Port GrayscalConverter to Arm

[stefannikolei]

Port Grayscaleconverter to ARM

Benchmarks:

|        Method |      Mean |    Error |   StdDev | Ratio | RatioSD |
|-------------- |----------:|---------:|---------:|------:|--------:|
|        Scalar | 118.76 ns | 0.499 ns | 0.443 ns |  1.00 |    0.00 |
| SimdVectorAvx |        NA |       NA |       NA |     ? |       ? |
| SimdVectorArm |  53.57 ns | 0.410 ns | 0.364 ns |  0.45 |    0.00 |

[JimBobSquarePants]

Great speed up! All these additions are going to be a gamechanger for ARM

[JimBobSquarePants]

Looks like something in #2401 has changed the result of some of the WebP output on ARM.

[brianpopow]

Looks like something in #2401 has changed the result of some of the WebP output on ARM.

Just stating the obvious: It seems impossible to me, that the changes from this PR could be responsible for WebP test failures.
@stefannikolei If you run the webp tests on you machine, do you see any test failing?
You can run only Webp tests with: dotnet test -c Release -f net6.0 --filter "Format=Webp"

[stefannikolei]

   WebpDecoderTests
    WebpDecoder_CanDecode_Lossless_WithThreeTransforms
     WebpDecoder_CanDecode_Lossless_WithThreeTransforms<Rgba32>(provider: Webp/bike_lossless.webp[Rgba32])
     WebpDecoder_CanDecode_Lossless_WithThreeTransforms<Rgba32>(provider: Webp/color_cache_bits_11.webp[Rgba32])
    WebpDecoder_Decode_Resize
     WebpDecoder_Decode_Resize<Rgba32>(provider: Webp/bike_lossless.webp[Rgba32])

Yeah this tests fail

[JimBobSquarePants]

Looks like something in #2401 has changed the result of some of the WebP output on ARM.

Just stating the obvious: It seems impossible to me, that the changes from this PR could be responsible for WebP test failures. @stefannikolei If you run the webp tests on you machine, do you see any test failing? You can run only Webp tests with: dotnet test -c Release -f net6.0 --filter "Format=Webp"

@brianpopow #2401 is the low hanging fruit PR. WebP is touched there. I'm doing another code review as we didn't label the PR to catch ARM issues.

[gfoidl]

I also have a look at this right now.

[gfoidl]

Just a FYI for easier location of error on x86-dev-machine: run the tests with DOTNET_EnableSSE=0 dotnet test -c Release -f net6.0 --filter "Format=Webp"

-- or --

Update the .runsettings to

<?xml version="1.0" encoding="utf-8" ?>
<RunSettings>
    <RunConfiguration>
        <!--Used in conjunction with ActiveIssueAttribute to skip tests with known issues-->
        <TestCaseFilter>category!=failing</TestCaseFilter>

        <EnvironmentVariables>
            <DOTNET_EnableSSE>0</DOTNET_EnableSSE>
        </EnvironmentVariables>
    </RunConfiguration>
</RunSettings>

[gfoidl]

Hm, so far I can't spot the bug.

State so far:

• AVX --> ✔️
• ARM --> 🐛
• SSE disabled --> scalar --> 🐛

So I assume the bug isn't specific to ARM, rather it's anywhere in the scalar-path (that is hit by ARM too, but not with SSE/AVX).
Will search and debug further to smash the bug down....

[gfoidl]

I found something, will send a PR in a moment (just checking if a similar bug doesn't exist elsewhere).

Edit: PR --> #2413

[JimBobSquarePants]

OK. Let's get this merged as the changes are covered. Thanks