can I use pinned memory?

[CUDAkk]

Hi All,

I have installed CUDA 2.3 driver , toolkit and SDK and i want to use pinned memory but when I see the cudaDeviceProp structure then one of its member value is

canMapHostMemory 1 .

I want to know what this value indicates?

[CUDAkk]

I wnat to know what this value indicates?

I think it indicates this property is on .

[CUDAkk]

Hi ,

I have written code for using pinned memory :-

char *device = NULL;

unsigned  int flags;

int device Count, idev = 0;

unsigned char*a ;     // Pinned memory allocated on the CPU

            unsigned char*d_a,  // Device pointers for mapped memory

cudaDeviceProp deviceProp;



cudaError err = cudaSetDevice(idev);

err = cudaGetDeviceProperties(&deviceProp, idev);

flags = cudaHostAllocMapped;

            err = cudaHostAlloc((void **)&a, dstSize, flags);

In my case :

dstSize = 17280000

      flag      = 2 ( cudaHostAllocMapped )

But after executing the line

err = cudaHostAlloc((void **)&a, dstSize, flags);

I got error cudaErrorUnknown .

can any one help on this ?

[CUDAkk]

I have seen SDK 2.3 project simpleZeroCopy code and I try to implement it in my code. But I have some confusion on using it.

Is following code should be only on main function or I can write it in any function :

int n, nelem, idev, deviceCount;

  char *device = NULL;

  unsigned int flags;

  size_t bytes;

  unsigned char *a, *b, *c;                      // Pinned memory allocated on the CPU

  unsigned char *d_a, *d_b, *d_c;                // Device pointers for mapped memory  

  float errorNorm, refNorm, ref, diff;

  cudaDeviceProp deviceProp;  

  /* Get the device selected by the user or default to 0, and then set it. */

  if(cutGetCmdLineArgumentstr(argc, (const char**)argv, "device", &device))

  {

	cudaGetDeviceCount(&deviceCount);

	idev = atoi(device);

	if(idev >= deviceCount || idev < 0)

	{

	  fprintf(stderr, "Invalid device number %d, using default device 0.\n",

			  idev);

	  idev = 0;

	}

  }

  else

  {

	idev = 0;

  }

  cudaError err1 = cudaSetDevice(idev);

  /* Verify the selected device supports mapped memory and set the device

	 flags for mapping host memory. */

  err1 = cudaGetDeviceProperties(&deviceProp, idev);

  err1 = cudaSetDeviceFlags(cudaDeviceMapHost);

  /* Allocate mapped CPU memory. */

  nelem = 17280000;

  bytes = nelem*sizeof(unsigned char);	

  flags = cudaHostAllocMapped;

  err1 = cudaHostAlloc((void **)&a, bytes, flags);

  err1 = cudaHostAlloc((void **)&b, bytes, flags);

  err1 = cudaHostAlloc((void **)&c, bytes, flags);

err1 = cudaHostGetDevicePointer((void **)&d_a, (void *)a, 0);

err1 = cudaHostGetDevicePointer((void **)&d_b, (void *)b, 0);

err1 = cudaHostGetDevicePointer((void **)&d_c, (void *)c, 0);

Because if I write it on main function then it works fine but if I write it any other function then it gives error. Please help…

[CUDAkk]

This problem is solved . We can write it any function . Actually I got this error because of prior error.

[arnaudSSI]

This problem is solved . We can write it any function . Actually I got this error because of prior error.

And what was your error?

I got also a “cudaErrorUnknown” when I try to allocate pinned memory with cudaHostAlloc().

Thanks

[allenporter]

And what was your error?

I got also a “cudaErrorUnknown” when I try to allocate pinned memory with cudaHostAlloc().

Thanks

Make sure you’re first calling:

cudaSetDeviceFlags(cudaDeviceMapHost);