make GpuElemwise work with empty tensor. Add tests for this.

theano/sandbox/cuda/elemwise.py

@@ -37,7 +37,7 @@ def get_str_list_logical_scalar(node, value_str='ii_i%i_value', data_str='ii_i%i
 class NaiveAlgo(object):
     verbose = 0 # 1, 2 or 3 for more verbose output.
     cache_version = ()
-    cache_version = ('debug', 9, verbose)
+    cache_version = ('debug', 10, verbose)
 
     def __init__(self, scalar_op, sync=True):
         """ 
@@ -787,6 +787,7 @@ nd_collapse_[i]=0;
                 """ %locals()
             else:
                 print >> sio, " return 0; " %locals()
+        print >> sio, "if(numEls==0) return 0;"
         print >> sio, "switch (nd_collapse==0?0:min(%(nd)s,nd_collapse)) {"%locals()
         print >> sio, "case 0: {"
         launch_Ccontiguous(nodename, id_self, scalar_op, self.sync)

theano/sandbox/cuda/tests/test_basic_ops.py

@@ -146,6 +146,20 @@ def test_reshape():
         return T.Reshape(2)(v, theano._asarray([2,3], dtype='int32'))
     utt.verify_grad(just_vals, [a_val])
 
+def test_elemwise_empty():
+    #test with 0 element
+    a = tcn.shared_constructor(theano._asarray(numpy.random.rand(0,0), dtype='float32'), 'a')
+
+    b = tensor.fmatrix()
+
+    f = pfunc([b], [], updates=[(a, a+b)], mode=mode_with_gpu)
+    f2 = pfunc([b], [], updates=[(a, a+b)], mode=mode_without_gpu)
+
+    a0 = a.value * 1.0
+    f(numpy.ones((0,0)))
+
+    assert numpy.all(a0 + 1.0 == a.value)
+
 def test_elemwise0():
 
     a = tcn.shared_constructor(theano._asarray(numpy.random.rand(4,4), dtype='float32'), 'a')