On Sunday, January 6, 2013 11:08:53 PM UTC-8, JT wrote: > Well if you cannot build it you can't, i did. There is no comparissons > only rules for 0=left 1=right and a recursive algorithm at digit > level. You do not need to understand because the once before you > didn't either.
Do you realize how expensive breaking out bits are? You want to build a tree at the bit level? Heck, the time to extract the bit exceeds the compare, then you've got to get an address and check to see if exists or not.
I'll play just a bit more.
How do you determine if you are at a branch or leaf node? Do you think there is no cost to this? Just retrieving a relative address and adding it to the base is expensive. Why do you think we went to superscalar processors with speculative execution? Even so most micros have less than 8M onboard cache so paging happens quite a bit. Paging is quite expensive.
With today's processors you have to worry about more than the number of instructions executed you have to worry about how close they are to each other and how to get as much done with as little paging as possible.
In the late '70 I was working on a B-1700. While I knew the IBM360 had microcode (Amdal lost the battle so went to work for Fuji then learned the advantage of volitile memory microcode when they came out with the 370) the B-1700 sported two distinct sets of microcodes in memory at the same time, one optomized for the OS and algol and a secod optomized for RPG and Cobol. I thought I'd like to work on microcode but I don't have the patience for the fine tuning. Neither do most of us. The CISC vs RISC fight is over. Sure there's RISC under the covers but most of us just don't care. We just want to write programs that do fancy stuff and not waste a bunch of time on the minutiae.
So go introduce your count sort. It's a variation of a binary insertion sort. You won't save any time but what the heck, if it's your passion then go for it and feel the pain of rejection and ridicule.