add IntSuperAccumulator

[BenWibking]

Summary

Adds an integer superaccumulator class that performs reproducible summation of {non-subnormal, non-NAN, non-Inf} non-negative single-precision floats on device.

GPU atomics code path is currently very slow for many summands.

Additional background

Partially addresses #4669.

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

[BenWibking]

bwibking@moth:~/amrex/Tests/Numerics/IntSuperAccumulator> time ./main3d.hip.FLOAT.HIP.ex
Initializing AMReX (22.03-1419-gd364b6bd6475)...
Initializing HIP...
HIP initialized with 1 device.
AMReX (22.03-1419-gd364b6bd6475) initialized
Range [1.175494351e-38, 2.350988562e-38] with count 32768: reference=5.774813904e-34 accumulator=5.774813904e-34
Range [4.253529587e+37, 8.507059173e+37] with count 4: reference=2.543284763e+38 accumulator=2.543284763e+38
Range [9.99999996e-13, 0.001000000047] with count 32768: reference=16.3573494 accumulator=16.3573494
Range [9.999999747e-05, 1] with count 65536: reference=32755.99805 accumulator=32755.99805
Range [1, 1000000] with count 65536: reference=3.265710694e+10 accumulator=3.265710694e+10
Range [0.009999999776, 100000000] with count 131072: reference=6.547726926e+12 accumulator=6.547726926e+12
Constant accumulation: reference=25000 accumulator=25000
Total GPU global memory (MB): 65520
Free  GPU global memory (MB): 16162
[The         Arena] max space allocated (MB): 49140
[The         Arena] max space used      (MB): 39
[The Managed Arena] max space allocated (MB): 8
[The Managed Arena] max space used      (MB): 0
[The  Pinned Arena] max space allocated (MB): 8
[The  Pinned Arena] max space used      (MB): 0
AMReX (22.03-1419-gd364b6bd6475) finalized

real	0m14.501s
user	0m13.866s
sys	0m0.579s
bwibking@moth:~/amrex/Tests/Numerics/IntSuperAccumulator> time ./main3d.gnu.FLOAT.ex
Initializing AMReX (22.03-1419-gd364b6bd6475)...
AMReX (22.03-1419-gd364b6bd6475) initialized
Range [1.175494351e-38, 2.350988562e-38] with count 32768: reference=5.774813904e-34 accumulator=5.774813904e-34
Range [4.253529587e+37, 8.507059173e+37] with count 4: reference=2.543284763e+38 accumulator=2.543284763e+38
Range [9.99999996e-13, 0.001000000047] with count 32768: reference=16.3573494 accumulator=16.3573494
Range [9.999999747e-05, 1] with count 65536: reference=32755.99805 accumulator=32755.99805
Range [1, 1000000] with count 65536: reference=3.265710694e+10 accumulator=3.265710694e+10
Range [0.009999999776, 100000000] with count 131072: reference=6.547726926e+12 accumulator=6.547726926e+12
Constant accumulation: reference=25000 accumulator=25000
AMReX (22.03-1419-gd364b6bd6475) finalized

real	0m0.016s
user	0m0.010s
sys	0m0.005s