Tune RLE on A100

[gevtushenko]

Description

Partially addresses #238

encode

A100 SXM4

MaxSegSize	I8	I16	I32	I64	I128	F32	F64
2^1	-30.23%	-28.87%	-17.80%	-6.43%	-0.92%	-34.61%	-5.49%
2^4	-48.77%	-41.89%	-40.58%	-23.53%	-13.27%	-37.00%	-20.40%
2^8	-51.13%	-44.11%	-42.55%	-21.86%	-20.44%	-40.23%	-19.05%

A100 PCIe

MaxSegSize	I8	I16	I32	I64	I128	F32	F64
2^1	-30.89%	-28.14%	-9.72%	1.02%	-1.02%	-30.77%	1.38%
2^4	-48.27%	-41.02%	-38.93%	-22.86%	-11.70%	-34.74%	-20.12%
2^8	-50.90%	-43.80%	-41.71%	-21.89%	-16.06%	-38.98%	-18.93%

non-trivial-runs

A100 SXM4

MaxSegSize	I8	I16	I32	I64	I128	F32	F64
2^1	-33.47%	-31.65%	-26.23%	-43.87%	-55.21%	-20.57%	-43.79%
2^4	-28.08%	-25.89%	-21.81%	-40.52%	-52.93%	-19.47%	-39.74%
2^8	-29.85%	-27.99%	-24.18%	-41.40%	-54.58%	-24.34%	-42.78%

A100 PCIe

MaxSegSize	I8	I16	I32	I64	I128	F32	F64
2^1	-32.88%	-31.10%	-24.48%	-35.71%	-47.02%	-18.16%	-34.66%
2^4	-28.03%	-24.45%	-19.89%	-34.64%	-44.50%	-17.44%	-33.55%
2^8	-29.40%	-27.70%	-22.61%	-34.95%	-46.34%	-22.09%	-34.17%

Checklist

• New or existing tests cover these changes.
• The documentation is up to date with these changes.