[BUILD] CUDA_QUANT_PREPROCESS off by default and Adjust CI

.github/workflows/linux_cuda_ci.yml

@@ -29,7 +29,21 @@ jobs:
       dockerfile_path: tools/ci_build/github/linux/docker/Dockerfile.manylinux2_28_cuda
       docker_build_args: '--build-arg BASEIMAGE=onnxruntimebuildcache.azurecr.io/internal/azureml/onnxruntime/build/cuda13_x64_almalinux8_gcc14:20251107.1'
       docker_image_repo: onnxruntimecuda13manylinuxbuild
-      extra_build_flags: '--use_binskim_compliant_compile_flags --build_wheel --parallel --nvcc_threads 4 --flash_nvcc_threads 4 --cuda_version=13.0 --cuda_home=/usr/local/cuda-13.0 --cudnn_home=/usr/local/cuda-13.0 --enable_cuda_profiling --build_java --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=86 onnxruntime_BUILD_UNIT_TESTS=ON onnxruntime_ENABLE_CUDA_EP_INTERNAL_TESTS=ON'
+      extra_build_flags: >-
+        --use_binskim_compliant_compile_flags
+        --build_wheel
+        --parallel
+        --build_java
+        --nvcc_threads 4
+        --flash_nvcc_threads 4
+        --cuda_version=13.0
+        --cuda_home=/usr/local/cuda-13.0
+        --cudnn_home=/usr/local/cuda-13.0
+        --enable_cuda_profiling
+        --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=86
+        --cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=ON
+        --cmake_extra_defines onnxruntime_ENABLE_CUDA_EP_INTERNAL_TESTS=ON
+        --cmake_extra_defines onnxruntime_BUILD_CUDA_QUANT_PREPROCESS=ON
       python_path_prefix: 'PATH=/opt/python/cp310-cp310/bin:$PATH'
       run_tests: false            # <<< Do not run tests in this job
       upload_build_output: true   # <<< Upload the build/Release directory
@@ -111,5 +125,15 @@ jobs:
           build_config: Release
           mode: 'test' # Set mode to test
           execution_providers: 'cuda'
-          extra_build_flags: '--use_binskim_compliant_compile_flags --cuda_version=13.0 --cuda_home=/usr/local/cuda-13.0 --cudnn_home=/usr/local/cuda-13.0 --enable_cuda_profiling --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=86 onnxruntime_BUILD_UNIT_TESTS=ON onnxruntime_ENABLE_CUDA_EP_INTERNAL_TESTS=ON'
+          extra_build_flags: >-
+            --use_binskim_compliant_compile_flags
+            --cuda_version=13.0
+            --cuda_home=/usr/local/cuda-13.0
+            --cudnn_home=/usr/local/cuda-13.0
+            --enable_cuda_profiling
+            --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=86
+            --cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=ON
+            --cmake_extra_defines onnxruntime_QUICK_BUILD=ON
+            --cmake_extra_defines onnxruntime_ENABLE_CUDA_EP_INTERNAL_TESTS=ON
+            --cmake_extra_defines onnxruntime_BUILD_CUDA_QUANT_PREPROCESS=ON
           python_path_prefix: 'PATH=/opt/python/cp310-cp310/bin:$PATH'

.github/workflows/linux_cuda_no_cudnn.yml

@@ -32,7 +32,19 @@ jobs:
       dockerfile_path: tools/ci_build/github/linux/docker/Dockerfile.manylinux2_28_cuda
       docker_build_args: '--build-arg BASEIMAGE=onnxruntimebuildcache.azurecr.io/internal/azureml/onnxruntime/build/cuda13_x64_almalinux8_gcc14:20251107.1'
       docker_image_repo: onnxruntimecuda13manylinuxbuild
-      extra_build_flags: '--use_binskim_compliant_compile_flags --build_wheel --parallel --nvcc_threads 4 --flash_nvcc_threads 4 --cuda_version=13.0 --cuda_home=/usr/local/cuda-13.0 --cudnn_home=/usr/local/cuda-13.0 --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=86 onnxruntime_BUILD_UNIT_TESTS=ON'
+      extra_build_flags: >-
+        --use_binskim_compliant_compile_flags
+        --build_wheel
+        --parallel
+        --nvcc_threads 4
+        --flash_nvcc_threads 4
+        --cuda_version=13.0
+        --cuda_home=/usr/local/cuda-13.0
+        --cudnn_home=/usr/local/cuda-13.0
+        --cmake_extra_defines CMAKE_CUDA_ARCHITECTURES=86
+        --cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=ON
+        --cmake_extra_defines onnxruntime_QUICK_BUILD=ON
+        --cmake_extra_defines onnxruntime_USE_FPA_INTB_GEMM=OFF
       python_path_prefix: 'PATH=/opt/python/cp312-cp312/bin:$PATH'
       run_tests: false
       upload_build_output: true

.github/workflows/linux_cuda_plugin_ci.yml

@@ -42,6 +42,7 @@ jobs:
         --cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=ON
         --cmake_extra_defines onnxruntime_QUICK_BUILD=ON
         --cmake_extra_defines onnxruntime_BUILD_CUDA_EP_AS_PLUGIN=ON
+        --cmake_extra_defines onnxruntime_BUILD_CUDA_QUANT_PREPROCESS=ON
       python_path_prefix: 'PATH=/opt/python/cp312-cp312/bin:$PATH'
       run_tests: false
       upload_build_output: true
@@ -145,4 +146,3 @@ jobs:
               cd /onnxruntime_src/onnxruntime/test/python/transformers
               python test_cuda_plugin_ep.py
             "
-

cmake/CMakeLists.txt

@@ -77,8 +77,8 @@ cmake_dependent_option(onnxruntime_ENABLE_CUDA_EP_INTERNAL_TESTS "Build with CUD
 
 cmake_dependent_option(onnxruntime_USE_CUDA_NHWC_OPS "Build CUDA with NHWC op support" ON "onnxruntime_USE_CUDA" OFF)
 cmake_dependent_option(onnxruntime_BUILD_CUDA_EP_AS_PLUGIN "Build CUDA EP as a separate plugin shared library instead of the legacy in-tree provider" OFF "onnxruntime_USE_CUDA" OFF)
-option(onnxruntime_BUILD_CUDA_QUANT_PREPROCESS "Build CUDA weight-packing module onnxruntime_cuda_quant_preprocess.so" ON)
-option(onnxruntime_CUDA_MINIMAL "Build CUDA without any operations apart from memcpy ops. Usefuel for a very minial TRT build" OFF)
+option(onnxruntime_BUILD_CUDA_QUANT_PREPROCESS "Build CUDA weight-packing module onnxruntime_cuda_quant_preprocess.so" OFF)
+option(onnxruntime_CUDA_MINIMAL "Build CUDA without any operations apart from memcpy ops. Useful for a very minimal TRT build" OFF)
 option(onnxruntime_ENABLE_CUDA_LINE_NUMBER_INFO "When building with CUDA support, generate device code line number information." OFF)
 option(onnxruntime_USE_OPENVINO "Build with OpenVINO support" OFF)
 option(onnxruntime_USE_COREML "Build with CoreML support" OFF)