Group build args (microsoft#24337)

1. Split build.py to two files, because currently the file is over 3000
lines. This PR moves 900 of them to a new file.
2. Put the build args into groups. It makes more explicit that "--x86",
"--arm", "--arm64" and "--arm64ec" args are for Windows only.
3. Remove the "--use_avx512" and "--gen-api-doc" build args, as they are
not referenced anywhere. "--gen-api-doc" was for generating documents
for pytorch frontend.
4. Remove MPI related build flags.
5. Delete tools/ci_build/github/pai/orttraining-ci.yml
6. Remove --use_preinstalled_eigen and --eigen_path. Now we have a more
unified approach for all ORT's dependencies (not just eigen). See
https://onnxruntime.ai/docs/build/dependencies.html for more
information.
7. Windows specific build options won't show up on non-Windows
platforms. The same for macOS.

Author: snnn

cmake/CMakeLists.txt

@@ -180,7 +180,6 @@ option(onnxruntime_ENABLE_TRAINING_OPS "Include training operators but no traini
 option(onnxruntime_ENABLE_TRAINING_E2E_TESTS "Enable training end-to-end tests." OFF)
 option(onnxruntime_ENABLE_CPU_FP16_OPS "Build with advanced instruction sets" ON)
 option(onnxruntime_USE_NCCL "Build with NCCL support" OFF)
-option(onnxruntime_USE_MPI "Build with MPI support" OFF)
 
 # WebAssembly options
 option(onnxruntime_BUILD_WEBASSEMBLY_STATIC_LIB "Enable this option to create WebAssembly static library" OFF)
@@ -1651,29 +1650,6 @@ if (onnxruntime_ENABLE_DLPACK)
 endif()
 
 if (UNIX OR onnxruntime_USE_NCCL)
-  # MPI is INDEPENDENT of NCCL for now. You can build NCLL without MPI and launch multi-GPU with your own launcher.
-  if (onnxruntime_USE_MPI)
-    if (EXISTS "${onnxruntime_MPI_HOME}")
-      set(MPI_HOME "${onnxruntime_MPI_HOME}")
-    elseif (EXISTS "/bert_ort/openmpi")
-      set(MPI_HOME "/bert_ort/openmpi")
-    endif()
-    find_package(MPI)
-
-    if (MPI_CXX_FOUND)
-      message( STATUS "MPI Version: ${MPI_CXX_VERSION}")
-      message( STATUS "MPI (include: ${MPI_CXX_INCLUDE_DIRS}, library: ${MPI_CXX_LIBRARIES})" )
-      mark_as_advanced(MPI_CXX_INCLUDE_DIRS MPI_CXX_LIBRARIES)
-      list(APPEND onnxruntime_EXTERNAL_LIBRARIES ${MPI_CXX_LIBRARIES} ${MPI_CXX_LINK_FLAGS})
-    else ()
-      message(
-            FATAL_ERROR
-            "MPI is not found. Please define onnxruntime_MPI_HOME to specify the path of MPI. Otherwise, NCCL will be disabled."
-            "or you can remove --use_mpi from build args to disable MPI."
-          )
-    endif()
-  endif()
-
   # Find NCCL
   if (onnxruntime_USE_NCCL)
     if (onnxruntime_USE_CUDA)
@@ -1743,13 +1719,9 @@ if (UNIX OR onnxruntime_USE_NCCL)
   endif()
 else()
   set(onnxruntime_USE_NCCL OFF)
-  set(onnxruntime_USE_MPI OFF)
-  message( WARNING "MPI and NCCL are disabled because build is on Windows or USE_NCCL is set to OFF." )
+  message( WARNING "NCCL is disabled because build is on Windows or USE_NCCL is set to OFF." )
 endif()
 
-if (onnxruntime_USE_MPI)
-  add_definitions(-DUSE_MPI=1)
-endif()
 
 # Default version parts for Microsoft.AI.MachineLearning.dll, onnxruntime.dll, onnxruntime_providers_openvino.dll and onnxruntime_providers_shared.dll in non-ADO pipeline local builds
 set(VERSION_MAJOR_PART   0 CACHE STRING "First part of numeric file/product version.")

cmake/onnxruntime_framework.cmake

@@ -77,9 +77,6 @@ if (onnxruntime_ENABLE_TRAINING_OPS)
     onnxruntime_add_include_to_target(onnxruntime_framework Python::Module dlpack::dlpack)
   endif()
 endif()
-if (onnxruntime_USE_MPI)
-  target_include_directories(onnxruntime_framework PUBLIC ${MPI_CXX_INCLUDE_DIRS})
-endif()
 
 if (onnxruntime_ENABLE_ATEN)
   onnxruntime_add_include_to_target(onnxruntime_framework dlpack::dlpack)

cmake/onnxruntime_providers_cpu.cmake

@@ -206,7 +206,7 @@ if (onnxruntime_ENABLE_TRAINING)
     onnxruntime_add_include_to_target(onnxruntime_providers Python::Module)
   endif()
 
-  if (onnxruntime_USE_NCCL OR onnxruntime_USE_MPI)
+  if (onnxruntime_USE_NCCL)
     target_include_directories(onnxruntime_providers PUBLIC ${MPI_CXX_INCLUDE_DIRS})
   endif()
 endif()

cmake/onnxruntime_providers_cuda.cmake

@@ -252,10 +252,6 @@
       target_include_directories(${target} PRIVATE ${ORTTRAINING_ROOT} ${MPI_CXX_INCLUDE_DIRS})
     endif()
 
-    if(onnxruntime_USE_MPI)
-      target_link_libraries(${target} PRIVATE ${MPI_LIBRARIES} ${MPI_CXX_LINK_FLAGS})
-    endif()
-
     if (onnxruntime_USE_NCCL)
       target_include_directories(${target} PRIVATE ${NCCL_INCLUDE_DIRS})
       target_link_libraries(${target} PRIVATE ${NCCL_LIBRARIES})

cmake/onnxruntime_providers_rocm.cmake

@@ -158,9 +158,6 @@
 
   if (onnxruntime_ENABLE_TRAINING)
     target_include_directories(onnxruntime_providers_rocm PRIVATE ${ORTTRAINING_ROOT} ${CMAKE_CURRENT_BINARY_DIR}/amdgpu/orttraining ${MPI_CXX_INCLUDE_DIRS})
-    if(onnxruntime_USE_MPI)
-      target_link_libraries(onnxruntime_providers_rocm PRIVATE ${MPI_LIBRARIES} ${MPI_CXX_LINK_FLAGS})
-    endif()
 
     # RCCL is enabled by default for ROCM builds
     #if (onnxruntime_USE_NCCL)