@neelakausik, essentially you will need to create a device pool, a pinned pool, and a unified memory pool per thread, then use those to create a resource per thread. The call to init_care_for_testing() above creates a GPU wide pool called "DEVICE_POOL" and another pool called "PINNED_POOL". Those should be the basis for the per thread device and pinned memory pools. On line 53 I create a GPU wide unified memory pool called "UM_POOL". You will need to create that manually in whatever test cases you have, then have that allocator be the basis for the per thread unified memory pool.

You should test out this functionality in standalone CARE. Write some tests and benchmarks that involve RAJA reducers, scans, and CARE loops.