Hello Dear all,
I have large image more than 8k*8k to copy from host to device for other processing. For better performance, is it good to copy it to global memory, texture memory or other types? What can I do if the memory in my device is not sufficient?
Thank you very much
It depends on what you do with the data. Global memory is large but slow if you want to use the data using a lot of blocks containing a lot of threads. If you want to edit/process the image you can copy it to the global memory and then copy a part of it in the shared memory of the threads of the same block in order to do your calculations, editing, etc much quicker. After finishing you can transfer the editted parts from the shared memory to the global one and then back to host.
Thank you for your reply. Is there an advantage of copying it to texture memory for applying after filtering operations?
- Badges
- Science topic
- Similar topics
- Computer Science
- Images