Eliminate passing the global and local thread group size to GpuPartition

The best global and local size is infered basd on input size.
In future, we should infer it based on available hardware and input
size.

#15: Find the optimal size for global and local number of threads 

Task-Url: http://github.com/meisam/spark/issues/issue/15

core/src/main/scala/org/apache/spark/rdd/GpuFilteredPartition.scala

@@ -14,10 +14,7 @@ class GpuFilteredPartition[T <: Product : ClassTag]
     val startSelectionTotalTime = System.nanoTime
 
     if (columnTypes(colIndex) == "INT") {
-      localSize = math.min(256, intData(colIndex).length)
-      globalSize = localSize * math.min(1 + (size - 1) / localSize, 2048)
-
-      val resultSize = compute(intData(colIndex), size.toLong, value, operation, globalSize, localSize)
+      val resultSize = compute(intData(colIndex), size.toLong, value, operation)
 
       size = resultSize
       intData.zipWithIndex.filter(_._1 != null).foreach({