Help with Device Memory and Host Memory

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi everybody,

I'm tried to do my first program in CUDA but I have some problems about Device and Host Memory.

My code is the follows:

void FunctionOne (Struct1 *HostContext, unsigned char *HostinBuf, unsigned int inLen)
{
UINT4 *in;
int mdi;
unsigned int i=0;

Struct1 *DeviceContext;
unsigned char *inBuf;

// Store Memory in the Device
CUDA_SAFE_CALL(cudaMalloc((void **)&DeviceContext, sizeof(Struct1) ));
CUDA_SAFE_CALL(cudaMemcpy(DeviceContext,HostContext, sizeof(Struct1 ), cudaMemcpyHostToDevice));

    CUDA_SAFE_CALL( cudaMalloc((void **)&inBuf, sizeof(unsigned char )*inLen ));
CUDA_SAFE_CALL(cudaMemcpy(inBuf,HostinBuf, sizeof(unsigned char )*inLen, cudaMemcpyHostToDevice));

// if I’m right the first step is pass the variable in the HOST Memory to Device memory

//… Working with the Device memory

while (inLen--) {
  /* add new character to buffer, increment mdi */
  DeviceContext->in[mdi++] = inBuf[i];   
 i++;
}

// Store the modification to the Host Memory
CUDA_SAFE_CALL(cudaMemcpy(HostContext, DeviceContext, sizeof(Struct1 ), cudaMemcpyDeviceToHost));

// free Device Memory
CUDA_SAFE_CALL(cudaFree(DeviceContext));
CUDA_SAFE_CALL(cudaFree(inBuf));

}

The problem is the HostContext data structure is not update with the DeviceContext data structure and I don’t know why.

I’m working with CUDA 2.0 and 9600GT graphic card.
I’m compiling as Release.

Best Regards,
GUAN

2

you try to use the device memory address on you cpu, that’s not gonna work… you have to start a kernel on the gpu to modify memory at the device.
have a look at the developer’s manual.

3

Thanks Ocire.

One question more.

For calling kernel function I make in my host something like this:

ActualizacionIn<<<1,16,0>>>(DeviceContext,in);

Can I uses this type of calls into kernel function too ?

I do but it is a little bit confuse

4

if you mean, how to call a device function from within a kernel: no, that is done like a normal function call.