Not install cuda on windows (#64083)

xuxinyi389 · web-flow · commit 449fbdfd4b58 · 2024-05-29T19:13:00.000+08:00
* load all once

* fix 32 bits dll error

* fix

* fix conflict with cuda env

* fix_cudnn

* fix version

* close switch
diff --git a/paddle/common/flags.cc b/paddle/common/flags.cc
@@ -1724,3 +1724,7 @@ PHI_DEFINE_EXPORTED_string(cusolver_dir,  // NOLINT
 PHI_DEFINE_EXPORTED_string(cusparse_dir,  // NOLINT
                            "",
                            "Specify path for loading libcusparse.so.*.");
+PHI_DEFINE_EXPORTED_string(
+    win_cuda_bin_dir,  // NOLINT
+    "",
+    "Specify path for loading *.dll about cuda on windows");
diff --git a/paddle/phi/backends/dynload/dynamic_loader.cc b/paddle/phi/backends/dynload/dynamic_loader.cc
@@ -14,6 +14,7 @@ limitations under the License. */
 #include "paddle/phi/backends/dynload/dynamic_loader.h"
 #include <dirent.h>
 
+#include <codecvt>
 #include <cstdlib>
 #include <string>
 #include <vector>
@@ -45,6 +46,7 @@ COMMON_DECLARE_string(cusparselt_dir);
 COMMON_DECLARE_string(curand_dir);
 COMMON_DECLARE_string(cusolver_dir);
 COMMON_DECLARE_string(cusparse_dir);
+COMMON_DECLARE_string(win_cuda_bin_dir);
 #ifdef PADDLE_WITH_HIP
 
 PHI_DEFINE_string(miopen_dir,
@@ -132,8 +134,12 @@ static constexpr char* win_cufft_lib =
 
 static inline std::string join(const std::string& part1,
                                const std::string& part2) {
-  // directory separator
+// directory separator
+#if defined(_WIN32)
+  const char sep = '\\';
+#else
   const char sep = '/';
+#endif
   if (!part2.empty() && part2.front() == sep) {
     return part2;
   }
@@ -263,6 +269,26 @@ static inline void* GetDsoHandleFromSearchPath(
 #else
   int dynload_flags = 0;
 #endif  // !_WIN32
+#if defined(_WIN32)
+  std::vector<std::wstring> cuda_bin_search_path = {
+      L"cublas",
+      L"cuda_nvrtc",
+      L"cuda_runtime",
+      L"cudnn",
+      L"cufft",
+      L"curand",
+      L"cusolver",
+      L"cusparse",
+      L"nvjitlink",
+  };
+  for (auto search_path : cuda_bin_search_path) {
+    std::wstring_convert<std::codecvt_utf8_utf16<wchar_t>> converter;
+    std::wstring win_path_wstring =
+        converter.from_bytes(FLAGS_win_cuda_bin_dir);
+    search_path = win_path_wstring + L"\\" + search_path + L"\\bin";
+    AddDllDirectory(search_path.c_str());
+  }
+#endif
   std::vector<std::string> dso_names = split(dso_name, ";");
   void* dso_handle = nullptr;
   for (auto const& dso : dso_names) {
@@ -324,8 +350,26 @@ void* GetCublasDsoHandle() {
 #if defined(__APPLE__) || defined(__OSX__)
   return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "libcublas.dylib");
 #elif defined(_WIN32) && defined(PADDLE_WITH_CUDA)
-  return GetDsoHandleFromSearchPath(
-      FLAGS_cuda_dir, win_cublas_lib, true, {cuda_lib_path});
+  if (CUDA_VERSION >= 11000 && CUDA_VERSION < 12000) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
+    return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "cublas64_11.dll");
+#else
+    return GetDsoHandleFromSearchPath(
+        FLAGS_cuda_dir, win_cublas_lib, true, {cuda_lib_path});
+#endif
+  } else if (CUDA_VERSION >= 12000 && CUDA_VERSION <= 12030) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
+    return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "cublas64_12.dll");
+#else
+    return GetDsoHandleFromSearchPath(
+        FLAGS_cuda_dir, win_cublas_lib, true, {cuda_lib_path});
+#endif
+  } else {
+    std::string warning_msg(
+        "Your CUDA_VERSION is less than 11 or greater than 12, paddle "
+        "temporarily no longer supports");
+    return nullptr;
+  }
 #elif defined(__linux__) && defined(PADDLE_WITH_CUDA)
   if (CUDA_VERSION >= 11000 && CUDA_VERSION < 12000) {
 #ifdef WITH_PIP_CUDA_LIBRARIES
@@ -403,8 +447,23 @@ void* GetCUDNNDsoHandle() {
       "Toolkit\\CUDA\\v10.0\n"
       "You should do this according to your CUDA installation directory and "
       "CUDNN version.");
-  return GetDsoHandleFromSearchPath(
-      FLAGS_cudnn_dir, win_cudnn_lib, true, {cuda_lib_path}, win_warn_meg);
+  if (CUDA_VERSION >= 11000 && CUDA_VERSION < 12030) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
+    return GetDsoHandleFromSearchPath(
+        FLAGS_cuda_dir, "cudnn64_8.dll", true, {cuda_lib_path}, win_warn_meg);
+#else
+    return GetDsoHandleFromSearchPath(
+        FLAGS_cuda_dir, win_cudnn_lib, true, {cuda_lib_path}, win_warn_meg);
+#endif
+  } else if (CUDA_VERSION >= 12030) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
+    return GetDsoHandleFromSearchPath(
+        FLAGS_cuda_dir, "cudnn64_9.dll", true, {cuda_lib_path}, win_warn_meg);
+#else
+    return GetDsoHandleFromSearchPath(
+        FLAGS_cuda_dir, win_cudnn_lib, true, {cuda_lib_path}, win_warn_meg);
+#endif
+  }
 #elif defined(PADDLE_WITH_HIP)
   return GetDsoHandleFromSearchPath(FLAGS_miopen_dir, "libMIOpen.so", false);
 #else
@@ -461,8 +520,13 @@ void* GetCurandDsoHandle() {
 #if defined(__APPLE__) || defined(__OSX__)
   return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "libcurand.dylib");
 #elif defined(_WIN32) && defined(PADDLE_WITH_CUDA)
+#ifdef WITH_PIP_CUDA_LIBRARIES
+  return GetDsoHandleFromSearchPath(
+      FLAGS_cuda_dir, "curand64_10.dll", true, {cuda_lib_path});
+#else
   return GetDsoHandleFromSearchPath(
       FLAGS_cuda_dir, win_curand_lib, true, {cuda_lib_path});
+#endif
 #elif defined(PADDLE_WITH_HIP)
   return GetDsoHandleFromSearchPath(FLAGS_rocm_dir, "libhiprand.so");
 #else
@@ -500,8 +564,13 @@ void* GetCusolverDsoHandle() {
 #if defined(__APPLE__) || defined(__OSX__)
   return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "libcusolver.dylib");
 #elif defined(_WIN32) && defined(PADDLE_WITH_CUDA)
+#ifdef WITH_PIP_CUDA_LIBRARIES
+  return GetDsoHandleFromSearchPath(
+      FLAGS_cuda_dir, "cusolver64_11.dll", true, {cuda_lib_path});
+#else
   return GetDsoHandleFromSearchPath(
       FLAGS_cuda_dir, win_cusolver_lib, true, {cuda_lib_path});
+#endif
 #else
 #ifdef WITH_PIP_CUDA_LIBRARIES
   return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "libcusolver.so.11");
@@ -515,8 +584,26 @@ void* GetCusparseDsoHandle() {
 #if defined(__APPLE__) || defined(__OSX__)
   return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "libcusparse.dylib");
 #elif defined(_WIN32) && defined(PADDLE_WITH_CUDA)
-  return GetDsoHandleFromSearchPath(
-      FLAGS_cuda_dir, win_cusparse_lib, true, {cuda_lib_path});
+  if (CUDA_VERSION >= 11000 && CUDA_VERSION < 12000) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
+    return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "cusparse64_11.dll");
+#else
+    return GetDsoHandleFromSearchPath(
+        FLAGS_cuda_dir, win_cusparse_lib, true, {cuda_lib_path});
+#endif
+  } else if (CUDA_VERSION >= 12000 && CUDA_VERSION <= 12030) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
+    return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "cusparse64_12.dll");
+#else
+    return GetDsoHandleFromSearchPath(
+        FLAGS_cuda_dir, win_cusparse_lib, true, {cuda_lib_path});
+#endif
+  } else {
+    std::string warning_msg(
+        "Your CUDA_VERSION is less than 11 or greater than 12, paddle "
+        "temporarily no longer supports");
+    return nullptr;
+  }
 #elif defined(__linux__) && defined(PADDLE_WITH_CUDA)
   if (CUDA_VERSION >= 11000 && CUDA_VERSION < 12000) {
 #ifdef WITH_PIP_CUDA_LIBRARIES
@@ -709,8 +796,26 @@ void* GetCUFFTDsoHandle() {
     return nullptr;
   }
 #elif defined(_WIN32) && defined(PADDLE_WITH_CUDA)
-  return GetDsoHandleFromSearchPath(
-      FLAGS_cuda_dir, win_cufft_lib, true, {cuda_lib_path});
+  if (CUDA_VERSION >= 11000 && CUDA_VERSION < 12000) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
+    return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "cufft64_10.dll");
+#else
+    return GetDsoHandleFromSearchPath(
+        FLAGS_cuda_dir, win_cufft_lib, true, {cuda_lib_path});
+#endif
+  } else if (CUDA_VERSION >= 12000 && CUDA_VERSION <= 12030) {
+#ifdef WITH_PIP_CUDA_LIBRARIES
+    return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "cufft64_11.dll");
+#else
+    return GetDsoHandleFromSearchPath(
+        FLAGS_cuda_dir, win_cufft_lib, true, {cuda_lib_path});
+#endif
+  } else {
+    std::string warning_msg(
+        "Your CUDA_VERSION is less than 11 or greater than 12, paddle "
+        "temporarily no longer supports");
+    return nullptr;
+  }
 #else
   return GetDsoHandleFromSearchPath(FLAGS_cuda_dir, "libcufft.so");
 #endif
diff --git a/paddle/phi/common/port.cc b/paddle/phi/common/port.cc
@@ -18,7 +18,6 @@
 #include <memory>
 #include <stdexcept>
 #include <string>
-
 #include "glog/logging.h"
 
 #if !defined(_WIN32)
@@ -42,7 +41,14 @@ void *dlsym(void *handle, const char *symbol_name) {
 
 void *dlopen(const char *filename, int flag) {
   std::string file_name(filename);
-  HMODULE hModule = LoadLibrary(file_name.c_str());
+  HMODULE hModule = nullptr;
+#ifdef WITH_PIP_CUDA_LIBRARIES
+  hModule =
+      LoadLibraryEx(file_name.c_str(), NULL, LOAD_LIBRARY_SEARCH_DEFAULT_DIRS);
+#endif
+  if (!hModule) {
+    hModule = LoadLibrary(file_name.c_str());
+  }
   if (!hModule) {
     if (flag) {
       throw std::runtime_error(file_name + " not found.");
@@ -72,7 +78,7 @@ int gettimeofday(struct timeval *tp, void *tzp) {
 
   return (0);
 }
-#endif              // !_WIN32
+#endif  // !_WIN32
 
 void ExecShellCommand(const std::string &cmd, std::string *message) {
   std::array<char, 128> buffer = {};
diff --git a/python/paddle/__init__.py b/python/paddle/__init__.py
@@ -577,6 +577,7 @@
     if os.path.exists(cuh_file):
         os.environ.setdefault('runtime_include_dir', runtime_include_dir)
 
+
 if is_compiled_with_cuda():
     import os
     import platform
@@ -608,6 +609,118 @@
         cupti_dir_lib_path = package_dir + "/.." + "/nvidia/cuda_cupti/lib"
         set_flags({"FLAGS_cupti_dir": cupti_dir_lib_path})
 
+    elif (
+        platform.system() == 'Windows'
+        and platform.machine() in ('x86_64', 'AMD64')
+        and paddle.version.with_pip_cuda_libraries == 'ON'
+    ):
+        package_dir = os.path.dirname(os.path.abspath(__file__))
+        win_cuda_bin_path = package_dir + "\\.." + "\\nvidia"
+        set_flags({"FLAGS_win_cuda_bin_dir": win_cuda_bin_path})
+
+        import sys
+
+        if sys.platform == 'win32':
+            pfiles_path = os.getenv('ProgramFiles', 'C:\\Program Files')
+            py_dll_path = os.path.join(sys.exec_prefix, 'Library', 'bin')
+            th_dll_path = os.path.join(os.path.dirname(__file__), 'libs')
+            site_cuda_base_path = os.path.join(
+                os.path.dirname(__file__), '..', 'nvidia'
+            )
+            site_cuda_list = [
+                "cublas",
+                "cuda_nvrtc",
+                "cuda_runtime",
+                "cudnn",
+                "cufft",
+                "curand",
+                "cusolver",
+                "cusparse",
+                "nvjitlink",
+            ]
+
+            if sys.exec_prefix != sys.base_exec_prefix:
+                base_py_dll_path = os.path.join(
+                    sys.base_exec_prefix, 'Library', 'bin'
+                )
+            else:
+                base_py_dll_path = ''
+
+            dll_paths = list(
+                filter(
+                    os.path.exists, [th_dll_path, py_dll_path, base_py_dll_path]
+                )
+            )
+            for site_cuda_package in site_cuda_list:
+                site_cuda_path = os.path.join(
+                    site_cuda_base_path, site_cuda_package, 'bin'
+                )
+                if os.path.exists(site_cuda_path):
+                    dll_paths.append(site_cuda_path)
+
+            import ctypes
+
+            kernel32 = ctypes.WinDLL('kernel32.dll', use_last_error=True)
+            with_load_library_flags = hasattr(kernel32, 'AddDllDirectory')
+            prev_error_mode = kernel32.SetErrorMode(0x0001)
+
+            kernel32.LoadLibraryW.restype = ctypes.c_void_p
+            if with_load_library_flags:
+                kernel32.LoadLibraryExW.restype = ctypes.c_void_p
+
+            for dll_path in dll_paths:
+                os.add_dll_directory(dll_path)
+
+            try:
+                ctypes.CDLL('vcruntime140.dll')
+                ctypes.CDLL('msvcp140.dll')
+                ctypes.CDLL('vcruntime140_1.dll')
+            except OSError:
+                print(
+                    '''Microsoft Visual C++ Redistributable is not installed, this may lead to the DLL load failure.
+                        It can be downloaded at https://aka.ms/vs/16/release/vc_redist.x64.exe'''
+                )
+            import glob
+
+            dlls = glob.glob(os.path.join(th_dll_path, '*.dll'))
+            for site_cuda_package in site_cuda_list:
+                site_cuda_path = os.path.join(
+                    site_cuda_base_path, site_cuda_package, 'bin'
+                )
+                if os.path.exists(site_cuda_path):
+                    dlls.extend(
+                        glob.glob(os.path.join(site_cuda_path, '*.dll'))
+                    )
+            # Not load 32 bit dlls in 64 bit python.
+            dlls = [dll for dll in dlls if '32_' not in dll]
+            path_patched = False
+            for dll in dlls:
+                is_loaded = False
+                print("dll:", dll)
+                if with_load_library_flags:
+                    res = kernel32.LoadLibraryExW(dll, None, 0x00001100)
+                    last_error = ctypes.get_last_error()
+                    if res is None and last_error != 126:
+                        err = ctypes.WinError(last_error)
+                        err.strerror += f' Error loading "{dll}" or one of its dependencies.'
+                        raise err
+                    elif res is not None:
+                        is_loaded = True
+                if not is_loaded:
+                    if not path_patched:
+                        prev_path = os.environ['PATH']
+                        os.environ['PATH'] = ';'.join(
+                            dll_paths + [os.environ['PATH']]
+                        )
+                        path_patched = True
+                    res = kernel32.LoadLibraryW(dll)
+                    if path_patched:
+                        os.environ['PATH'] = prev_path
+                    if res is None:
+                        err = ctypes.WinError(ctypes.get_last_error())
+                        err.strerror += f' Error loading "{dll}" or one of its dependencies.'
+                        raise err
+            kernel32.SetErrorMode(prev_error_mode)
 
 disable_static()
 
diff --git a/python/setup.py.in b/python/setup.py.in
diff --git a/setup.py b/setup.py
