Re: sub-optimal code for packed boolean arrays -- bug or inherent limitation

["Jeffrey R. Carter" <spam.jrcarter.not@acm.nospam.org>]

Alinabi wrote:
> 
> Now, concerning efficiency: these basic operations on bitboards
> are used 1e8 to 1e9 times every time the program tries to decide
> what to move next. Even the smallest improvement in speed can
> mean the difference between searching 10 or 11 ply deep, which
> can mean an improvement of over 50 Elo points in strength.

I'm sure that's true. However, you have not demonstrated that there's a 
speed difference between the 2 versions. Even if there is, it doesn't 
necessarily mean that there will be a difference in the # of plys that 
can be searched. The old rule, "1st get it right, then make it fast," 
still applies. Once you have it finished you can easily see if modifying 
this single procedure actually makes a difference. Until then, you're 
wasting a lot of effort on something you don't know is important.

You should also be aware that you're sending the compiler conflicting 
messages. Packing the array indicates that you want the compiler to 
minimize storage, even at the expense of speed, while "pragma Optimize 
(Time);" indicates that you want to minimize time, even at the expense 
of storage. Obtaining speed often requires wasting storage; the fastest 
implementation might be to not pack the array. I don't know how many of 
these you have, but with the large memories on modern machines, you 
might be better off with the additional 56 bytes per array. Again, you 
won't know for sure until you have a working version you can measure.

-- 
Jeff Carter
"Gentlemen, you can't fight in here. This is the War Room!"
Dr. Strangelove
30