Array Encoding/Decoding performance improvements (continued)

[danielmarbach]

This is a work in progress to show how array encoding/decoding of
primitive types can be improved.

Before

BenchmarkDotNet=v0.13.0, OS=Windows 10.0.19042.985 (20H2/October2020Update)
AMD Ryzen 9 3950X, 1 CPU, 32 logical and 16 physical cores
.NET SDK=6.0.100-preview.4.21255.9
  [Host]   : .NET 5.0.5 (5.0.521.16609), X64 RyuJIT
  ShortRun : .NET 5.0.5 (5.0.521.16609), X64 RyuJIT

Job=ShortRun  Runtime=.NET 5.0  IterationCount=3  
LaunchCount=1  WarmupCount=3  

Method	Mean	Error	StdDev	Gen 0	Gen 1	Gen 2	Allocated	Code Size
ArrayAmqpSymbolDecode_1M_MAA	535.44 μs	149.503 μs	8.195 μs	76.1719	26.3672	-	647,048 B	276 B
ArrayAmqpSymbolDecode_1K_MAA	52.14 μs	8.871 μs	0.486 μs	7.6904	0.9155	-	64,480 B	276 B
ArrayAmqpSymbolEncode_100K_MAA	1,496.76 μs	192.350 μs	10.543 μs	97.6563	-	-	819,048 B	226 B
ArrayAmqpSymbolEncode_1K_MAA	146.51 μs	40.250 μs	2.206 μs	9.7656	-	-	81,928 B	226 B
Bytes_Encode_MAA	23.43 μs	0.445 μs	0.024 μs	-	-	-	-	326 B
Bytes_Decode_MAA	320.12 μs	605.266 μs	33.177 μs	2.9297	2.9297	2.9297	1,048,600 B	555 B
ArrayInt32Encode_MAA_1M	101,466.66 μs	28,031.501 μs	1,536.501 μs	6000.0000	-	-	50,331,760 B	320 B
ArrayInt32Encode_MAA_1K	98.50 μs	16.658 μs	0.913 μs	5.8594	-	-	49,264 B	320 B
ArrayInt32Decode_1M_MAA	51,164.37 μs	4,089.352 μs	224.151 μs	3000.0000	-	-	29,361,027 B	276 B
ArrayInt32Decode_1K_MAA	54.88 μs	3.563 μs	0.195 μs	3.4180	-	-	28,696 B	276 B
ArrayUInt32Encode_MAA_1M	105,372.92 μs	32,508.110 μs	1,781.879 μs	6000.0000	-	-	50,331,760 B	320 B
ArrayUInt32Encode_MAA_1K	97.29 μs	19.274 μs	1.056 μs	5.8594	-	-	49,264 B	320 B
ArrayUInt32Decode_1M_MAA	57,301.39 μs	3,303.465 μs	181.074 μs	3000.0000	-	-	29,361,148 B	276 B
ArrayUInt32Decode_1K_MAA	58.39 μs	6.783 μs	0.372 μs	3.4180	-	-	28,696 B	276 B
ArrayInt64Encode_MAA_1M	105,115.89 μs	27,232.659 μs	1,492.714 μs	6000.0000	-	-	50,331,760 B	320 B
ArrayInt64Encode_MAA_1K	98.41 μs	43.838 μs	2.403 μs	5.8594	-	-	49,264 B	320 B
ArrayInt64Decode_1M_MAA	51,023.07 μs	17,808.911 μs	976.166 μs	3000.0000	-	-	33,554,456 B	276 B
ArrayInt64Decode_1K_MAA	47.36 μs	3.820 μs	0.209 μs	3.9063	0.0610	-	32,792 B	276 B
ArrayUInt64Encode_MAA_1M	102,957.17 μs	18,212.982 μs	998.315 μs	6000.0000	-	-	50,331,760 B	320 B
ArrayUInt64Encode_MAA_1K	96.94 μs	59.427 μs	3.257 μs	5.8594	-	-	49,264 B	320 B
ArrayUInt64Decode_1M_MAA	54,329.93 μs	30,664.377 μs	1,680.818 μs	3000.0000	-	-	33,554,456 B	276 B
ArrayUInt64Decode_1K_MAA	50.64 μs	24.464 μs	1.341 μs	3.9063	0.0610	-	32,792 B	276 B
ArrayBoolEncode_MAA_1M	104,928.10 μs	29,551.077 μs	1,619.794 μs	6000.0000	-	-	50,331,760 B	320 B
ArrayBoolEncode_MAA_1K	96.20 μs	16.166 μs	0.886 μs	5.8594	-	-	49,264 B	320 B
ArrayBoolDecode_1M_MAA	44,793.29 μs	2,699.893 μs	147.990 μs	3000.0000	-	-	26,214,424 B	276 B
ArrayBoolDecode_1K_MAA	44.57 μs	4.078 μs	0.224 μs	3.0518	-	-	25,624 B	276 B
ArrayDecimalEncode_MAA_1M	132,444.94 μs	73,831.727 μs	4,046.966 μs	18000.0000	-	-	150,995,072 B	320 B
ArrayDecimalEncode_MAA_1K	129.31 μs	70.490 μs	3.864 μs	17.5781	-	-	147,584 B	320 B
ArrayDecimalDecode_1M_MAA	112,627.93 μs	13,433.443 μs	736.332 μs	9000.0000	-	-	92,274,712 B	276 B
ArrayDecimalDecode_1K_MAA	108.87 μs	10.047 μs	0.551 μs	10.7422	0.6104	-	90,136 B	276 B
ArrayDoubleEncode_MAA_1M	101,110.83 μs	22,499.287 μs	1,233.262 μs	6000.0000	-	-	50,331,760 B	320 B
ArrayDoubleEncode_MAA_1K	98.77 μs	36.658 μs	2.009 μs	5.8594	-	-	49,264 B	320 B
ArrayDoubleDecode_1M_MAA	50,673.58 μs	18,513.259 μs	1,014.774 μs	3000.0000	-	-	33,554,456 B	276 B
ArrayDoubleDecode_1K_MAA	45.88 μs	2.758 μs	0.151 μs	3.9063	0.0610	-	32,792 B	276 B
ArrayFloatEncode_MAA_1M	100,870.10 μs	14,169.486 μs	776.677 μs	6000.0000	-	-	50,331,760 B	320 B
ArrayFloatEncode_MAA_1K	100.00 μs	20.726 μs	1.136 μs	5.8594	-	-	49,264 B	320 B
ArrayFloatDecode_1M_MAA	50,928.70 μs	10,357.976 μs	567.756 μs	3000.0000	-	-	29,360,227 B	276 B
ArrayFloatDecode_1K_MAA	49.35 μs	3.641 μs	0.200 μs	3.4180	-	-	28,696 B	276 B

After

BenchmarkDotNet=v0.13.0, OS=Windows 10.0.19042.985 (20H2/October2020Update)
AMD Ryzen 9 3950X, 1 CPU, 32 logical and 16 physical cores
.NET SDK=6.0.100-preview.4.21255.9
  [Host]   : .NET 5.0.5 (5.0.521.16609), X64 RyuJIT
  ShortRun : .NET 5.0.5 (5.0.521.16609), X64 RyuJIT

Job=ShortRun  Runtime=.NET 5.0  IterationCount=3  
LaunchCount=1  WarmupCount=3  

Method	Mean	Error	StdDev	Gen 0	Gen 1	Gen 2	Allocated	Code Size
ArrayAmqpSymbolDecode_1M_MAA	337,431.5 ns	70,315.51 ns	3,854.23 ns	47.3633	15.6250	-	401,288 B	276 B
ArrayAmqpSymbolDecode_1K_MAA	31,422.2 ns	2,298.39 ns	125.98 ns	4.7607	0.4883	-	39,904 B	276 B
ArrayAmqpSymbolEncode_100K_MAA	600,549.1 ns	22,404.81 ns	1,228.08 ns	-	-	-	-	233 B
ArrayAmqpSymbolEncode_1K_MAA	58,396.3 ns	2,661.36 ns	145.88 ns	-	-	-	-	233 B
Bytes_Encode_MAA	22,882.8 ns	2,215.89 ns	121.46 ns	-	-	-	-	411 B
Bytes_Decode_MAA	342,778.9 ns	890,252.17 ns	48,797.73 ns	2.9297	2.9297	2.9297	1,048,601 B	555 B
ArrayInt32Encode_MAA_1M	851,541.3 ns	39,990.25 ns	2,192.00 ns	-	-	-	-	327 B
ArrayInt32Encode_MAA_1K	950.1 ns	51.39 ns	2.82 ns	-	-	-	-	327 B
ArrayInt32Decode_1M_MAA	2,019,578.2 ns	745,896.04 ns	40,885.08 ns	11.7188	11.7188	11.7188	4,194,332 B	276 B
ArrayInt32Decode_1K_MAA	972.9 ns	106.33 ns	5.83 ns	0.4921	-	-	4,120 B	276 B
ArrayUInt32Encode_MAA_1M	848,060.4 ns	12,011.16 ns	658.37 ns	-	-	-	-	327 B
ArrayUInt32Encode_MAA_1K	940.0 ns	115.65 ns	6.34 ns	-	-	-	-	327 B
ArrayUInt32Decode_1M_MAA	2,040,378.3 ns	564,903.81 ns	30,964.28 ns	11.7188	11.7188	11.7188	4,194,332 B	276 B
ArrayUInt32Decode_1K_MAA	1,010.7 ns	94.37 ns	5.17 ns	0.4921	-	-	4,120 B	276 B
ArrayInt64Encode_MAA_1M	907,390.7 ns	89,501.36 ns	4,905.87 ns	-	-	-	-	327 B
ArrayInt64Encode_MAA_1K	952.1 ns	115.76 ns	6.35 ns	-	-	-	-	327 B
ArrayInt64Decode_1M_MAA	3,383,721.9 ns	3,818,522.36 ns	209,306.10 ns	23.4375	23.4375	23.4375	8,388,640 B	276 B
ArrayInt64Decode_1K_MAA	1,115.4 ns	77.21 ns	4.23 ns	0.9804	-	-	8,216 B	276 B
ArrayUInt64Encode_MAA_1M	929,740.7 ns	614,016.69 ns	33,656.33 ns	-	-	-	-	327 B
ArrayUInt64Encode_MAA_1K	946.0 ns	35.88 ns	1.97 ns	-	-	-	-	327 B
ArrayUInt64Decode_1M_MAA	3,450,412.7 ns	1,869,324.49 ns	102,463.98 ns	23.4375	23.4375	23.4375	8,388,640 B	276 B
ArrayUInt64Decode_1K_MAA	1,106.5 ns	90.17 ns	4.94 ns	0.9804	-	-	8,216 B	276 B
ArrayBoolEncode_MAA_1M	1,012,373.1 ns	53,120.62 ns	2,911.72 ns	-	-	-	1 B	327 B
ArrayBoolEncode_MAA_1K	1,063.6 ns	44.50 ns	2.44 ns	-	-	-	-	327 B
ArrayBoolDecode_1M_MAA	1,198,136.6 ns	191,227.15 ns	10,481.81 ns	-	-	-	1,048,601 B	276 B
ArrayBoolDecode_1K_MAA	1,073.1 ns	145.35 ns	7.97 ns	0.1240	-	-	1,048 B	276 B
ArrayDecimalEncode_MAA_1M	22,354,117.7 ns	7,959,300.14 ns	436,276.10 ns	5000.0000	-	-	41,943,049 B	327 B
ArrayDecimalEncode_MAA_1K	20,883.4 ns	1,987.14 ns	108.92 ns	4.8828	-	-	40,960 B	327 B
ArrayDecimalDecode_1M_MAA	57,522,114.8 ns	12,834,405.35 ns	703,497.06 ns	-	-	-	16,777,391 B	276 B
ArrayDecimalDecode_1K_MAA	52,401.2 ns	1,246.96 ns	68.35 ns	1.9531	-	-	16,408 B	276 B
ArrayDoubleEncode_MAA_1M	1,053,240.8 ns	823,822.43 ns	45,156.49 ns	-	-	-	1 B	327 B
ArrayDoubleEncode_MAA_1K	1,064.5 ns	39.81 ns	2.18 ns	-	-	-	-	327 B
ArrayDoubleDecode_1M_MAA	3,339,528.0 ns	2,736,238.70 ns	149,982.48 ns	23.4375	23.4375	23.4375	8,388,640 B	276 B
ArrayDoubleDecode_1K_MAA	1,092.7 ns	250.16 ns	13.71 ns	0.9804	-	-	8,216 B	276 B
ArrayFloatEncode_MAA_1M	1,256,692.6 ns	45,829.78 ns	2,512.08 ns	-	-	-	1 B	327 B
ArrayFloatEncode_MAA_1K	1,315.5 ns	44.92 ns	2.46 ns	-	-	-	-	327 B
ArrayFloatDecode_1M_MAA	2,094,115.0 ns	3,181,908.62 ns	174,411.15 ns	7.8125	7.8125	7.8125	4,194,331 B	276 B
ArrayFloatDecode_1K_MAA	1,096.8 ns	282.48 ns	15.48 ns	0.4921	-	-	4,120 B	276 B

[danielmarbach]

@xinchen10 I think I'm missing one primitive here and running the latest results in the benchmark (which is still quite a bit of effort)

Can you have a look and tell me what you think or if there is anything missing from your side?

[xinchen10]

@danielmarbach I took a quick look at the changes and I think they look good. I will look more closely at the changes later.