nvcc_compile now links against cuda_ndarray.so using -lcuda_ndarray

theano/sandbox/cuda/__init__.py

@@ -71,6 +71,17 @@ date = max(stat_times)
 cuda_ndarray_loc = os.path.join(config.compiledir, 'cuda_ndarray')
 cuda_ndarray_so = os.path.join(cuda_ndarray_loc,
                                'cuda_ndarray.' + get_lib_extension())
+libcuda_ndarray_so = os.path.join(cuda_ndarray_loc,
+                               'libcuda_ndarray.' + get_lib_extension())
+
+
+# Add the theano cache directory's cuda_ndarray subdirectory to the list of
+# places that are hard-coded into compiled modules' runtime library search
+# list.  This works in conjunction with nvcc_compiler.nvcc_module_compile_str
+# which adds this folder during compilation with -L and also adds -lcuda_ndarray
+# when compiling modules.
+nvcc_compiler.add_standard_rpath(cuda_ndarray_loc)
+
 compile_cuda_ndarray = True
 
 if os.path.exists(cuda_ndarray_so):
@@ -106,7 +117,14 @@ try:
                     code,
                     location=cuda_ndarray_loc,
                     include_dirs=[cuda_path], libs=['cublas'])
-
+            # If necessary, 
+            # create a symlink called libcuda_ndarray.so
+            # which nvcc_module_compile_str uses when linking
+            # any module except "cuda_ndarray" itself.
+            try:
+                open(libcuda_ndarray_so).close()
+            except IOError:
+                os.symlink(cuda_ndarray_so, libcuda_ndarray_so)
             from cuda_ndarray.cuda_ndarray import *
 except Exception, e:
     error( "Failed to compile cuda_ndarray.cu: %s" % str(e))

theano/sandbox/cuda/nvcc_compiler.py

@@ -191,10 +191,10 @@ def nvcc_module_compile_str(
     cmd.extend('-I%s'%idir for idir in include_dirs)
     cmd.extend(['-o',lib_filename])
     cmd.append(os.path.split(cppfilename)[-1])
-    if module_name != 'cuda_ndarray':
-        cmd.append(os.path.join(os.path.split(cppfilename)[0],'..','cuda_ndarray','cuda_ndarray.'+get_lib_extension()))
     cmd.extend(['-L%s'%ldir for ldir in lib_dirs])
     cmd.extend(['-l%s'%l for l in libs])
+    if module_name != 'cuda_ndarray':
+        cmd.append("-lcuda_ndarray")
     if sys.platform == 'darwin':
         cmd.extend(darwin_python_lib.split())