SIMD implementation of transforms (merge request !2695) seems to break compilation at least in some non-X86 environments. For example, compilation fails with long list of errors when using cmake 3.27.7, clang compiler 15.0.0.15000040 and default settings of XCode 15.0.1 in MacOS.

Reason for the failure appears to be that the new TrQuantX86.h #includes an x86 SIMD header file even if the target architecture is not supporting such. Encapsulating the contents of that file inside "#ifdef TARGET_SIMD_X86" seems to fix the issue.

Interestingly the VTM git's MacOS pipeline tests don't capture this problem. Wonder if one of those can be configured so that this kind of errors would be identified automatically?

When nnpfc_chroma_loc_info_present_flag is zero, nnpfc_chroma_sample_loc_type_frame is not present in the NNPFC SEI. Regardless, the value of nnpfc_chroma_sample_loc_type_frame is checked at the decoder. The value of nnpfc_chroma_sample_loc_type_frame should only be checked if the syntax element is present in the NNPFC SEI.

In EncAdaptiveLoopFilter.cpp, function mergeClasses contains following copy

std::copy_n(clipMerged[numRemaining - 1][0], MAX_NUM_ALF_CLASSES * MAX_NUM_ALF_LUMA_COEFF,

clipMerged[numRemaining - 2][0]);

Encoder crashed here, problem seems to be that 1D copy is performed instead of 2D. Address may not be contiguous. This issue is present in latest VTM22.0.