szabadka

Could you be using an older version of testdata? It seems that all of those tests fail when attempting to read the fuzzer test case from testdata/oss-fuzz

I was thinking that if we add a new function for this, it could address both the ordering and single plane case. The function would take JxlPixelFormat, which has num_channels,...

Could you provide concrete examples? In the form of original input, guetzli output, and something that has better quality (according to butteraugli) and smaller file size.

One reason that this could happen is if the endianness is not set correctly in JxlDecoderSetImageOutCallback(). Could you try setting ``` d.m_pixelFormat.endianness = JXL_NATIVE_ENDIAN; ``` in the code snippet above...

I am currently working on this. As a first step, I will add a testing framework that runs the PackedPixelFile -> Encoder API -> JXL bitstream -> Decoder API ->...

Is there a way to get access to one of these Debian build machines to debug the issue? Or is there a way to set up ppc64/i386 runners on GitHub?

> > Is there a way to get access to one of these Debian build machines to debug the issue? Or is there a way to set up ppc64/i386 runners...

In the attention layer the array length is the context size, which is typically not a multiple of 16 (unless we are past the local-attention window in recurrent gemma).

Could you check if the created JXL file also fails to decode with djxl?

I could not reproduce it using clang-14, does it only occur with clang 11? I tried the following: (cd build-asan/ && ninja && ./tools/jxlinfo ~/Downloads/github/issue3724/crash_1.jxl.jpg) output: JPEG XL file format...