Future Plans for Thread Block Cluster Optimization?

Hi team,

I’ve been exploring the ​​Tile Lang​​ project and find its approach very promising. While reviewing the examples, I noticed that the warp-specialized GEMM implementation doesn’t appear to use ​​TMA (Tensor Memory Accelerator) broadcast optimization​​.

After checking the codebase, I saw that `/src/transform/cluster_planning.cc` includes a pass for thread block cluster optimizations, but it doesn’t seem to be enabled in the current compilation pipeline.

​​Questions:​​

1. Are there plans to support ​​thread block cluster optimizations​​, such as ​​TMA broadcast​​ and ​​Distributed Shared Memory​​?
2. If this is a planned feature, I’d be interested in contributing to its development.

This optimization aligns with my interests, and I’d appreciate any guidance on how to help. Looking forward to your insights!:)

Start by reading /src/transform/cluster_planning.cc and tracing whether its thread block cluster optimization pass is connected to the current compilation pipeline. Review the warp-specialized GEMM examples for TMA broadcast usage. Done would require a clear project decision or implementation plan covering TMA broadcast and Distributed Shared Memory support.

Written by the indexing model from the issue text.