fix merged conflicts

.pipelines/ci.yml

@@ -121,12 +121,12 @@ stages:
       inputs:
         versionSpec: '3.x'
         disableDownloadFromRegistry: true
-        addToPath: false
+        addToPath: true
         architecture: 'x64'
 
     - script: |
         python -m pip install --upgrade setuptools pip
-        python -m pip install numpy
+        python -m pip install 'numpy < 2.0.0'
         export OCOS_NO_OPENCV=1
         export OCOS_SCB_DEBUG=1
         CPU_NUMBER=8 python -m pip install -e .
@@ -322,6 +322,7 @@ stages:
         python -m pip install --upgrade pip
         python -m pip install --upgrade setuptools
         python -m pip install --upgrade wheel
+        python -m pip install 'numpy < 2.0.0'
         python -m pip install onnxruntime==$(ort.version)
       displayName: Install requirements
 
@@ -507,13 +508,13 @@ stages:
       inputs:
         versionSpec: '3.x'
         disableDownloadFromRegistry: true
-        addToPath: false
+        addToPath: true
         architecture: 'x64'
       displayName: Use ADO python task
 
     - script: |
         python -m pip install --upgrade setuptools pip
-        python -m pip install numpy
+        python -m pip install "numpy < 2.0.0"
         set OCOS_NO_OPENCV=1
         set OCOS_SCB_DEBUG=1
         python -m pip install -v -e .
@@ -570,7 +571,9 @@ stages:
 
       - script: |
           set CUDA_PATH=$(Agent.TempDirectory)\v11.8
-          call .\build.bat -T cuda="%CUDA_PATH%" -DOCOS_ENABLE_CTEST=ON -DOCOS_USE_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=70;86^
+          call .\build.bat -T cuda="%CUDA_PATH%" -DOCOS_ENABLE_CTEST=ON^
+            -DCMAKE_CUDA_FLAGS_INIT=-allow-unsupported-compiler^
+            -DOCOS_USE_CUDA=ON -DCMAKE_CUDA_ARCHITECTURES=70;86^
             -DOCOS_ONNXRUNTIME_VERSION="$(ORT_VERSION)" -DONNXRUNTIME_PKG_DIR=.\onnxruntime-win-x64-gpu-$(ORT_VERSION)
         displayName: build the customop library with onnxruntime
 
@@ -583,14 +586,14 @@ stages:
         inputs:
           versionSpec: '3.x'
           disableDownloadFromRegistry: true
-          addToPath: false
+          addToPath: true
           architecture: 'x64'
         displayName: Use ADO python task
 
       - script: |
           set CUDA_PATH=$(Agent.TempDirectory)\v11.8
           python -m pip install --upgrade setuptools pip
-          python -m pip install numpy coloredlogs flatbuffers packaging protobuf sympy
+          python -m pip install "numpy < 2.0.0" coloredlogs flatbuffers packaging protobuf sympy
           python -m pip install onnxruntime-gpu==$(ORT_VERSION)
           python -m pip install -v --config-settings "ortx-user-option=use-cuda,cuda_archs=70;86" .
         displayName: Build and install onnxruntime-extensions CUDA package.

.pipelines/ci_optional.yml

@@ -132,7 +132,7 @@ stages:
       inputs:
         versionSpec: '3.x'
         disableDownloadFromRegistry: true
-        addToPath: false
+        addToPath: true
         architecture: 'x64'
       displayName: Use ADO python task
 
@@ -172,7 +172,7 @@ stages:
         inputs:
           versionSpec: '3.x'
           disableDownloadFromRegistry: true
-          addToPath: false
+          addToPath: true
           architecture: 'x64'
         displayName: Use ADO python task
 

.pipelines/wheels_linux.yml

@@ -2,7 +2,7 @@ parameters:
 - name: ExtraEnv
   displayName: 'Extra env variable set to CIBW_ENVIRONMENT, in form of "A=1 B=2 C=3"'
   type: string
-  default: 'OCOS_ENABLE_AZURE=1'
+  default: 'OCOS_ENABLE_AZURE=0'
 
 jobs:
 - job: linux_x86_64

.pyproject/cmdclass.py

@@ -226,6 +226,8 @@ def build_cmake(self, extension):
                 if sys.platform == "win32":
                     cuda_path = os.environ.get("CUDA_PATH")
                     cmake_args += [f'-T cuda={cuda_path}']
+                    # TODO: temporarily add a flag for MSVC 19.40
+                    cmake_args += ['-DCMAKE_CUDA_FLAGS_INIT=-allow-unsupported-compiler']
                 f_ver = ext_fullpath.parent / "_version.py"
                 with f_ver.open('a') as _f:
                     _f.writelines(["\n", f"cuda = \"{cuda_ver}\"", "\n"])
@@ -235,7 +237,8 @@ def build_cmake(self, extension):
                 else:
                     smi = _load_nvidia_smi()
                     if not smi:
-                        raise RuntimeError(f"Cannot detect the CUDA archs from your machine, please specify it by yourself.")
+                        raise RuntimeError(
+                            "Cannot detect the CUDA archs from your machine, please specify it manually.")
                     cmake_args += ['-DCMAKE_CUDA_ARCHITECTURES=' + smi]
 
         # CMake lets you override the generator - we need to check this.
@@ -274,7 +277,6 @@ def build_cmake(self, extension):
                 cmake_args += [
                     "-DCMAKE_OSX_ARCHITECTURES={}".format(";".join(archs))]
 
-
         # overwrite the Python module info if the auto-detection doesn't work.
         # export Python3_INCLUDE_DIRS=/opt/python/cp38-cp38
         # export Python3_LIBRARIES=/opt/python/cp38-cp38
@@ -292,14 +294,18 @@ def build_cmake(self, extension):
             '--parallel' + ('' if cpu_number is None else ' ' + cpu_number)
         ]
         cmake_exe = 'cmake'
-        # unlike Linux/macOS, cmake pip package on Windows fails to build some 3rd party dependencies.
-        # so we have to use the cmake installed from Visual Studio.
-        if os.environ.get(VSINSTALLDIR_NAME):
-            cmake_exe = os.environ[VSINSTALLDIR_NAME] + \
-                        'Common7\\IDE\\CommonExtensions\\Microsoft\\CMake\\CMake\\bin\\cmake.exe'
-            # Add this cmake directory into PATH to make sure the child-process still find it.
-            os.environ['PATH'] = os.path.dirname(
-                cmake_exe) + os.pathsep + os.environ['PATH']
+        # if sys.platform == "win32":
+        #     # unlike Linux/macOS, cmake pip package on Windows fails to build some 3rd party dependencies.
+        #     # so we have to use the cmake from a standalone installation or the one from Visual Studio.
+        #     standalone_cmake = os.path.join(os.environ.get("ProgramFiles"), "\\CMake\\bin\\cmake.exe")
+        #     if os.path.exists(standalone_cmake):
+        #         cmake_exe = standalone_cmake
+        #     elif os.environ.get(VSINSTALLDIR_NAME):
+        #         cmake_exe = os.environ[VSINSTALLDIR_NAME] + \
+        #                     'Common7\\IDE\\CommonExtensions\\Microsoft\\CMake\\CMake\\bin\\cmake.exe'
+        #     # Add this cmake directory into PATH to make sure the child-process still find it.
+        #     os.environ['PATH'] = os.path.dirname(
+        #         cmake_exe) + os.pathsep + os.environ['PATH']
 
         self.spawn([cmake_exe, '-S', str(project_dir),
                     '-B', str(build_temp)] + cmake_args)

CMakeLists.txt

@@ -36,6 +36,11 @@ if(CMAKE_VERSION VERSION_GREATER_EQUAL "3.24.0")
   cmake_policy(SET CMP0077 NEW)
 endif()
 
+# Avoid warning of Calling FetchContent_Populate(GSL) is deprecated
+if(CMAKE_VERSION VERSION_GREATER_EQUAL "3.30.0")
+  cmake_policy(CMP0169 OLD)
+endif()
+
 # Needed for Java
 set(CMAKE_C_STANDARD 99)
 
@@ -90,12 +95,11 @@ set(OCOS_ONNXRUNTIME_PKG_URI "" CACHE STRING
   "Specify the onnxruntime C++ shared library zip package path, like ./onnxruntime-win-x64-1.16.0.zip")
 set(OCOS_BUILD_PRESET "" CACHE STRING
   "Specify the build preset cmake settings file path, like 'token_api_only' which includes ./cmake/presets/token_api_only.cmake")
-# TODO: Remove the following statements if AzureOp build is enabled by default.
-# If build_buildid environment varaible is set, which means this is a CI build, then always enable AzureOp.
-# or it is enabled when OCOS_ENABLE_AZURE is set, which means the user explicitly enables it.
-if ((DEFINED ENV{BUILD_BUILDID}) OR (DEFINED ENV{OCOS_ENABLE_AZURE}))
+
+# AzureOp can be enabled by environment varaible OCOS_ENABLE_AZURE == 1
+if (DEFINED ENV{OCOS_ENABLE_AZURE})
   set(OCOS_ENABLE_AZURE ON CACHE INTERNAL "" FORCE)
-  message(STATUS "=> AzureOp is enabled by default.")
+  message(STATUS "=> AzureOp is enabled env variable.")
 endif()
 
 function(disable_all_operators)

build.bat

@@ -1,12 +1,21 @@
 @ECHO OFF
 SETLOCAL ENABLEDELAYEDEXPANSION
+
+IF NOT EXIST "%ProgramFiles%\CMake\bin\cmake.exe" GOTO :FIND_VS
+set cmake_exe="%ProgramFiles%\CMake\bin\cmake.exe"
+
+:FIND_VS
 IF DEFINED VSINSTALLDIR GOTO :VSDEV_CMD
 set _VSFINDER=%~dp0tools\get_vsdevcmd.ps1
 for /f "tokens=* USEBACKQ" %%i in (
     `powershell -NoProfile -ExecutionPolicy Bypass -File "%_VSFINDER%"`) do call "%%i"
 
 IF NOT DEFINED VSINSTALLDIR GOTO :NOT_FOUND
 
+IF DEFINED cmake_exe GOTO :CMAKE_DEF
+set cmake_exe="%VSINSTALLDIR%Common7\IDE\CommonExtensions\Microsoft\CMake\CMake\bin\cmake.exe"
+
+:CMAKE_DEF
 IF "%1" == "-A" GOTO :VSDEV_CMD
 set GEN_PLATFORM=-A x64
 
@@ -16,8 +25,8 @@ IF "%VisualStudioVersion:~0,2%" == "16" GOTO :START_BUILD
 set GENERATOR="Visual Studio 17 2022"
 
 :START_BUILD
-set cmake_exe="%VSINSTALLDIR%Common7\IDE\CommonExtensions\Microsoft\CMake\CMake\bin\cmake.exe"
 mkdir .\out\Windows\ 2>NUL
+ECHO %cmake_exe% -G %GENERATOR% %GEN_PLATFORM% %* -B out\Windows -S .
 %cmake_exe% -G %GENERATOR% %GEN_PLATFORM% %* -B out\Windows -S .
 IF %ERRORLEVEL% NEQ 0 EXIT /B %ERRORLEVEL%
 %cmake_exe% --build out\Windows --config RelWithDebInfo

cmake/externals/opencv-no-rtti.patch

@@ -39,6 +39,19 @@ index d95e5db163..db185453df 100644
  include(cmake/OpenCVCompilerOptions.cmake)
 
  ocv_cmake_hook(POST_COMPILER_OPTIONS)
+diff --git a/cmake/OpenCVDetectCXXCompiler.cmake b/cmake/OpenCVDetectCXXCompiler.cmake
+index 7f229cde96..92e204a5b9 100644
+--- a/cmake/OpenCVDetectCXXCompiler.cmake
++++ b/cmake/OpenCVDetectCXXCompiler.cmake
+@@ -171,7 +171,7 @@ elseif(MSVC)
+     set(OpenCV_RUNTIME vc15)
+   elseif(MSVC_VERSION MATCHES "^192[0-9]$")
+     set(OpenCV_RUNTIME vc16)
+-  elseif(MSVC_VERSION MATCHES "^193[0-9]$")
++  elseif(MSVC_VERSION MATCHES "^19[34][0-9]$")
+     set(OpenCV_RUNTIME vc17)
+   else()
+     message(WARNING "OpenCV does not recognize MSVC_VERSION \"${MSVC_VERSION}\". Cannot set OpenCV_RUNTIME")
 diff --git a/modules/core/include/opencv2/core/ocl.hpp b/modules/core/include/opencv2/core/ocl.hpp
 index 4503fa00dd..642b0508d0 100644
 --- a/modules/core/include/opencv2/core/ocl.hpp

docs/custom_ops.md

@@ -1312,7 +1312,7 @@ expect(node, inputs=[text, pattern, rewrite], outputs=[y],
 
 
 ## Azure operators
-
+Starting from onnxruntime-extensions 0.12, these Azure operators will be removed from the official onnxruntime-extensions packages. However, they can still be built from source using `cmake -DOCOS_ENABLE_AZURE=ON ...`.
 ### OpenAIAudioToText
 
 <details>

operators/cuda/cuda_ops.cc

@@ -8,13 +8,13 @@
 #include "cuda/fast_gelu.h"
 #include "cuda/mul_sigmoid.h"
 #include "cuda/negxplus1.h"
+#include "cuda/replace_zero.h"
 #include "cuda/rotary.h"
 #include "cuda/scatter_nd_of_shape.h"
 #include "cuda/transpose_cast.h"
 #endif
 
 FxLoadCustomOpFactory LoadCustomOpClasses_Contrib = []() -> CustomOpArray& {
-
   using AddSharedInputFloat32Type = typename contrib::AddOrMulSharedInput<float, true>;
   using MulSharedInputFloat32Type = typename contrib::AddOrMulSharedInput<float, false>;
 
@@ -25,7 +25,6 @@ FxLoadCustomOpFactory LoadCustomOpClasses_Contrib = []() -> CustomOpArray& {
   using Transpose2DCastFloat16ToFloat32Type = typename contrib::Transpose2DCast<ortc::MFloat16, float>;
 #endif
 
-
   static OrtOpLoader op_loader(
       []() { return nullptr; }
 #ifdef USE_CUDA
@@ -37,6 +36,7 @@ FxLoadCustomOpFactory LoadCustomOpClasses_Contrib = []() -> CustomOpArray& {
       CustomCudaStructV2("MulMulSigmoid", contrib::MulMulSigmoid<float>),
       CustomCudaStructV2("MulSigmoid", contrib::MulSigmoid<float>),
       CustomCudaStructV2("NegXPlus1", contrib::NegXPlus1<float>),
+      CustomCudaStructV2("ReplaceZero", contrib::ReplaceZero<float>),
       CustomCudaStructV2("Rotary", contrib::Rotary<float>),
       CustomCudaStructV2("ScatterNDOfShape", contrib::ScatterNDOfShape<float>),
 #if ORT_API_VERSION >= 16
@@ -49,6 +49,7 @@ FxLoadCustomOpFactory LoadCustomOpClasses_Contrib = []() -> CustomOpArray& {
       CustomCudaStructV2("MulMulSigmoid", contrib::MulMulSigmoid<ortc::MFloat16>),
       CustomCudaStructV2("MulSigmoid", contrib::MulSigmoid<ortc::MFloat16>),
       CustomCudaStructV2("NegXPlus1", contrib::NegXPlus1<ortc::MFloat16>),
+      CustomCudaStructV2("ReplaceZero", contrib::ReplaceZero<ortc::MFloat16>),
       CustomCudaStructV2("Rotary", contrib::Rotary<ortc::MFloat16>),
       CustomCudaStructV2("ScatterNDOfShape", contrib::ScatterNDOfShape<ortc::MFloat16>),
       CustomCudaStructV2("Transpose2DCastFP16", Transpose2DCastFloat32ToFloat16Type),

operators/cuda/negxplus1.h

@@ -8,6 +8,9 @@
 
 namespace contrib {
 
+/**
+* NegXPlus1(X) = 1 - X
+*/
 template <typename T>
 struct NegXPlus1 {
   template <typename TDict>

operators/cuda/replace_zero.h

@@ -0,0 +1,51 @@
+// Copyright (c) Microsoft Corporation. All rights reserved.
+// Licensed under the MIT License.
+
+#pragma once
+#include "ocos.h"
+#include "replace_zero_impl.cuh"
+#include "ortx_common.h"
+
+namespace contrib {
+
+/**
+* Y = ReplaceZero(X, by=c) is equivalent to:
+*
+* Y = X.copy()
+* X[X == 0] = c
+*
+* This operation usually appears when a tensor is updated with an operator Equal and Where.
+* This kernel avoids the creation of one null tensor.
+*/
+template <typename T>
+struct ReplaceZero {
+  template <typename TDict>
+  OrtxStatus OnModelAttach(const TDict& dict) {
+    float default_value=0;
+    by_ = dict.TryToGetAttributeWithDefault("by", default_value);
+    return {};
+  }
+  OrtxStatus Compute(Ort::Custom::CUDAKernelContext* ctx,
+                       const ortc::Tensor<T>& input,
+                       ortc::Tensor<T>& output) const {
+    const T* input_data = input.Data();
+    auto input_shape = input.Shape();
+    T* output_data = output.Allocate(input_shape);
+    auto input_length = input.NumberOfElement();
+    if (0 == input_length) {
+      return {};
+    }
+
+    LaunchReplaceZeroKernel<T>(reinterpret_cast<cudaStream_t>(ctx->GetCudaStream()),
+                             input_length,
+                             input_data,
+                             output_data,
+                             by_);
+    return {};
+  }
+
+  private:
+  float by_;
+};
+
+}  // namespace contrib