Handling Type casting in CUDA.

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I have a some code :

__device__get_value(float i,float j)
{
// some manipulations .
}

device fun_table(float alpha, int* Array)
{
int i = 0, j = 0;
for(i=0 ;i<430;++i)
for(j=0 ; j<640 ;++j)
{
//body here.
get_value(i, j) ;

}
}
This device function called from some kernel.

my problem is that :

  • here actual parameters in function call of get_value(i, j) is int types but the formal parameters are float type.
    this implecit typecasting takes lot of time .
    Is any optimum way to handle this problem.

Thanks :
Kundan