invalid device symbol

Accelerated Computing CUDA CUDA Programming and Performance
1

global.cu

__device__ __constant__ MyParam cParams;

main.cpp

#include "update.h"

int main()

{

MyParam mp;

Host_UpdateParam(&mp);

}

update.h

void Host_UpdateParam(MyParam* mp);

update.cu

"#include "update.h"

"#include "global.cu"

void Host_UpdateParam(MyParam* mp)

{

CUDA_SAFE_CALL(cudaMemcpyToSymbol (cParams, mp, sizeof(MyParam) ) );

CUDA_SAFE_CALL(cudaThreadSynchronize());

}

I got cuda error in file ‘update.cu’ at line with

CUDA_SAFE_CALL(cudaMemcpyToSymbol (cParams, mp, sizeof(MyParam) ) );

Any suggestion? Thanks!

2

according to
[url=“The Official NVIDIA Forums | NVIDIA”]http://forums.nvidia.com/index.php?showtopic=104029[/url]
(constant seems to be only file scope!)

And here’s some experiment
[url=“http://forums.nvidia.com/index.php?showtopic=108893”]The Official NVIDIA Forums | NVIDIA
(Summary: Unable to set constant when working with separate .cu and .cpp files. When using single .cu program everything is fine. In emulation everything is fine.)

um… is it impossible?

3

according to
[url=“The Official NVIDIA Forums | NVIDIA”]The Official NVIDIA Forums | NVIDIA
(constant seems to be only file scope!)

And here’s some experiment
[url=“http://forums.nvidia.com/index.php?showtopic=108893”]The Official NVIDIA Forums | NVIDIA
(Summary: Unable to set constant when working with separate .cu and .cpp files. When using single .cu program everything is fine. In emulation everything is fine.)

um… is it impossible?

4

You r right bout the scope. But surely theres another way to get around this… As for emulation: dont depend on it! It was removed in 3.1 cause code that runs fine in emulation could fail in device mode and the other way round.

5

You r right bout the scope. But surely theres another way to get around this… As for emulation: dont depend on it! It was removed in 3.1 cause code that runs fine in emulation could fail in device mode and the other way round.