Any type casting for sort int.

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi All,

Is any alternate type casting for int as we have __float_as_int in CUDA for int from float.
But it is giving some garbage results.

suppose int a = int(600/500); then I got the a value is 1.
but when I use int a = float_as_int (600/500); then I got 15463728 (some garbage value).

why is it so?

Thanks
kundan

2

__float_as_int does not convert values. It leaves the bit pattern unchanged and reinterprets those bits as an integer. Since float and int have very different representations, the results will be “strange”. See page 90 of the programming guide. And [url=“Single-precision floating-point format - Wikipedia”]http://en.wikipedia.org/wiki/Single_precision[/url]

3

Hi Jamie K,

Thanks for reply.That’s makes things more understandable .

Could you send the complete name of programming guide where i will get more informations?

Once again lot of thanks

Kundan

4

He is referrring to the “Programming guide” that comes with CUDA installation.

Check out the “doc” sub-directory of your CUDA install directory,

5

Hi All,

I want to reduce or remove type casting in CUDA program.

[b] for(a = 0; a<640 ; ++a)

       i = int(Array[a] + 0.5f);[/b]

Is any way to reduce the overhead of typecasting ?

thanks

Kundan