sorting hashes with opencl

i have lines of 3 hashes (ie md5, 128bit). actually plenty of them. think billions, thus they wont fit in main memory. they are in a file and need to get sorted. using gnu sort it takes a long time obviously, but it works.

i think it could be possibly worth to split them ie into a vector of 6 64bit ints and sort them in batches with opencl, then mergejoin them. i have a radeon hd 6950 with 2gb at hand.

but i have no experience with opencl.

so the questions:

• which opencl datastructure would i want to use for this task?

• 开发者_Python百科

  which sorting algo would i use

• could the mergejoin also be accelerated?

Since it is on disk I would just use STLXXL.

http://stxxl.sourceforge.net/

There is OpenCL code.... but try this first :)