Question about Mapped Memory

Accelerated Computing CUDA CUDA Programming and Performance
1

I am using cudaHostAlloc to allocate mapped memory. I know that that i need to use cudaThreadSynchronize() so that the changes to mapped memory can be seen after I call a GPU function.

The question is do i need to use cudaThreadSynchronize after i change values on the mapped memory using the CPU so it is updated on the GPU

ie:

int* mapped;
int* mapped_device;

cudaHostAlloc(mapped … )
cudaHostGetDevicePointer(mapped_device, mapped, … )
mapped[0] = 1;

cudaThreadSynchronize(); <--------- Do I need this

kernel<<<… , … >>>(mapped_device);

thanks

2

So long as there isn’t anything running accessing that region, no.