[POC] Fix editor screenshots with HDR enabled (linear 2 sRGB on half-floats).

[bruvzg]

Same as #119013, but doing linear-to-sRGB transform before converting to 8-bit.

Opening it for reference only, code probably need optimization to work on low-end systems.

This PR:

#119013:

[allenwp]

This, to me, is a whole new feature added to the Image class -- and one that I've wanted for a while! Thanks! It's good that it updates the internal editor screenshot functionality, but I think it would be great to consider adding this as a new feature of the Image class in 4.8. #119013 can still be merged as the basic bugfix for 4.7.

[allenwp]

Haven't tested this yet, but sometime before this is ready for review, docs will need to be updated in at least these three spots:

1. https://docs.godotengine.org/en/stable/classes/class_image.html#class-image-method-srgb-to-linear
2. https://docs.godotengine.org/en/stable/classes/class_image.html#class-image-method-linear-to-srgb
3. https://docs.godotengine.org/en/stable/classes/class_viewporttexture.html#description
   • The order of the two lines in the example script needs to be changed, just like the changes made to editor_node

[allenwp]

Here is a list of other places that call these updated functions that might benefit from reworking in the same way as EditorNode:

1. texture_loader_ktx.cpp: static Ref<Image> load_from_file_access(Ref<FileAccess> f, Error *r_error)
2. image_loader_svg.cpp: ImageLoaderSVG::load_image(Ref<Image> p_image, Ref<FileAccess> p_fileaccess, BitField<ImageFormatLoader::LoaderFlags> p_flags, float p_scale)
3. scene_debugger.cpp: SceneDebugger::_msg_rq_screenshot(const Array &p_args)
4. movie_writer.cpp: MovieWriter::add_frame()

[bruvzg]

Here is a list of other places that call these updated functions that might benefit from reworking in the same way as EditorNode:

Seems like 1 and 2 does not work with any half-float formats, so won't be affected or need any changes, 3 is already changed to do convention to RGBA8 after linear-to-srgb, added the same for the 4.

docs will need to be updated in at least these three spots

Updated.

[bruvzg]

For the reference current half-float code is about 10 times slower than RGBA8 lookup table code, it's still acceptable for screenshots (takes about 0.5 sec for 8K image in debug build), but likely is not OK for MoveWriter.

Edit: if parallelized, it's slightly faster than RGBA8, but offloading it to GPU would likely be a better option.