Optimize jpeg decoder based on traces.

[JimBobSquarePants]

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 matches 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

I ran a trace against the decoder after reading this benchmark here (which we're doing really well with IMO) and thought I'd clean up some low hanging fruit. Shaved a few percent of the decoding benchmarks.

Decode

Adds hardware accelerated rounding to SimdUtils.FastRound.

Before and after performance traces.

Method	TestImage	Mean	Error	StdDev	Ratio	RatioSD	Gen 0	Gen 1	Gen 2	Allocated
'Decode Jpeg - System.Drawing'	Jpg/b(...)e.jpg [21]	5.122 ms	1.3978 ms	0.0766 ms	1.00	0.00	-	-	-	176 B
'Decode Jpeg - ImageSharp'	Jpg/b(...)e.jpg [21]	11.991 ms	0.2514 ms	0.0138 ms	2.34	0.03	-	-	-	15816 B
'Decode Jpeg - System.Drawing'	Jpg/b(...)f.jpg [28]	14.943 ms	1.8410 ms	0.1009 ms	1.00	0.00	-	-	-	176 B
'Decode Jpeg - ImageSharp'	Jpg/b(...)f.jpg [28]	29.759 ms	1.5452 ms	0.0847 ms	1.99	0.01	-	-	-	16824 B
'Decode Jpeg - System.Drawing'	Jpg/i(...)e.jpg [43]	388.229 ms	382.8946 ms	20.9877 ms	1.00	0.00	-	-	-	216 B
'Decode Jpeg - ImageSharp'	Jpg/i(...)e.jpg [43]	276.490 ms	195.5104 ms	10.7166 ms	0.71	0.01	-	-	-	36022368 B

Scan Only
Slight refactor of HuffmanScanBuffer.DecodeHuffman to remove double lookup on slow path. There's now nothing that stands out here.

Method	Job	Runtime	TestImage	Mean	Error	StdDev	Ratio	Gen 0	Gen 1	Gen 2	Allocated
'System.Drawing FULL'	Job-HITJFX	.NET 4.7.2	Jpg/b(...)e.jpg [21]	5.828 ms	0.9885 ms	0.0542 ms	1.00	46.8750	-	-	211566 B
JpegDecoderCore.ParseStream	Job-HITJFX	.NET 4.7.2	Jpg/b(...)e.jpg [21]	5.833 ms	0.2923 ms	0.0160 ms	1.00	-	-	-	12416 B
'System.Drawing FULL'	Job-WPSKZD	.NET Core 2.1	Jpg/b(...)e.jpg [21]	6.018 ms	2.1374 ms	0.1172 ms	1.00	46.8750	-	-	210768 B
JpegDecoderCore.ParseStream	Job-WPSKZD	.NET Core 2.1	Jpg/b(...)e.jpg [21]	4.382 ms	0.9009 ms	0.0494 ms	0.73	-	-	-	12360 B
'System.Drawing FULL'	Job-ZLSNRP	.NET Core 3.1	Jpg/b(...)e.jpg [21]	5.714 ms	0.4078 ms	0.0224 ms	1.00	-	-	-	176 B
JpegDecoderCore.ParseStream	Job-ZLSNRP	.NET Core 3.1	Jpg/b(...)e.jpg [21]	4.239 ms	1.0943 ms	0.0600 ms	0.74	-	-	-	12406 B

[JimBobSquarePants]

Updated to match the full decoding benchmark so we can more easily see the difference.

[codecov]

Codecov Report

Merging #1303 into master will decrease coverage by 0.02%.
The diff coverage is 65.00%.

@@            Coverage Diff             @@
##           master    #1303      +/-   ##
==========================================
- Coverage   82.79%   82.77%   -0.03%     
==========================================
  Files         689      689              
  Lines       30731    30733       +2     
  Branches     3472     3473       +1     
==========================================
- Hits        25444    25439       -5     
- Misses       4581     4587       +6     
- Partials      706      707       +1     

Flag	Coverage Δ
#unittests	82.77% <65.00%> (-0.03%)	⬇️

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

Impacted Files	Coverage Δ
src/ImageSharp/Common/Helpers/SimdUtils.cs	65.90% <36.36%> (-14.58%)	⬇️
...rmats/Jpeg/Components/Decoder/HuffmanScanBuffer.cs	94.73% <100.00%> (-0.07%)	⬇️

---

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 521fae3...1eaaf1a. Read the comment docs.

[antonfirsov]

Found a possible issue that is easy to fix, but otherwise looks good. Cool stuff!