Merge pull request #5502 from ReyhaneAskari/io_toposort_5042

doc/tutorial/shape_info.txt

@@ -53,9 +53,9 @@ can lead to errors. Consider this example:
 >>> theano.printing.debugprint(f) # doctest: +NORMALIZE_WHITESPACE
 MakeVector{dtype='int64'} [id A] ''   4
  |Elemwise{Add}[(0, 0)] [id B] ''   3
- | |Shape_i{0} [id C] ''   1
+ | |Shape_i{0} [id C] ''   2
  | | |x [id D]
- | |Shape_i{0} [id E] ''   2
+ | |Shape_i{0} [id E] ''   1
  |   |y [id F]
  |Shape_i{1} [id G] ''   0
    |x [id D]

theano/compile/tests/test_profiling.py

@@ -56,8 +56,8 @@ class Test_profiling(unittest.TestCase):
             lines1 = [l for l in the_string.split("\n") if "Max if linker" in l]
             lines2 = [l for l in the_string.split("\n") if "Minimum peak" in l]
             if theano.config.device == 'cpu':
-                assert "CPU: 4112KB (8204KB)" in the_string, (lines1, lines2)
-                assert "CPU: 8204KB (12296KB)" in the_string, (lines1, lines2)
+                assert "CPU: 4112KB (4104KB)" in the_string, (lines1, lines2)
+                assert "CPU: 8204KB (8196KB)" in the_string, (lines1, lines2)
                 assert "CPU: 8208KB" in the_string, (lines1, lines2)
                 assert "Minimum peak from all valid apply node order is 4104KB" in the_string, (
                     lines1, lines2)

theano/gof/graph.py

@@ -608,6 +608,8 @@ def stack_search(start, expand, mode='bfs', build_inv=False):
     expand : callable
         When we get to a node, add expand(node) to the list of nodes to visit.
         This function should return a list, or None.
+    mode : string
+        'bfs' or 'dfs' for breath first search or depth first search.
 
     Returns
     -------
@@ -632,7 +634,7 @@ def stack_search(start, expand, mode='bfs', build_inv=False):
         start_pop = start.popleft
     else:
         start_pop = start.pop
-    expand_inv = {}
+    expand_inv = {}  # var: clients
     while start:
         l = start_pop()
         if id(l) not in rval_set:
@@ -878,7 +880,7 @@ def clone_get_equiv(inputs, outputs, copy_inputs_and_orphans=True, memo=None):
     return memo
 
 
-def general_toposort(r_out, deps, debug_print=False,
+def general_toposort(outputs, deps, debug_print=False,
                      compute_deps_cache=None, deps_cache=None,
                      clients=None):
     """
@@ -932,9 +934,9 @@ def general_toposort(r_out, deps, debug_print=False,
                 return deps_cache[io]
     assert deps_cache is not None
 
-    assert isinstance(r_out, (tuple, list, deque))
+    assert isinstance(outputs, (tuple, list, deque))
 
-    reachable, _clients = stack_search(deque(r_out), compute_deps_cache,
+    reachable, _clients = stack_search(deque(outputs), compute_deps_cache,
                                        'dfs', True)
     if clients is not None:
         clients.update(_clients)
@@ -948,9 +950,9 @@ def general_toposort(r_out, deps, debug_print=False,
             rlist.append(node)
             rset.add(node)
             for client in _clients.get(node, []):
-                deps_cache[client] = [a for a in deps_cache[client]
-                                      if a is not node]
-                if not deps_cache[client]:
+                d = [a for a in deps_cache[client] if a is not node]
+                deps_cache[client] = d
+                if not d:
                     sources.append(client)
 
     if len(rlist) != len(reachable):
@@ -980,17 +982,37 @@ def io_toposort(inputs, outputs, orderings=None, clients=None):
         node->clients for each node in the subgraph that is sorted
 
     """
-    # the inputs are used only here in the function that decides what 'predecessors' to explore
-    iset = set(inputs)
-
-    # We build 2 functions as a speed up
-    deps_cache = {}
+    if not orderings and clients is None:  # ordering can be None or empty dict
+        # Specialized function that is faster when more then ~10 nodes
+        # when no ordering.
+
+        # Do a new stack implementation with the vm algo.
+        # This will change the order returned.
+        computed = set(inputs)
+        todo = [o.owner for o in reversed(outputs) if o.owner]
+        order = []
+        while todo:
+            cur = todo.pop()
+            # We suppose that all outputs are always computed
+            if cur.outputs[0] in computed:
+                continue
+            if all([i in computed or i.owner is None for i in cur.inputs]):
+                computed.update(cur.outputs)
+                order.append(cur)
+            else:
+                todo.append(cur)
+                todo.extend(i.owner for i in cur.inputs if i.owner)
+        return order
 
     compute_deps = None
     compute_deps_cache = None
-    if not orderings:  # can be None or empty dict
+    iset = set(inputs)
+    deps_cache = {}
+
+    if not orderings:  # ordering can be None or empty dict
         # Specialized function that is faster when no ordering.
         # Also include the cache in the function itself for speed up.
+
         def compute_deps_cache(obj):
             if obj in deps_cache:
                 return deps_cache[obj]
@@ -1013,6 +1035,9 @@ def io_toposort(inputs, outputs, orderings=None, clients=None):
                 deps_cache[obj] = rval
             return rval
     else:
+
+        # the inputs are used only here in the function that decides what
+        # 'predecessors' to explore
         def compute_deps(obj):
             rval = []
             if obj not in iset:
@@ -1023,7 +1048,7 @@ def io_toposort(inputs, outputs, orderings=None, clients=None):
                     rval = list(obj.inputs)
                 rval.extend(orderings.get(obj, []))
             else:
-                assert not orderings.get(obj, [])
+                assert not orderings.get(obj, None)
             return rval
 
     topo = general_toposort(outputs, deps=compute_deps,

theano/gof/tests/test_graph.py

@@ -212,7 +212,7 @@ class TestToposort:
         o0 = MyOp.make_node(r1, r2)
         o1 = MyOp.make_node(r3, r4)
         all = io_toposort([r1, r2, r3, r4], o0.outputs + o1.outputs)
-        assert all == [o1, o0]
+        assert all == [o1, o0] or all == [o0, o1]
 
     def test_4(self):
         """Test inputs and outputs mixed together in a chain graph"""

theano/gof/tests/test_link.py

@@ -153,7 +153,7 @@ class TestWrapLinker(unittest.TestCase):
         i[0].data = 1
         i[1].data = 2
         fn()
-        assert nodes == [div, add, mul]
+        assert nodes == [div, add, mul] or nodes == [add, div, mul]
         assert o[0].data is None
 
     def test_1(self):
@@ -171,7 +171,7 @@ class TestWrapLinker(unittest.TestCase):
         i[0].data = 1
         i[1].data = 2
         fn()
-        assert nodes == [div, add, mul]
+        assert nodes == [div, add, mul] or nodes == [add, div, mul]
         assert o[0].data == 1.5
 
 

theano/sparse/tests/test_basic.py

@@ -1572,12 +1572,13 @@ class UsmmTests(unittest.TestCase):
                 # Usmm is tested at the same time in debugmode
                 # Check if the optimization local_usmm and local_usmm_csx is
                 # applied
-                assert isinstance(topo[0].op,
-                                  theano.sparse.basic.CSMProperties)
-                assert isinstance(topo[1].op, theano.tensor.DimShuffle)
-                assert isinstance(topo[2].op, theano.tensor.Subtensor)
-                assert topo[3].op == theano.tensor.neg
-                assert isinstance(topo[4].op, UsmmCscDense)
+                def check_once(x):
+                    assert sum([isinstance(n.op, x) for n in topo]) == 1
+                check_once(theano.sparse.basic.CSMProperties)
+                check_once(theano.tensor.DimShuffle)
+                check_once(theano.tensor.Subtensor)
+                check_once(UsmmCscDense)
+                check_once(theano.tensor.Elemwise)
                 if inplace:
                     assert topo[4].op.inplace
             elif not fast_compile:

theano/tensor/elemwise.py

@@ -1629,7 +1629,7 @@ for(int i=0;i<PyArray_NDIM(%(iname)s);i++){
 
     def c_code_cache_version_apply(self, node):
         # the version corresponding to the c code in this Op
-        version = [6]
+        version = [7]
 
         # now we insert versions for the ops on which we depend...
         scalar_node = Apply(

theano/tensor/elemwise_cgen.py

@@ -100,13 +100,13 @@ def make_checks(loop_orders, dtypes, sub):
             check += """
             if (%%(lv%(j0)s)s_n%(x0)s != %%(lv%(j)s)s_n%(x)s)
             {
-                PyErr_Format(PyExc_ValueError, "Input dimension mis-match. (input[%%%%i].shape[%%%%i] = %%%%i, input[%%%%i].shape[%%%%i] = %%%%i)",
+                PyErr_Format(PyExc_ValueError, "Input dimension mis-match. (input[%%%%i].shape[%%%%i] = %%%%lli, input[%%%%i].shape[%%%%i] = %%%%lli)",
                    %(j0)s,
                    %(x0)s,
-                   %%(lv%(j0)s)s_n%(x0)s,
+                   (long long int) %%(lv%(j0)s)s_n%(x0)s,
                    %(j)s,
                    %(x)s,
-                   %%(lv%(j)s)s_n%(x)s
+                   (long long int) %%(lv%(j)s)s_n%(x)s
                 );
                 %%(fail)s
             }

theano/tensor/nnet/tests/test_sigm.py

@@ -256,8 +256,10 @@ class T_sigmoid_opts(unittest.TestCase):
             [x, y],
             (sigmoid(x) * sigmoid(-y) * -tensor.exp(-x) *
                 tensor.exp(x * y) * tensor.exp(y)), mode=m)
-        match(f, [sigmoid, tensor.mul, tensor.neg, tensor.exp, sigmoid,
-                  tensor.mul])
+        topo = f.maker.fgraph.toposort()
+        for op, nb in [(sigmoid, 2), (tensor.mul, 2),
+                       (tensor.neg, 1), (tensor.exp, 1)]:
+            assert sum([n.op == op for n in topo]) == nb
         # assert check_stack_trace(f, ops_to_check=[sigmoid, tensor.mul,
         #                                           tensor.exp])
 

theano/tensor/tests/test_opt.py

@@ -1568,14 +1568,19 @@ def test_log1p():
     y = fmatrix()
     f = function([x, y], T.log(tensor.fill(y, 1) + (x)), mode=m)
     # the first three ops are Shape_i, Shape_i, and Dimshuffle
-    assert [node.op for node in f.maker.fgraph.toposort()][3:] == [
-        T.log1p, tensor.alloc]
+    topo = f.maker.fgraph.toposort()
+    assert topo[-1].op == tensor.alloc
+    assert T.log1p in [node.op for node in topo]
+
     f = function([x, y], T.log(0 + (x) + tensor.fill(y, 1.0)), mode=m)
-    assert [node.op for node in f.maker.fgraph.toposort()][3:] == [
-        T.log1p, tensor.alloc]
+    topo = f.maker.fgraph.toposort()
+    assert topo[-1].op == tensor.alloc
+    assert T.log1p in [node.op for node in topo]
+
     f = function([x, y], T.log(2 + (x) - tensor.fill(y, 1.0)), mode=m)
-    assert ([node.op for node in f.maker.fgraph.toposort()][3:] ==
-            [T.log1p, tensor.alloc])
+    topo = f.maker.fgraph.toposort()
+    assert topo[-1].op == tensor.alloc
+    assert T.log1p in [node.op for node in topo]
 
     f([1e-7, 10], [[0, 0], [0, 0]])  # debugmode will verify values
 
@@ -2207,8 +2212,9 @@ class test_local_subtensor_lift(unittest.TestCase):
         assert isinstance(prog[0].op, tensor.DimShuffle)
         assert isinstance(prog[1].op.scalar_op, theano.scalar.
                           Composite)  # Composite{add,exp}
-        assert prog[2].op == tensor.add
-        assert isinstance(prog[3].op, tensor.Subtensor)  # first subtensor
+        assert prog[2].op == tensor.add or prog[3].op == tensor.add
+        # first subtensor
+        assert isinstance(prog[2].op, tensor.Subtensor) or isinstance(prog[3].op, tensor.Subtensor)
         assert len(prog) == 4
         f([[0, 1], [2, 3]], [4, 5])  # let debugmode test something
 

theano/tests/test_printing.py

@@ -252,11 +252,11 @@ def test_debugprint():
     s = s.getvalue()
     # The additional white space are needed!
     reference = '\n'.join([
-        "Elemwise{add,no_inplace} [id A] ''   0 clients:[('[id B]', 1), ('output', '')]",
+        "Elemwise{add,no_inplace} [id A] ''   0 clients:[('output', ''), ('[id C]', 1)]",
         " |A [id D]",
         " |B [id E]",
-        "Elemwise{sub,no_inplace} [id B] ''   1",
-        " |Elemwise{add,no_inplace} [id A] ''   0 clients:[('[id B]', 1), ('output', '')]",
+        "Elemwise{sub,no_inplace} [id C] ''   1",
+        " |Elemwise{add,no_inplace} [id A] ''   0 clients:[('output', ''), ('[id C]', 1)]",
         " |D [id F]",
     ]) + '\n'
     if s != reference: