set Accumulator type to double to ensure deterministic results

src/GksGpu/Definitions/AccumulatorDataType.h

@@ -3,8 +3,18 @@
 
 namespace GksGpu {
 
-typedef float realAccumulator;
-//typedef double realAccumulator;
+// This file is used to control the data type of accumulator variables.
+// Accumulator variables are variables, where cell values are written 
+// during the flux computation, which is per face. Since the face evaluation 
+// order on GPUs is arbitrary, the cutoff errors for these accumulators are non 
+// deterministic. This deficiency can be solved for single precision calculations
+// by setting the accumulator data type to double. The deviations are then 
+// so small, that they are cut off during the downcast to single.
+// using double precision accumulators has some performance implications, 
+// especially on consumer hardware.
+
+//typedef float realAccumulator;
+typedef double realAccumulator;
 
 } // namespace GksGpu