Fix error in Advanced indexing with > 2D

theano/gpuarray/subtensor.py

@@ -552,7 +552,7 @@ class GpuAdvancedSubtensor(HideC, tensor.AdvancedSubtensor):
         # build the strides
         strides = [1]
         for i in range(p - 1, 0, -1):
-            stride = x.shape[i] * strides[-1]
+            stride = x.shape[i] * strides[0]
             strides.insert(0, stride)
 
         # build the indices and use it

theano/tensor/tests/test_subtensor.py

@@ -1603,6 +1603,26 @@ class TestAdvancedSubtensor(unittest.TestCase):
         cmd = f2(0, 1, 2) == aa[[0, 1, 2], :, 0:2]
         self.assertTrue(cmd.all())
 
+    def test_adv_sub_3d(self):
+        # Reported in https://github.com/Theano/Theano/issues/5674
+        X = tensor.tensor3("X")
+
+        xx = numpy.zeros((3, 2, 2), config.floatX)
+        for i in range(3):
+            for j in range(2):
+                for k in range(2):
+                    xx[i, j, k] = 100 * i + 10 * j + k
+
+        b_idx = numpy.zeros((2, 2), 'int32')
+        b_idx[0, 1] = 1
+        b_idx[1, 1] = 2
+
+        r_idx = numpy.arange(xx.shape[1])[numpy.newaxis, :]
+        c_idx = numpy.arange(xx.shape[2])[numpy.newaxis, :]
+
+        out = X[b_idx, r_idx, c_idx].eval({X: xx})
+        utt.assert_allclose(out, xx[b_idx, r_idx, c_idx])
+
     def test_grad(self):
         ones = numpy.ones((1, 3), dtype=self.dtype)
         n = self.shared(ones * 5, broadcastable=(True, False))