Comments on: Optimizing for Loops: Reverse Loops

By: Robert

Robert — Mon, 12 Apr 2010 09:31:07 +0000

seems that web page removes a – in the text..

I did meant minus minus ‘i’
—i
i—

By: Robert

Robert — Mon, 12 Apr 2010 09:30:00 +0000

I noticed this niffy for loop and checked assemble compilation for it. I must say

// is still fastest with no cmp done. due to –i order of index there will be more clear how the for loop works

for(int i=x-1; i>=0; –i)
{
printf() //causes a call..
}
x = 5 => loop 4,3,2,1,0
in the case of
for(i=x; i–; )
{
printf() //casues a call! but will cause the i to break and result in a cmp/test due to its order and thus not being the fastest allternative.
}

x = 5 => loop 4,3,2,1,0

looks good in code though…
i– => cmp i and then dec…
–i => dec and cmp in this case branch directly due to zero flag is set default. hence one instr less..

By: Jessie Delpino

Jessie Delpino — Sat, 13 Feb 2010 14:44:44 +0000

Thanks for taking the time to discuss this, I feel strongly about it and love learning more on this topic. If possible, as you gain expertise, would you mind updating your blog with more information? It is enormously helpful and beneficial to your readers.

By: Guy

Guy — Sun, 16 Dec 2007 16:05:07 +0000

The basic idea behind it which also appears when compiling without the optimization (optimization greatly depends on the compiler) is that x86 based CPUs have the zero-flag in their ALU. For every arithmetic operation this zerfo-flag is check so by changing the order of the loop we can utilize this zero-flag to omit additional comparison operation for each loop. This should work like the loop with CX in assembly.

By: Kyku

Kyku — Sun, 16 Dec 2007 12:39:54 +0000

I don’t see any loops generated when your program is compiled the with -O2. So the the compiler must be smart enough to calculate the final result.
You’d better move the value of loop limit out of compiler’s visibility (get it from argv, for example). What’s more if I put the second loop first, it’s still slower, so your benchmark methodology seems wrong.

By: Guy

Guy — Wed, 07 Nov 2007 18:05:19 +0000

You are right, it should be i--, I fixed the typo.
Thanks for pointing it out.

By: Bryan E

Bryan E — Wed, 07 Nov 2007 15:49:14 +0000

Should be “i–” in the second loop….