Taichi 2.0

[jarmitage]

@yuanming-hu: Our current challenge (and opportunity) is in determining the next significant milestones for Taichi 2.0. While the future path is not fully defined, we are excited about the possibilities and welcome your thoughts on potential directions. (One possibility is more support for differentiable graphics, given GenAI, leveraging differentiable graphics, nowadays have increasing impact.) – #8506 (comment)

What are your thoughts, ideas, hopes and dreams for Taichi 2.0?

[jarmitage]

Targeting WebGPU would be great for many use cases cc @AmesingFlank #3781

[bgailleton]

If this is still active, I'd love to see in 2.0:

• some sort of memory pool (basically being able to deallocate/reallocate diffe rent fields on the go, freeing memory or keeping it for another use - basically like pytorch - without having to reinit)

• multiple runtimes

• fields, data oriented class in aot

• some sort of thread lock (like cuda atomicCAS)

Cheers!