Vectorize Png encoder filters

[TechPizzaDev]

Prerequisites

• I have written a descriptive pull-request title
• I have verified that there are no overlapping pull-requests open
• I have verified that I am following the existing coding patterns and practice as demonstrated in the repository. These follow strict Stylecop rules 👮.
• I have provided test coverage for my change (where applicable)

Description

[CLAassistant]

All committers have signed the CLA.

[TechPizzaDev]

And here are some excessively detailed benchmarks on random RGB (and Alpha 255) data.

Environment

BenchmarkDotNet=v0.12.1, OS=Windows 10.0.19042
Intel Core i7-4720HQ CPU 2.60GHz (Haswell), 1 CPU, 8 logical and 4 physical cores
.NET Core SDK=5.0.202
  [Host] : .NET Core 5.0.5 (CoreCLR 5.0.521.16609, CoreFX 5.0.521.16609), X64 RyuJIT

Job=MediumRun  EvaluateOverhead=True  Toolchain=InProcessEmitToolchain  
IterationCount=15  LaunchCount=1  WarmupCount=10  

Average

Type	Method	Size	Mean	Error	StdDev	Ratio
Average	Average	64	38.834 μs	0.1914 μs	0.1494 μs	1.000
Average	Ssse3Average	64	5.135 μs	0.0210 μs	0.0186 μs	0.132
Average	Sse2Average	64	6.175 μs	0.3104 μs	0.2903 μs	0.159
Average	Avx2Average	64	3.599 μs	0.0302 μs	0.0267 μs	0.093
Average	Average	256	672.200 μs	25.7805 μs	22.8537 μs	1.000
Average	Ssse3Average	256	85.646 μs	1.5865 μs	1.3248 μs	0.127
Average	Sse2Average	256	92.483 μs	1.2481 μs	1.0422 μs	0.138
Average	Avx2Average	256	57.248 μs	1.3243 μs	1.2388 μs	0.085
Average	Average	1024	9,968.715 μs	59.5136 μs	55.6691 μs	1.000
Average	Ssse3Average	1024	1,546.775 μs	8.0654 μs	7.1497 μs	0.155
Average	Sse2Average	1024	1,652.449 μs	7.1568 μs	5.9763 μs	0.166
Average	Avx2Average	1024	1,268.928 μs	18.3060 μs	17.1234 μs	0.127
Average	Average	4096	177,359.313 μs	1,886.8009 μs	1,764.9148 μs	1.000
Average	Ssse3Average	4096	27,926.817 μs	226.7537 μs	189.3496 μs	0.157
Average	Sse2Average	4096	29,114.548 μs	176.3376 μs	164.9463 μs	0.164
Average	Avx2Average	4096	19,337.158 μs	206.2467 μs	182.8324 μs	0.109

Paeth

Type	Method	Size	Mean	Error	StdDev	Ratio
Paeth	Paeth	64	147.512 μs	1.0109 μs	0.8961 μs	1.000
Paeth	VectorPaeth	64	17.381 μs	0.0630 μs	0.0492 μs	0.118
Paeth	UnsafeVectorPaeth	64	16.269 μs	0.0782 μs	0.0653 μs	0.110
Paeth	Paeth	256	2,541.548 μs	12.2208 μs	10.8334 μs	1.000
Paeth	VectorPaeth	256	271.942 μs	1.6199 μs	1.4360 μs	0.107
Paeth	UnsafeVectorPaeth	256	255.983 μs	0.8942 μs	0.7927 μs	0.101
Paeth	Paeth	1024	40,888.074 μs	142.4349 μs	133.2337 μs	1.000
Paeth	VectorPaeth	1024	4,403.322 μs	11.5166 μs	10.2092 μs	0.108
Paeth	UnsafeVectorPaeth	1024	4,241.534 μs	76.9621 μs	68.2249 μs	0.104
Paeth	Paeth	4096	664,667.564 μs	2,478.8314 μs	2,197.4195 μs	1.000
Paeth	VectorPaeth	4096	73,365.707 μs	360.9557 μs	319.9778 μs	0.110
Paeth	UnsafeVectorPaeth	4096	67,785.956 μs	1,067.5039 μs	946.3144 μs	0.102

Up

Type	Method	Size	Mean	Error	StdDev	Ratio
Up	Up	64	28.109 μs	0.0926 μs	0.0774 μs	1.000
Up	VectorUp	64	3.464 μs	0.0113 μs	0.0106 μs	0.123
Up	UnsafeVectorUp	64	3.208 μs	0.0186 μs	0.0165 μs	0.114
Up	Up	256	449.780 μs	2.2934 μs	2.1453 μs	1.000
Up	VectorUp	256	59.635 μs	0.2284 μs	0.2025 μs	0.133
Up	UnsafeVectorUp	256	52.384 μs	0.2513 μs	0.2098 μs	0.116
Up	Up	1024	7,226.976 μs	33.3938 μs	29.6027 μs	1.000
Up	VectorUp	1024	1,174.367 μs	14.2758 μs	12.6551 μs	0.162
Up	UnsafeVectorUp	1024	1,179.356 μs	9.2726 μs	8.2199 μs	0.163
Up	Up	4096	137,018.553 μs	696.9427 μs	651.9207 μs	1.000
Up	VectorUp	4096	18,564.646 μs	164.0998 μs	128.1183 μs	0.135
Up	UnsafeVectorUp	4096	18,764.654 μs	153.3329 μs	135.9256 μs	0.137

Sub

Type	Method	Size	Mean	Error	StdDev	Ratio
Sub	Sub	64	31.704 μs	0.2713 μs	0.2405 μs	1.000
Sub	VectorSub	64	3.358 μs	0.0158 μs	0.0132 μs	0.106
Sub	UnsafeVectorSub	64	3.166 μs	0.0057 μs	0.0051 μs	0.100
Sub	Sub	256	504.137 μs	1.5142 μs	1.4164 μs	1.000
Sub	VectorSub	256	56.196 μs	0.2222 μs	0.1855 μs	0.111
Sub	UnsafeVectorSub	256	50.632 μs	0.6273 μs	0.5561 μs	0.100
Sub	Sub	1024	8,162.503 μs	26.4349 μs	22.0743 μs	1.000
Sub	VectorSub	1024	923.093 μs	4.9075 μs	4.5905 μs	0.113
Sub	UnsafeVectorSub	1024	922.966 μs	13.2451 μs	11.7415 μs	0.113
Sub	Sub	4096	155,133.122 μs	971.6967 μs	908.9257 μs	1.000
Sub	VectorSub	4096	14,224.630 μs	119.9329 μs	93.6357 μs	0.092
Sub	UnsafeVectorSub	4096	14,952.815 μs	82.6097 μs	77.2732 μs	0.096

[codecov]

Codecov Report

Merging #1630 (4cfc701) into master (a8cb711) will increase coverage by 0.08%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1630      +/-   ##
==========================================
+ Coverage   83.67%   83.75%   +0.08%     
==========================================
  Files         749      749              
  Lines       33111    33275     +164     
  Branches     3714     3736      +22     
==========================================
+ Hits        27707    27871     +164     
  Misses       4682     4682              
  Partials      722      722              

Flag	Coverage Δ
unittests	83.75% <100.00%> (+0.08%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/ImageSharp/Common/Helpers/Numerics.cs	97.23% <100.00%> (+0.23%)	⬆️
...rc/ImageSharp/Formats/Png/Filters/AverageFilter.cs	100.00% <100.00%> (ø)
src/ImageSharp/Formats/Png/Filters/PaethFilter.cs	100.00% <100.00%> (ø)
src/ImageSharp/Formats/Png/Filters/SubFilter.cs	100.00% <100.00%> (ø)
src/ImageSharp/Formats/Png/Filters/UpFilter.cs	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 993f96e...4cfc701. Read the comment docs.

[brianpopow]

@TechnologicalPizza thank you for the contribution, the performance gains look very promising 👍

Unit Tests for this could be done with the help of the FeatureTestRunner

See TransposeInto tests in Block8x8FTests.cs:

ImageSharp/tests/ImageSharp.Tests/Formats/Jpg/Block8x8FTests.cs

Line 167 in 993f96e

public void TransposeInto()

[JimBobSquarePants]

I cannot adequately describe how much joy it brings me to see such fantastic collaboration here! Really great to see everyone pitching in with ideas. 😄

[TechPizzaDev]

Environment

BenchmarkDotNet=v0.12.1, OS=Windows 10.0.19042
Intel Core i7-4720HQ CPU 2.60GHz (Haswell), 1 CPU, 8 logical and 4 physical cores
.NET Core SDK=5.0.202
  [Host] : .NET Core 5.0.5 (CoreCLR 5.0.521.16609, CoreFX 5.0.521.16609), X64 RyuJIT

Job=MediumRun  EvaluateOverhead=True  Toolchain=InProcessEmitToolchain  
IterationCount=15  LaunchCount=1  WarmupCount=10

Paeth 🥳

Method	Size	Mean	Error	StdDev	Ratio	Code Size
Paeth	256	2,533.20 μs	5.760 μs	4.810 μs	1.00	386 B
VectorPaeth	256	256.20 μs	0.683 μs	0.570 μs	0.10	386 B
Avx2Paeth	256	65.25 μs	0.238 μs	0.223 μs	0.03	386 B
Paeth	1024	40,717.35 μs	62.354 μs	52.068 μs	1.00	386 B
VectorPaeth	1024	3,999.67 μs	18.097 μs	16.928 μs	0.10	386 B
Avx2Paeth	1024	1,335.79 μs	60.784 μs	53.884 μs	0.03	386 B
Paeth	4096	660,246.26 μs	1,574.296 μs	1,472.598 μs	1.00	386 B
VectorPaeth	4096	68,048.32 μs	206.054 μs	192.743 μs	0.10	386 B
Avx2Paeth	4096	19,532.50 μs	84.610 μs	79.145 μs	0.03	386 B

Average

Method	Size	Mean	Error	StdDev	Ratio
Average	256	620.15 μs	5.117 μs	4.786 μs	1.00
Ssse3Average	256	64.73 μs	0.177 μs	0.165 μs	0.10
Avx2Average	256	33.87 μs	0.072 μs	0.060 μs	0.05
Average	1024	9,897.43 μs	46.592 μs	41.303 μs	1.00
Ssse3Average	1024	1,325.84 μs	7.321 μs	6.848 μs	0.13
Avx2Average	1024	1,055.26 μs	35.832 μs	33.517 μs	0.11
Average	4096	176,043.52 μs	400.279 μs	354.837 μs	1.00
Ssse3Average	4096	23,177.05 μs	162.236 μs	135.474 μs	0.13
Avx2Average	4096	17,104.39 μs	188.302 μs	166.925 μs	0.10

Up

Method	Size	Mean	Error	StdDev	Ratio
Up	256	447.54 μs	0.267 μs	0.208 μs	1.00
VectorUp	256	52.68 μs	0.184 μs	0.163 μs	0.12
Avx2Up	256	28.81 μs	0.117 μs	0.097 μs	0.06
Up	1024	7,239.53 μs	34.194 μs	31.985 μs	1.00
VectorUp	1024	1,225.86 μs	12.761 μs	11.312 μs	0.17
Avx2Up	1024	1,001.42 μs	13.296 μs	11.103 μs	0.14
Up	4096	136,730.78 μs	495.719 μs	439.442 μs	1.00
VectorUp	4096	18,613.86 μs	127.692 μs	99.693 μs	0.14
Avx2Up	4096	16,700.44 μs	177.679 μs	166.201 μs	0.12

Sub

Method	Size	Mean	Error	StdDev	Ratio
Sub	256	502.37 μs	1.479 μs	1.311 μs	1.00
VectorSub	256	50.32 μs	0.315 μs	0.294 μs	0.10
Avx2Sub	256	24.05 μs	0.117 μs	0.098 μs	0.05
Sub	1024	8,061.44 μs	29.533 μs	27.625 μs	1.00
VectorSub	1024	982.06 μs	15.100 μs	13.386 μs	0.12
Avx2Sub	1024	595.18 μs	25.069 μs	23.450 μs	0.07
Sub	4096	153,740.68 μs	385.423 μs	341.667 μs	1.00
VectorSub	4096	14,872.55 μs	111.238 μs	104.052 μs	0.10
Avx2Sub	4096	11,971.57 μs	34.938 μs	29.175 μs	0.08

[JimBobSquarePants]

Excellent stuff, let's get it merged! 🚀