basis_universal: Update to 1.60

[Chubercik]

• Source: https://github.com/BinomialLLC/basis_universal/releases/tag/v1_60
• Changes: BinomialLLC/basis_universal@v1_50_0...v1_60
• Release notes: https://github.com/BinomialLLC/basis_universal/wiki/Release-Notes

---

Due to the update changing a plethora of things, here's a recap:

• Compression parameters were renamed, as seen in modules/basis_universal/image_compress_basisu.cpp
• Because libktx makes use of the bundled basis_universal library, I had to change the type of one variable, as seen in thirdparty/libktx/patches/0003-basisu-1.60.patch (new patch for libktx)
• thirdparty/basis_universal/encoder/basisu_astc_hdr_6x6_enc.h, which is a new file, had a superscript two symbol in a comment - this is all good and nice, but due to, I guess, UTF-8 non-complience, it generated a warning that was treated as an error by one Windows CI job; I changed the symbol to ^2, as seen in thirdparty/basis_universal/patches/0005-windows-illegal-character.patch (also see: Change ² (U+00B2) to ^2 BinomialLLC/basis_universal#394)
• clang sanitizers didn't like the ambiguity of 1ULL, as it could be treated as a 32-bit or a 64-bit integer, due to an implicit conversion; patched in thirdparty/basis_universal/patches/0006-ambiguous-calls.patch (also see: Fix ambiguous calls to safe_shift_left() in basisu_containers.h BinomialLLC/basis_universal#387)
• Other patches have been regenerated, notably another change to jpeg_decoder flags had to be added in thirdparty/basis_universal/patches/0002-external-jpgd.patch

---

Speaking of patches, wouldn't simply prepending function bodies in thirdparty/basis_universal/patches/0004-remove-tinydds-qoi.patch with return false; be an easier change that achieves the same result? Or would something in the build system throw a tantrum over unreachable code?

[BlueCube3310]

Tested with https://polyhaven.com/a/cambridge (8192x4096, 13 mipmaps, compression = BasisUniversal):

4.4-stable	This PR
16.5s	4.2s

The encoding process is about 4x faster. I wasn't able to spot any regressions in terms of quality either

[fire]

Can you help upstream? This is non blocking for approving this pr.

[Calinou]

Tested locally, it works as expected.

Benchmark

PC specifications

• CPU: Intel Core i9-13900K
• GPU: NVIDIA GeForce RTX 4090
• RAM: 64 GB (2×32 GB DDR5-5800 C30)
• SSD: Solidigm P44 Pro 2 TB
• OS: Linux (Fedora 42)

Using an optimized editor build (production=yes lto=full). BasisU encode times for https://polyhaven.com/a/cambridge in .hdr format with Basis Universal compression (mipmap generation disabled):

	1K	2K	4K	8K	16K	17K*
master	143 ms	622 ms	2282 ms	9563 ms	9406 ms	36502 ms
This PR	37 ms	166 ms	645 ms	2488 ms	7845 ms	10158 ms

*: While testing this, I noticed the downscaling logic has an issue (at least with HDRIs or Basis Universal compression). The 17K HDRI will be downsized to 16K, but the resulting texture is pure black. This occurs both before and after this PR.

WARNING: res://cambridge_17k.hdr: Texture was downsized on import as its width (17858 pixels) exceeded the importable size limit (16384 pixels).
     at: import (./editor/import/resource_importer_texture.cpp:599)

[akien-mga]

Thirdparty code integration wise, this looks good to me.

[akien-mga]

thirdparty/basis_universal/encoder/basisu_astc_hdr_6x6_enc.h, which is a new file, had a superscript two symbol in a comment - this is all good and nice, but due to, I guess, UTF-8 non-complience, it generated a warning that was treated as an error by one Windows CI job; I changed the symbol to ^2, as seen in thirdparty/basis_universal/patches/0005-windows-illegal-character.patch

This one would be worth PR'ing upstream too.

[Repiteo]

Got a PR that'll fix the CI issue, #105756

[Repiteo]

Thanks!

[DanielKinsman]

After this was merged the patches in thirdparty/basis_universal/patches do not get us to the current state of master, when starting from the basis universal commit mentioned in thirdparty/README. So we've made some changes but have no idea what they are. The build fails without these changes.

It makes rebasing my current PR on master which touches the same code difficult.

[akien-mga]

The problem seems to be the "illegal character" patch, which is not cross-platform. It likely works on Windows but fails for me on Linux.

[akien-mga]

• basis_universal: Rediff patches, remove Windows encoding fix #105856