git"
-        },
-        "comments": "git submodule at cmake/external/coremltools"
-      }
-    },
     {
       "component": {
         "type": "git",
@@ -174,7 +164,7 @@
       "component": {
         "type": "git",
         "git": {
-          "commitHash": "a3d65c80d32c3e584b7aab41d516a0043b2a5e84",
+          "commitHash": "3acac70a551c321574732e5bfd67930244bb7151",
           "repositoryUrl": "https://github.com/emscripten-core/emsdk.git"
         },
         "comments": "git submodule at cmake/external/emsdk"
@@ -215,7 +205,7 @@
         "type": "git",
         "git": {
           "commitHash": "db78ac1d7716f56fc9f1b030b715f872f93964e4",
-          "repositoryUrl": "https://github.com/nlohmann/json"
+          "repositoryUrl": "https://github.com/nlohmann/json.git"
         },
         "comments": "git submodule at cmake/external/json"
       }
@@ -234,7 +224,7 @@
       "component": {
         "type": "git",
         "git": {
-          "commitHash": "2d54553b7a78c7c35620b827e7e5ab2228ecb495",
+          "commitHash": "f412df7a2b64421e1f1d61fde6055a6ea288e8f5",
           "repositoryUrl": "https://github.com/microsoft/mimalloc.git"
         },
         "comments": "git submodule at cmake/external/mimalloc"
@@ -255,7 +245,7 @@
         "type": "git",
         "git": {
           "commitHash": "436617053d0f39a1019a371c3a9aa599b3cb2cea",
-          "repositoryUrl": "https://github.com/google/nsync"
+          "repositoryUrl": "https://github.com/google/nsync.git"
         },
         "comments": "git submodule at cmake/external/nsync"
       }
@@ -265,7 +255,7 @@
         "type": "git",
         "git": {
           "commitHash": "be76ca7148396176784ba8733133b9fb1186ea0d",
-          "repositoryUrl": "https://github.com/onnx/onnx"
+          "repositoryUrl": "https://github.com/onnx/onnx.git"
         },
         "comments": "git submodule at cmake/external/onnx"
       }
@@ -294,7 +284,7 @@
       "component": {
         "type": "git",
         "git": {
-          "commitHash": "1f416bb462689f3ef9e3f1057a113d9c6aba6972",
+          "commitHash": "e9456d57605c883cdf985e634ab483e2c1500bb1",
           "repositoryUrl": "https://github.com/onnx/onnx-tensorrt.git"
         },
         "comments": "git submodule at cmake/external/onnx-tensorrt"
@@ -304,7 +294,7 @@
       "component": {
         "type": "git",
         "git": {
-          "commitHash": "553df22c67bee5f0fe6599cff60f1afc6748c635",
+          "commitHash": "994c6181247d7b419b28889fc57d5817e2089419",
           "repositoryUrl": "https://github.com/onnx/onnx.git"
         },
         "comments": "git submodule at cmake/external/onnx-tensorrt/third_party/onnx"
@@ -324,22 +314,12 @@
       "component": {
         "type": "git",
         "git": {
-          "commitHash": "09f082940113661256310e3f4811aa7261a9fa05",
+          "commitHash": "59a2ac2745d8a57ac94c6accced73620d59fb844",
           "repositoryUrl": "https://github.com/pybind/pybind11.git"
         },
         "comments": "git submodule at cmake/external/onnx-tensorrt/third_party/onnx/third_party/pybind11"
       }
     },
-    {
-      "component": {
-        "type": "git",
-        "git": {
-          "commitHash": "6a00cbc4a9b8e68b71caf7f774b3f9c753ae84d5",
-          "repositoryUrl": "https://github.com/wjakob/clang-cindex-python3"
-        },
-        "comments": "git submodule at cmake/external/onnx-tensorrt/third_party/onnx/third_party/pybind11/tools/clang"
-      }
-    },
     {
       "component": {
         "type": "git",
@@ -410,62 +390,12 @@
         "comments": "git submodule at cmake/external/tensorboard"
       }
     },
-    {
-      "component": {
-        "type": "git",
-        "git": {
-          "commitHash": "9ec2b92d180dff8877e402018b97baa574031b8b",
-          "repositoryUrl": "https://github.com/microsoft/onnxruntime-tvm.git"
-        },
-        "comments": "git submodule at cmake/external/tvm"
-      }
-    },
-    {
-      "component": {
-        "type": "git",
-        "git": {
-          "commitHash": "b257a9221ee1e5180d994b3488ddcc259b0ac157",
-          "repositoryUrl": "https://github.com/dmlc/HalideIR"
-        },
-        "comments": "git submodule at cmake/external/tvm/3rdparty/HalideIR"
-      }
-    },
-    {
-      "component": {
-        "type": "git",
-        "git": {
-          "commitHash": "5c792cef3aee54ad8b7000111c9dc1797f327b59",
-          "repositoryUrl": "https://github.com/dmlc/dlpack"
-        },
-        "comments": "git submodule at cmake/external/tvm/3rdparty/dlpack"
-      }
-    },
-    {
-      "component": {
-        "type": "git",
-        "git": {
-          "commitHash": "d07fb7a443b5db8a89d65a15a024af6a425615a5",
-          "repositoryUrl": "https://github.com/dmlc/dmlc-core"
-        },
-        "comments": "git submodule at cmake/external/tvm/3rdparty/dmlc-core"
-      }
-    },
-    {
-      "component": {
-        "type": "git",
-        "git": {
-          "commitHash": "cabe04d6d6b05356fa8f9741704924788f0dd762",
-          "repositoryUrl": "https://github.com/agauniyal/rang"
-        },
-        "comments": "git submodule at cmake/external/tvm/3rdparty/rang"
-      }
-    },
     {
       "component": {
         "type": "git",
         "git": {
           "commitHash": "e8c599bca6c56c44b6730ad93f6abbc9ecd60fc1",
-          "repositoryUrl": "https://github.com/microsoft/wil"
+          "repositoryUrl": "https://github.com/microsoft/wil.git"
         },
         "comments": "git submodule at cmake/external/wil"
       }

cmake/CMakeLists.txt

@@ -120,8 +120,8 @@ option(onnxruntime_EXTENDED_MINIMAL_BUILD "onnxruntime_MINIMAL_BUILD with suppor
 option(onnxruntime_MINIMAL_BUILD_CUSTOM_OPS "Add custom operator kernels support to a minimal build." OFF)
 option(onnxruntime_REDUCED_OPS_BUILD "Reduced set of kernels are registered in build via modification of the kernel registration source files." OFF)
 option(onnxruntime_DISABLE_EXTERNAL_INITIALIZERS "Don't allow models to load external data" OFF)
-cmake_dependent_option(onnxruntime_ENABLE_RUNTIME_OPTIMIZATION_REPLAY_IN_MINIMAL_BUILD
-                       "Enable runtime graph optimization replay for ORT format models in an extended minimal build."
+cmake_dependent_option(onnxruntime_ENABLE_RUNTIME_OPTIMIZATION_IN_MINIMAL_BUILD
+                       "Enable runtime graph optimizations for ORT format models in an extended minimal build."
                        OFF "onnxruntime_EXTENDED_MINIMAL_BUILD" OFF)
 
 #A special option just for debugging and sanitize check. Please do not enable in option in retail builds.
@@ -347,8 +347,8 @@ if (onnxruntime_MINIMAL_BUILD)
     # enable EPs that compile kernels at runtime
     add_compile_definitions(ORT_EXTENDED_MINIMAL_BUILD)
 
-    if (onnxruntime_ENABLE_RUNTIME_OPTIMIZATION_REPLAY_IN_MINIMAL_BUILD)
-      add_compile_definitions(ORT_ENABLE_RUNTIME_OPTIMIZATION_REPLAY_IN_MINIMAL_BUILD)
+    if (onnxruntime_ENABLE_RUNTIME_OPTIMIZATION_IN_MINIMAL_BUILD)
+      add_compile_definitions(ORT_ENABLE_RUNTIME_OPTIMIZATION_IN_MINIMAL_BUILD)
     endif()
   endif()
 
@@ -491,6 +491,14 @@ if (onnxruntime_CROSS_COMPILING)
   endif()
 endif()
 
+if (CMAKE_CXX_COMPILER_VERSION VERSION_LESS 9.0)
+  check_cxx_compiler_flag(-Wno-error HAS_NOERROR)
+  if (HAS_NOERROR)
+    string(APPEND CMAKE_CXX_FLAGS " -Wno-error=attributes")
+    string(APPEND CMAKE_C_FLAGS " -Wno-error=attributes")
+  endif()
+endif()
+
 # Mark symbols to be invisible, for macOS/iOS target only
 # Due to many dependencies have different symbol visibility settings, set global compile flags here.
 if (${CMAKE_SYSTEM_NAME} MATCHES "Darwin|iOS")

cmake/external/onnx_minimal.cmake

@@ -31,10 +31,10 @@ else()
 
   if(HAS_UNUSED_BUT_SET_VARIABLE)
     target_compile_options(onnx_proto PRIVATE "-Wno-unused-but-set-variable")
-  endif()   
+  endif()
 endif()
 
-# For reference, this would be the full ONNX source include. We only need data_type_utils.* in this build.
+# For reference, this would be the full ONNX source include. We only need data_type_utils in this build.
 # file(GLOB_RECURSE onnx_src CONFIGURE_DEPENDS
 #     "${ONNX_SOURCE_ROOT}/onnx/*.h"
 #     "${ONNX_SOURCE_ROOT}/onnx/*.cc"
@@ -46,10 +46,11 @@ endif()
 #     "${ONNX_SOURCE_ROOT}/onnx/test/*"
 #     "${ONNX_SOURCE_ROOT}/onnx/cpp2py_export.cc"
 # )
-# list(REMOVE_ITEM onnx_src ${onnx_exclude_src})  
-file(GLOB onnx_src CONFIGURE_DEPENDS
-"${ONNX_SOURCE_ROOT}/onnx/common/common.h"
-"${ONNX_SOURCE_ROOT}/onnx/defs/data_type_utils.*"
+# list(REMOVE_ITEM onnx_src ${onnx_exclude_src})
+set(onnx_src
+  "${ONNX_SOURCE_ROOT}/onnx/common/common.h"
+  "${ONNX_SOURCE_ROOT}/onnx/defs/data_type_utils.h"
+  "${ONNX_SOURCE_ROOT}/onnx/defs/data_type_utils.cc"
 )
 
 add_library(onnx ${onnx_src})
@@ -75,7 +76,7 @@ if (WIN32)
           /EHsc   # exception handling - C++ may throw, extern "C" will not
       )
     endif()
-    
+
     target_compile_options(onnx_proto PRIVATE
         /wd4244 # 'argument' conversion from 'google::protobuf::int64' to 'int', possible loss of data
     )

cmake/onnxruntime.cmake

@@ -7,7 +7,7 @@ if(UNIX)
     set(OUTPUT_STYLE xcode)
   else()
     set(OUTPUT_STYLE gcc)
-  endif()  
+  endif()
 else()
   set(SYMBOL_FILE ${CMAKE_CURRENT_BINARY_DIR}/onnxruntime_dll.def)
   set(OUTPUT_STYLE vc)
@@ -157,6 +157,8 @@ if(CMAKE_SYSTEM_NAME STREQUAL "Android" AND onnxruntime_BUILD_JAVA)
   endforeach()
 endif()
 
+# This list is a reversed topological ordering of library dependencies.
+# Earlier entries may depend on later ones. Later ones should not depend on earlier ones.
 set(onnxruntime_INTERNAL_LIBRARIES
   onnxruntime_session
   ${onnxruntime_libs}
@@ -174,10 +176,10 @@ set(onnxruntime_INTERNAL_LIBRARIES
   ${onnxruntime_winml}
   onnxruntime_optimizer
   onnxruntime_providers
-  onnxruntime_util
   ${onnxruntime_tvm_libs}
   onnxruntime_framework
   onnxruntime_graph
+  onnxruntime_util
   ${ONNXRUNTIME_MLAS_LIBS}
   onnxruntime_common
   onnxruntime_flatbuffers

cmake/onnxruntime_framework.cmake

@@ -53,7 +53,7 @@ if (onnxruntime_ENABLE_TRAINING OR onnxruntime_ENABLE_TRAINING_OPS)
     onnxruntime_add_include_to_target(onnxruntime_framework Python::Module)
     target_include_directories(onnxruntime_framework PRIVATE ${PROJECT_SOURCE_DIR}/external/dlpack/include)
   endif()
-  if (onnxruntime_USE_NCCL OR onnxruntime_USE_MPI)  
+  if (onnxruntime_USE_NCCL OR onnxruntime_USE_MPI)
     target_include_directories(onnxruntime_framework PUBLIC ${MPI_CXX_INCLUDE_DIRS})
   endif()
 endif()
@@ -95,4 +95,8 @@ if (onnxruntime_DEBUG_NODE_INPUTS_OUTPUTS_ENABLE_DUMP_TO_SQLDB)
   target_compile_definitions(onnxruntime_framework PRIVATE DEBUG_NODE_INPUTS_OUTPUTS_ENABLE_DUMP_TO_SQLDB)
 endif()
 
+if (WIN32)
+  target_compile_definitions(onnxruntime_framework PRIVATE _SCL_SECURE_NO_WARNINGS)
+endif()
+
 install(DIRECTORY ${PROJECT_SOURCE_DIR}/../include/onnxruntime/core/framework  DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/onnxruntime/core)

cmake/onnxruntime_graph.cmake

@@ -17,6 +17,7 @@ if (onnxruntime_MINIMAL_BUILD)
     "${ONNXRUNTIME_ROOT}/core/graph/schema_registry.cc"
     "${ONNXRUNTIME_ROOT}/core/graph/contrib_ops/*defs.h"
     "${ONNXRUNTIME_ROOT}/core/graph/contrib_ops/*defs.cc"
+    "${ONNXRUNTIME_ROOT}/core/graph/contrib_ops/onnx_deprecated_operators.cc"
     "${ONNXRUNTIME_ROOT}/core/graph/contrib_ops/onnx_function_util.h"
     "${ONNXRUNTIME_ROOT}/core/graph/contrib_ops/onnx_function_util.cc"
   )

cmake/onnxruntime_optimizer.cmake

@@ -21,12 +21,14 @@ if (onnxruntime_MINIMAL_BUILD)
       "${ONNXRUNTIME_ROOT}/core/optimizer/qdq_transformer/qdq_util.cc"
       "${ONNXRUNTIME_ROOT}/core/optimizer/initializer.h"
       "${ONNXRUNTIME_ROOT}/core/optimizer/initializer.cc"
+      "${ONNXRUNTIME_ROOT}/core/optimizer/selectors_actions/helpers.h"
+      "${ONNXRUNTIME_ROOT}/core/optimizer/selectors_actions/helpers.cc"
       "${ONNXRUNTIME_ROOT}/core/optimizer/utils.h"
       "${ONNXRUNTIME_ROOT}/core/optimizer/utils.cc"
     )
   endif()
 
-  if (onnxruntime_ENABLE_RUNTIME_OPTIMIZATION_REPLAY_IN_MINIMAL_BUILD)
+  if (onnxruntime_ENABLE_RUNTIME_OPTIMIZATION_IN_MINIMAL_BUILD)
     list(APPEND onnxruntime_optimizer_src_patterns
       "${ONNXRUNTIME_INCLUDE_DIR}/core/optimizer/graph_transformer_utils.h"
       "${ONNXRUNTIME_ROOT}/core/optimizer/graph_transformer_utils.cc"
@@ -36,6 +38,9 @@ if (onnxruntime_MINIMAL_BUILD)
       "${ONNXRUNTIME_ROOT}/core/optimizer/qdq_transformer/selectors_actions/*.cc"
       "${ONNXRUNTIME_ROOT}/core/optimizer/selectors_actions/*.h"
       "${ONNXRUNTIME_ROOT}/core/optimizer/selectors_actions/*.cc"
+
+      "${ONNXRUNTIME_ROOT}/core/optimizer/transpose_optimizer/*.h"
+      "${ONNXRUNTIME_ROOT}/core/optimizer/transpose_optimizer/*.cc"
     )
   endif()
 else()

cmake/onnxruntime_providers.cmake

@@ -784,7 +784,7 @@ if (onnxruntime_USE_COREML)
 
   # Compile CoreML proto definition to ${CMAKE_CURRENT_BINARY_DIR}/coreml
   if (CMAKE_SYSTEM_NAME STREQUAL "Darwin" OR CMAKE_SYSTEM_NAME STREQUAL "iOS")
-    set(COREML_PROTO_ROOT ${PROJECT_SOURCE_DIR}/external/coremltools/mlmodel/format)
+    set(COREML_PROTO_ROOT ${PROJECT_SOURCE_DIR}/../onnxruntime/core/providers/coreml/mlmodel_format)
     file(GLOB coreml_proto_srcs
       "${COREML_PROTO_ROOT}/*.proto"
     )

cmake/onnxruntime_python.cmake

@@ -311,6 +311,9 @@ if (onnxruntime_ENABLE_TRAINING)
   file(GLOB onnxruntime_python_ortmodule_torch_cpp_ext_fused_ops_srcs CONFIGURE_DEPENDS
     "${ORTTRAINING_SOURCE_DIR}/python/training/ortmodule/torch_cpp_extensions/cuda/fused_ops/*"
   )
+  file(GLOB onnxruntime_python_utils_data_srcs CONFIGURE_DEPENDS
+  "${ORTTRAINING_SOURCE_DIR}/python/training/utils/data/*"
+  )
 else()
   file(GLOB onnxruntime_python_capi_training_srcs CONFIGURE_DEPENDS
     "${ONNXRUNTIME_ROOT}/python/training/*.py"
@@ -541,6 +544,7 @@ if (onnxruntime_ENABLE_TRAINING)
     COMMAND ${CMAKE_COMMAND} -E make_directory $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/ortmodule/torch_cpp_extensions/cpu/torch_interop_utils
     COMMAND ${CMAKE_COMMAND} -E make_directory $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/ortmodule/torch_cpp_extensions/cuda/torch_gpu_allocator
     COMMAND ${CMAKE_COMMAND} -E make_directory $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/ortmodule/torch_cpp_extensions/cuda/fused_ops
+    COMMAND ${CMAKE_COMMAND} -E make_directory $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/utils/data/
     COMMAND ${CMAKE_COMMAND} -E copy
         ${onnxruntime_python_capi_training_srcs}
         $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/capi/training/
@@ -580,6 +584,9 @@ if (onnxruntime_ENABLE_TRAINING)
     COMMAND ${CMAKE_COMMAND} -E copy
         ${onnxruntime_python_ortmodule_torch_cpp_ext_fused_ops_srcs}
         $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/ortmodule/torch_cpp_extensions/cuda/fused_ops/
+    COMMAND ${CMAKE_COMMAND} -E copy
+        ${onnxruntime_python_utils_data_srcs}
+        $<TARGET_FILE_DIR:${build_output_target}>/onnxruntime/training/utils/data/
   )
 endif()