Fix gpu lazy opt warning

theano/sandbox/cuda/opt.py

@@ -497,12 +497,19 @@ def local_gpu_lazy_ifelse(node):
             # Should not happen, but just in case
             if isinstance(c.type, CudaNdarrayType):
                 c = host_from_gpu(c)
+            if all([isinstance(o.type, CudaNdarrayType) or o.dtype != 'float32'
+                    for o in outs]):
+                return
 
             for i in range(len(outs)):
-                if not isinstance(outs[i], CudaNdarrayType):
+                if (not isinstance(outs[i].type, CudaNdarrayType) and
+                        outs[i].dtype == 'float32'):
                     outs[i] = gpu_from_host(outs[i])
-            return [host_from_gpu(out) for out in
-                    gpu_ifelse.make_node(c, *outs).outputs]
+            outs = gpu_ifelse(c, *outs, return_list=True)
+            for i in range(len(outs)):
+                if isinstance(outs[i].type, CudaNdarrayType):
+                    outs[i] = host_from_gpu(outs[i])
+            return outs
 
     if isinstance(node.op, GpuFromHost):
         host_input = node.inputs[0]
@@ -522,11 +529,14 @@ def local_gpu_lazy_ifelse(node):
             # Should not happen, but just in case
             if isinstance(c.type, CudaNdarrayType):
                 c = host_from_gpu(c)
+            if all([isinstance(o.type, CudaNdarrayType) or o.dtype != 'float32'
+                    for o in outs]):
+                return
 
             for i in range(len(outs)):
-                if not isinstance(outs[i], CudaNdarrayType):
+                if (not isinstance(outs[i].type, CudaNdarrayType) and
+                        outs[i].dtype == 'float32'):
                     outs[i] = gpu_from_host(outs[i])
-
             outs = gpu_ifelse.make_node(c, *outs).outputs
             return outs
 

theano/tests/test_ifelse.py

@@ -51,6 +51,33 @@ class test_ifelse(unittest.TestCase, utt.TestOptimizationMixin):
         assert numpy.allclose(vx, f(1, vx, vy))
         assert numpy.allclose(vy, f(0, vx, vy))
 
+    def test_mixed_dtype(self):
+        x1 = tensor.vector('x1', dtype='int32')
+        x2 = tensor.vector('x2', dtype=self.dtype)
+        y1 = tensor.vector('y1', dtype='int32')
+        y2 = tensor.vector('y2', dtype=self.dtype)
+        c = tensor.iscalar('c')
+        f = theano.function([c, x1, x2, y1, y2],
+                            ifelse(c, (x1, x2), (y1, y2)), mode=self.mode)
+        self.assertFunctionContains1(f, self.get_ifelse(2))
+        rng = numpy.random.RandomState(utt.fetch_seed())
+
+        xlen = rng.randint(200)
+        ylen = rng.randint(200)
+
+        vx1 = numpy.asarray(rng.uniform(size=(xlen,))*3, 'int32')
+        vx2 = numpy.asarray(rng.uniform(size=(xlen,)), self.dtype)
+        vy1 = numpy.asarray(rng.uniform(size=(ylen,))*3, 'int32')
+        vy2 = numpy.asarray(rng.uniform(size=(ylen,)), self.dtype)
+
+        o1, o2 = f(1, vx1, vx2, vy1, vy2)
+        assert numpy.allclose(vx1, o1)
+        assert numpy.allclose(vx2, o2)
+
+        o1, o2 = f(0, vx1, vx2, vy1, vy2)
+        assert numpy.allclose(vy1, o1)
+        assert numpy.allclose(vy2, o2)
+
     def test_lazy_if_on_generics(self):
         x = theano.generic()
         y = theano.generic()