[BUG] Optimize GPU error message file search path (#54180)

cmake/third_party.cmake

@@ -333,23 +333,21 @@ if(WITH_GPU)
     ${URL} "externalError" MD5 a712a49384e77ca216ad866712f7cafa
   )# download file externalErrorMsg.tar.gz
   if(WITH_TESTING)
-    # copy externalErrorMsg.pb, just for unittest can get error message correctly.
+    # copy externalErrorMsg.pb for UnitTest
     set(SRC_DIR ${THIRD_PARTY_PATH}/externalError/data)
-    if(WIN32 AND (NOT "${CMAKE_GENERATOR}" STREQUAL "Ninja"))
-      set(DST_DIR1
-          ${CMAKE_BINARY_DIR}/paddle/fluid/third_party/externalError/data)
-    else()
-      set(DST_DIR1 ${CMAKE_BINARY_DIR}/paddle/third_party/externalError/data)
-    endif()
-    set(DST_DIR2
+    # for python UT 'test_exception.py'
+    set(DST_DIR1
         ${CMAKE_BINARY_DIR}/python/paddle/include/third_party/externalError/data
     )
+    # for C++ UT 'enforce_test'
+    set(DST_DIR2 ${CMAKE_BINARY_DIR}/paddle/third_party/externalError/data)
     add_custom_command(
       TARGET download_externalError
       POST_BUILD
       COMMAND ${CMAKE_COMMAND} -E copy_directory ${SRC_DIR} ${DST_DIR1}
       COMMAND ${CMAKE_COMMAND} -E copy_directory ${SRC_DIR} ${DST_DIR2}
-      COMMENT "copy_directory from ${SRC_DIR} to ${DST_DIR}")
+      COMMENT "copy_directory from ${SRC_DIR} to ${DST_DIR1}"
+      COMMENT "copy_directory from ${SRC_DIR} to ${DST_DIR2}")
   endif()
 endif()
 

paddle/fluid/pybind/pybind.cc

@@ -1924,6 +1924,11 @@ All parameter, weight, gradient are variables in Paddle.
   m.def("_cuda_synchronize", [](const platform::CUDAPlace &place) {
     platform::DeviceContextPool::Instance().Get(place)->Wait();
   });
+  m.def("_test_enforce_gpu_success", []() {
+#if defined(PADDLE_WITH_CUDA)
+    PADDLE_ENFORCE_GPU_SUCCESS(cudaErrorInsufficientDriver);
+#endif
+  });
 
   m.def("get_float_stats", []() {
     std::vector<paddle::platform::ExportedStatValue<float>> float_stats;

paddle/phi/core/enforce.cc

@@ -270,30 +270,27 @@ std::string GetExternalErrorMsg(T status) {
   bool _initSucceed = false;
   phi::proto::ExternalErrorDesc externalError;
   if (externalError.ByteSizeLong() == 0) {
-    std::string filePath;
+    std::string search_path_1;
+    std::string search_path_2;
+    std::string search_path_3;
 #if !defined(_WIN32)
     Dl_info info;
     if (dladdr(reinterpret_cast<void*>(GetCurrentTraceBackString), &info)) {
-      std::string strModule(info.dli_fname);
-      const size_t last_slash_idx = strModule.find_last_of("/");
-      std::string compare_path = strModule.substr(strModule.length() - 6);
+      std::string phi_so_path(info.dli_fname);
+      const size_t last_slash_idx = phi_so_path.find_last_of("/");
       if (std::string::npos != last_slash_idx) {
-        strModule.erase(last_slash_idx, std::string::npos);
-      }
-      // TODO(lizhiyu02): I don't know what the 'compare_path.compare("avx.so")
-      // == 0' means, while
-      //  'compare_path.find("dist-packages") != std::string::npos' means that
-      //  after using 'pip install paddle'.
-      if (compare_path.compare("avx.so") == 0 ||
-          strModule.find("dist-packages") != std::string::npos) {
-        filePath =
-            strModule +
-            "/../include/third_party/externalError/data/externalErrorMsg.pb";
-      } else {
-        // Just for unittest
-        filePath = strModule +
-                   "/../third_party/externalError/data/externalErrorMsg.pb";
+        phi_so_path.erase(last_slash_idx, std::string::npos);
       }
+      // due to 'phi_so_path' may be 'site-packages/paddle/libs/libphi.so' or
+      // 'build/paddle/phi/libphi.so', we have different search path
+      search_path_1 =
+          phi_so_path +
+          "/../include/third_party/externalError/data/externalErrorMsg.pb";
+      search_path_2 = phi_so_path +
+                      "/../third_party/externalError/data/externalErrorMsg.pb";
+      search_path_3 =
+          phi_so_path +
+          "/../../third_party/externalError/data/externalErrorMsg.pb";
     }
 #else
     char buf[512];
@@ -303,24 +300,34 @@ std::string GetExternalErrorMsg(T status) {
             ? (HMODULE)mbi.AllocationBase
             : NULL;
     GetModuleFileName(h_module, buf, 512);
-    std::string strModule(buf);
-    const size_t last_slash_idx = strModule.find_last_of("\\");
-    std::string compare_path = strModule.substr(strModule.length() - 7);
+    std::string exe_path(buf);
+    const size_t last_slash_idx = exe_path.find_last_of("\\");
     if (std::string::npos != last_slash_idx) {
-      strModule.erase(last_slash_idx, std::string::npos);
-    }
-    if (strModule.find("dist-packages") != std::string::npos) {
-      filePath = strModule +
-                 "\\..\\include\\third_"
-                 "party\\externalerror\\data\\externalErrorMsg.pb";
-    } else {
-      filePath = strModule +
-                 "\\..\\..\\third_party"
-                 "\\externalerror\\data\\externalErrorMsg.pb";
+      exe_path.erase(last_slash_idx, std::string::npos);
     }
+    // due to 'exe_path' may be 'site-packages\\paddle\\fluid\\libpaddle.pyd' or
+    // 'build\\paddle\\fluid\\platform\\enforce_test.exe', we have different
+    // search path
+    search_path_1 =
+        exe_path +
+        "\\..\\include\\third_party\\externalError\\data\\externalErrorMsg.pb";
+    search_path_2 =
+        exe_path +
+        "\\..\\third_party\\externalError\\data\\externalErrorMsg.pb";
+    search_path_3 =
+        exe_path +
+        "\\..\\..\\third_party\\externalError\\data\\externalErrorMsg.pb";
 #endif
-    std::ifstream fin(filePath, std::ios::in | std::ios::binary);
+    std::ifstream fin(search_path_1, std::ios::in | std::ios::binary);
     _initSucceed = externalError.ParseFromIstream(&fin);
+    if (!_initSucceed) {
+      std::ifstream fin(search_path_2, std::ios::in | std::ios::binary);
+      _initSucceed = externalError.ParseFromIstream(&fin);
+    }
+    if (!_initSucceed) {
+      std::ifstream fin(search_path_3, std::ios::in | std::ios::binary);
+      _initSucceed = externalError.ParseFromIstream(&fin);
+    }
   }
   using __CUDA_STATUS_TYPE__ = decltype(status);
   phi::proto::ApiType proto_type =

python/paddle/fluid/core.py

@@ -298,6 +298,7 @@ try:
     from .libpaddle import _set_paddle_lib_path
     from .libpaddle import _create_loaded_parameter
     from .libpaddle import _cuda_synchronize
+    from .libpaddle import _test_enforce_gpu_success
     from .libpaddle import _is_compiled_with_heterps
     from .libpaddle import _promote_types_if_complex_exists
     from .libpaddle import _set_cached_executor_build_strategy

python/paddle/fluid/tests/unittests/test_exception.py

@@ -32,6 +32,23 @@ class TestException(unittest.TestCase):
 
         self.assertIsNotNone(exception)
 
+    def test_gpu_success(self):
+        if not paddle.is_compiled_with_cuda():
+            return
+
+        try:
+            core._test_enforce_gpu_success()
+        except Exception as e:
+            self.assertTrue(isinstance(e, OSError))
+            self.assertIn(
+                "CUDA error(35), CUDA driver version is insufficient for CUDA runtime version.",
+                str(e),
+            )
+            self.assertIn(
+                "[Hint: 'cudaErrorInsufficientDriver'. This indicates that the installed NVIDIA CUDA driver is older than the CUDA runtime library. This is not a supported configuration.Users should install an updated NVIDIA display driver to allow the application to run.]",
+                str(e),
+            )
+
 
 class TestExceptionNoCStack(unittest.TestCase):
     def setUp(self):