[onnxruntime/build] Add CI testing for ORT build with generic interface

cmake/CMakeLists.txt

@@ -835,7 +835,7 @@ endif()
 set(ONNXRUNTIME_PROVIDER_NAMES cpu)
 
 set(ORT_PROVIDER_FLAGS)
-set(ORT_PROVIDER_CMAKE_FLAGS)
+set(ORT_EXTRA_INTERFACE_FLAGS)
 
 if (onnxruntime_USE_CUDA)
   enable_language(CUDA)
@@ -870,74 +870,87 @@ else()
   set(onnxruntime_USE_MEMORY_EFFICIENT_ATTENTION OFF)
 endif()
 
-if (onnxruntime_USE_CUDA OR onnxruntime_USE_CUDA_INTERFACE)
+if (onnxruntime_USE_CUDA)
     list(APPEND ORT_PROVIDER_FLAGS -DUSE_CUDA=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_CUDA=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES cuda)
 
     if (onnxruntime_USE_FLASH_ATTENTION)
       message( STATUS "Enable flash attention for CUDA EP")
       list(APPEND ORT_PROVIDER_FLAGS -DUSE_FLASH_ATTENTION=1)
-      list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_FLASH_ATTENTION=1)
     endif()
 
     if (onnxruntime_USE_LEAN_ATTENTION)
       message( STATUS "Enable lean attention for CUDA EP")
       list(APPEND ORT_PROVIDER_FLAGS -DUSE_LEAN_ATTENTION=1)
-      list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_LEAN_ATTENTION=1)
     endif()
 
     if (onnxruntime_USE_MEMORY_EFFICIENT_ATTENTION)
       message( STATUS "Enable memory efficient attention for CUDA EP")
       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 OR onnxruntime_USE_VITISAI_INTERFACE)
+if (onnxruntime_USE_CUDA_INTERFACE AND (NOT onnxruntime_USE_CUDA))
+    list(APPEND ORT_EXTRA_INTERFACE_FLAGS -DUSE_CUDA=1)
+endif()
+
+if (onnxruntime_USE_VITISAI)
     list(APPEND ORT_PROVIDER_FLAGS -DUSE_VITISAI=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_VITISAI=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES vitisai)
 endif()
+
+if (onnxruntime_USE_VITISAI_INTERFACE AND (NOT onnxruntime_USE_VITISAI))
+    list(APPEND ORT_EXTRA_INTERFACE_FLAGS -DUSE_VITISAI=1)
+endif()
+
 if (onnxruntime_USE_DNNL)
     list(APPEND ORT_PROVIDER_FLAGS -DUSE_DNNL=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES dnnl)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_DNNL=1)
 endif()
-if (onnxruntime_USE_OPENVINO OR onnxruntime_USE_OPENVINO_INTERFACE)
+
+if (onnxruntime_USE_OPENVINO)
     list(APPEND ORT_PROVIDER_FLAGS -DUSE_OPENVINO=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES openvino)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_OPENVINO=1)
 endif()
-if (onnxruntime_USE_TENSORRT OR onnxruntime_USE_TENSORRT_INTERFACE)
+
+if (onnxruntime_USE_OPENVINO_INTERFACE AND (NOT onnxruntime_USE_OPENVINO))
+    list(APPEND ORT_EXTRA_INTERFACE_FLAGS -DUSE_OPENVINO=1)
+endif()
+
+if (onnxruntime_USE_TENSORRT)
     list(APPEND ORT_PROVIDER_FLAGS -DUSE_TENSORRT=1)
     #TODO: remove the following line and change the test code in onnxruntime_shared_lib_test to use the new EP API.
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES tensorrt)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_TENSORRT=1)
 endif()
+
+if (onnxruntime_USE_TENSORRT_INTERFACE AND (NOT onnxruntime_USE_TENSORRT))
+    list(APPEND ORT_INTERFACE_FLAGS -DUSE_TENSORRT=1)
+endif()
+
 if (onnxruntime_USE_RKNPU)
     list(APPEND ORT_PROVIDER_FLAGS -DUSE_RKNPU=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_RKNPU=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES rknpu)
 endif()
 if (onnxruntime_USE_VSINPU)
     list(APPEND ORT_PROVIDER_FLAGS -DUSE_VSINPU=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_VSINPU=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES vsinpu)
 endif()
 if (onnxruntime_USE_NNAPI_BUILTIN)
     list(APPEND ORT_PROVIDER_FLAGS -DUSE_NNAPI=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_NNAPI_BUILTIN=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES nnapi)
 endif()
 if (onnxruntime_USE_JSEP)
     list(APPEND ORT_PROVIDER_FLAGS -DUSE_JSEP=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_JSEP=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES js)
 endif()
 if (onnxruntime_USE_QNN OR onnxruntime_USE_QNN_INTERFACE)
-    list(APPEND ORT_PROVIDER_FLAGS -DUSE_QNN=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_QNN=1)
+
+    if(onnxruntime_USE_QNN)
+        list(APPEND ORT_PROVIDER_FLAGS -DUSE_QNN=1)
+    else()
+        list(APPEND ORT_EXTRA_INTERFACE_FLAGS -DUSE_QNN=1) 
+    endif()
+
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES qnn)
     if (NOT QNN_ARCH_ABI)
       string(TOLOWER ${onnxruntime_target_platform} GEN_PLATFORM)
@@ -963,7 +976,7 @@ if (onnxruntime_USE_QNN OR onnxruntime_USE_QNN_INTERFACE)
       endif()
     endif()
 
-    if ((NOT onnxruntime_USE_QNN_INTERFACE) AND (MSVC OR ${CMAKE_SYSTEM_NAME} STREQUAL "Linux"))
+    if (onnxruntime_USE_QNN AND (MSVC OR ${CMAKE_SYSTEM_NAME} STREQUAL "Linux"))
       file(GLOB QNN_LIB_FILES LIST_DIRECTORIES false "${onnxruntime_QNN_HOME}/lib/${QNN_ARCH_ABI}/libQnn*.so"
 	                                     "${onnxruntime_QNN_HOME}/lib/${QNN_ARCH_ABI}/Qnn*.dll"
                                              "${onnxruntime_QNN_HOME}/lib/${QNN_ARCH_ABI}/libHtpPrepare.so"
@@ -980,59 +993,48 @@ endif()
 if (onnxruntime_USE_SNPE)
     list(APPEND ORT_PROVIDER_FLAGS -DUSE_SNPE=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES snpe)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_SNPE=1)
 endif()
 if (onnxruntime_USE_WINML)
   list(APPEND ORT_PROVIDER_FLAGS  -DUSE_WINML=1)
-  list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_WINML=1)
   list(APPEND ONNXRUNTIME_PROVIDER_NAMES winml)
 endif()
 if (onnxruntime_USE_ACL)
     list(APPEND ORT_PROVIDER_FLAGS  -DUSE_ACL=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_ACL=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES acl)
 endif()
 if (onnxruntime_USE_DML)
     list(APPEND ORT_PROVIDER_FLAGS  -DUSE_DML=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_DML=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES dml)
     if(onnxruntime_ENABLE_NPU_ADAPTER_ENUMERATION)
       list(APPEND ORT_PROVIDER_FLAGS -DENABLE_NPU_ADAPTER_ENUMERATION=1)
     endif()
 endif()
 if (onnxruntime_USE_MIGRAPHX)
     list(APPEND ORT_PROVIDER_FLAGS  -DUSE_MIGRAPHX=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_MIGRAPHX=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES migraphx)
 endif()
 if (onnxruntime_USE_ARMNN)
     list(APPEND ORT_PROVIDER_FLAGS  -DUSE_ARMNN=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_ARMNN=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES armnn)
 endif()
 if (onnxruntime_USE_ROCM)
     list(APPEND ORT_PROVIDER_FLAGS  -DUSE_ROCM=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_ROCM=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES rocm)
 endif()
 if (onnxruntime_USE_COREML)
     list(APPEND ORT_PROVIDER_FLAGS  -DUSE_COREML=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_COREML=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES coreml)
 endif()
 if (onnxruntime_USE_XNNPACK)
   list(APPEND ORT_PROVIDER_FLAGS -DUSE_XNNPACK=1)
-  list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_XNNPACK=1)
   list(APPEND ONNXRUNTIME_PROVIDER_NAMES xnnpack)
 endif()
 if (onnxruntime_USE_WEBNN)
   list(APPEND ORT_PROVIDER_FLAGS -DUSE_WEBNN=1)
-  list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_WEBNN=1)
   list(APPEND ONNXRUNTIME_PROVIDER_NAMES webnn)
 endif()
 if (onnxruntime_USE_WEBGPU)
   list(APPEND ORT_PROVIDER_FLAGS -DUSE_WEBGPU=1)
-  list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_WEBGPU=1)
   list(APPEND ONNXRUNTIME_PROVIDER_NAMES webgpu)
   if (onnxruntime_BUILD_DAWN_MONOLITHIC_LIBRARY)
     list(APPEND ORT_PROVIDER_FLAGS -DBUILD_DAWN_MONOLITHIC_LIBRARY=1)
@@ -1049,12 +1051,10 @@ if (onnxruntime_USE_WEBGPU)
 endif()
 if (onnxruntime_USE_CANN)
     list(APPEND ORT_PROVIDER_FLAGS  -DUSE_CANN=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_CANN=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES cann)
 endif()
 if (onnxruntime_USE_AZURE)
     list(APPEND ORT_PROVIDER_FLAGS  -DUSE_AZURE=1)
-    list(APPEND ORT_PROVIDER_CMAKE_FLAGS -Donnxruntime_USE_AZURE=1)
     list(APPEND ONNXRUNTIME_PROVIDER_NAMES azure)
 endif()
 if (onnxruntime_USE_LOCK_FREE_QUEUE)
@@ -1188,6 +1188,13 @@ function(onnxruntime_set_compile_flags target_name)
     foreach(ORT_FLAG ${ORT_PROVIDER_FLAGS})
       target_compile_definitions(${target_name} PRIVATE ${ORT_FLAG})
     endforeach()
+
+    if("${target_name}" STREQUAL "onnxruntime")
+        foreach(ORT_EXTRA_FLAG ${ORT_EXTRA_INTERFACE_FLAGS})
+          target_compile_definitions(${target_name} PRIVATE ${ORT_EXTRA_FLAG})
+        endforeach()
+    endif()
+
     if (HAS_DEPRECATED_COPY)
       #too many such errors in eigen
       target_compile_options(${target_name} PRIVATE "$<$<COMPILE_LANGUAGE:CUDA>:SHELL:--compiler-options -Wno-deprecated-copy>" "$<$<COMPILE_LANGUAGE:CXX>:-Wno-deprecated-copy>")
@@ -1422,7 +1429,7 @@ if (onnxruntime_ENABLE_TRAINING_APIS)
   )
 endif()
 
-if (onnxruntime_USE_OPENVINO OR onnxruntime_USE_OPENVINO_INTERFACE)
+if (onnxruntime_USE_OPENVINO)
 
   add_definitions(-DUSE_OPENVINO=1)
 
@@ -1435,7 +1442,7 @@ if (onnxruntime_USE_OPENVINO OR onnxruntime_USE_OPENVINO_INTERFACE)
     add_definitions(-DOPENVINO_CONFIG_GPU=1)
   endif()
 
-  if (onnxruntime_USE_OPENVINO_CPU OR onnxruntime_USE_OPENVINO_INTERFACE) # OpenVino CPU interface is default built.
+  if (onnxruntime_USE_OPENVINO_CPU)
     add_definitions(-DOPENVINO_CONFIG_CPU=1)
   endif()
 

tools/ci_build/build.py

@@ -1543,12 +1543,6 @@ def generate_build_tree(
             "-Donnxruntime_USE_FULL_PROTOBUF=ON",
         ]
 
-    # When this flag is enabled, that means we only build ONNXRuntime shared library, expecting some compatible EP
-    # shared lib being build in a seperate process. So we skip the test for now as ONNXRuntime shared lib built under
-    # this flag is not expected to work alone
-    if args.enable_generic_interface:
-        cmake_args += ["-Donnxruntime_BUILD_UNIT_TESTS=OFF"]
-
     if args.enable_lazy_tensor:
         import torch
 
@@ -2669,7 +2663,12 @@ def main():
         # Disable ONNX Runtime's builtin memory checker
         args.disable_memleak_checker = True
 
-    if args.enable_generic_interface:
+    # When this flag is enabled, it is possible ONNXRuntime shared library is build separately, expecting some compatible EP
+    # shared lib being build in a seperate process. So we skip the testing if none of the primary EPs are built with ONNXRuntime
+    # shared lib
+    if args.enable_generic_interface and not (
+        args.use_tensorrt or args.use_openvino or args.use_vitisai or (args.use_qnn and args.use_qnn != "static_lib")
+    ):
         args.test = False
 
     # If there was no explicit argument saying what to do, default

tools/ci_build/github/azure-pipelines/win-qnn-arm64-ci-pipeline.yml

@@ -50,8 +50,13 @@ jobs:
     matrix:
       SHARED_LIB:
         QnnLibKind: 'shared_lib'
+        qnn_build_args: '--use_qnn'
       STATIC_LIB:
         QnnLibKind: 'static_lib'
+        qnn_build_args: '--use_qnn'
+      SHARED_LIB_GENERIC_INTERFACE:
+        QnnLibKind: 'shared_lib'
+        qnn_build_args: '--use_qnn --use_generic_interface'
   steps:
 
   - script: |