Khronos KTX Files

Changes since v4.2.0 (by part)

libktx

• Fix memory leak of input_image in ktxTexture2_CompressAstcEx (fa1fe4d7) (@null)

Version 4.2.0

Overview

v4.2.0 has few user-facing changes. Most of the work has been behind the scenes improving the build system and fixing warnings across the many supported compilers. User-facing changes are detailed below.

New Features in v4.2.0

• Install packages for GNU/Linux on Arm64 have been added.

• The Java wrapper is now included in the Windows Arm64 install package.

Significant Changes since v4.1.0

• The following behavioral changes have been made to toktx:

  • If the input PNG file has a gAMA chunk with a value 45460 the image data is now converted to the sRGB transfer function intead of just assigning sRGB as the transfer function of the output file.
  • If the gAMA chunk has a value other than 45640 or 100000 toktx will now exit with an error. Previously it used heuristics to decide whether to transform the input to linear or sRGB. Use --convert_oetf or --assign_oetf to specified the desired behavior.

• The Khronos Data Format header file KHR/khr_df.h has been added to the install packages and is included in ktx.h. A new transfer function query ktxTexture2_GetOETF_e that returns a khr_df_transfer_e replaces ktxTexture2_GetOETF that returned a ktx_uint32_t. The latter is still available for backward compatibility, A new ktxTexture2_GetColorModel_e query has been added returning a khr_df_model_e.

Known Issues in v4.2.0.

• Some image bits in output files encoded to ASTC, ETC1S/Basis-LZ or UASTC on arm64 devices may differ from those encoded from the same input images on x86_64 devices. The differences will not be human visible and will only show up in bit-exact comparisons.

• toktx will not read JPEG files with a width or height > 32768 pixels.

• toktx will not read 4-component JPEG files such as those sometimes created by Adobe software where the 4th component can be used to re-create a CMYK image.

• Users making Basisu encoded or block compressed textures for WebGL must be aware of WebGL restrictions with regard to texture size and may need to resize images appropriately using the --resize feature of toktx. In general the dimensions of block compressed textures must be a multiple of the block size and for WebGL 1.0 must be a power of 2. For portability glTF's KHR_texture_basisu extension requires texture dimensions to be a multiple of 4, the block size of the Universal texture formats.

• Basis Universal encoding results (both ETC1S/LZ and UASTC) are non-deterministic across platforms. Results are valid but level sizes and data will differ slightly. See issue [#60] in the basis_universal repository.

• UASTC RDO results differ from run to run unless multi-threading or RDO multi-threading is disabled. In toktx use --threads 1 for the former or --uastc_rdo_m for the latter. As with the preceeding issue results are valid but level sizes will differ slightly. See issue [#151] in the basis_universal repository.

• Neither the Vulkan nor GL loaders support depth/stencil textures.

Notice

• Building with Visual Studio 2015 and 2017 is no longer supported.

Changes since v4.1.0 (by part)

libktx

• Pull upstream ASTC encoder for FP option setting fixes. (#713) (8e68fe04) (@MarkCallow)

• Pull upstream ASTC for fixes building with GCC 11 for arm64, (#700) (514051ca) (@MarkCallow)

• Update Vulkan SDK for macOS CI. (#688) (f57dc8fa) (@MarkCallow)

• CI and Build Improvements (#687) (38f48586) (@MarkCallow)

Tools

• Pull upstream ASTC for fixes building with GCC 11 for arm64, (#700) (514051ca) (@MarkCallow)

• Reimplement image input handling for toktx. (#702) (1646c4d0) (@MarkCallow)

• Fix normalization when the result overflows (#701) (f81330b5) (@wasimabbas-arm)

Java Wrapper

• Fix outdated references to master. (e724f180) (@MarkCallow)

• Miscellaneous CI script and build fixes (#692) (fefd4a65) (@MarkCallow)

• Remove pinned buffer list in JNI wrapper to avoid segmentation faults (#697) (9b084d50) (@ShukantPal)