make GpuSum work correctly with tensor of 0 element.

theano/sandbox/cuda/basic_ops.py

@@ -684,6 +684,9 @@ class GpuSum(Op):
     def c_code_reduce_ccontig(self, sio, node, name, x, z, fail):
         print >> sio, """
         {
+          if(CudaNdarray_SIZE(%(x)s)==0){
+            cudaMemset(CudaNdarray_DEV_DATA(%(z)s),0,sizeof(float));
+          }else{
             int verbose = 0;
             dim3 n_threads(
                     std::min(CudaNdarray_SIZE(%(x)s),
@@ -710,6 +713,7 @@ class GpuSum(Op):
                     n_threads.z);
                 %(fail)s;
             }
+         }
         }
         """ %locals()
 
@@ -1091,7 +1095,7 @@ class GpuSum(Op):
         """ %locals()
 
     def c_code_cache_version(self):
-        return (12,)
+        return (13,)
 
 
     def c_support_code_apply(self, node, nodename):

theano/sandbox/cuda/tests/test_basic_ops.py

@@ -34,8 +34,7 @@ def test_sum():
     test sum pattern 1, 11, 10, 01, 100, 110, 011, 001, 111, 0011, 0111, 1011, 1111
     TODO: test with broadcast
     """
-
-    for shape, pattern in [((5,),[0]),
+    for shape, pattern in [((0,),[0]),((5,),[0]),
                            ((5,4),[0,1]),((33,31),[0,1]),((5,4),[1]),((5,4),[0]),#need something bigger then 32 for some opt test.
                            ((5,4,3),[0]),((5,4,3),[0,1]),((5,4,3),[2]),((5,4,3),[1,2]),((5,4,3),[0,1,2]),
                            ((5,4,3,20),[2,3]), ((5,4,3,2),[0,1,2,3]), ((5,4,3,2),[0,2,3]),((5,4,3,2),[1,2,3])]:
@@ -49,7 +48,10 @@ def test_sum():
         f2 = theano.function([a],b, mode=mode_without_gpu)
         assert tcn.GpuSum in [x.op.__class__ for x in f.maker.env.toposort()]
         assert T.Sum in [x.op.__class__ for x in f2.maker.env.toposort()]
-        assert numpy.allclose(f2(val),f(val))
+        if val.size==0:
+            assert f2(val)==f(val)
+        else:
+            assert numpy.allclose(f2(val),f(val))
         
 
         #test with dimshuffle