Tune AMD GCN warptiles for large and small tiles#11

Open

rasbid wants to merge 1 commit intomasterfrom

codex/locate-and-update-chip-tuning-for-vulkan

Owner

rasbid commented Oct 13, 2025

Summary

extend the AMD GCN tuning block to override large and small matmul warptiles alongside the medium configuration
ensure the new warptiles use 64-lane multiples and satisfy shared-memory limits for quantized and ID kernels

Testing

python - <<'PY' ...

https://chatgpt.com/codex/tasks/task_e_68ebeeb11e4c8330bf9d2b3c868610a8


          Tune AMD GCN warptiles

314d596

rasbid added the codex label

— with

ChatGPT Codex Connector

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

Labels