fix use nvidia cuda libraries bug (#62425)

risemeup1 · web-flow · commit 19a5ae5b652a · 2024-03-06T13:23:00.000+08:00
* fix

* fix

* fix
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -65,7 +65,8 @@ option(WITH_SETUP_INSTALL "Compile PaddlePaddle with setup.py" OFF)
 option(WITH_SHARED_PHI "Compile PaddlePaddle with SHARED LIB of PHI" ON)
 option(CINN_ONLY "Compile CINN only in Paddle" OFF)
 option(CINN_WITH_CUDNN "Compile CINN with CUDNN support" ON)
-
+option(WITH_PIP_CUDA_LIBRARIES
+       "Paddle uses the CUDA library provided by NVIDIA" OFF)
 find_package(Git REQUIRED)
 
 # config GIT_URL with github mirrors to speed up dependent repos clone
@@ -97,11 +98,16 @@ endif()
 
 if(WITH_GPU AND NOT APPLE)
   #(Note risemeup1): The cudart dynamic library libcudart.so is used by set CUDA_USE_STATIC_CUDA_RUNTIME and CMAKE_CUDA_FLAGS
-  if(LINUX)
+  if(CMAKE_SYSTEM_NAME STREQUAL "Linux" AND CMAKE_SYSTEM_PROCESSOR STREQUAL
+                                            "x86_64")
     set(CUDA_USE_STATIC_CUDA_RUNTIME
         OFF
         CACHE BOOL "" FORCE)
     set(CMAKE_CUDA_FLAGS "--cudart shared")
+    if(WITH_PIP_CUDA_LIBRARIES)
+      #(Note risemeup1): Flag 'WITH_PIP_CUDA_LIBRARIES' will be used in dynamic_loader.cc to search for CUDA-related .so files through the Python libraries provided by NVIDIA.
+      add_definitions(-DWITH_PIP_CUDA_LIBRARIES)
+    endif()
   endif()
   enable_language(CUDA)
   message(STATUS "CUDA compiler: ${CMAKE_CUDA_COMPILER}, version: "
diff --git a/paddle/phi/backends/dynload/dynamic_loader.cc b/paddle/phi/backends/dynload/dynamic_loader.cc
@@ -289,9 +289,17 @@ void* GetCublasDsoHandle() {
       FLAGS_cuda_dir, win_cublas_lib, true, {cuda_lib_path});
 #elif defined(__linux__) && defined(PADDLE_WITH_CUDA)
   if (CUDA_VERSION >= 11000 && CUDA_VERSION < 12000) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
     return GetDsoHandleFromSearchPath(FLAGS_cublas_dir, "libcublas.so.11");
+#else
+    return GetDsoHandleFromSearchPath(FLAGS_cublas_dir, "libcublas.so");
+#endif
   } else if (CUDA_VERSION >= 12000 && CUDA_VERSION <= 12030) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
     return GetDsoHandleFromSearchPath(FLAGS_cublas_dir, "libcublas.so.12");
+#else
+    return GetDsoHandleFromSearchPath(FLAGS_cublas_dir, "libcublas.so");
+#endif
   } else {
     std::string warning_msg(
         "Your CUDA_VERSION is less than 11 or greater than 12, paddle "
@@ -309,9 +317,17 @@ void* GetCublasLtDsoHandle() {
 // APIs available after CUDA 10.1
 #if defined(__linux__) && defined(PADDLE_WITH_CUDA)
   if (CUDA_VERSION >= 11000 && CUDA_VERSION < 12000) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
     return GetDsoHandleFromSearchPath(FLAGS_cublas_dir, "libcublasLt.so.11");
+#else
+    return GetDsoHandleFromSearchPath(FLAGS_cublas_dir, "libcublasLt.so");
+#endif
   } else if (CUDA_VERSION >= 12000 && CUDA_VERSION <= 12030) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
     return GetDsoHandleFromSearchPath(FLAGS_cublas_dir, "libcublasLt.so.12");
+#else
+    return GetDsoHandleFromSearchPath(FLAGS_cublas_dir, "libcublasLt.so");
+#endif
   } else {
     std::string warning_msg(
         "Your CUDA_VERSION is less than 11 or greater than 12, paddle "
@@ -353,8 +369,13 @@ void* GetCUDNNDsoHandle() {
 #elif defined(PADDLE_WITH_HIP)
   return GetDsoHandleFromSearchPath(FLAGS_miopen_dir, "libMIOpen.so", false);
 #else
+#ifdef WITH_PIP_CUDA_LIBRARIES
   return GetDsoHandleFromSearchPath(
       FLAGS_cudnn_dir, "libcudnn.so.8", false, {cuda_lib_path});
+#else
+  return GetDsoHandleFromSearchPath(
+      FLAGS_cudnn_dir, "libcudnn.so", false, {cuda_lib_path});
+#endif
 #endif
 }
 
@@ -364,11 +385,22 @@ void* GetCUPTIDsoHandle() {
       FLAGS_cupti_dir, "libcupti.dylib", false, {cupti_lib_path});
 #elif defined(__linux__) && defined(PADDLE_WITH_CUDA)
   if (CUDA_VERSION >= 11000 && CUDA_VERSION < 12000) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
     return GetDsoHandleFromSearchPath(
-        FLAGS_cupti_dir, "libcupti.so.11.7", false, {cupti_lib_path});
+        FLAGS_cupti_dir, "libcupti.so.11.8", false, {cupti_lib_path});
+#else
+    return GetDsoHandleFromSearchPath(
+        FLAGS_cupti_dir, "libcupti.so", false, {cupti_lib_path});
+#endif
+
   } else if (CUDA_VERSION >= 12000 && CUDA_VERSION < 12030) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
     return GetDsoHandleFromSearchPath(
         FLAGS_cupti_dir, "libcupti.so.12", false, {cupti_lib_path});
+#else
+    return GetDsoHandleFromSearchPath(
+        FLAGS_cupti_dir, "libcupti.so", false, {cupti_lib_path});
+#endif
   } else {
     std::string warning_msg(
         "Your CUDA_VERSION is less than 11 or greater than 12, paddle "
@@ -377,7 +409,7 @@ void* GetCUPTIDsoHandle() {
   }
 #else
   return GetDsoHandleFromSearchPath(
-      FLAGS_cupti_dir, "libcupti.so.11.7", false, {cupti_lib_path});
+      FLAGS_cupti_dir, "libcupti.so", false, {cupti_lib_path});
 #endif
 }
 
@@ -390,7 +422,12 @@ void* GetCurandDsoHandle() {
 #elif defined(PADDLE_WITH_HIP)
   return GetDsoHandleFromSearchPath(FLAGS_rocm_dir, "libhiprand.so");
 #else
+#ifdef WITH_PIP_CUDA_LIBRARIES
   return GetDsoHandleFromSearchPath(FLAGS_curand_dir, "libcurand.so.10");
+#else
+  return GetDsoHandleFromSearchPath(FLAGS_curand_dir, "libcurand.so");
+#endif
+
 #endif
 }
 
@@ -422,7 +459,11 @@ void* GetCusolverDsoHandle() {
   return GetDsoHandleFromSearchPath(
       FLAGS_cuda_dir, win_cusolver_lib, true, {cuda_lib_path});
 #else
+#ifdef WITH_PIP_CUDA_LIBRARIES
   return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "libcusolver.so.11");
+#else
+  return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "libcusolver.so");
+#endif
 #endif
 }
 
@@ -434,9 +475,17 @@ void* GetCusparseDsoHandle() {
       FLAGS_cuda_dir, win_cusparse_lib, true, {cuda_lib_path});
 #elif defined(__linux__) && defined(PADDLE_WITH_CUDA)
   if (CUDA_VERSION >= 11000 && CUDA_VERSION < 12000) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
     return GetDsoHandleFromSearchPath(FLAGS_cusparse_dir, "libcusparse.so.11");
+#else
+    return GetDsoHandleFromSearchPath(FLAGS_cusparse_dir, "libcusparse.so");
+#endif
   } else if (CUDA_VERSION >= 12000 && CUDA_VERSION <= 12030) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
     return GetDsoHandleFromSearchPath(FLAGS_cusparse_dir, "libcusparse.so.12");
+#else
+    return GetDsoHandleFromSearchPath(FLAGS_cusparse_dir, "libcusparse.so");
+#endif
   } else {
     std::string warning_msg(
         "Your CUDA_VERSION is less than 11 or greater than 12, paddle "
@@ -536,8 +585,14 @@ void* GetNCCLDsoHandle() {
   return GetDsoHandleFromSearchPath(
       FLAGS_rccl_dir, "librccl.so", true, {}, warning_msg);
 #else
+#ifdef WITH_PIP_CUDA_LIBRARIES
   return GetDsoHandleFromSearchPath(
       FLAGS_nccl_dir, "libnccl.so.2", true, {}, warning_msg);
+#else
+  return GetDsoHandleFromSearchPath(
+      FLAGS_nccl_dir, "libnccl.so", true, {}, warning_msg);
+#endif
+
 #endif
 }
 
@@ -592,8 +647,12 @@ void* GetCUFFTDsoHandle() {
   return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "libcufft.dylib");
 #elif defined(__linux__) && defined(PADDLE_WITH_CUDA)
   if (CUDA_VERSION >= 11000 && CUDA_VERSION < 12000) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
     return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "libcufft.so.10");
-  } else if (CUDA_VERSION >= 12000 && CUDA_VERSION < 13000) {
+#else
+    return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "libcufft.so");
+#endif
+  } else if (CUDA_VERSION >= 12000 && CUDA_VERSION < 12030) {
     return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "libcufft.so.11");
   } else {
     std::string warning_msg(
@@ -639,6 +698,5 @@ void* GetXPTIDsoHandle() {
   return nullptr;
 #endif
 }
-
 }  // namespace dynload
 }  // namespace phi
diff --git a/python/env_dict.py.in b/python/env_dict.py.in
@@ -90,5 +90,6 @@ env_dict={
     'CINN_SOURCE_DIR':'@CINN_SOURCE_DIR@',
     'WITH_CPP_DIST':'@WITH_CPP_DIST@',
     'PADDLE_INSTALL_DIR':'@PADDLE_INSTALL_DIR@',
-    'PADDLE_LIB_TEST_DIR':'@PADDLE_LIB_TEST_DIR@'
+    'PADDLE_LIB_TEST_DIR':'@PADDLE_LIB_TEST_DIR@',
+    'WITH_PIP_CUDA_LIBRARIES':'@WITH_PIP_CUDA_LIBRARIES@'
 }
diff --git a/python/setup.py.in b/python/setup.py.in
@@ -407,10 +407,7 @@ write_distributed_training_mode_py(filename='@PADDLE_BINARY_DIR@/python/paddle/i
 
 def get_paddle_extra_install_requirements():
     #(Note risemeup1): Paddle will install the pypi cuda package provided by Nvidia, which includes the cuda runtime, cudnn, and cublas, thereby making the operation of 'pip install paddle' no longer dependent on the installation of cuda and cudnn.
-    paddle_cuda_install_requirements = os.getenv(
-        "PADDLE_CUDA_INSTALL_REQUIREMENTS", None
-    )
-    if paddle_cuda_install_requirements == "ON":
+    if '@WITH_PIP_CUDA_LIBRARIES@' == 'ON':
         PADDLE_CUDA_INSTALL_REQUIREMENTS = {
             "V11": (
                 "nvidia-cuda-runtime-cu11==11.8.89; platform_system == 'Linux' and platform_machine == 'x86_64' | "
diff --git a/setup.py b/setup.py
@@ -936,10 +936,7 @@ def get_setup_requires():
 
 def get_paddle_extra_install_requirements():
     # (Note risemeup1): Paddle will install the pypi cuda package provided by Nvidia, which includes the cuda runtime, cudnn, and cublas, thereby making the operation of 'pip install paddle' no longer dependent on the installation of cuda and cudnn.
-    paddle_cuda_install_requirements = os.getenv(
-        "PADDLE_CUDA_INSTALL_REQUIREMENTS", None
-    )
-    if paddle_cuda_install_requirements == "ON":
+    if env_dict.get("WITH_PIP_CUDA_LIBRARIES") == "ON":
         PADDLE_CUDA_INSTALL_REQUIREMENTS = {
             "V11": (
                 "nvidia-cuda-runtime-cu11==11.8.89; platform_system == 'Linux' and platform_machine == 'x86_64' | "

Original file line number	Diff line number	Diff line change
`@@ -90,5 +90,6 @@ env_dict={`
`90`	`90`	`'CINN_SOURCE_DIR':'@CINN_SOURCE_DIR@',`
`91`	`91`	`'WITH_CPP_DIST':'@WITH_CPP_DIST@',`
`92`	`92`	`'PADDLE_INSTALL_DIR':'@PADDLE_INSTALL_DIR@',`
`93`		`- 'PADDLE_LIB_TEST_DIR':'@PADDLE_LIB_TEST_DIR@'`
	`93`	`+ 'PADDLE_LIB_TEST_DIR':'@PADDLE_LIB_TEST_DIR@',`
	`94`	`+ 'WITH_PIP_CUDA_LIBRARIES':'@WITH_PIP_CUDA_LIBRARIES@'`
`94`	`95`	`}`