file i/o

Accelerated Computing CUDA CUDA Programming and Performance
#1

Im wondering how to deal with large files. Im guessing theres 2 ways

  1. allocate space on the video card to read to or write from and then do the I/O
  2. create a buffer and used that to transfer to and from the video card, duplicating everything

Is there a noticeable difference between the 2?
Thanks

#2

Actually there is only one way: You have to make a buffer on the CPU, and make a same size buffer on the GPU. Fill the CPU buffer, then you cudaMemcpy() from the CPU to the GPU (“host to device”). Or are you asking about something else?

#3

this post should be ignored/deleted. sorry