MUD-Dev
mailing list archive
[ Other Periods
| Other mailing lists
| Search
]
Date:
[ Previous
| Next
]
Thread:
[ Previous
| Next
]
Index:
[ Author
| Date
| Thread
]
Re: [MUD-Dev] Naming and Directories?
Nathan F Yospe wrote:
> of looking up words. My implementation was similar, except that I had an
> extra level of switching when the wordbag got small... a bag that was at
> least seven links full had the 26 branches, but the smaller ones had the
> optimization of holding no empty links, and rapid-checking the links. In
> testing, I'd found this became faster than the array-indexing at smaller
> sizes, for some reason. I still haven't a clue as to why this would come
> to be...
You didn't happen to test this with a tight loop? If you did, then it
most likely was caused by the the non-array implementation being held in
the primary cache which is usually small (8K on a Pentium). Caches are
very sensitive to non-continuous hits. This does of course not matter at
all if the cache was wiped by some intermediate code anyway. Something I
would expect to be the case for real life user-name lookups.
I don't have any good documentation right here, but if I am not mistaken
this would be bad for a Pentium cache...
for(int i=0; i<257; i++) x += bytetable[i*32];
for(int i=0; i<256; i++) y -= bytetable[i*32];
Actually it is more complicated than this... I believe accessing a
fewer bytes could be bad if you are are unlucky with the address
locations...
--
Ola Fosheim Groestad,Norway http://www.stud.ifi.uio.no/~olag/
_______________________________________________
MUD-Dev maillist - MUD-Dev#kanga,nu
http://www.kanga.nu/lists/listinfo/mud-dev
- Thread context:
- Re: [MUD-Dev] Naming and Directories?, (continued)
[ Other Periods
| Other mailing lists
| Search
]