Tune unique by key on A100

[gevtushenko]

Description

Partially addresses #238

A100 SXM

Max Segment Size 2

KeyT{ct}	I8	I16	I32	I64	F32	F64
I8	-32.11%	-27.03%	-19.81%	-5.60%	-19.73%	-5.43%
I16	-27.58%	-32.65%	-14.29%	-4.88%	-13.83%	-4.80%
I32	-25.13%	-14.87%	-10.32%	-5.10%	-10.41%	-5.12%
I64	-28.08%	-27.72%	-22.54%	-13.01%	-22.59%	-13.11%

Max Segment Size 16

KeyT{ct}	I8	I16	I32	I64	F32	F64
I8	-31.92%	-29.20%	-22.21%	-19.24%	-22.23%	-19.29%
I16	-32.47%	-38.83%	-32.07%	-19.74%	-32.12%	-19.77%
I32	-32.64%	-34.54%	-27.59%	-15.71%	-27.47%	-15.76%
I64	-42.60%	-45.65%	-40.30%	-30.18%	-40.34%	-30.21%

Max Segment Size 256

KeyT{ct}	I8	I16	I32	I64	F32	F64
I8	-32.79%	-29.72%	-21.35%	-20.26%	-21.44%	-20.46%
I16	-33.44%	-40.40%	-34.44%	-24.35%	-34.40%	-24.33%
I32	-33.00%	-36.08%	-29.31%	-21.51%	-29.41%	-21.56%
I64	-44.41%	-47.76%	-46.25%	-36.61%	-46.26%	-36.42%

A100 PCIe

Max Segment Size 2

KeyT{ct}	I8	I16	I32	I64	F32	F64
I8	-31.24%	-25.57%	-14.77%	-1.57%	-15.61%	-2.06%
I16	-26.50%	-26.36%	-8.97%	-2.35%	-10.30%	-2.39%
I32	-19.70%	-10.64%	-5.45%	-2.65%	-5.45%	-2.63%
I64	-20.59%	-18.57%	-12.92%	-6.58%	-12.91%	-6.63%

Max Segment Size 16

KeyT{ct}	I8	I16	I32	I64	F32	F64
I8	-31.61%	-27.97%	-21.09%	-14.15%	-21.29%	-14.17%
I16	-31.70%	-37.15%	-26.75%	-13.71%	-26.90%	-13.66%
I32	-31.22%	-29.08%	-20.86%	-8.89%	-20.83%	-8.88%
I64	-39.26%	-38.84%	-31.85%	-20.88%	-31.82%	-20.83%

Max Segment Size 256

KeyT{ct}	I8	I16	I32	I64	F32	F64
I8	-32.39%	-29.16%	-21.12%	-16.78%	-21.11%	-16.75%
I16	-32.88%	-38.95%	-30.44%	-17.48%	-29.92%	-17.42%
I32	-32.18%	-32.17%	-24.15%	-12.24%	-23.96%	-12.21%
I64	-42.86%	-43.47%	-37.58%	-26.41%	-37.56%	-26.42%

Checklist

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