This is about 20-30% faster than slice threading
a0c0900 ffv1dec: Support frame threading with gop > 1
libavcodec/ffv1.h | 1 +
libavcodec/ffv1dec.c | 106 +++++++++++++++++++++++++++++++++++++++++++++-----
libavcodec/pthread.c | 2 +-
3 files changed, 98 insertions(+), 11 deletions(-)