CrossentropySoftmax1HotWithBiasDx + local_useless_incsubtensor_alloc

nnet.py.diff

diff --git a/theano/tensor/nnet/nnet.py b/theano/tensor/nnet/nnet.py
index 788213d..e468385 100644
--- a/theano/tensor/nnet/nnet.py
+++ b/theano/tensor/nnet/nnet.py
@@ -1099,7 +1099,7 @@ class CrossentropySoftmax1HotWithBiasDx (gof.Op):
         return [g_dy, g_sm, g_y_idx]
 
     def c_code_cache_version(self):
-        return (3,)
+        return (4,)
 
     def c_code(self, node, name, inp, out, sub):
         dnll, sm, y_idx = inp
@@ -1128,7 +1128,7 @@ class CrossentropySoftmax1HotWithBiasDx (gof.Op):
             PyErr_SetString(PyExc_ValueError, "rank error");
             %(fail)s;
         }
-        if (PyArray_DIMS(%(dnll)s)[0] != PyArray_DIMS(%(sm)s)[0])
+        if (PyArray_DIMS(%(dnll)s)[0] != PyArray_DIMS(%(sm)s)[0] && PyArray_DIMS(%(dnll)s)[0] != 1)
         {
             PyErr_Format(PyExc_ValueError,
                          "dnll.shape[0] (%%ld) != sm.shape[0] (%%ld)",
@@ -1136,7 +1136,7 @@ class CrossentropySoftmax1HotWithBiasDx (gof.Op):
                          (long int)PyArray_DIMS(%(sm)s)[0]);
             %(fail)s;
         }
-        if (PyArray_DIMS(%(dnll)s)[0] != PyArray_DIMS(%(y_idx)s)[0])
+        if (PyArray_DIMS(%(dnll)s)[0] != PyArray_DIMS(%(y_idx)s)[0] && PyArray_DIMS(%(dnll)s)[0] != 1)
         {
             PyErr_Format(PyExc_ValueError,
                          "dnll.shape[0] (%%ld) != y_idx.shape[0] (%%ld)",
@@ -1161,7 +1161,7 @@ class CrossentropySoftmax1HotWithBiasDx (gof.Op):
 
         for (size_t i = 0; i < PyArray_DIMS(%(dx)s)[0]; ++i)
         {
-            const dtype_%(dnll)s dnll_i = ((dtype_%(dnll)s*)(PyArray_BYTES(%(dnll)s) + PyArray_STRIDES(%(dnll)s)[0] * i))[0];
+            const dtype_%(dnll)s dnll_i = ((dtype_%(dnll)s*)(PyArray_BYTES(%(dnll)s) + PyArray_STRIDES(%(dnll)s)[0] * (PyArray_DIMS(%(dnll)s)[0] > 1 ? i : 0)))[0];
 
             const %(y_idx_type) s y_i = ((%(y_idx_type)s*)(PyArray_BYTES(%(y_idx)s) + PyArray_STRIDES(%(y_idx)s)[0] * i))[0];
 
@@ -1736,7 +1736,7 @@ def local_advanced_indexing_crossentropy_onehot_grad(node):
             # if the graph is valid, they have the same shape, so we
             # also know that z has the right shape.
 
-            if incr.type not in (dvector, fvector):
+            if incr.type not in (dvector, fvector) and not all(incr.broadcastable):
                 return
 
             # here we know that we are incrementing some part of matrix z by a vector