D3D pipeline possible inconsistencies

[Kevin Rushforth]

As you note, there are a few different issues here. To answer your 
questions as best I can:

1 & 3. We should document that self-illum maps and specular only use the 
rgb components and that alpha should be ignored. It's possible (although 
I don't know for sure) that the image is being treated as a non- 
premultiplied color format, and is subsequently converted to a 
premultiplied format; if so, this could explain the color darkening.

2. This also needs to be documented. The diffuse component should have 
an alpha that applies whether from the diffuse color or from a diffuse 
map. I agree with you that the pixel fragment should not be discarded 
just because the diffuse component is transparent. A specular highlight 
should be possible on a fully transparent object.

4. It does seem that the result should be the same regardless of whether 
the color comes from a specular map or color, but I'd need to dig further.

Do all of the above issues happen with the OpenGL shaders, too?

-- Kevin


On 4/26/2022 11:41 AM, Nir Lisker wrote:
> I found a comment [1] on JBS stating that specular and self-Illumination
> alphas should be ignored, so it seems like there's at least 2 bugs here
> already.
>
> https://bugs.openjdk.java.net/browse/JDK-8090548?focusedCommentId=13771150&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13771150
>
> On Tue, Apr 26, 2022 at 4:25 AM Nir Lisker <nlisker at gmail.com> wrote:
>
>> Hi,
>>
>> Using the updated lighting test sample [1], I found some odd behavior with
>> regards to PhongMaterial:
>>
>> 1. The effect of the opacity (alpha channel) of a self-illumination map is
>> not documented, but lowering its value makes the object darker. I looked at
>> the pixel shader [2] and only the rgb components are sampled, so I'm a bit
>> confused here. What is the intended behavior?
>>
>> 2. The opacity of the object is controlled in the shader by both the
>> diffuse color and diffuse map. This is also not documented (although it
>> might be obvious for some). In the shader, the pixel (fragment) is
>> discarded only if the map is fully transparent (line 55), but not the
>> color. This leads to a situation where the object completely disappears
>> when the map is transparent, but not when the color is. In the shader, the
>> pixel should be transparent because of the multiplication of the alpha, but
>> it's not, so this is also confusing. Should they both have the same
>> contribution? Shouldn't it be valid to have a transparent diffuse but still
>> have specular reflections?
>>
>> 3. The specular map and color behave differently in regards to the
>> opacity. There is no documented behavior here. The alpha on the color is
>> ignored (it's used for the specular power), but not on the map - it
>> controls the reflection's strength, probably by making its color darker. In
>> the shader, lines 76-84 indeed ignore the alpha of the color, but take the
>> alpha of the map, although later in line 93 it's not used, so again I'm
>> confused. What's the intended behavior?
>>
>> 4. The specular map and color also behave differently in regards to the
>> reflection's strength. In the shader, this happens in line 78: the specular
>> power is corrected with NTSC_Gray if there is a map (with or without
>> color), but not if there's only a color. Shouldn't the contributions be the
>> same? Is the NTSC_Gray correction correct in this case?
>>
>> Thanks,
>>   Nir
>>
>> [1] https://github.com/openjdk/jfx/pull/787
>> [2]
>> https://github.com/openjdk/jfx/blob/master/modules/javafx.graphics/src/main/native-prism-d3d/hlsl/Mtl1PS.hlsl
>>