@SebAaltonen How to implement the abstract interface Barrier in Metal? Because automatic resource tracking i use empty barrier function in my RHI. It feels strange. 0.0
@Kirby0Louise Microarchitecture does‘t consider the utilization for complex computing, such as dynamic branching. WaveSize is also very large, Wave64 or Wave128.
@gavkar In my understanding, the hardware acceleration mesh shader feature is to use this new unified cache to allocate more LDS to make payloads less prone to overflow to main memory, in order to alleviate latency and improve performance. I don't know if it's roughly like this?
@handleym99@gavkar I have an idea, I don't know if it's possible.
Maybe the meaning is that GPU can allocate more warps in real-time by taking out resources such as wasted registers from dynamic branches and tail effects?