Sliced-threads: do hpel and deblock after returning
authorFiona Glaser <fiona@x264.com>
Fri, 24 Feb 2012 21:34:39 +0000 (13:34 -0800)
committerFiona Glaser <fiona@x264.com>
Wed, 7 Mar 2012 02:11:32 +0000 (18:11 -0800)
commita155572ed547a3627ef00ca70ab804ff452147cd
tree2dfef03557d8c3fb7fcaeccde901643ed223ebd0
parent90408ecab16a06ceaa181ff2e495b8f1a9d170fa
Sliced-threads: do hpel and deblock after returning
Lowers encoding latency around 14% in sliced threads mode with preset superfast.
Additionally, even if there is no waiting time between frames, this improves parallelism, because hpel+deblock are done during the (singlethreaded) lookahead.
For ease of debugging, dump-yuv forces all of the threads to wait and finish instead of setting b_full_recon.
common/common.h
common/deblock.c
common/frame.c
common/frame.h
common/macroblock.c
encoder/encoder.c