Ensure that indices are suitable for numpy.take.

theano/misc/safe_asarray.py

@@ -39,16 +39,18 @@ def _asarray(a, dtype, order=None):
     if rval.dtype.num != dtype.num:
         # Type mismatch between the data type we asked for, and the one
         # returned by numpy.asarray.
-        if (dtype.num == numpy.dtype(numpy.int32).num or
-                dtype.num == numpy.dtype(numpy.int64).num):
+        # If both types have the same string description (byte order, basic
+        # type, and number of bytes), then it is safe to return a view.
+        if (dtype.str == rval.dtype.str):
             # Silent fix.
             return rval.view(dtype=dtype)
         else:
             # Unexpected mismatch: better know what is going on!
             raise TypeError('numpy.array did not return the data type we '
-                    'asked for (%s #%s), instead it returned type %s #%s: function '
-                    'theano._asarray may need to be extended to handle this '
-                    'data type as well.' %
-                    (dtype, dtype.num, rval.dtype, rval.dtype.num))
+                    'asked for (%s %s #%s), instead it returned type '
+                    '%s %s #%s: function '
+                    'theano._asarray may need to be modified to handle this '
+                    'data type.' %
+                    (dtype, dtype.str, dtype.num, rval.dtype, rval.str, rval.dtype.num))
     else:
         return rval

theano/tensor/basic.py

@@ -4909,8 +4909,13 @@ class AdvancedSubtensor1(Op):
         else:
             o = None
 
-        # I have read that using clip or wrap mode make it a lot faster when
-        # the output is provided
+        # If i.dtype is more precise than numpy.intc (int32 on 32-bit machines,
+        # int64 on 64-bit machines), numpy may raise the following error:
+        # TypeError: array cannot be safely cast to required type.
+        # Since we will probably not have an array with more than 2**31 items
+        # on a 32-bit arch, I suppose it is safe to cast i into intc.
+        i = theano._asarray(i, dtype=numpy.intc)
+
         out[0] = x.take(i, axis=0, out=o)
 
     def grad(self, inputs, grads):