Inconsistent results between R8_UNORM and RGBA

[jandk]

Hello,

I'm writing a library to decompress BCn textures, and while getting reference from BC4S by decoding through texconv, I noticed the values are not the same between R8_UNORM and RGBA.

Looking at the code it seems that for R8G8B8A8_UNORM you do add the 0.5/255.0 bias, but not for R8_UNORM.

Is this how it's supposed to be, or is this an issue? I couldn't find anything in the documentation, maybe you could provide more insight?

Greetings
Jan

[walbourn]

Can you provide the specific format input and output? I can take a look at the code paths.

[jandk]

Hi, thanks for the quick response. I did some testing and it gets even more confusing...

If I run texconv.exe -ft png -f R8_UNORM .\bc4s.dds, it calls ConvertScanline with inFormat DXGI_FORMAT_R8_SNORM and outFormat DXGI_FORMAT_R8_UNORM which ends up going with XMVectorMultiplyAdd(v, g_XMOneHalf, g_XMOneHalf), which seems completely normal. But then the call to StoreScanline ends up in the switch case DXGI_FORMAT_R8_UNORM, which does not have the extra bit bias.

If I run texconv.exe -ft png -f RGBA .\bc4s.dds it does the same until the StoreScanline where it ends up in the DXGI_FORMAT_R8G8B8A8_UNORM branch, which does add the g_8BitBias.

But both of them don't seem to line up with what I expect. If I open the dds file in paint.net (which uses the ddsfiletype-plus plugin, which uses DirectXTex under the hood), which ends up in this method:
https://github.com/0xC0000054/pdn-ddsfiletype-plus/blob/master/src/DdsFileTypePlusIO/DdsFileTypePlusIO.cpp#L243

If I do this, it does end up calling the Convert function with what seems to be the same DXGI_FORMAT_R8G8B8A8_UNORM. Except this time the result actually seems to be the steps I follow. Unpack SNORM, do the bc4 signed lerps, and pack snorm to unorm.

I've attached the dds file, and the three results as well.

images.zip