HDR Support for libktx and tools

[ViNeek]

The pull request adds initial support for HDR data in libktx and tools

In particular,

1. The CLI ktx tool is updated so that:

• the ktx info command supports UASTC HDR payload formats as defined in the KTX Specification.
• the ktx validate command implements new validation clauses related to the UASTC HDR payload formats
• the ktx encode command now accepts uastc-hdr-4x4 and uastc-hdr-6x6i as codec strings.
• These codecs require the input file format to be either R16G16B16_SFLOAT or R16G16B16A16_SFLOAT.
• the ktx create command when the encode parameter is used, accepts uastc-hdr-4x4 and uastc-hdr-6x6i for the codec parameter
• if the input is raw, these codecs require the format parameter to be either R16G16B16_SFLOAT or R16G16B16A16_SFLOAT;
• if the input is not raw, the input files must be EXR with half-float data type
• the ktx transcode supports the two new codecs for input files:
• uastc-hdr-4x4 inputs must accept three targets: rgba16f, astc-hdr-4x4, and bc6hu
• uastc-hdr-6x6i inputs must accept three targets: rgba16f, astc-hdr-6x6, and bc6hu
• if an input file is neither of these two formats, the targets are rejected,
• the ktx extract command supports the two new codecs for input files
• uastc-hdr-4x4 inputs may be extracted as raw, or transcoded as in ktx transcode
• uastc-hdr-6x6i inputs require transcoding as in ktx transcode when used with this command
• the ktx compare command, when used with UASTC HDR 4x4 or UASTC HDR 6x6 data and pixel comparison is requested, the image is transcoded to uncompressed half-float values prior to comparison

2. The existing JavaScript bindings is updated to support the new functionality, specifically transcoding of UASTC HDR payloads

Incidentally fixes #1109.

[MarkCallow]

I just reviewed the test updates again. Now I see that there are no tests of ktx validate on the new formats. This must be tested. Please add suitable tests.

[MarkCallow]

Please add a check to ktxTexture2_TranscodeBasis so that it returns immediately if outputFormat is KTX_TTF_ASTC_HDR_4x4_RGBA and the texture's colorModel is KHR_DF_MODEL_UASTC_HDR_4X4.

[MarkCallow]

I am not liking having NXN in KHR_DF_MODEL and KTX_SS_UASTC_HDR and NxN everywhere else including the Vulkan and OpenGL format names and KTX_PACK_ASTC_BLOCK_DIMENSION. Please change the KTX_SS_UASTC_HDR names. I can probably tolerate the KHR names being different but would ideally like to have those changed too.

[MarkCallow]

khr_df.h is not installed by, at least the macOS, installer.

[ViNeek]

khr_df.h is not installed by, at least the macOS, installer.

Testing on an M4 Mac with mac os Tahoe 26.2 with the standard cmake commands.

cmake -S . -B build -DKTX_FEATURE_TOOLS=ON
cmake --build build
cmake --install build --prefix ktxInstallation

The installation directory structure gets created correctly. Can you please share more information about your setup?

[MarkCallow]

I built and installed the package target. khr_df.h is not included in the installer. This is either because it is not included as a public header of the ktx target or it needs special handling because of the KHR subdirectory.

[MarkCallow]

I have just realized that support for KTXmapRange is not implemented. It needs to be displayed by ktx info, validated by ktx validate and settable in ktx create. This new metadata item is part of the HDR additions.

I am sorry that I failed to notice the omission before.

[MarkCallow]

Re khr_df.h, an install like this one in dfdutils.h

install(DIRECTORY
        ${CMAKE_CURRENT_SOURCE_DIR}/KHR/
        DESTINATION include/KHR
)

needs a line, COMPONENT dev, to add it to the installer. However I'm not sure the dev component will be known to dfdutils/CMakeLists.txt. You might have to add this to ktx/CMakeLists.txt. There are old install attempts in the latter file that can probably be deleted. See LL 206-210 and LL 722-735.

I think dfdutils/CMakeLists.txt should install only KHR/khr_df.h when EMBED is set, otherwise it should install it, dfdutils.h and the library. I can do that myself later, if you don't want to do it.

[MarkCallow]

I have opened an issue for each of the newly discovered items. I will merge this because the number of comments in this PR makes dealing with any more unwieldy, and time consuming, and what is here is working.

Please continue discussion on the new items in the relevant issue.