Altered the if-statements a bit for fair competition..

I did my own tests in C# .NET 4.0 Core i5 3.2 Ghz

const long count = 100000000;

var operands = new uint[]
{0×80000000,0xffffffff,1, 2, 3, 4, 8, 6,5,7,9,10,11,111};
foreach (var operand in operands)
{
var start = DateTime.Now;
for (int i = 0; i < count; i++)
{
if ((i & operand) == 0) ;
}
var milliseconds = (DateTime.Now -start).Milliseconds;
Console.WriteLine("{0}: \t{1} ms", operand, milliseconds);
}

All results was between 400~600ms. So I thing it would be JIT which optimize performance for difficult patterns. There are not so big variances for particular patterns.

if (rare boundary cases does not hold)
do common stuff
else
handle it

Instead of

if (rare boundary case)
handle it
else
do common stuff

W.r.t predication is useful in some cases but creates data dependencies that may stall the pipeline, as a predicated instruction waits for its condition’s evaluation to complete. A successfully predicted branch breaks such “dependency chains”. Predication works best when it replaces hard-to-predict branches.

From what people say, this is the best book on the detais of CPU and memory. I didn’t get to it myself yet, but that’s the book to look at if you want to read more on this topic.