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

…yguo/maccatalyst-build-ios-package

.github/workflows/labeler.yml

@@ -3,6 +3,9 @@ on:
   issues:
     types: [opened, edited]
 
+permissions:
+  issues: write
+
 jobs:
   triage:
     runs-on: ubuntu-latest

.github/workflows/stale.yml

@@ -13,7 +13,7 @@ jobs:
       issues: write
       pull-requests: write
     steps:
-      - uses: actions/stale@v9.0.0
+      - uses: actions/stale@v8
         with:
           # Comma separated list of labels that can be assigned to issues to exclude them from being marked as stale
           exempt-issue-labels: contributions welcome, feature request, regression

.lintrunner.toml

@@ -132,6 +132,7 @@ exclude_patterns = [
     'onnxruntime/core/flatbuffers/schema/*.fbs.h', # Generated code
     'onnxruntime/core/graph/contrib_ops/quantization_defs.cc',
     'onnxruntime/core/mlas/**', # Contains assembly code
+    'onnxruntime/core/mickey/cutlass_ext/**', # CUTLASS lib recommends NO automatic code formatting
     'winml/lib/Api.Image/shaders/**',  # Contains data chunks
 ]
 command = [

.pipelines/windowsai-steps.yml

@@ -80,11 +80,11 @@ jobs:
 
     # must call vsdevcmd first to add cmake to PATH
     - script: |
-        curl -O -L https://github.com/Kitware/CMake/releases/download/v3.26.3/cmake-3.26.3-windows-x86_64.zip
-        7z x cmake-3.26.3-windows-x86_64.zip
+        curl -O -L https://github.com/Kitware/CMake/releases/download/v3.28.3/cmake-3.28.3-windows-x86_64.zip
+        7z x cmake-3.28.3-windows-x86_64.zip
         set PYTHONHOME=$(Build.BinariesDirectory)\${{ parameters.PythonPackageName }}.3.9.7\tools
         set PYTHONPATH=$(Build.BinariesDirectory)\${{ parameters.PythonPackageName }}.3.9.7\tools
-        $(Build.BinariesDirectory)\${{ parameters.PythonPackageName }}.3.9.7\tools\python.exe "$(Build.SourcesDirectory)\tools\ci_build\build.py" --build_dir $(Build.BinariesDirectory) --parallel --use_binskim_compliant_compile_flags --build_shared_lib --enable_onnx_tests --ms_experimental --use_dml --use_winml --cmake_generator "Visual Studio 17 2022" --update --config RelWithDebInfo --enable_lto --use_telemetry --disable_rtti --enable_wcos $(BuildFlags) --cmake_extra_defines "CMAKE_EXE_LINKER_FLAGS_RELWITHDEBINFO=/PROFILE" "CMAKE_SHARED_LINKER_FLAGS_RELWITHDEBINFO=/PROFILE" CMAKE_SYSTEM_VERSION=10.0.19041.0 --cmake_path $(Build.BinariesDirectory)\cmake-3.26.3-windows-x86_64\bin\cmake.exe --ctest_path $(Build.BinariesDirectory)\cmake-3.26.3-windows-x86_64\bin\ctest.exe
+        $(Build.BinariesDirectory)\${{ parameters.PythonPackageName }}.3.9.7\tools\python.exe "$(Build.SourcesDirectory)\tools\ci_build\build.py" --build_dir $(Build.BinariesDirectory) --parallel --use_binskim_compliant_compile_flags --build_shared_lib --enable_onnx_tests --ms_experimental --use_dml --use_winml --cmake_generator "Visual Studio 17 2022" --update --config RelWithDebInfo --enable_lto --use_telemetry --disable_rtti --enable_wcos --windows_sdk_version "10.0.22621.0" $(BuildFlags) --cmake_extra_defines "CMAKE_EXE_LINKER_FLAGS_RELWITHDEBINFO=/PROFILE" "CMAKE_SHARED_LINKER_FLAGS_RELWITHDEBINFO=/PROFILE"  --cmake_path $(Build.BinariesDirectory)\cmake-3.28.3-windows-x86_64\bin\cmake.exe --ctest_path $(Build.BinariesDirectory)\cmake-3.28.3-windows-x86_64\bin\ctest.exe
       workingDirectory: '$(Build.BinariesDirectory)'
       displayName: 'Generate cmake config'
 

.vscode/settings.json

@@ -21,5 +21,8 @@
     "cpplint.filters": [
         "-build/include_subdir",
         "-runtime/references"
-    ]
+    ],
+    "files.associations": {
+        "span": "cpp"
+    }
 }

CITATION.cff

@@ -3,8 +3,7 @@ title: ONNX Runtime
 message: "Please use this information to cite ONNX Runtime in
   research or other publications."
 authors:
-  - affiliation: Microsoft Corporation
-    given-names: ONNX Runtime developers
+  - name: ONNX Runtime developers
 date-released: 2018-11-29
 url: "https://onnxruntime.ai"
 repository-code: "https://github.com/microsoft/onnxruntime"

cgmanifests/generate_cgmanifest.py

@@ -115,8 +115,8 @@ def normalize_path_separators(path):
 submodule_lines = proc.stdout.splitlines()
 for submodule_line in submodule_lines:
     (absolute_path, url, commit) = submodule_line.split(" ")
-    git_deps[GitDep(commit, url)] = "git submodule at {}".format(
-        normalize_path_separators(os.path.relpath(absolute_path, REPO_DIR))
+    git_deps[GitDep(commit, url)] = (
+        f"git submodule at {normalize_path_separators(os.path.relpath(absolute_path, REPO_DIR))}"
     )
 
 with open(os.path.join(SCRIPT_DIR, "..", "cmake", "deps.txt")) as f:

cgmanifests/generated/cgmanifest.json

@@ -86,7 +86,7 @@
       "component": {
         "type": "git",
         "git": {
-          "commitHash": "6df40a2471737b27271bdd9b900ab5f3aec746c7",
+          "commitHash": "0100f6a5779831fa7a651e4b67ef389a8752bd9b",
           "repositoryUrl": "https://github.com/google/flatbuffers.git"
         },
         "comments": "flatbuffers"
@@ -116,7 +116,7 @@
       "component": {
         "type": "git",
         "git": {
-          "commitHash": "361e8d1cfe0c6c36d30b39f1b61302ece5507320",
+          "commitHash": "344117638c8ff7e239044fd0fa7085839fc03021",
           "repositoryUrl": "https://github.com/google/benchmark.git"
         },
         "comments": "google_benchmark"

cmake/CMakeLists.txt

@@ -118,7 +118,6 @@ option(onnxruntime_GCOV_COVERAGE "Compile with options necessary to run code cov
 option(onnxruntime_DONT_VECTORIZE "Do not vectorize operations in Eigen" OFF)
 
 option(onnxruntime_USE_FULL_PROTOBUF "Link to libprotobuf instead of libprotobuf-lite when this option is ON" OFF)
-option(tensorflow_C_PACKAGE_PATH "Path to tensorflow C package installation dir")
 option(onnxruntime_ENABLE_LANGUAGE_INTEROP_OPS "Enable operator implemented in language other than cpp" OFF)
 option(onnxruntime_DEBUG_NODE_INPUTS_OUTPUTS "Dump debug information about node inputs and outputs when executing the model." OFF)
 cmake_dependent_option(onnxruntime_DEBUG_NODE_INPUTS_OUTPUTS_ENABLE_DUMP_TO_SQLDB "Build dump debug information about node inputs and outputs with support for sql database." OFF "onnxruntime_DEBUG_NODE_INPUTS_OUTPUTS" OFF)
@@ -324,17 +323,29 @@ if (onnxruntime_USE_ROCM)
   endif()
 
   # replicate strategy used by pytorch to get ROCM_VERSION
-  # https://github.com/pytorch/pytorch/blob/8eb21488fdcdb8b0e6fa2e46179b5fa6c42e75af/cmake/public/LoadHIP.cmake#L153-L173
-  file(READ "${onnxruntime_ROCM_HOME}/.info/version-dev" ROCM_VERSION_DEV_RAW)
-  string(REGEX MATCH "^([0-9]+)\.([0-9]+)\.([0-9]+)-.*$" ROCM_VERSION_DEV_MATCH ${ROCM_VERSION_DEV_RAW})
-  if (ROCM_VERSION_DEV_MATCH)
+  # https://github.com/pytorch/pytorch/blob/5c5b71b6eebae76d744261715231093e62f0d090/cmake/public/LoadHIP.cmake
+  # with modification
+  if (EXISTS "${onnxruntime_ROCM_HOME}/.info/version")
+    file(READ "${onnxruntime_ROCM_HOME}/.info/version" ROCM_VERSION_DEV_RAW)
+    string(REGEX MATCH "^([0-9]+)\.([0-9]+)\.([0-9]+)-.*$" ROCM_VERSION_MATCH ${ROCM_VERSION_DEV_RAW})
+  elseif (EXISTS "${onnxruntime_ROCM_HOME}/include/rocm_version.h")
+    file(READ "${onnxruntime_ROCM_HOME}/include/rocm_version.h" ROCM_VERSION_H_RAW)
+    string(REGEX MATCH "\"([0-9]+)\.([0-9]+)\.([0-9]+).*\"" ROCM_VERSION_MATCH ${ROCM_VERSION_H_RAW})
+  elseif (EXISTS "${onnxruntime_ROCM_HOME}/include/rocm-core/rocm_version.h")
+    file(READ "${onnxruntime_ROCM_HOME}/include/rocm-core/rocm_version.h" ROCM_VERSION_H_RAW)
+    string(REGEX MATCH "\"([0-9]+)\.([0-9]+)\.([0-9]+).*\"" ROCM_VERSION_MATCH ${ROCM_VERSION_H_RAW})
+  endif()
+
+  if (ROCM_VERSION_MATCH)
     set(ROCM_VERSION_DEV_MAJOR ${CMAKE_MATCH_1})
     set(ROCM_VERSION_DEV_MINOR ${CMAKE_MATCH_2})
     set(ROCM_VERSION_DEV_PATCH ${CMAKE_MATCH_3})
     set(ROCM_VERSION_DEV "${ROCM_VERSION_DEV_MAJOR}.${ROCM_VERSION_DEV_MINOR}.${ROCM_VERSION_DEV_PATCH}")
     math(EXPR ROCM_VERSION_DEV_INT "(${ROCM_VERSION_DEV_MAJOR}*10000) + (${ROCM_VERSION_DEV_MINOR}*100) + ${ROCM_VERSION_DEV_PATCH}")
+  else()
+    message(FATAL_ERROR "Cannot determine ROCm version string")
   endif()
-  message("\n***** ROCm version from ${onnxruntime_ROCM_HOME}/.info/version-dev ****\n")
+  message("\n***** ROCm version from ${onnxruntime_ROCM_HOME}/.info/version ****\n")
   message("ROCM_VERSION_DEV: ${ROCM_VERSION_DEV}")
   message("ROCM_VERSION_DEV_MAJOR: ${ROCM_VERSION_DEV_MAJOR}")
   message("ROCM_VERSION_DEV_MINOR: ${ROCM_VERSION_DEV_MINOR}")
@@ -715,6 +726,9 @@ if (onnxruntime_USE_CUDA)
     set(onnxruntime_USE_FLASH_ATTENTION OFF)
     set(onnxruntime_USE_MEMORY_EFFICIENT_ATTENTION OFF)
   endif()
+  if (CMAKE_CUDA_COMPILER_VERSION VERSION_LESS 11.4)
+    message( FATAL_ERROR "Failed build due to CUDA compiler version < 11.4")
+  endif()
 else()
   set(onnxruntime_USE_FLASH_ATTENTION OFF)
   set(onnxruntime_USE_MEMORY_EFFICIENT_ATTENTION OFF)
@@ -735,8 +749,8 @@ if (onnxruntime_USE_CUDA)
       list(APPEND ORT_PROVIDER_FLAGS -DUSE_MEMORY_EFFICIENT_ATTENTION=1)
       list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_MEMORY_EFFICIENT_ATTENTION=1)
     endif()
-
 endif()
+
 if (onnxruntime_USE_VITISAI)
     list(APPEND ORT_PROVIDER_FLAGS -DUSE_VITISAI=1)
     list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_VITISAI=1)
@@ -1239,17 +1253,15 @@ if (onnxruntime_USE_TVM)
     $<TARGET_PROPERTY:tvm,INTERFACE_INCLUDE_DIRECTORIES>)
 
   set(onnxruntime_tvm_libs onnxruntime_providers_tvm)
-
-  # needs to link with stdc++fs in Linux
-  if (UNIX)
-    if (NOT APPLE)
-      set(FS_STDLIB stdc++fs)
-    endif()
-  endif()
-  list(APPEND onnxruntime_EXTERNAL_LIBRARIES tvm ${FS_STDLIB})
+  list(APPEND onnxruntime_EXTERNAL_LIBRARIES tvm)
   list(APPEND onnxruntime_EXTERNAL_DEPENDENCIES tvm)
 endif()
 
+# needs to link with stdc++fs in Linux
+if (UNIX AND "${CMAKE_C_COMPILER_ID}" STREQUAL "GNU" AND CMAKE_C_COMPILER_VERSION VERSION_LESS 9)
+  set(FS_STDLIB stdc++fs)
+endif()
+list(APPEND onnxruntime_EXTERNAL_LIBRARIES ${FS_STDLIB})
 
 # onnxruntime-extensions
 if (onnxruntime_USE_EXTENSIONS)
@@ -1259,11 +1271,7 @@ endif()
 #Dependencies end. In the next we'll enable "treat warning as error"
 
 #Adjust warning flags
-if (onnxruntime_USE_CUDA)
-  set_msvc_c_cpp_compiler_warning_level(3)
-else()
-  set_msvc_c_cpp_compiler_warning_level(4)
-endif()
+set_msvc_c_cpp_compiler_warning_level(4)
 
 set(onnxruntime_DELAYLOAD_FLAGS "")
 
@@ -1400,6 +1408,10 @@ endif()
 if (onnxruntime_USE_CUDA)
   set(CMAKE_CUDA_RUNTIME_LIBRARY Shared)
   set(CMAKE_CUDA_STANDARD 17)
+  if(onnxruntime_CUDA_HOME)
+    file(TO_CMAKE_PATH CUDAToolkit_ROOT ${onnxruntime_CUDA_HOME})
+  endif()
+  find_package(CUDAToolkit REQUIRED)
   if(onnxruntime_CUDNN_HOME)
     file(TO_CMAKE_PATH ${onnxruntime_CUDNN_HOME} onnxruntime_CUDNN_HOME)
   endif()
@@ -1600,7 +1612,7 @@ if (UNIX AND onnxruntime_USE_NCCL)
 else()
   set(onnxruntime_USE_NCCL OFF)
   set(onnxruntime_USE_MPI OFF)
-message( WARNING "MPI and NCCL disabled on Win build." )
+  message( WARNING "MPI and NCCL are disabled because build is on Windows or USE_NCCL is set to OFF." )
 endif()
 
 if (onnxruntime_USE_MPI)
@@ -1729,14 +1741,12 @@ if(onnxruntime_BUILD_KERNEL_EXPLORER)
 endif()
 
 # When GDK_PLATFORM is set then WINAPI_FAMILY is defined in gdk_toolchain.cmake (along with other relevant flags/definitions).
-if (WIN32 AND NOT GDK_PLATFORM)
+if (WIN32 AND NOT GDK_PLATFORM AND NOT CMAKE_CROSSCOMPILING)
   if (NOT CMAKE_CXX_STANDARD_LIBRARIES MATCHES kernel32.lib)
     # On onecore, link to the onecore build of the MSVC runtime
     get_filename_component(msvc_path "${CMAKE_C_COMPILER}/../../../.." ABSOLUTE)
     link_directories(BEFORE "${msvc_path}/lib/onecore/${onnxruntime_target_platform}")
-    # The .lib files in the MSVC runtime have a DEFAULITLIB entry for onecore.lib, which in turn links to reverse forwarders.
-    # We ignore that entry and use onecore_apiset.lib instead, since system components must not rely on reverse forwarders.
-    add_link_options("/NODEFAULTLIB:onecore.lib")
+    # The .lib files in the MSVC runtime have a DEFAULITLIB entry for onecore.lib, but it shold not cause any conflict with onecoreuap.lib
   endif()
 endif()
 

cmake/adjust_global_compile_flags.cmake

@@ -101,13 +101,8 @@ if (onnxruntime_MINIMAL_BUILD)
   endif()
 endif()
 
-# Enable stream for all the non-minimal build, except for DML. There's currently a bug
-# in the allocation planner when reusing buffers and more than one streams are used that
-# make it possible (although rarely) to reach a reference count of 0 for a buffer that is
-# still being used. Since DML doesn't benefit from multiple streams, disabling it is the
-# safest option for now.
-# https://github.com/microsoft/onnxruntime/issues/19480
-if (NOT onnxruntime_MINIMAL_BUILD AND NOT onnxruntime_USE_DML)
+# Enable stream for all the non-minimal build
+if (NOT onnxruntime_MINIMAL_BUILD)
   add_compile_definitions(ORT_ENABLE_STREAM)
 endif()
 
@@ -219,7 +214,7 @@ endif()
 
 
 macro(check_nvcc_compiler_flag _FLAG _RESULT)
-    execute_process(COMMAND ${onnxruntime_CUDA_HOME}/bin/nvcc "${_FLAG}" RESULT_VARIABLE NVCC_OUT ERROR_VARIABLE NVCC_ERROR)
+    execute_process(COMMAND ${CUDAToolkit_BIN_DIR}/nvcc "${_FLAG}" RESULT_VARIABLE NVCC_OUT ERROR_VARIABLE NVCC_ERROR)
     message("NVCC_ERROR = ${NVCC_ERROR}")
     message("NVCC_OUT = ${NVCC_OUT}")
     if ("${NVCC_OUT}" MATCHES "0")

cmake/deps.txt

@@ -23,10 +23,10 @@ dlpack;https://github.com/dmlc/dlpack/archive/refs/tags/v0.6.zip;4d565dd2e5b3132
 # Until the 3.4.1 release this is the best option we have.
 # Issue link: https://gitlab.com/libeigen/eigen/-/issues/2744
 eigen;https://gitlab.com/libeigen/eigen/-/archive/e7248b26a1ed53fa030c5c459f7ea095dfd276ac/eigen-e7248b26a1ed53fa030c5c459f7ea095dfd276ac.zip;be8be39fdbc6e60e94fa7870b280707069b5b81a
-flatbuffers;https://github.com/google/flatbuffers/archive/refs/tags/v1.12.0.zip;ba0a75fd12dbef8f6557a74e611b7a3d0c5fe7bf
+flatbuffers;https://github.com/google/flatbuffers/archive/refs/tags/v23.5.26.zip;59422c3b5e573dd192fead2834d25951f1c1670c
 fp16;https://github.com/Maratyszcza/FP16/archive/0a92994d729ff76a58f692d3028ca1b64b145d91.zip;b985f6985a05a1c03ff1bb71190f66d8f98a1494
 fxdiv;https://github.com/Maratyszcza/FXdiv/archive/63058eff77e11aa15bf531df5dd34395ec3017c8.zip;a5658f4036402dbca7cebee32be57fb8149811e1
-google_benchmark;https://github.com/google/benchmark/archive/refs/tags/v1.7.0.zip;e97c368b176e8614e3f1bf13dd9abcf6a7ad9908
+google_benchmark;https://github.com/google/benchmark/archive/refs/tags/v1.8.3.zip;bf9870756ee3f8d2d3b346b24ee3600a41c74d3d
 google_nsync;https://github.com/google/nsync/archive/refs/tags/1.26.0.zip;5e7c00ef6bf5b787386fc040067903ec774e2752
 googletest;https://github.com/google/googletest/archive/530d5c8c84abd2a46f38583ee817743c9b3a42b4.zip;5e3a61db2aa975cfd0f97ba92c818744e7fa7034
 googlexnnpack;https://github.com/google/XNNPACK/archive/0da379fc4808f9601faef392352018c741c0f297.zip;663883491e380b628e0a5b162b5f2658032fae73
@@ -37,8 +37,8 @@ mimalloc;https://github.com/microsoft/mimalloc/archive/refs/tags/v2.1.1.zip;d5ee
 mp11;https://github.com/boostorg/mp11/archive/refs/tags/boost-1.82.0.zip;9bc9e01dffb64d9e0773b2e44d2f22c51aace063
 neural_speed;https://github.com/intel/neural-speed/archive/refs/tags/bestlav0.1.1.zip;65b0f7a0d04f72f0d5a8d48af70f0366f2ab3939
 onnx;https://github.com/onnx/onnx/archive/refs/tags/v1.15.0.zip;54c3f960a0541c5d8d3e60c2933e11f5d3688a11
-#use the commit of supporting all the plugins and TRT 8.6-GA (https://github.com/onnx/onnx-tensorrt/commit/0462dc31ae78f48744b6141ae376df1f96d3f459)
-onnx_tensorrt;https://github.com/onnx/onnx-tensorrt/archive/a43ce67187bab219520fd80f21af8bbd4354bc8c.zip;572535aefef477050f86744dfab1fef840198035
+#use the commit of Final DDS removal. DDS output is now supported by ORT TRT.
+onnx_tensorrt;https://github.com/onnx/onnx-tensorrt/archive/bacfaaa951653cd4e72efe727a543567cb38f7de.zip;26434329612e804164ab7baa6ae629ada56c1b26
 protobuf;https://github.com/protocolbuffers/protobuf/archive/refs/tags/v21.12.zip;7cf2733949036c7d52fda017badcab093fe73bfa
 protoc_win64;https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-win64.zip;b4521f7ada5b260380f94c4bd7f1b7684c76969a
 protoc_win32;https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-win32.zip;3688010318192c46ce73213cdfb6b3e5656da874