Suggestion Description

Port the backend of Iris to Gluon. TODOs:

1. Use @aggregate to encapsulate the Iris backend struct so that we don't need to pass the heap bases around. Instead we use the Iris object.
2. Port the entire iris.py to Gluon
3. Port the producer consumer example only to Gluon

Some resources to look at:
https://github.com/triton-lang/triton/tree/main/python/tutorials/gluon
https://github.com/triton-lang/triton/tree/main/python/examples/gluon
https://raw.githubusercontent.com/triton-lang/triton/refs/heads/main/python/examples/gluon/01-attention-forward.py
https://raw.githubusercontent.com/triton-lang/triton/refs/heads/main/python/tutorials/gluon/01-intro.py
https://raw.githubusercontent.com/triton-lang/triton/refs/heads/main/python/tutorials/gluon/02-layouts.py
https://raw.githubusercontent.com/triton-lang/triton/refs/heads/main/python/tutorials/gluon/03-async-copy.py
https://raw.githubusercontent.com/triton-lang/triton/refs/heads/main/python/tutorials/gluon/04-tma.py
https://raw.githubusercontent.com/triton-lang/triton/refs/heads/main/python/tutorials/gluon/05-wgmma.py
https://raw.githubusercontent.com/triton-lang/triton/refs/heads/main/python/tutorials/gluon/06-tcgen05.py
https://raw.githubusercontent.com/triton-lang/triton/refs/heads/main/python/tutorials/gluon/07-persistence.py
https://raw.githubusercontent.com/triton-lang/triton/refs/heads/main/python/tutorials/gluon/08-warp-specialization.py

Operating System

No response

GPU

No response

ROCm Component

No response