fix nvenc on win32:

* use WinDLL to avoid calling convention error "called with not enough arguments"
* "libcuda.dll" is now "nvcuda.dll"..
* ship the cuda compiled kernels into a /CUDA directory

git-svn-id: https://xpra.org/svn/Xpra/trunk@9999 3bb7dfac-3a0b-4e04-842a-767bc560f471

src/setup.py

@@ -1926,8 +1926,10 @@ def cython_add(*args, **kwargs):
             print(stdout or "")
             print(stderr or "")
             sys.exit(1)
-    add_data_files("share/xpra/cuda",
-                   ["xpra/codecs/cuda_common/%s.fatbin" % x for x in kernels])
+    CUDA_BIN = "share/xpra/cuda"
+    if WIN32:
+        CUDA_BIN = "CUDA"
+    add_data_files(CUDA_BIN, ["xpra/codecs/cuda_common/%s.fatbin" % x for x in kernels])
     for nvenc_version, _nvenc_version_enabled in {4 : nvenc4_ENABLED,
                                                   5 : nvenc5_ENABLED}.items():
         if not _nvenc_version_enabled:

src/xpra/codecs/cuda_common/cuda_context.py

@@ -105,7 +105,8 @@ def init_all_devices():
                 if i==0:
                     #we print the list info "header" from inside the loop
                     #so that the log output is bunched up together
-                    log.info("CUDA %s / PyCUDA %s, found %s device(s):", ".".join([str(x) for x in driver.get_version()]), pycuda.VERSION_TEXT, ngpus)
+                    log.info("CUDA %s / PyCUDA %s, found %s device%s:",
+                             ".".join([str(x) for x in driver.get_version()]), pycuda.VERSION_TEXT, ngpus, ["","s"][int(ngpus!=1)])
                 DEVICES.append(i)
                 log.info("  + %s (memory: %s%% free, compute: %s.%s)", device_info(device), 100*free/total, SMmajor, SMminor)
             finally:

src/xpra/codecs/nvenc4/encoder.pyx

@@ -798,17 +798,19 @@ cuCtxGetCurrent = None
 def init_nvencode_library():
     global NvEncodeAPICreateInstance, cuCtxGetCurrent
     IF NV_WINDOWS:
+        load = ctypes.WinDLL
         nvenc_libname = "nvencodeapi.dll"
-        cuda_libname = "libcuda.dll"
+        cuda_libname = "nvcuda.dll"
     ELSE:
         #assert os.name=="posix"
+        load = ctypes.cdll.LoadLibrary
         nvenc_libname = "libnvidia-encode.so"
         cuda_libname = "libcuda.so"
     #CUDA:
     log("init_nvencode_library() will try to load %s", cuda_libname)
     try:
-        x = ctypes.cdll.LoadLibrary(cuda_libname)
-        log("init_nvencode_library() %s=%s", cuda_libname, x)
+        x = load(cuda_libname)
+        log("init_nvencode_library() %s(%s)=%s", load, cuda_libname, x)
     except Exception as e:
         raise Exception("nvenc: the required library %s cannot be loaded: %s" % (cuda_libname, e))
     cuCtxGetCurrent = x.cuCtxGetCurrent
@@ -818,8 +820,8 @@ def init_nvencode_library():
     #nvidia-encode:
     log("init_nvencode_library() will try to load %s", nvenc_libname)
     try:
-        x = ctypes.cdll.LoadLibrary(nvenc_libname)
-        log("init_nvencode_library() %s=%s", nvenc_libname, x)
+        x = load(nvenc_libname)
+        log("init_nvencode_library() %s(%s)=%s", nvenc_libname, x)
     except Exception as e:
         raise Exception("nvenc: the required library %s cannot be loaded: %s" % (nvenc_libname, e))
     NvEncodeAPICreateInstance = x.NvEncodeAPICreateInstance

src/xpra/codecs/nvenc5/encoder.pyx

@@ -795,17 +795,19 @@ cuCtxGetCurrent = None
 def init_nvencode_library():
     global NvEncodeAPICreateInstance, cuCtxGetCurrent
     IF NV_WINDOWS:
+        load = ctypes.WinDLL
         nvenc_libname = "nvencodeapi.dll"
-        cuda_libname = "libcuda.dll"
+        cuda_libname = "nvcuda.dll"
     ELSE:
         #assert os.name=="posix"
+        load = ctypes.cdll.LoadLibrary
         nvenc_libname = "libnvidia-encode.so"
         cuda_libname = "libcuda.so"
     #CUDA:
     log("init_nvencode_library() will try to load %s", cuda_libname)
     try:
-        x = ctypes.cdll.LoadLibrary(cuda_libname)
-        log("init_nvencode_library() %s=%s", cuda_libname, x)
+        x = load(cuda_libname)
+        log("init_nvencode_library() %s(%s)=%s", load, cuda_libname, x)
     except Exception as e:
         raise Exception("nvenc: the required library %s cannot be loaded: %s" % (cuda_libname, e))
     cuCtxGetCurrent = x.cuCtxGetCurrent
@@ -815,8 +817,8 @@ def init_nvencode_library():
     #nvidia-encode:
     log("init_nvencode_library() will try to load %s", nvenc_libname)
     try:
-        x = ctypes.cdll.LoadLibrary(nvenc_libname)
-        log("init_nvencode_library() %s=%s", nvenc_libname, x)
+        x = load(nvenc_libname)
+        log("init_nvencode_library() %s(%s)=%s", load, nvenc_libname, x)
     except Exception as e:
         raise Exception("nvenc: the required library %s cannot be loaded: %s" % (nvenc_libname, e))
     NvEncodeAPICreateInstance = x.NvEncodeAPICreateInstance