DepositCurrent: atomicAdd -> lockAdd

[WeiqunZhang]

The new amrex::BaseFab::lockAdd function is an optimized version of atomicAdd for OpenMP. In my testing on Frontier CPUs, it's up to 10x faster. I did not test the new function using hipace. But I used https://github.com/WeiqunZhang/amrex-devtests/tree/main/fab_atomicAdd for testing.

• Small enough (< few 100s of lines), otherwise it should probably be split into smaller PRs
• Tested (describe the tests in the PR description)
• Runs on GPU (basic: the code compiles and run well with the new module)
• Contains an automated test (checksum and/or comparison with theory)
• Documented: all elements (classes and their members, functions, namespaces, etc.) are documented
• Constified (All that can be const is const)
• Code is clean (no unwanted comments, )
• Style and code conventions are respected at the bottom of https://github.com/Hi-PACE/hipace
• Proper label and GitHub project, if applicable

[AlexanderSinn]

Hipace has AMREX_SPACEDIM == 3 but the array for DepositCurrent is only 2D with a length in the z direction of one. I think lockAdd would then only have one lock which probably would result in bad performance.

[WeiqunZhang]

Yes, I overlooked that. Let me try to handle that in amrex.

[AlexanderSinn]

I tested HiPACE++ running on two 48-core CPUs with 4095*4095*100 cells and a tile size of 256^2. While the lockAdd version is faster, particularly at lower thread counts, it does not solve the bad scaling past 48 threads.