Use size_t for shape

theano/sandbox/cuda/basic_ops.py

@@ -3342,13 +3342,15 @@ class GpuAlloc(GpuOp):
         }
         if (%(memset_0)s && CudaNdarray_is_c_contiguous(%(out)s))
         {
-            if (cudaSuccess != cudaMemset(%(out)s->devdata, 0,
-                                          CudaNdarray_SIZE(%(out)s) * 4))
+            cudaError_t err = cudaMemset(%(out)s->devdata, 0,
+                                         CudaNdarray_SIZEt(%(out)s) * 4);
+            if (cudaSuccess != err)
             {
                 PyErr_Format(PyExc_MemoryError,
-                             "GpuAlloc: Error memsetting %%d"
-                             " bytes of device memory.",
-                             CudaNdarray_SIZE(%(out)s) * 4);
+                             "GpuAlloc: Error memsetting %%ld"
+                             " bytes of device memory. %%s",
+                             (long)(CudaNdarray_SIZEt(%(out)s) * 4),
+                             cudaGetErrorString(err));
                 Py_XDECREF(%(out)s);
                 %(out)s = NULL;
                 %(fail)s;
@@ -3372,7 +3374,7 @@ class GpuAlloc(GpuOp):
         return [None for i in inputs]
 
     def c_code_cache_version(self):
-        return (7,)
+        return (9,)
 
     def do_constant_folding(self, node):
         for client in node.outputs[0].clients:

theano/sandbox/cuda/cuda_ndarray.cu

@@ -5117,6 +5117,17 @@ CudaNdarray_SIZE(const CudaNdarray *self)
     }
     return size;
 }
+size_t
+CudaNdarray_SIZEt(const CudaNdarray *self)
+{
+    if (self->nd == -1) return 0;
+    size_t size = 1;
+    for (int i = 0; i < self->nd; ++i)
+    {
+        size *= CudaNdarray_HOST_DIMS(self)[i];
+    }
+    return size;
+}
 
 PyObject *
 CudaNdarray_SIZE_Object(const CudaNdarray *self, void *closure)

theano/sandbox/cuda/cuda_ndarray.cuh

@@ -286,6 +286,9 @@ DllExport float *CudaNdarray_DEV_DATA(const CudaNdarray * self);
  * Return the number of elements in the ndarray (product of the dimensions)
  */
 DllExport int CudaNdarray_SIZE(const CudaNdarray *self);
+// Useful as many cuda function use size_t as input. This make sure we use the
+// most precission and not int.
+DllExport size_t CudaNdarray_SIZEt(const CudaNdarray *self);
 
 static PyObject *CudaNdarray_SIZE_Object(const CudaNdarray *self, void *closure);