Author: Dave Gomboc
Date: 11:21:58 06/17/99
Go up one level in this thread
On June 17, 1999 at 13:51:09, Eugene Nalimov wrote: >On June 17, 1999 at 13:24:50, Dave Gomboc wrote: > >>On June 17, 1999 at 03:13:08, David Blackman wrote: >> >>>On June 16, 1999 at 23:01:38, Dave Gomboc wrote: >>> >>>>On June 16, 1999 at 22:56:02, Greg Lazarou wrote: >>>> >>>>>Hi James, >>>>> >>>>>I don't have the straight answer to your question but if x is an integer then >>>>>multiplications by 2 and divisions by 2 (the 4th and 2nd cases respectively) >>>>>should be very fast left and right shifts. Now you can let the compiler figure >>>>>it out (and you may be able to help the compiler figure it out by expressing >>>>>x*=.5 as x /= 2) or you can code it as a shift yourself... >>>> >>>>N.B. (x /= 2) == (x >> 1) iff (x is unsigned) >>>> >>>>Dave >>> >>>But for signed numbers (x >> 1) is almost always what you really want. And if >>>you want to divide by something other than a power of two you curse hardware and >>>language designers for not making signed division work properly. >> >>This is true, if you really want "x >> 1", you'd better type it in yourself, >>because AFAIK optimizers explicitly do not change x/=2 into x>>1 when x is >>signed. >> >>Dave > >I already posted it here, but once again: for signed division by a power of 2, >you can use something like that (if I remember it correctly): > >y = x/2 --> y = (x >> 1) + (x == -1) >y = x/(2**N) --> y = (x >> N) + (((unsigned) (x + 2**N-1)) < 2**N-1) > >And at least one of the optimizing compilers do that - I wrote it myself before >I came to MS. > >Eugene I did see your other post. Those are pretty cool optimizations! Of course, they're still not quite as fast as "x >> n" though. Dave
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.