Simonson, Lucanus J wrote:
Phil wrote:
Having said all that, personally I have little interest in an algorithm
that has no benefit for fewer than millions of items. What do other people think about that aspect?
I frequently sort gigabytes, tens of gigabytes or even hundreds of gigabytes of geometry data as a pre-requisite to scanline over the data. However, a faster sort would only benefit me if the speedup were reliably 2X or more. A 30% speedup in sort would disappear with the effect of Amdahl's Law to something under 10%. I get better speedup just by adopting each new version of the compiler. For me, the modest size and tenuous nature (depends upon architecture specific constants) of the speedup does not justify any effort to integrate and carry the dependency to the library that provides it. I have been following the thread only out of academic interest.
hmmm - have you looked at postman's sort. This was the subject of an article in 1992 C user's journal and subsequently available as a commercial product. see www.rrsd.com. Robert Ramey