Adding Type Casting Functions to VS2012 Project CUDA Toolkit 5.5

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi! I have a problem during compilation process. I have been receiving next error:
Error 2 error : identifier “__double2float_rn” is undefined
I use VS2012 and CUDA Toolkit 5.5 on Windows 7 32.
There are headers I attached to the project:
#include <stdio.h>
#include <stdlib.h>
#include <cuda.h>
#include <cuda_runtime.h>
#include <device_functions.h>

Everything compiles absolutely normal before I’m trying to add this function to my code.

Thanks for any help.

2

Sounds like you’re not targeting an architecture greater than sm_12.

Check your Property Pages for your project and make sure this is at least “compute_13,sm_13”:

3

There is “compute_10,sm_10” value in this string. Could you tell me how this value appears? Is it appears automatically depend on type of videocard?

4

You should check which model of CUDA capable device do you have. Then figure out which compute capability does it have from this webpage [url]https://developer.nvidia.com/cuda-gpus[/url]. The Code Generation values should correspond to the compute capability of your card.

5

Sorry for not answering. And thanks for your answers. I checked my GPU (GeForce 9800 GT). Actually I have other problem may be depend on the capabilities of my GPU (I’m not sure). The problem is that I have to do a lot of double precision calculations on GPU and then convert them to float values. Sure it is possible because I checked this on the same task for CPU. But when I use GPU an unpredictable behaviour appears.
There is an example of the code (not the actual code, but the same):

float Pf; // float input/output parameter
Formulaf(t); // double precision device function
double Pd = Pf +  Formulaf(t);
Pf = Pd + ...;

Unfortunately, due to the character of the solution and GPU features I can’t do all operations with single or double precision values.
The first thing I had thinking over was the type casting. So I tried to use __double2float_rn function.