add Kokkos / CUDA / ROCM / oneAPI terminology table for hierarchical parallelism

[rbourgeois33]

Kokkos	CUDA (Nvidia)	ROCM (AMD)	oneAPI/DPC++ (Intel)
Team (of threads)	Block	Workgroup/Block	Workgroup
Scratch pad memory (of a team)	Shared Memory	Shared Memory	Local Memory
League size	Number of Blocks	Number of Workgroups/Blocks	Number of Workgroups
Team size	Number of Threads per Block	Number of Threads per Workgroup/Block	Number of Threads per Workgroup
League rank	Index of the Block	Index of the Workgroup/Block	Index of the Workgroup
Team rank	Index of the Thread within a Block	Index of the Thread within a Workgroup	Index of the Thread within a Workgroup
team member	thread	thread	?
Team Barrier	__syncthreads()	__syncthreads()	item.barrier()
Core clusters	Streaming Multiprocessors (SMs)	Compute Units (CUs)	I am unsure

[pzehner]

That's a cool guide to add somewhere in the cheat sheet. Maybe that could make a third sheet "Terminologies cheat sheet."

I think we would need to differentiate/complete:

• Memories (global memory, shared memory, local memory, register, UVM);
• Execution chips (streaming multiprocessor, streaming processor, warp, thread);
• Hierarchical parallelism itself.

[rbourgeois33]

Agreed, let's start simple with only CUDA/Kokkos then (as this will help out most users)

[rbourgeois33]

https://docs.google.com/spreadsheets/d/11aH1Dk1JOKUyF7ro__qtqBpenLCZ5I-Gv5_cDEU8FbU/edit?usp=sharing

I've started a cleaner set of table there (markdown is not ideal)