Fix issue when writing lossy webp with alpha data

[brianpopow]

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

Fix for issue #2171: The alpha chunk data was written completely (the size of the image in pixels) and not only the compressed data when writing a lossy webp image.

[brianpopow]

There are new failing tests with .Net7.0, this time on macos only. 222 are failing:

Error Message:
System.PlatformNotSupportedException : Operation is not supported on this platform.
Stack Trace:
at System.Runtime.Intrinsics.Vector256.Create(Single value)
at SixLabors.ImageSharp.Formats.Jpeg.Components.FastFloatingPointDCT.<IDCT8x8_Avx>g__IDCT8x8_1D_Avx|16_0(Block8x8F& block) in /_/src/ImageSharp/Formats/Jpeg/Components/FastFloatingPointDCT.Intrinsic.cs:line 141
at SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.JpegComponentPostProcessor.CopyBlocksToColorBuffer(Int32 spectralStep) in /_/src/ImageSharp/Formats/Jpeg/Components/Decoder/JpegComponentPostProcessor.cs:line 70

[JimBobSquarePants]

That’s concerning. @tannergooding sorry to page but are there changes coming to .NET 7 we should know about?

[tannergooding]

System.Runtime.Intrinsics.Vector256.Create(Single value) should work on all platforms, it might just not be accelerated on certain ones (x64 machines without AVX and Arm64 machines in general).

Do you have a simple repro? I don't immediately see any changes to the .NET 7 version or to any code newly using Vector256 in this PR.

[brianpopow]

System.Runtime.Intrinsics.Vector256.Create(Single value) should work on all platforms, it might just not be accelerated on certain ones (x64 machines without AVX and Arm64 machines in general).

Do you have a simple repro? I don't immediately see any changes to the .NET 7 version or to any code newly using Vector256 in this PR.

Hi @tannergooding. I am not 100% sure if this will be enough, but from the error messages in this CI run I would assume just calling:

var scale = Vector256.Create(200.0f);

on a Mac should be enough to trigger this. Since I do not have a Mac, I cannot verify that.
@JimBobSquarePants Maybe you have a chance to test, if that is indeed enough?

[tannergooding]

This might be an issue specifically around a machine with AVX but no AVX2 support. I'll see if I can get a repro

[JimBobSquarePants]

Thanks @tannergooding I've been unable to reproduce this on my old 2.7 GHz Dual-Core Intel Core i5

[JimBobSquarePants]

I got some environmental info via #2175

I'm at a loss to explain the cause.

OS=macOS Big Sur 11.6.7 (20G630) [Darwin 20.6.0]
Intel Xeon CPU E5-1650 v2 3.50GHz (Max: 3.34GHz), 1 CPU, 3 logical and 3 physical cores
.NET SDK=7.0.100-preview.6.22352.1
[Host] : .NET 7.0.0 (7.0.22.32404), X64 RyuJIT

[brianpopow]

I got some environmental info via #2175

I'm at a loss to explain the cause.

OS=macOS Big Sur 11.6.7 (20G630) [Darwin 20.6.0]
Intel Xeon CPU E5-1650 v2 3.50GHz (Max: 3.34GHz), 1 CPU, 3 logical and 3 physical cores
.NET SDK=7.0.100-preview.6.22352.1
[Host] : .NET 7.0.0 (7.0.22.32404), X64 RyuJIT

I think Tanner is on the right track with his guess that it's maybe related to a machine with AVX but not AVX2. Intel Xeon CPU E5-1650 only has AVX from what I can tell.

[antonfirsov]

Edit: Missed your update @brianpopow 😄

This might be an issue specifically around a machine with AVX but no AVX2 support.

Seems correct, E5-1650 v2 is Ivy Bridge.

Maybe we can workaround these issues to get back clean CI until there is an upstream fix? We can disable Avx path on .NET7 when Avx2.IsSupported == false.

ImageSharp/src/ImageSharp/Formats/Jpeg/Components/FastFloatingPointDCT.cs

Lines 113 to 116 in e54219a

	if(Avx.IsSupported)
	{
	IDCT8x8_Avx(refblock);
	}

[JimBobSquarePants]

Maybe we can workaround these issues to get back clean CI until there is an upstream fix? We can disable Avx path on .NET7 when Avx2.IsSupported == false.

You mean for this instance? I think we're just hitting this one cos it comes first. We use Vector256.Create(single) in a few places. Theoretically we could turn off the support by setting the correct environmental variable during CI but I think I'd rather just disable .NET 7 Mac tests for now.

[tannergooding]

Ok, I can reproduce the issue and will work on getting a fix up on the dotnet/runtime side.

[JimBobSquarePants]

AWESOME!

[antonfirsov]

I'd rather just disable .NET 7 Mac tests for now.

Yeah, makes sense.

[tannergooding]

@JimBobSquarePants, the issue discussed above should be fixed as of dotnet/runtime#72522 and will land in .NET 7 RC1.

-- Also, sorry for the ping after this has been closed for two weeks; wasn't sure of a better place to relay this small note.

[JimBobSquarePants]

@tannergooding That's fantastic news, thanks for letting me know.

That's a very interesting conversation regarding AVX/AVX2 issues. I wonder if any of that is causing our NaN handling issues in #2117 . I've struggled to find the specs for Intel Xeon Platinum 8272CL