SVC inter layer prediction
In both files with spatial scalability av1-1-b8-22-svc-L2T1.ivf and av1-1-b8-22-svc-L2T2.ivf we see L0 predicting from L1. According to AV1 spec, this should not be allowed:
Coded video data of a temporal level with temporal_id T and spatial level with spatial_id S are only allowed to reference previously coded video data of temporal_id T’ and spatial_id S’, where T’ <= T and S’ <= S.
Interestingly, the VQAnalyzer does not complain about it and decoders seem also fine. So I wonder if the specification text needs to be refined or if the files need to be fixed. Something to clarify within the AOM maybe.