remove WITH_DSO compile option (#25444)

CMakeLists.txt

@@ -79,7 +79,6 @@ find_package(Threads REQUIRED)
 include(simd)
 
 ################################ Exposed Configurations #######################################
-option(WITH_DSO         "Compile PaddlePaddle with dynamic linked CUDA" ON)
 option(WITH_AVX         "Compile PaddlePaddle with AVX intrinsics"      ${AVX_FOUND})
 option(WITH_PYTHON      "Compile PaddlePaddle with python interpreter"  ON)
 option(WITH_TESTING     "Compile PaddlePaddle with unit testing"        OFF)

cmake/configure.cmake

@@ -16,10 +16,6 @@ if(NOT WITH_PYTHON)
     add_definitions(-DPADDLE_NO_PYTHON)
 endif(NOT WITH_PYTHON)
 
-if(WITH_DSO)
-    add_definitions(-DPADDLE_USE_DSO)
-endif(WITH_DSO)
-
 if(WITH_TESTING)
     add_definitions(-DPADDLE_WITH_TESTING)
 endif(WITH_TESTING)

cmake/cuda.cmake

@@ -188,12 +188,6 @@ endif()
 
 add_definitions("-DPADDLE_CUDA_BINVER=\"${CUDA_VERSION_MAJOR}${CUDA_VERSION_MINOR}\"")
 
-if(NOT WITH_DSO)
-    if(WIN32)
-      set_property(GLOBAL PROPERTY CUDA_MODULES ${CUDNN_LIBRARY} ${CUDA_CUBLAS_LIBRARIES} ${CUDA_curand_LIBRARY} ${CUDA_cusolver_LIBRARY})
-    endif(WIN32)
-endif(NOT WITH_DSO)
-
 # setting nvcc arch flags
 select_nvcc_arch_flags(NVCC_FLAGS_EXTRA)
 set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} ${NVCC_FLAGS_EXTRA}")

cmake/hip.cmake

@@ -13,10 +13,6 @@ include_directories("/opt/rocm/thrust")
 
 set(HIP_HCC_FLAGS "${HIP_HCC_FLAGS} -fPIC -DPADDLE_WITH_HIP -std=c++11" )
 
-if(WITH_DSO)
-  set(HIP_HCC_FLAGS "${HIP_HCC_FLAGS} -DPADDLE_USE_DSO")
-endif(WITH_DSO)
-
 if(WITH_TESTING)
   set(HIP_HCC_FLAGS "${HIP_HCC_FLAGS} -DPADDLE_WITH_TESTING")
 endif(WITH_TESTING)

cmake/tensorrt.cmake

@@ -37,16 +37,10 @@ find_library(TENSORRT_LIBRARY NAMES ${TR_INFER_LIB} ${TR_INFER_RT}
     DOC "Path to TensorRT library.")
 
 if(TENSORRT_INCLUDE_DIR AND TENSORRT_LIBRARY)
-    if(WITH_DSO)
     set(TENSORRT_FOUND ON)
-    endif(WITH_DSO)
 else()
     set(TENSORRT_FOUND OFF)
-    if(WITH_DSO)
-        message(WARNING "TensorRT is NOT found when WITH_DSO is ON.")
-    else(WITH_DSO)
-        message(STATUS "TensorRT is disabled because WITH_DSO is OFF.")
-    endif(WITH_DSO)
+    message(STATUS "TensorRT is disabled.")
 endif()
 
 if(TENSORRT_FOUND)

paddle/fluid/platform/CMakeLists.txt

@@ -94,10 +94,10 @@ if(WITH_GPU)
 endif()
 
 if(WIN32)
-    if(WITH_GPU AND NOT WITH_DSO)
+    if(WITH_GPU)
         get_property(cuda_modules GLOBAL PROPERTY CUDA_MODULES)
         target_link_libraries(device_context ${cuda_modules})
-    endif(WITH_GPU AND NOT WITH_DSO)
+    endif(WITH_GPU)
 endif(WIN32)
 
 nv_test(device_context_test SRCS device_context_test.cu DEPS device_context gpu_info)

paddle/fluid/platform/dynload/cublas.h

@@ -36,7 +36,6 @@ extern void *cublas_dso_handle;
  *
  * note: default dynamic linked libs
  */
-#ifdef PADDLE_USE_DSO
 #define DECLARE_DYNAMIC_LOAD_CUBLAS_WRAP(__name)                             \
   struct DynLoad__##__name {                                                 \
     using FUNC_TYPE = decltype(&::__name);                                   \
@@ -50,16 +49,6 @@ extern void *cublas_dso_handle;
     }                                                                        \
   };                                                                         \
   extern DynLoad__##__name __name
-#else
-#define DECLARE_DYNAMIC_LOAD_CUBLAS_WRAP(__name)     \
-  struct DynLoad__##__name {                         \
-    template <typename... Args>                      \
-    inline cublasStatus_t operator()(Args... args) { \
-      return ::__name(args...);                      \
-    }                                                \
-  };                                                 \
-  extern DynLoad__##__name __name
-#endif
 
 #define CUBLAS_BLAS_ROUTINE_EACH(__macro) \
   __macro(cublasSaxpy_v2);                \

paddle/fluid/platform/dynload/cuda_driver.cc

@@ -25,14 +25,10 @@ void* cuda_dso_handle = nullptr;
 
 CUDA_ROUTINE_EACH(DEFINE_WRAP);
 
-#ifdef PADDLE_USE_DSO
 bool HasCUDADriver() {
   std::call_once(cuda_dso_flag, []() { cuda_dso_handle = GetCUDADsoHandle(); });
   return cuda_dso_handle != nullptr;
 }
-#else
-bool HasCUDADriver() { return false; }
-#endif
 
 }  // namespace dynload
 }  // namespace platform

paddle/fluid/platform/dynload/cuda_driver.h

@@ -27,8 +27,6 @@ extern std::once_flag cuda_dso_flag;
 extern void* cuda_dso_handle;
 extern bool HasCUDADriver();
 
-#ifdef PADDLE_USE_DSO
-
 #define DECLARE_DYNAMIC_LOAD_CUDA_WRAP(__name)                           \
   struct DynLoad__##__name {                                             \
     template <typename... Args>                                          \
@@ -43,19 +41,6 @@ extern bool HasCUDADriver();
   };                                                                     \
   extern struct DynLoad__##__name __name
 
-#else
-
-#define DECLARE_DYNAMIC_LOAD_CUDA_WRAP(__name) \
-  struct DynLoad__##__name {                   \
-    template <typename... Args>                \
-    inline auto operator()(Args... args) {     \
-      return ::__name(args...);                \
-    }                                          \
-  };                                           \
-  extern DynLoad__##__name __name
-
-#endif
-
 /**
  * include all needed cuda driver functions
  **/

paddle/fluid/platform/dynload/cudnn.cc

@@ -50,7 +50,6 @@ CUDNN_DNN_ROUTINE_EACH_R7(DEFINE_WRAP);
 CUDNN_DNN_ROUTINE_EACH_AFTER_R7(DEFINE_WRAP);
 #endif
 
-#ifdef PADDLE_USE_DSO
 bool HasCUDNN() {
   std::call_once(cudnn_dso_flag,
                  []() { cudnn_dso_handle = GetCUDNNDsoHandle(); });
@@ -64,9 +63,6 @@ void EnforceCUDNNLoaded(const char* fn_name) {
           "Cannot load cudnn shared library. Cannot invoke method %s.",
           fn_name));
 }
-#else
-bool HasCUDNN() { return true; }
-#endif
 
 }  // namespace dynload
 }  // namespace platform

paddle/fluid/platform/dynload/cudnn.h

@@ -28,8 +28,6 @@ extern std::once_flag cudnn_dso_flag;
 extern void* cudnn_dso_handle;
 extern bool HasCUDNN();
 
-#ifdef PADDLE_USE_DSO
-
 extern void EnforceCUDNNLoaded(const char* fn_name);
 #define DECLARE_DYNAMIC_LOAD_CUDNN_WRAP(__name)                            \
   struct DynLoad__##__name {                                               \
@@ -46,19 +44,6 @@ extern void EnforceCUDNNLoaded(const char* fn_name);
   };                                                                       \
   extern struct DynLoad__##__name __name
 
-#else
-
-#define DECLARE_DYNAMIC_LOAD_CUDNN_WRAP(__name) \
-  struct DynLoad__##__name {                    \
-    template <typename... Args>                 \
-    inline auto operator()(Args... args) {      \
-      return ::__name(args...);                 \
-    }                                           \
-  };                                            \
-  extern DynLoad__##__name __name
-
-#endif
-
 /**
  * include all needed cudnn functions in HPPL
  * different cudnn version has different interfaces

paddle/fluid/platform/dynload/cupti.h

@@ -36,7 +36,6 @@ extern void *cupti_dso_handle;
  *
  * note: default dynamic linked libs
  */
-#ifdef PADDLE_USE_DSO
 #define DECLARE_DYNAMIC_LOAD_CUPTI_WRAP(__name)                            \
   struct DynLoad__##__name {                                               \
     template <typename... Args>                                            \
@@ -50,16 +49,6 @@ extern void *cupti_dso_handle;
     }                                                                      \
   };                                                                       \
   extern DynLoad__##__name __name
-#else
-#define DECLARE_DYNAMIC_LOAD_CUPTI_WRAP(__name)            \
-  struct DynLoad__##__name {                               \
-    template <typename... Args>                            \
-    inline CUptiResult CUPTIAPI operator()(Args... args) { \
-      return __name(args...);                              \
-    }                                                      \
-  };                                                       \
-  extern DynLoad__##__name __name
-#endif
 
 #define CUPTI_ROUTINE_EACH(__macro)           \
   __macro(cuptiActivityEnable);               \

paddle/fluid/platform/dynload/curand.h

@@ -25,7 +25,7 @@ namespace platform {
 namespace dynload {
 extern std::once_flag curand_dso_flag;
 extern void *curand_dso_handle;
-#ifdef PADDLE_USE_DSO
+
 #define DECLARE_DYNAMIC_LOAD_CURAND_WRAP(__name)                             \
   struct DynLoad__##__name {                                                 \
     template <typename... Args>                                              \
@@ -39,16 +39,6 @@ extern void *curand_dso_handle;
     }                                                                        \
   };                                                                         \
   extern DynLoad__##__name __name
-#else
-#define DECLARE_DYNAMIC_LOAD_CURAND_WRAP(__name) \
-  struct DynLoad__##__name {                     \
-    template <typename... Args>                  \
-    curandStatus_t operator()(Args... args) {    \
-      return ::__name(args...);                  \
-    }                                            \
-  };                                             \
-  extern DynLoad__##__name __name
-#endif
 
 #define CURAND_RAND_ROUTINE_EACH(__macro)      \
   __macro(curandCreateGenerator);              \

paddle/fluid/platform/dynload/cusolver.h

@@ -26,7 +26,7 @@ namespace platform {
 namespace dynload {
 extern std::once_flag cusolver_dso_flag;
 extern void *cusolver_dso_handle;
-#ifdef PADDLE_USE_DSO
+
 #define DECLARE_DYNAMIC_LOAD_CUSOLVER_WRAP(__name)                   \
   struct DynLoad__##__name {                                         \
     template <typename... Args>                                      \
@@ -41,16 +41,6 @@ extern void *cusolver_dso_handle;
     }                                                                \
   };                                                                 \
   extern DynLoad__##__name __name
-#else
-#define DECLARE_DYNAMIC_LOAD_CUSOLVER_WRAP(__name) \
-  struct DynLoad__##__name {                       \
-    template <typename... Args>                    \
-    cusolverStatus_t operator()(Args... args) {    \
-      return ::__name(args...);                    \
-    }                                              \
-  };                                               \
-  extern DynLoad__##__name __name
-#endif
 
 #define CUSOLVER_ROUTINE_EACH(__macro)  \
   __macro(cusolverDnCreate);            \

paddle/fluid/platform/dynload/nccl.h

@@ -26,8 +26,6 @@ namespace dynload {
 extern std::once_flag nccl_dso_flag;
 extern void* nccl_dso_handle;
 
-#ifdef PADDLE_USE_DSO
-
 #define DECLARE_DYNAMIC_LOAD_NCCL_WRAP(__name)                           \
   struct DynLoad__##__name {                                             \
     template <typename... Args>                                          \
@@ -41,16 +39,6 @@ extern void* nccl_dso_handle;
     }                                                                    \
   };                                                                     \
   extern DynLoad__##__name __name
-#else
-#define DECLARE_DYNAMIC_LOAD_NCCL_WRAP(__name) \
-  struct DynLoad__##__name {                   \
-    template <typename... Args>                \
-    ncclResult_t operator()(Args... args) {    \
-      return __name(args...);                  \
-    }                                          \
-  };                                           \
-  extern DynLoad__##__name __name
-#endif
 
 #define NCCL_RAND_ROUTINE_EACH(__macro) \
   __macro(ncclCommInitAll);             \

paddle/fluid/platform/dynload/nvrtc.cc

@@ -25,15 +25,11 @@ void* nvrtc_dso_handle = nullptr;
 
 NVRTC_ROUTINE_EACH(DEFINE_WRAP);
 
-#ifdef PADDLE_USE_DSO
 bool HasNVRTC() {
   std::call_once(nvrtc_dso_flag,
                  []() { nvrtc_dso_handle = GetNVRTCDsoHandle(); });
   return nvrtc_dso_handle != nullptr;
 }
-#else
-bool HasNVRTC() { return false; }
-#endif
 
 }  // namespace dynload
 }  // namespace platform

paddle/fluid/platform/dynload/nvrtc.h

@@ -27,8 +27,6 @@ extern std::once_flag nvrtc_dso_flag;
 extern void* nvrtc_dso_handle;
 extern bool HasNVRTC();
 
-#ifdef PADDLE_USE_DSO
-
 #define DECLARE_DYNAMIC_LOAD_NVRTC_WRAP(__name)                            \
   struct DynLoad__##__name {                                               \
     template <typename... Args>                                            \
@@ -43,19 +41,6 @@ extern bool HasNVRTC();
   };                                                                       \
   extern struct DynLoad__##__name __name
 
-#else
-
-#define DECLARE_DYNAMIC_LOAD_NVRTC_WRAP(__name) \
-  struct DynLoad__##__name {                    \
-    template <typename... Args>                 \
-    inline auto operator()(Args... args) {      \
-      return ::__name(args...);                 \
-    }                                           \
-  };                                            \
-  extern DynLoad__##__name __name
-
-#endif
-
 /**
  * include all needed nvrtc functions
  **/

paddle/fluid/platform/dynload/tensorrt.h

@@ -30,8 +30,6 @@ namespace dynload {
 extern std::once_flag tensorrt_dso_flag;
 extern void* tensorrt_dso_handle;
 
-#ifdef PADDLE_USE_DSO
-
 #define DECLARE_DYNAMIC_LOAD_TENSORRT_WRAP(__name)                      \
   struct DynLoad__##__name {                                            \
     template <typename... Args>                                         \
@@ -49,17 +47,6 @@ extern void* tensorrt_dso_handle;
   };                                                                    \
   extern DynLoad__##__name __name
 
-#else
-#define DECLARE_DYNAMIC_LOAD_TENSORRT_WRAP(__name) \
-  struct DynLoad__##__name {                       \
-    template <typename... Args>                    \
-    tensorrtResult_t operator()(Args... args) {    \
-      return __name(args...);                      \
-    }                                              \
-  };                                               \
-  extern DynLoad__##__name __name
-#endif
-
 #define TENSORRT_RAND_ROUTINE_EACH(__macro) \
   __macro(createInferBuilder_INTERNAL);     \
   __macro(createInferRuntime_INTERNAL);

paddle/scripts/paddle_build.sh

@@ -193,7 +193,6 @@ function cmake_base() {
     Configuring cmake in /paddle/build ...
         -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE:-Release}
         ${PYTHON_FLAGS}
-        -DWITH_DSO=ON
         -DWITH_GPU=${WITH_GPU:-OFF}
         -DWITH_AMD_GPU=${WITH_AMD_GPU:-OFF}
         -DWITH_DISTRIBUTE=${distibuted_flag}
@@ -222,7 +221,6 @@ EOF
     cmake .. \
         -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE:-Release} \
         ${PYTHON_FLAGS} \
-        -DWITH_DSO=ON \
         -DWITH_GPU=${WITH_GPU:-OFF} \
         -DWITH_AMD_GPU=${WITH_AMD_GPU:-OFF} \
         -DWITH_DISTRIBUTE=${distibuted_flag} \