Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

what should I do for porting? #6

Open
gofreelee opened this issue Dec 31, 2022 · 1 comment
Open

what should I do for porting? #6

gofreelee opened this issue Dec 31, 2022 · 1 comment
Assignees

Comments

@gofreelee
Copy link

Hello, I want to try port the reef to 6900XT GPU. What general work needs to be done? I see that the DNN model in resource is all the generated code of specific architecture, so do I mainly need this part of code of porting?
Thanks in advance~

@zombie12138
Copy link

zombie12138 commented Feb 24, 2024

I have ported the reef to 6900xt in my repo.

  1. modify some Makefile flags
  2. Some assembly is not suitable for 6900, and you need to fix it.
  3. some issue will occur when you get the address of functions using d2d hipmemcpy in 6900xt, but using a kernel to copy works.
  4. modify the resource counts in code, which will make reef much faster
  5. generate new profile for each model, but I have NOT do that.

Besides, I also do same modify in a newer tvm for hip code generation.

Some of my code in my repo may be vague, but reef does work on my 6900xt. Some models sometimes have deadlocks while running. But I haven't investigated in depth.

The reasult for inference test may be a little different because of the different wavefront size

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants