GPGPU Programming for Games and Science

Sample Applications 351

float z0 = ∗( float∗)&encoding0 ;

float z1 = ∗( float∗)&encoding1 ;

float f0 = MyFunction( z0 );

float f1 = MyFunction( z1 );

if (f0∗ f1 <=0.0f)

{

roots . insert ( std :: abs( f0) <=std::abs(f1) ? z0 : z1);

}

z0 = −z0 ;

z1 = −z1 ;

f0 = MyFunction( z0 );

f1 = MyFunction( z1 );

if (f0∗ f1 <=0.0f)

{

roots . insert ( std :: abs( f0) <=std::abs(f1) ? z0 : z1);

}

}

}

}

LISTING 7.5: Root finding on the CPU using an exhaustive search with a

single thread.

The same approach for

double roots is a lot slower. Do not try this at home.

You have 2

32

times more numbers to process, so the total execution time is

on the order of 1430 years!

7.2.4.2 CPU Root Finding Using Multiple Threads

The exhaustive processing for finite

float numbers may be reduced by dis-

tributing ...