Assertion failed: (i < m_size), function operator[], file basisu_containers.h, line 1752

[MarkCallow]

This assert happens with certain input data in v1_60 and v2_0_2 whenever -hdr_6x6i_level is greater than 7. It does not happen in v2_1_0.

Here is a .exr file that triggers it.

R16G16B16_SFLOAT_2D_8x8.exr.zip

The command I am using is

bin/basisu -hdr_6x6i_level 8 R16G16B16_SFLOAT_2D_8x8.exr

One of our CI tests triggers this assert with the fix for KhronosGroup/KTX-Software#1146, as the value specified for hdr_6x6i_level is now being passed to the encoder. This can be seen in the build logs for PR KhronosGroup/KTX-Software#1149

Please inform us what to backport to v1_60 to fix this as we prefer not to update to v2.1 so close to our planned HDR release.

[richgel999]

Thank you, looking at it. Visual Studio with our latest code on github (v2.1) doesn't have the issue, but I'll try an earlier version and Linux.

Command line: -hdr_6x6i_level 8 R16G16B16_SFLOAT_2D_8x8.exr

DEBUG or _DEBUG defined
NDEBUG is NOT defined
Basis Universal LDR/HDR GPU Texture Supercompression System v2.10.0 (x64)
Copyright (C) 2019-2026 Binomial LLC, All rights reserved
Using SSE 4.1: 1, Multithreading: 1, Zstandard support: 1, OpenCL: 0
Processing 1 total file(s)
Processing source file "R16G16B16_SFLOAT_2D_8x8.exr"
Read source image "R16G16B16_SFLOAT_2D_8x8.exr", 8x8
One or more input pixels was NaN, setting to 0.
One or more input pixels was negative -- setting these pixel components to 0 because ASTC HDR doesn't support signed values.
Warning: clean_astc_hdr_pixels() had to modify the input image to encode to ASTC HDR - see previous warning(s).
Total slices: 1
Slice: 0, alpha: 0, orig width/height: 8x8, width/height: 8x8, first_block: 0, image_index: 0, mip_level: 0, iframe: 0
Mode: ASTC 6x6 HDR Intermediate, Base Effort Level (0-4): 2, Highest Effort Level (0-4): 3, Lambda: 0.000000, REC 2020: false
Writing v1.60 compatible UASTC HDR 6x6i bitstream
Blurring image
Transforming to ITP
Wrote compressed output file "R16G16B16_SFLOAT_2D_8x8.ktx2"
Compression succeeded to file "R16G16B16_SFLOAT_2D_8x8.ktx2" size 286 bytes in 0.216 secs, 35.750 bits/texel
Total successes: 1 failures: 0

C:\dev\bu_3_18\basis_universal\bin\basisuD.exe (process 32428) exited with code 0 (0x0).
Press any key to close this window . . .

[richgel999]

I've reproduced it using this snapshot:
https://github.com/BinomialLLC/basis_universal/releases/tag/v2_00_final_snapshot

This is a bug in encode_block_3_subsets() that was fixed recently here:
c606cef

Should be very easy to merge. I'll make sure this fixes it.

[richgel999]

This was found via automated fuzz testing and fixed recently. You can merge these specific changes from this commit into your repo - it should be a trivial merge because it's the same version of the library you're using:

BinomialLLC/KTX-Software-Binomial-Fork@2b76b66

You specifically need:

Critical fixes (for extreme inputs/outputs):
external/basis_universal/encoder/basisu_astc_hdr_6x6_enc.cpp
external/basis_universal/transcoder/basisu_transcoder.cpp

Compiler warning:
external/basis_universal/encoder/basisu_astc_hdr_common.cpp

[richgel999]

There are also two other back ported fixes both impacting external/basis_universal/transcoder/basisu_transcoder.cpp - both trivial merges:

KTX2 header parsing uint64 fix (discovered recently):
BinomialLLC/KTX-Software-Binomial-Fork@8bb258d

init fix (you're already working around this):
BinomialLLC/KTX-Software-Binomial-Fork@cbb819d

[MarkCallow]

Thanks for the quick answer. Starting work on merging these fixes now.

[MarkCallow]

Fixes are working. Thanks.