Skip to content

Improve dav1d_parse_sequence_header()

Henrik Gramner requested to merge gramner/dav1d:parse_seq_header_improvements into master

This avoids initializing a separate decoder instance, which not only gets rid of a bunch of unnecessary short-lived memory allocations but also makes things three orders of magnitude faster:

2494821 decicycles in dav1d_parse_sequence_header, 62 runs, 2 skips

1551 decicycles in dav1d_parse_sequence_header, 62 runs, 2 skips

Merge request reports

VideoLAN code repository instance