local opt: Alloc(Alloc(m, x, 1), x, y) -> Alloc(m, x, y)

theano/tensor/opt.py

@@ -5887,3 +5887,32 @@ register_canonicalize(gof.OpRemove(theano.gradient.disconnected_grad_),
 def local_grad_clip(node):
     if isinstance(node.op, theano.gradient.GradClip):
         return node.inputs
+
+
+@register_canonicalize
+@register_stabilize
+@register_specialize
+@gof.local_optimizer([T.Alloc])
+def local_merge_alloc(node):
+    # This opt takes care of several cases:
+    # Alloc(Alloc(m, x, 1, 1, 1), x, y, z, w) -> Alloc(m, x, y, z, w)
+    # Alloc(Alloc(m, y, 1, 1), x, y, z, w) -> Alloc(m, x, y, z, w)
+    if not isinstance(node.op, T.Alloc):
+        return False
+    if not node.inputs[0].owner or not isinstance(
+            node.inputs[0].owner.op, T.Alloc):
+        return False
+    inputs_outer = node.inputs
+    inputs_inner = node.inputs[0].owner.inputs
+    dims_outer = inputs_outer[1:]
+    dims_inner = inputs_inner[1:]
+    dims_outer_rev = dims_outer[::-1]
+    dims_inner_rev = dims_inner[::-1]
+    # check if the pattern of broadcasting is matched, in the reversed ordering
+    for dim_inner, dim_outer in zip(dims_inner_rev, dims_outer_rev):
+        if dim_inner != dim_outer:
+            if isinstance(dim_inner, Constant) and dim_inner.data == 1:
+                pass
+            else:
+                return False
+    return [T.alloc(inputs_inner[0], *dims_outer)]

theano/tensor/tests/test_opt.py

@@ -5443,6 +5443,32 @@ class TestIntDivByOne(unittest.TestCase):
         assert len(divs) == 0
 
 
+def test_local_merge_alloc():
+    x = T.iscalar('x')
+    y = T.iscalar('y')
+    z = T.iscalar('z')
+    w = T.iscalar('w')
+    m = T.fscalar('m')
+    # case 1
+    # Alloc(Alloc(m, x, 1, 1, 1), x, y, z, w) -> Alloc(m, x, y, z, w)
+    output = T.alloc(T.alloc(m, 1, y, 1, 1), x, y, z, w)
+    f = theano.function([m, x, y, z, w], output)
+    topo = f.maker.fgraph.toposort()
+    assert len(topo) == 1
+    assert isinstance(topo[0].op, T.Alloc)
+    o = f(0., 1, 2, 3, 4)
+    assert o.shape == (1, 2, 3, 4)
+
+    # case 2
+    # Alloc(Alloc(m, y, 1, 1), x, y, z, w) -> Alloc(m, x, y, z, w)
+    output = T.alloc(T.alloc(m, y, 1, 1), x, y, z, w)
+    f = theano.function([m, x, y, z, w], output)
+    topo = f.maker.fgraph.toposort()
+    assert len(topo) == 1
+    assert isinstance(topo[0].op, T.Alloc)
+    o = f(0., 1, 2, 3, 4)
+    assert o.shape == (1, 2, 3, 4)
+
 if __name__ == '__main__':
     t = TestMakeVector('setUp')
     t.setUp()