Simplify offset and address calculation #3

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.

Sign up for GitHub

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

Jump to bottom

Open

yongwonshin wants to merge 1 commit into SAITPublic:develop from yongwonshin:develop

yongwonshin commented Dec 18, 2023

Motivation

Address calculation and offset logic are quite complex to understand.
Some transformed thread indices (w_idx, gidx, offset) are to help address mapping, which unnecessarily complicates the code.

Key changes

I created addr_gen_s utility function which accepts any offset and column value.
Address (in-bank offset) calculation is now flattened following the CUDA convention.

Test

All pass OpenCLPimIntegrationTests.

To Reviewers

Test passed using OpenCL and NVIDIA GPU, but I'm not sure it would pass in the real PIM hardware.
addr_gen_s calls addr_gen_ which adds slight overhead.


          Simplify offset and address calculation

313a391

- addr_gen_s accepts any size of offset and col
- flatten and inline address calculation

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet