Convert 16bit colors to 32bit colors on MetalANGLE/Simulator

[Berstanio]

Currently the iOS simulator wont work with MetalANGLE in all cases because of kakashidinho/metalangle#68.
This is a first fix for this by converting 16bit colors manually to 32bit colors when running on simulator.
The fix can be confirmed by running the "BitmapFontAtlasRegionTest" test.

The only thing I'm atm unsure is, whether it is wrong to expect a ByteBuffer. But as far as I know, a direct buffer needs to be passed, and all direct Buffer extend from ByteBuffer.

[Tom-Ski]

Does this problem occur on different macs in the same way? For example M1 vs non M1?

This would also be required to do in the GL30 interfaces too no?

[Berstanio]

Does this problem occur on different macs in the same way? For example M1 vs non M1?

I think so, but I can only test on a M1.

This would also be required to do in the GL30 interfaces too no?

You mean e.g. glTexImage3D? Yeah, probably, I oversaw that.

[Tom-Ski]

I think so, but I can only test on a M1.

Me too atm, would be good if someone can test to see if this has adverse effects on non M1

You mean e.g. glTexImage3D? Yeah, probably, I oversaw that.

Yeah

[Berstanio]

Does this problem occur on different macs in the same way? For example M1 vs non M1?

https://developer.apple.com/documentation/metal/developing_metal_apps_that_run_in_simulator?language=objc
Seems so according to the docs:

"Don't use the following pixel formats: MTLPixelFormatR8Unorm_sRGB, MTLPixelFormatB5G6R5Unorm, MTLPixelFormatA1BGR5Unorm, MTLPixelFormatABGR4Unorm, MTLPixelFormatBGR5A1Unorm, or any XR10 or YUV formats."

[obigu]

Tested running BitmapFontAtlasRegionTest on EXCRUCIATINGLY SLOW Intel Mac and it crashes without the PR. With the PR test runs ok (even if it's quite an ugly test).

[Tom-Ski]

Ok lets have this in then for now and see how it goes