Delete ROCM EP (#25181)

1. Delete ROCM EP, because there is no active development and we have
another AMD GPU EP(migraphx) to use.
2. Delete WASM64 build option, because the feature was incomplete.
Likely we will need to reimplement it later. But, we will delete it for
now( I already discussed it with @fs-eire) .
3. Delete the kernel explorer python extension, which was solely used by
the ROCM EP
4. Delete the triton related build options, which wasn't really put into
use.
5. Add a pull request pipeline for Migraphx EP.

The following cmake options are removed:

- onnxruntime_USE_ROCM
- onnxruntime_ENABLE_WEBASSEMBLY_MEMORY64
- onnxruntime_ENABLE_TRITON
- onnxruntime_USE_COMPOSABLE_KERNEL
- onnxruntime_USE_COMPOSABLE_KERNEL_CK_TILE
- onnxruntime_USE_ROCBLAS_EXTENSION_API
- onnxruntime_USE_TRITON_KERNEL
- onnxruntime_BUILD_KERNEL_EXPLORER
- onnxruntime_BUILD_CACHE
- MSVC_Z7_OVERRIDE

Author: snnn

.github/workflows/linux-dnnl.yml

@@ -33,7 +33,8 @@ jobs:
       architecture: x64
       dockerfile_path: tools/ci_build/github/linux/docker/Dockerfile.manylinux2_28_cpu
       docker_image_repo: onnxruntimecpubuildpythonx64
-      extra_build_flags: '--use_binskim_compliant_compile_flags --build_wheel --build_nuget --use_dnnl'
+      extra_build_flags: '--use_binskim_compliant_compile_flags --build_wheel --build_nuget'
       python_path_prefix: 'PATH=/opt/python/cp310-cp310/bin:$PATH'
+      execution_providers: 'dnnl'
     secrets:
       GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/linux_migraphx_ci.yml

@@ -0,0 +1,40 @@
+# This workflow builds and tests the ONNX Runtime for Linux for migraphx EP
+# It leverages a reusable workflow (`reusable_linux_build.yml`) to handle the core build and test logic
+# within Docker containers, ensuring a consistent environment.
+# This file is very similar to linux_ci.yml, but much simpler
+
+
+name: Linux MigraphX CI
+
+on:
+  push:
+    branches: [main, 'rel-*']
+  pull_request:
+    branches: [main, 'rel-*']
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event_name == 'pull_request' && github.ref || github.sha }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read
+  packages: write
+  attestations: write
+  id-token: write
+
+jobs:
+  build-linux-x64-release-migraphx:
+    name: Build Linux x64 Release (migraphx EP)
+    uses: ./.github/workflows/reusable_linux_build.yml
+    with:
+      pool_name: "onnxruntime-github-Ubuntu2204-AMD-CPU"
+      build_config: Release
+      architecture: x64
+      dockerfile_path: tools/ci_build/github/linux/docker/migraphx-ci-pipeline-env.Dockerfile
+      docker_image_repo: onnxruntimetrainingmigraphx-cibuild-rocm
+      extra_build_flags: '--enable_training --cmake_extra_defines CMAKE_HIP_COMPILER=/opt/rocm/llvm/bin/clang++ --rocm_version=6.4 --rocm_home /opt/rocm --nccl_home /opt/rocm --enable_nccl --skip_submodule_sync'
+      run_tests: false
+      execution_providers: 'migraphx'
+    secrets:
+      GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}

cmake/CMakeLists.txt

@@ -131,7 +131,6 @@ option(onnxruntime_ENABLE_INSTRUMENT "Enable Instrument with Event Tracing for W
 option(onnxruntime_USE_TELEMETRY "Build with Telemetry" OFF)
 cmake_dependent_option(onnxruntime_USE_MIMALLOC "Override new/delete and arena allocator with mimalloc" OFF "WIN32;NOT onnxruntime_USE_CUDA;NOT onnxruntime_USE_OPENVINO" OFF)
 option(onnxruntime_USE_CANN "Build with CANN support" OFF)
-option(onnxruntime_USE_ROCM "Build with AMD GPU support" OFF)
 option(onnxruntime_USE_XNNPACK "Build with XNNPACK support. Provides an alternative math library on ARM, WebAssembly and x86." OFF)
 option(onnxruntime_USE_WEBNN "Build with WebNN support. Enable hardware acceleration in web browsers." OFF)
 option(onnxruntime_USE_WEBGPU "Build with WebGPU support. Enable WebGPU via C/C++ interface." OFF)
@@ -193,7 +192,6 @@ option(onnxruntime_WEBASSEMBLY_RUN_TESTS_IN_BROWSER "Enable this option to run t
 option(onnxruntime_ENABLE_WEBASSEMBLY_DEBUG_INFO "Enable this option to turn on DWARF format debug info" OFF)
 option(onnxruntime_ENABLE_WEBASSEMBLY_PROFILING "Enable this option to turn on WebAssembly profiling and preserve function names" OFF)
 option(onnxruntime_ENABLE_WEBASSEMBLY_OUTPUT_OPTIMIZED_MODEL "Enable this option to allow WebAssembly to output optimized model" OFF)
-option(onnxruntime_ENABLE_WEBASSEMBLY_MEMORY64 "Enable this option to allow WebAssembly to use 64bit memory" OFF)
 option(onnxruntime_ENABLE_WEBASSEMBLY_RELAXED_SIMD "Enable WebAssembly Relaxed SIMD" OFF)
 
 # Enable bitcode for iOS
@@ -223,7 +221,6 @@ option(onnxruntime_PREBUILT_PYTORCH_PATH "Path to pytorch installation dir")
 option(onnxruntime_EXTERNAL_TRANSFORMER_SRC_PATH "Path to external transformer src dir")
 
 option(onnxruntime_ENABLE_CUDA_PROFILING "Enable CUDA kernel profiling" OFF)
-option(onnxruntime_ENABLE_ROCM_PROFILING "Enable ROCM kernel profiling" OFF)
 
 option(onnxruntime_ENABLE_CPUINFO "Enable cpuinfo" ON)
 
@@ -232,21 +229,6 @@ option(onnxruntime_ENABLE_ATEN "Enable ATen fallback" OFF)
 
 # dlpack support
 cmake_dependent_option(onnxruntime_ENABLE_DLPACK "Enable dlpack" ON "onnxruntime_ENABLE_TRAINING OR onnxruntime_ENABLE_ATEN OR onnxruntime_ENABLE_PYTHON" OFF)
-
-# Triton support
-option(onnxruntime_ENABLE_TRITON "Enable Triton" OFF)
-
-# composable kernel is managed automatically, unless user want to explicitly disable it, it should not be manually set
-option(onnxruntime_USE_COMPOSABLE_KERNEL "Enable composable kernel for ROCm EP" ON)
-cmake_dependent_option(onnxruntime_USE_COMPOSABLE_KERNEL_CK_TILE "Enable ck_tile for composable kernel" ON "onnxruntime_USE_COMPOSABLE_KERNEL" OFF)
-option(onnxruntime_USE_ROCBLAS_EXTENSION_API "Enable rocblas tuning for ROCm EP" OFF)
-option(onnxruntime_USE_TRITON_KERNEL "Enable triton compiled kernel" OFF)
-option(onnxruntime_BUILD_KERNEL_EXPLORER "Build Kernel Explorer for testing and profiling GPU kernels" OFF)
-
-option(onnxruntime_BUILD_CACHE "onnxruntime build with cache" OFF)
-# https://zeux.io/2010/11/22/z7-everything-old-is-new-again/
-cmake_dependent_option(MSVC_Z7_OVERRIDE "replacing /Zi and /ZI with /Z7 when using MSVC with CCache" ON "onnxruntime_BUILD_CACHE; MSVC" OFF)
-
 option(onnxruntime_USE_AZURE "Build with azure inferencing support" OFF)
 option(onnxruntime_USE_LOCK_FREE_QUEUE "Build with lock-free task queue for threadpool." OFF)
 option(onnxruntime_FORCE_GENERIC_ALGORITHMS "Disable optimized arch-specific algorithms. Use only for testing and debugging generic algorithms." OFF)
@@ -292,176 +274,6 @@ if (onnxruntime_ENABLE_TRAINING_APIS)
   endif()
 endif()
 
-if (onnxruntime_USE_ROCM)
-  if (WIN32)
-    message(FATAL_ERROR "ROCM does not support build in Windows!")
-  endif()
-  if (onnxruntime_USE_CUDA)
-    message(FATAL_ERROR "ROCM does not support build with CUDA!")
-  endif()
-
-  # replicate strategy used by pytorch to get ROCM_VERSION
-  # https://github.com/pytorch/pytorch/blob/5c5b71b6eebae76d744261715231093e62f0d090/cmake/public/LoadHIP.cmake
-  # with modification
-  if (EXISTS "${onnxruntime_ROCM_HOME}/.info/version")
-    message("\n***** ROCm version from ${onnxruntime_ROCM_HOME}/.info/version ****\n")
-    file(READ "${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")
-    message("\n***** ROCm version from ${onnxruntime_ROCM_HOME}/include/rocm_version.h ****\n")
-    file(READ "${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")
-    message("\n***** ROCm version from ${onnxruntime_ROCM_HOME}/include/rocm-core/rocm_version.h ****\n")
-    file(READ "${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}")
-
-    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}")
-    message("ROCM_VERSION_DEV_PATCH: ${ROCM_VERSION_DEV_PATCH}")
-    message("ROCM_VERSION_DEV_INT:   ${ROCM_VERSION_DEV_INT}")
-  else()
-    message(FATAL_ERROR "Cannot determine ROCm version string")
-  endif()
-
-
-  if (NOT CMAKE_HIP_COMPILER)
-    set(CMAKE_HIP_COMPILER "${onnxruntime_ROCM_HOME}/llvm/bin/clang++")
-  endif()
-
-  if (NOT CMAKE_HIP_ARCHITECTURES)
-    if (ROCM_VERSION_DEV VERSION_LESS "6.2")
-      message(FATAL_ERROR "CMAKE_HIP_ARCHITECTURES is not set when ROCm version < 6.2")
-    else()
-      set(CMAKE_HIP_ARCHITECTURES "gfx908;gfx90a;gfx1030;gfx1100;gfx1101;gfx940;gfx941;gfx942;gfx1200;gfx1201")
-    endif()
-  endif()
-
-  file(GLOB rocm_cmake_components ${onnxruntime_ROCM_HOME}/lib/cmake/*)
-  list(APPEND CMAKE_PREFIX_PATH ${rocm_cmake_components})
-  # Force cmake to accept the configured HIP compiler. Because the configured CMAKE_PREFIX_PATH does not work during
-  # enable_language(HIP), we might need to move configuring of CMAKE_PREFIX_PATH to build.py (in the future).
-  set(CMAKE_HIP_COMPILER_FORCED ON)
-
-  enable_language(HIP)
-  # NOTE: Flags -mllvm -amdgpu-early-inline-all=true are critical for gpu kernel code performance. -mllvm passes the
-  # next flag to underlying LLVM instead of clang and -amdgpu-early-inline-all=true is the optimization flag for LLVM.
-  # With CMake's enable_language(HIP), additional flags including the proceeding one are propagated from
-  # hip-lang::device library. But in some weird cases, the hip-lang::device target may not be properly configured, for
-  # example, the CMAKE_PREFIX_PATH might be improperly configured.
-  if(NOT DEFINED _CMAKE_HIP_DEVICE_RUNTIME_TARGET)
-    message(FATAL_ERROR "HIP Language is not properly configured.")
-  endif()
-  add_compile_options("$<$<COMPILE_LANGUAGE:HIP>:SHELL:-x hip>")
-
-  if (NOT onnxruntime_HIPIFY_PERL)
-    find_path(HIPIFY_PERL_PATH
-      NAMES hipify-perl
-      HINTS
-      ${onnxruntime_ROCM_HOME}/bin
-      ${onnxruntime_ROCM_HOME}/hip/bin)
-    if (HIPIFY_PERL_PATH-NOTFOUND)
-      MESSAGE(FATAL_ERROR "hipify-perl not found")
-    endif()
-    MESSAGE("HIPIFY PATH: ${HIPIFY_PERL_PATH}/hipify-perl")
-    set(onnxruntime_HIPIFY_PERL ${HIPIFY_PERL_PATH}/hipify-perl)
-  endif()
-
-  # replicate strategy used by pytorch to get ROCM_VERSION
-  # https://github.com/pytorch/pytorch/blob/1a10751731784942dcbb9c0524c1369a29d45244/cmake/public/LoadHIP.cmake#L45-L109
-  # with modification
-  set(ROCM_INCLUDE_DIRS "${onnxruntime_ROCM_HOME}/include")
-  set(PROJECT_RANDOM_BINARY_DIR "${CMAKE_BINARY_DIR}")
-  set(file "${CMAKE_BINARY_DIR}/detect_rocm_version.cc")
-
-  # Find ROCM version for checks
-  # ROCM 5.0 and later will have header api for version management
-  if(EXISTS ${ROCM_INCLUDE_DIRS}/rocm_version.h)
-    file(WRITE ${file} ""
-      "#include <rocm_version.h>\n"
-      )
-  elseif(EXISTS ${ROCM_INCLUDE_DIRS}/rocm-core/rocm_version.h)
-    file(WRITE ${file} ""
-      "#include <rocm-core/rocm_version.h>\n"
-      )
-  else()
-    message(FATAL_ERROR "********************* rocm_version.h couldnt be found ******************\n")
-  endif()
-
-  file(APPEND ${file} ""
-    "#include <cstdio>\n"
-
-    "#ifndef ROCM_VERSION_PATCH\n"
-    "#define ROCM_VERSION_PATCH 0\n"
-    "#endif\n"
-    "#define STRINGIFYHELPER(x) #x\n"
-    "#define STRINGIFY(x) STRINGIFYHELPER(x)\n"
-    "int main() {\n"
-    "  printf(\"%d.%d.%s\", ROCM_VERSION_MAJOR, ROCM_VERSION_MINOR, STRINGIFY(ROCM_VERSION_PATCH));\n"
-    "  return 0;\n"
-    "}\n"
-    )
-
-  try_run(run_result compile_result ${PROJECT_RANDOM_BINARY_DIR} ${file}
-    CMAKE_FLAGS "-DINCLUDE_DIRECTORIES=${ROCM_INCLUDE_DIRS}"
-    RUN_OUTPUT_VARIABLE rocm_version_from_header
-    COMPILE_OUTPUT_VARIABLE output_var
-    )
-  # We expect the compile to be successful if the include directory exists.
-  if(NOT compile_result)
-    message(FATAL_ERROR "ROCM: Couldn't determine version from header: " ${output_var})
-  endif()
-  message(STATUS "ROCM: Header version is: " ${rocm_version_from_header})
-  set(ROCM_VERSION_DEV_RAW ${rocm_version_from_header})
-
-  string(REGEX MATCH "^([0-9]+)\.([0-9]+)\.([0-9]+).*$" ROCM_VERSION_DEV_MATCH ${ROCM_VERSION_DEV_RAW})
-
-  if (ROCM_VERSION_DEV_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 rocm_version.h ****\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}")
-  message("ROCM_VERSION_DEV_PATCH: ${ROCM_VERSION_DEV_PATCH}")
-  message("ROCM_VERSION_DEV_INT:   ${ROCM_VERSION_DEV_INT}")
-  message("\n***** HIP LANGUAGE CONFIG INFO ****\n")
-  message("CMAKE_HIP_COMPILER:      ${CMAKE_HIP_COMPILER}")
-  message("CMAKE_HIP_ARCHITECTURES: ${CMAKE_HIP_ARCHITECTURES}")
-  message("CMAKE_HIP_FLAGS:         ${CMAKE_HIP_FLAGS}")
-  string(TOUPPER ${CMAKE_BUILD_TYPE} BUILD_TYPE)
-  message("CMAKE_HIP_FLAGS_${BUILD_TYPE}: ${CMAKE_HIP_FLAGS_${BUILD_TYPE}}")
-  add_definitions(-DROCM_VERSION=${ROCM_VERSION_DEV_INT})
-
-  if (onnxruntime_USE_COMPOSABLE_KERNEL AND ROCM_VERSION_DEV VERSION_LESS "5.3")
-    message(WARNING "composable kernel is only supported on ROCm >= 5.3")
-    set(onnxruntime_USE_COMPOSABLE_KERNEL OFF)
-    set(onnxruntime_USE_COMPOSABLE_KERNEL_CK_TILE OFF)
-  endif()
-  if (onnxruntime_USE_COMPOSABLE_KERNEL_CK_TILE AND ROCM_VERSION_DEV VERSION_LESS "6.0")
-    message(WARNING "ck_tile can only be enabled on ROCm >= 6.0 due to compatibility and compilation speed, disable automatically")
-    set(onnxruntime_USE_COMPOSABLE_KERNEL_CK_TILE OFF)
-  endif()
-  if (onnxruntime_USE_COMPOSABLE_KERNEL_CK_TILE AND CMAKE_BUILD_TYPE STREQUAL "Debug")
-    message(WARNING "ck_tile hits compiler error in Debug build, disable automatically")
-    set(onnxruntime_USE_COMPOSABLE_KERNEL_CK_TILE OFF)
-  endif()
-endif()
 
 
 
@@ -1048,10 +860,6 @@ if (onnxruntime_USE_ARMNN)
     list(APPEND ORT_PROVIDER_FLAGS  -DUSE_ARMNN=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES armnn)
 endif()
-if (onnxruntime_USE_ROCM)
-    list(APPEND ORT_PROVIDER_FLAGS  -DUSE_ROCM=1)
-    list(APPEND ONNXRUNTIME_PROVIDER_NAMES rocm)
-endif()
 if (onnxruntime_USE_COREML)
     list(APPEND ORT_PROVIDER_FLAGS  -DUSE_COREML=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES coreml)
@@ -1292,32 +1100,6 @@ function(onnxruntime_set_compile_flags target_name)
           target_compile_options(${target_name} PRIVATE "$<$<COMPILE_LANGUAGE:CXX>:-Wno-strict-aliasing>")
       endif()
     endif()
-    if (onnxruntime_USE_ROCM)
-      # flags are detected with CXX language mode, some flags are not supported with hipclang
-      # because we may mix gcc and hipclang
-      set(ORT_HIP_WARNING_FLAGS ${ORT_WARNING_FLAGS})
-      list(REMOVE_ITEM ORT_HIP_WARNING_FLAGS -Wno-nonnull-compare)
-      # Unsupported by Clang 18 yet.
-      list(REMOVE_ITEM ORT_HIP_WARNING_FLAGS -Wno-dangling-reference)
-
-      list(REMOVE_ITEM ORT_HIP_WARNING_FLAGS -Wno-interference-size)
-      # float16.h:90:12: error: ‘tmp’ is used uninitialized
-      list(APPEND ORT_HIP_WARNING_FLAGS -Wno-uninitialized)
-      list(APPEND ORT_HIP_WARNING_FLAGS -Wno-deprecated-copy)
-
-      # some #pragma unroll will fail, do not treat them as error
-      # #warning must not be treated as error
-      list(APPEND ORT_HIP_WARNING_FLAGS -Wno-error=pass-failed "-Wno-error=#warnings")
-
-      # otherwise error: builtin __has_trivial_assign is deprecated; use __is_trivially_assignable instead
-      if (ROCM_VERSION_DEV VERSION_GREATER_EQUAL "5.4")
-        list(APPEND ORT_HIP_WARNING_FLAGS "-Wno-deprecated-builtins")
-      endif()
-
-      foreach(FLAG ${ORT_HIP_WARNING_FLAGS})
-        target_compile_options(${target_name} PRIVATE "$<$<COMPILE_LANGUAGE:HIP>:SHELL:${FLAG}>")
-      endforeach()
-    endif()
 endfunction()
 
 function(onnxruntime_set_source_file_properties target_name)
@@ -1346,14 +1128,6 @@ function(onnxruntime_configure_target target_name)
     set_target_properties(${target_name} PROPERTIES INTERPROCEDURAL_OPTIMIZATION_MINSIZEREL TRUE)
   endif()
 
-  if (onnxruntime_BUILD_KERNEL_EXPLORER)
-    get_target_property(target_type ${target_name} TYPE)
-    if (target_type STREQUAL "MODULE_LIBRARY" OR target_type STREQUAL "SHARED_LIBRARY")
-      set_property(TARGET ${target_name}
-        APPEND_STRING PROPERTY LINK_FLAGS " -Xlinker --version-script=${ONNXRUNTIME_ROOT}/python/tools/kernel_explorer/version_script.lds ")
-    endif()
-  endif()
-
   # Keep BinSkim happy
   if(MSVC AND NOT onnxruntime_target_platform MATCHES "ARM")
     target_link_options(${target_name} PRIVATE "/CETCOMPAT")
@@ -1652,10 +1426,6 @@ if (onnxruntime_ENABLE_CUDA_PROFILING)
   add_compile_definitions(ENABLE_CUDA_PROFILING)
 endif()
 
-if (onnxruntime_ENABLE_ROCM_PROFILING)
-  add_compile_definitions(ENABLE_ROCM_PROFILING)
-endif()
-
 if (onnxruntime_ENABLE_TRAINING)
   add_compile_definitions(ENABLE_TRAINING_CORE)
   add_compile_definitions(ENABLE_STRIDED_TENSORS)
@@ -1678,7 +1448,7 @@ if (UNIX OR onnxruntime_USE_NCCL)
   if (onnxruntime_USE_NCCL)
     if (onnxruntime_USE_CUDA)
       set(NCCL_LIBNAME "nccl")
-    elseif (onnxruntime_USE_ROCM OR onnxruntime_USE_MIGRAPHX)
+    elseif (onnxruntime_USE_MIGRAPHX)
       set(NCCL_LIBNAME "rccl")
     endif()
     find_path(NCCL_INCLUDE_DIR
@@ -1871,11 +1641,6 @@ if (CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
   endif()
 endif()
 
-if(onnxruntime_BUILD_KERNEL_EXPLORER)
-  message(STATUS "Kernel Explorer Build is enabled")
-  list(APPEND ONNXRUNTIME_CMAKE_FILES onnxruntime_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 AND NOT CMAKE_CROSSCOMPILING)
   if (NOT CMAKE_CXX_STANDARD_LIBRARIES MATCHES kernel32.lib)

cmake/adjust_global_compile_flags.cmake

@@ -50,11 +50,6 @@ if (CMAKE_SYSTEM_NAME STREQUAL "Emscripten")
     string(APPEND CMAKE_CXX_FLAGS " -s DISABLE_EXCEPTION_CATCHING=0")
   endif()
 
-  if (onnxruntime_ENABLE_WEBASSEMBLY_MEMORY64)
-    string(APPEND CMAKE_C_FLAGS " -DORT_WASM64")
-    string(APPEND CMAKE_CXX_FLAGS " -DORT_WASM64")
-  endif()
-
   # Build WebAssembly with multi-threads support.
   if (onnxruntime_ENABLE_WEBASSEMBLY_THREADS)
     string(APPEND CMAKE_C_FLAGS " -pthread -Wno-pthreads-mem-growth")

cmake/deps.txt

@@ -53,7 +53,6 @@ safeint;https://github.com/dcleblanc/SafeInt/archive/refs/tags/3.0.28.zip;23f252
 tensorboard;https://github.com/tensorflow/tensorboard/archive/373eb09e4c5d2b3cc2493f0949dc4be6b6a45e81.zip;67b833913605a4f3f499894ab11528a702c2b381
 cutlass;https://github.com/NVIDIA/cutlass/archive/refs/tags/v3.9.2.zip;b7f8dc4a879765127ce31dfeabd31c556c80ec79
 extensions;https://github.com/microsoft/onnxruntime-extensions/archive/c24b7bab0c12f53da76d0c31b03b9f0f8ec8f3b4.zip;239063aee4946a9af147b473a4c3da78ba7413b4
-composable_kernel;https://github.com/ROCmSoftwarePlatform/composable_kernel/archive/204da9c522cebec5220bba52cd3542ebcaf99e7a.zip;1827348efd47831c13074245274d41b7cae8a557
 directx_headers;https://github.com/microsoft/DirectX-Headers/archive/refs/tags/v1.613.1.zip;47653509a3371eabb156360f42faf582f314bf2e
 cudnn_frontend;https://github.com/NVIDIA/cudnn-frontend/archive/refs/tags/v1.12.0.zip;7e733cfdc410d777b76122d64232499205589a96
 dawn;https://github.com/google/dawn/archive/9733be39e18186961d503e064874afe3e9ceb8d1.zip;2a4017c32892b90d072a9102eba90ae691fae36d