Late Move Reductions as popularized by Fruit and Glaurung are used in Spark with different implementation details. Since splitting is a cheap operation in Spark, avoiding an expensive copy of the board and associated data, the search tree can be split at any depth, maximizing the use of the otherwise idle cores.
Like its earlier 0x88 relative Bright, Spark is able to search in parallel supporting up to 16 CPU cores, basically applying the Young Brothers Wait Concept.
0 kommentar(er)
