Must the result of cublasSdot be host variable?

When I invoke the cublasSdot function with the result alloced on device, it gave me an error.
But it successed with the result on host.
the code as follow:

#include <cublas.h>
#include <cublas_v2.h>

#include <stdio.h>
#include <stdlib.h>

int main(void)
{
	int n = 10;

	float *h_x = (float *)calloc(n,sizeof(float));
	float *h_y = (float *)calloc(n,sizeof(float));
	float *h_result = (float *)calloc(1,sizeof(float));

	for (int k = 0; k < n; ++k)
	{
		h_x[k] = 1.0f;
		h_y[k] = 1.0f;
	}	

	cublasHandle_t handle;
	cublasCreate(&handle);
	float *d_x;
	float *d_y;
	float *d_result;

	cublasAlloc(n,sizeof(float),(void **)&d_x);
	cublasAlloc(n,sizeof(float),(void **)&d_y);
	cublasAlloc(1,sizeof(float),(void **)&d_result);

	cublasSetVector(n,sizeof(float),h_x,1,d_x,1);
	cublasSetVector(n,sizeof(float),h_y,1,d_y,1);

	cublasSdot(handle,n,d_x,1,d_y,1,d_result);

	cublasGetVector(1,sizeof(float),d_result,1,h_result,1);

	printf("%f\n", h_result);

	cublasFree(d_x);
	cublasFree(d_y);
	cublasFree(d_result);
	cublasDestory(handle);

	free(h_x);
	free(h_y);
	free(h_result);
	return 0;
}

Thanks for your help!!!

read the cublas documentation to set the scalar parameter pointer mode.

http://docs.nvidia.com/cuda/cublas/index.html#scalar-parameters

http://docs.nvidia.com/cuda/cublas/index.html#cublaspointermode_t

aside:

I’m curious what this is:

cublasDestory(handle);

That’s not valid CUBLAS code. Did you actually type this code into the browser? I don’t understand how this could be a copy-paste error, and this code won’t compile.

EXCELLENT!!
Thanks a lot! Yesterday I was coding on my home pc without nv gpu. So the code has an error.
Now I could put the result on the device! I fix the code as follow. This code could run by win7+vs2012+cuda7.5.

#include <cublas.h>
#include <cublas_v2.h>

#include <stdio.h>
#include <stdlib.h>

int main(void)
{
	int n = 10;

	float *h_x = (float *)calloc(n,sizeof(float));
	float *h_y = (float *)calloc(n,sizeof(float));
	float *h_result = (float *)calloc(1,sizeof(float));

	for (int k = 0; k < n; ++k)
	{
		h_x[k] = 1.0f;
		h_y[k] = 1.0f;
	}	

	cublasHandle_t handle;
	cublasCreate(&handle);

	cublasSetPointerMode(handle,CUBLAS_POINTER_MODE_DEVICE); // set here!!!

	float *d_x;
	float *d_y;
	float *d_result;

	cublasAlloc(n,sizeof(float),(void **)&d_x);
	cublasAlloc(n,sizeof(float),(void **)&d_y);
	cublasAlloc(1,sizeof(float),(void **)&d_result);

	cublasSetVector(n,sizeof(float),h_x,1,d_x,1);
	cublasSetVector(n,sizeof(float),h_y,1,d_y,1);

	cublasSdot(handle,n,d_x,1,d_y,1,d_result);

	cublasGetVector(1,sizeof(float),d_result,1,h_result,1);

	printf("%f\n", *h_result); // fixed

	cublasFree(d_x);
	cublasFree(d_y);
	cublasFree(d_result);
	// cublasDestory(handle);
	cublasDestroy(handle);

	free(h_x);
	free(h_y);
	free(h_result);
	return 0;
}

cublasDestory is incorrect. It should be cublasDestroy

[url]http://docs.nvidia.com/cuda/cublas/index.html#cublasdestroy[/url]

so I don’t think this is the code you are actually running. cublasDestory should not compile correctly

Thank you very much!
I am sorry for that I wrote a wrong code, and did not check it.
It is a waste of time that pointing out my spelling mistake.