Author: Matt Taylor
Date: 12:41:26 02/26/03
Go up one level in this thread
On February 26, 2003 at 14:04:16, Uri Blass wrote:
>I had the following code in my program
>
>do
>{
>...
>if (info[target]==EMPTA)
> target+=8;
>else
> target=64;
>}
>while (target<64)
>
>I found that changing it to else break did not do my program faster.
>
>I think that with break I have branches and without break the compiler can
>translate it to
>
>target=((info[target]==EMPTA)?target+8:64);
>
>My questions are
>1)How can I write the code as branchless code without target=64 that is a waste
>of time?
>
>2)I understood from previous discussion that for some reason ? is considered
>branchless code.
>What is the reason for it.
>
>It seems to me that using ? is the same as using if.
>If it is not the case then what is the difference?
>
>Uri
In this case, I don't think so. I'd need to look at assembly output to tell you.
The bottom of the loop itself is a branch, so you're going to branch one way or
another. If you use the break statement, you can make the loop more efficient by
making the condition 1 (always true), and the compiler will change this into an
unconditional branch. An unconditional branch never gets mispredicted. When the
compiler sees your break statement, it will translate that into a jump out of
the loop.
Usually the ternary conditional construct can be converted into branchless code
because usually the expressions are arithmetic, constants, and other branchless
code. Also, they are small. The construct (condition ? 1 : 0) is particularly
efficient since it translates into a single instruction.
Branchless code isn't -always- more efficient, and any program more complex than
"Hello world" requires at least some branches (for function calls and loops).
Branchless code is preferred when each side of the branch only does a little bit
of computation -- the branch itself may cost more than the computation it
avoids.
-Matt
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.