Host Memory mapping to GPU

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I am new to CUDA, and was learning about memory mapping of host to GPU using cudaHostAlloc().
please correct me if am wrong but wat i think about cudaHostAlloc() is that it allocates memory on host and then map it to the GPU memory.
but there is no actual memory allocated on GPU.

so wat i was thinking was that insted if i just use simple C language to allocate memory on host and then use the same procedure like calling cudgetDevicePointer()
for getting pointer of GPU to be mapped to the host memory… will that work . i tried trying that but i received ‘invalid argument’ as error.
so am i doing something wrong and alos is there any other way to map the host memory to GPU expect cudaHostAlloc().

hoping to get reply soon.
Thanks.

2

Consider reading section 3.2.4 of CUDA C Programmin Guide, p. 29.

That might help.

Regards,

MK

3

Consider reading section 3.2.4 of CUDA C Programmin Guide, p. 29.
That might help.

Regards,
MK

MK ,
Thanks…i read that part but now am in new doubt. i didnt get the logic of ‘Write-Combining Memory’ . can anyone please help me in this context?

4

Google produced something like this (old but bold): Intel Write-Combining Memory Implementation Guidelines

This may help too: Andy Glew’s comp-arch.net wiki

Bless You,

MK