make sur we select the good value for n_thread.{y,z} for the new GpuSum pattern

theano/sandbox/cuda/basic_ops.py

@@ -784,7 +784,7 @@ class GpuSum(Op):
         strides_dim = ",".join(["CudaNdarray_HOST_STRIDES(%(x)s)[%(i)s]"%locals() for i in range(N+1)])
         threads_y = """
             //get as many y threads as we can fit
-            while (n_threads.x * n_threads.y < NUM_VECTOR_OP_THREADS_PER_BLOCK)
+            while (n_threads.x * (n_threads.y+1) <= NUM_VECTOR_OP_THREADS_PER_BLOCK)
             {
                 if (n_threads.y < CudaNdarray_HOST_DIMS(%(x)s)[%(N)s-1])
                     n_threads.y += 1;
@@ -794,13 +794,13 @@ class GpuSum(Op):
 """%locals()
         threads_z = """
             //get as many z threads as we can fit
-            while (n_threads.x * n_threads.y * n_threads.z <= NUM_VECTOR_OP_THREADS_PER_BLOCK)
+            while (n_threads.x * n_threads.y * (n_threads.z+1) <= NUM_VECTOR_OP_THREADS_PER_BLOCK)
             {
-                if (n_threads.z > CudaNdarray_HOST_DIMS(%(x)s)[%(N)s-2])
+                if (n_threads.z < CudaNdarray_HOST_DIMS(%(x)s)[%(N)s-2])
+                    n_threads.z += 1;
+                else
                     break;
-                n_threads.z += 1;
             }
-            n_threads.z -= 1;
 """%locals()
         if len(self.reduce_mask)==2:
             threads_y = ''