tis 2022-06-14 klockan 17:04 +0200 skrev Andreas Rheinhardt: > Tomas Härdin: > > > > > > Why call it at all? Why not just add a new parameter to > ff_jpeg2000_init_component that is always set to 1 when called from > the > encoder and 0 when called from the decoder?
Oh yeah that's even simpler > (And is this really a bottleneck?) Callgrind certainly thinks so. It's called hundreds of thousands of times per frame. Remember that this is in the serial part of the code so any savings there get amplified -threads fold fps-wise. init_tile() accounts for a mere 0.8 seconds out of 59.8 partly thanks to this. Here's a rough breakdown for the curious with -threads 64: 0.8 everything up to and including init_tiles() 23.0 jpeg2000_read_bitstream_packets() 0.1 jpeg2000_setup_cbs() 24.1 jpeg2000_decode_cb() 9.8 jpeg2000_idwt() 2.0 jpeg2000_mct_write_frame() 0.0 jpeg2000_dec_cleanup() jpeg2000_read_bitstream_packets() is obviously the main thing to focus on for anyone wanting to bump the speed up even more. But it's nasty. Maybe it could be tile-threaded, but it takes some doing.. /Tomas _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
