Summary

There are two very different types of functions called amrex::launch. One gives raw access to a GPU block to be used with shared memory. The other type iterates over threads more like ParallelFor but gives a Box to the functor, not an IntVect. For GPU the box will only have one element, for CPU this is the same box that was passed into launch. This was used to quickly port amrex CPU code to GPU. However, I find this quite confusing.

This PR replaces the uses of the second type of launch inside amrex with ParallelFor and removes the section in the documentation that covered it. This avoids having to write the 4D loop inside every kernel.

TODO (in a future PR?): replace ...LAMBDA_RANGE macros used in LinearSolvers.

Additional background

Checklist

The proposed changes:

• fix a bug or incorrect behavior in AMReX
• add new capabilities to AMReX
• changes answers in the test suite to more than roundoff level
• are likely to significantly affect the results of downstream AMReX users
• include documentation in the code and/or rst files, if appropriate