[Feature Request] [Pass] InjectDmaBarrier #35
Description
Required prerequisites
- I have searched the Issue Tracker that this hasn't already been reported. (comment there if it has.)
Motivation
InjectDmaBarrier automatily insert mbarriers that protect a Dma's src/dest buffers before the Dma operation is completed. We can adapt the existing InjectTmaBarrier pass structure to write our own InjectDmaBarrier pass:
- Remove the thread logic. Remove thread extent tracking and warp specialization checks
- Collect DMA operations that need synchronization
- Map them to barrier IDs
Solution
No response
Alternatives
No response
Additional context
No response