Tune scan on A100

[gevtushenko]

Description

Partially addresses #238

A100 SXM

I8	I16	I32	I64	I128	F32	F64
-7.38%	-12.47%	-7.07%	-12.34%	-39.34%	-6.22%	-9.67%

T{ct}	OffsetT{ct}	Elements{io}	Ref Time	Ref Noise	Cmp Time	Cmp Noise	Diff	%Diff	Status
I8	I32	2^16	11.274 us	1.62%	11.206 us	1.24%	-0.068 us	-0.60%	PASS
I8	I32	2^20	15.281 us	1.11%	18.514 us	38.02%	3.232 us	21.15%	FAIL
I8	I32	2^24	66.328 us	4.75%	68.364 us	1.90%	2.035 us	3.07%	FAIL
I8	I32	2^28	985.373 us	0.50%	912.611 us	0.28%	-72.762 us	-7.38%	FAIL
I16	I32	2^16	11.454 us	1.74%	11.843 us	1.89%	0.390 us	3.40%	FAIL
I16	I32	2^20	16.281 us	1.14%	18.805 us	29.97%	2.524 us	15.50%	FAIL
I16	I32	2^24	75.730 us	3.34%	72.374 us	2.28%	-3.356 us	-4.43%	FAIL
I16	I32	2^28	1.096 ms	0.62%	959.182 us	0.36%	-136.606 us	-12.47%	FAIL
I32	I32	2^16	12.001 us	1.93%	11.339 us	1.52%	-0.662 us	-5.52%	FAIL
I32	I32	2^20	17.548 us	1.27%	17.540 us	1.71%	-0.007 us	-0.04%	PASS
I32	I32	2^24	101.450 us	2.59%	97.099 us	2.18%	-4.351 us	-4.29%	FAIL
I32	I32	2^28	1.461 ms	0.93%	1.358 ms	0.61%	-103.249 us	-7.07%	FAIL
I64	I32	2^16	12.204 us	1.42%	12.424 us	1.27%	0.219 us	1.80%	FAIL
I64	I32	2^20	23.132 us	6.37%	23.322 us	2.01%	0.189 us	0.82%	PASS
I64	I32	2^24	193.714 us	3.63%	174.254 us	1.29%	-19.459 us	-10.05%	FAIL
I64	I32	2^28	2.918 ms	1.05%	2.558 ms	0.50%	-360.115 us	-12.34%	FAIL
I128	I32	2^16	17.732 us	2.08%	19.449 us	0.91%	1.716 us	9.68%	FAIL
I128	I32	2^20	67.733 us	1.93%	66.599 us	2.08%	-1.134 us	-1.67%	PASS
I128	I32	2^24	901.361 us	0.40%	573.754 us	0.35%	-327.608 us	-36.35%	FAIL
I128	I32	2^28	14.352 ms	0.30%	8.706 ms	0.18%	-5645.326 us	-39.34%	FAIL
F32	I32	2^16	11.611 us	2.22%	11.010 us	2.15%	-0.601 us	-5.18%	FAIL
F32	I32	2^20	17.474 us	1.39%	17.918 us	1.48%	0.444 us	2.54%	FAIL
F32	I32	2^24	101.294 us	2.91%	97.482 us	2.07%	-3.812 us	-3.76%	FAIL
F32	I32	2^28	1.461 ms	1.01%	1.371 ms	0.59%	-90.847 us	-6.22%	FAIL
F64	I32	2^16	12.154 us	1.40%	11.932 us	1.62%	-0.223 us	-1.83%	FAIL
F64	I32	2^20	23.647 us	1.53%	23.833 us	6.12%	0.185 us	0.78%	PASS
F64	I32	2^24	193.145 us	4.23%	177.964 us	1.27%	-15.181 us	-7.86%	FAIL
F64	I32	2^28	2.911 ms	1.25%	2.630 ms	0.50%	-281.643 us	-9.67%	FAIL
C64	I32	2^16	14.414 us	2.30%	14.516 us	1.81%	0.101 us	0.70%	PASS
C64	I32	2^20	32.692 us	1.77%	32.494 us	1.72%	-0.197 us	-0.60%	PASS
C64	I32	2^24	313.287 us	0.68%	313.224 us	0.67%	-0.063 us	-0.02%	PASS
C64	I32	2^28	4.894 ms	0.33%	4.893 ms	0.35%	-0.713 us	-0.01%	PASS

A100 PCIe

I8	I16	I32	I64	I128	F32	F64
-8.56%	-13.04%	-4.8%	-7.01%	-39.26%	-3.27%	-4.93%

T{ct}	OffsetT{ct}	Elements{io}	Ref Time	Ref Noise	Cmp Time	Cmp Noise	Diff	%Diff	Status
I8	I32	2^16	10.568 us	13.61%	10.430 us	12.75%	-0.138 us	-1.30%	PASS
I8	I32	2^20	14.421 us	1.28%	16.937 us	34.15%	2.517 us	17.45%	FAIL
I8	I32	2^24	66.860 us	3.65%	68.554 us	1.82%	1.694 us	2.53%	FAIL
I8	I32	2^28	1.009 ms	0.50%	922.707 us	0.33%	-86.393 us	-8.56%	FAIL
I16	I32	2^16	10.321 us	1.95%	10.345 us	1.65%	0.023 us	0.23%	PASS
I16	I32	2^20	15.722 us	1.23%	16.695 us	16.36%	0.974 us	6.19%	FAIL
I16	I32	2^24	79.256 us	3.20%	75.415 us	2.42%	-3.841 us	-4.85%	FAIL
I16	I32	2^28	1.171 ms	0.91%	1.018 ms	0.60%	-152.608 us	-13.04%	FAIL
I32	I32	2^16	10.494 us	1.93%	10.458 us	1.93%	-0.036 us	-0.35%	PASS
I32	I32	2^20	17.621 us	1.61%	17.774 us	1.55%	0.153 us	0.87%	PASS
I32	I32	2^24	116.338 us	1.84%	112.834 us	1.23%	-3.504 us	-3.01%	FAIL
I32	I32	2^28	1.712 ms	0.73%	1.629 ms	0.50%	-82.209 us	-4.80%	FAIL
I64	I32	2^16	11.498 us	2.00%	11.362 us	1.59%	-0.136 us	-1.18%	PASS
I64	I32	2^20	24.167 us	1.30%	24.570 us	1.59%	0.403 us	1.67%	FAIL
I64	I32	2^24	221.281 us	1.40%	210.492 us	0.73%	-10.789 us	-4.88%	FAIL
I64	I32	2^28	3.411 ms	0.81%	3.172 ms	0.50%	-239.059 us	-7.01%	FAIL
I128	I32	2^16	17.267 us	2.24%	18.614 us	1.27%	1.347 us	7.80%	FAIL
I128	I32	2^20	68.209 us	1.74%	68.623 us	1.81%	0.414 us	0.61%	PASS
I128	I32	2^24	927.940 us	0.41%	584.699 us	0.42%	-343.241 us	-36.99%	FAIL
I128	I32	2^28	14.797 ms	0.33%	8.988 ms	0.17%	-5809.738 us	-39.26%	FAIL
F32	I32	2^16	10.120 us	1.96%	10.131 us	2.72%	0.010 us	0.10%	PASS
F32	I32	2^20	17.474 us	1.35%	17.849 us	1.63%	0.375 us	2.15%	FAIL
F32	I32	2^24	116.126 us	2.10%	113.764 us	1.13%	-2.362 us	-2.03%	FAIL
F32	I32	2^28	1.696 ms	0.71%	1.641 ms	0.50%	-55.495 us	-3.27%	FAIL
F64	I32	2^16	11.283 us	1.96%	11.031 us	1.52%	-0.252 us	-2.24%	FAIL
F64	I32	2^20	24.829 us	1.17%	25.073 us	9.31%	0.243 us	0.98%	PASS
F64	I32	2^24	220.856 us	2.02%	211.677 us	0.70%	-9.178 us	-4.16%	FAIL
F64	I32	2^28	3.355 ms	0.73%	3.190 ms	0.50%	-165.506 us	-4.93%	FAIL
C64	I32	2^16	13.720 us	2.14%	13.400 us	2.11%	-0.321 us	-2.34%	FAIL
C64	I32	2^20	33.843 us	1.58%	33.702 us	1.60%	-0.141 us	-0.42%	PASS
C64	I32	2^24	334.440 us	0.63%	334.209 us	0.67%	-0.231 us	-0.07%	PASS
C64	I32	2^28	5.239 ms	0.40%	5.238 ms	0.43%	-0.412 us	-0.01%	PASS

Checklist

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

[elstehle]

Impressively large tile sizes for __[u]int128 😮