About zeor-cpoy between host and GPU

Autonomous Machines Jetson & Embedded Systems Jetson AGX Orin
1

hi, NV experts, I have some question about zeor-cpoy between host and GPU on ORIN platform:

  1. I want to avoid cudaMemcpy for input and output between host and GPU, is there any other method except unified-memory ?
  2. I want to apply DMA-BUF between different process to transfer data on ORIN, but, I found GPU cannot accept my dma_buf_fd, So, is there any method to solve my trouble?

Is there anyone would like to tell me the secret?

3

Hi,

1. Pinned memory is also zero-copy.
2. This can be done by EGL mapping.
Please check our example below:

/usr/src/jetson_multimedia_api/samples/v4l2cuda

Here is the CUDA document for Jetson memory for your reference:

Thanks.

4

Hi, AastaLLL,
thanks for your explanation,
I’ll try your suggestion

8

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.