Closed Austin-Hu closed 4 years ago
With the PR applied, encoding performance could be increased. But memory consumption would be increased as its side effect, due to the increased number of Output FIFO objects, which take memory calculated by: EB_OUTPUTSTREAMBUFFERSIZE_MACRO * number_of_output_fifo, where EB_OUTPUTSTREAMBUFFERSIZE_MACRO depends on input resolution.
Some typical encoding performance and memory consumption comparison: | Command | Data Type | Without PR #512 | With PR #512 |
---|---|---|---|---|
./SvtHevcEncApp -i bbb_1920x1080_420p.yuv -w 1920 -h 1080 -n 5000 -b 0.265 | Encoding Performance (FPS) | 160 | 327 | |
Memory (GB) | 2 | 2.4 | ||
./SvtHevcEncApp -i bbb_1920x1080_420p.yuv -w 1920 -h 1080 -encMode 0 -intra-period 100 -scd 0 -rc 0 -q 22 -n 1000 -b 0.265 | Encoding Performance (FPS) | 3.73 | 6.9 | |
Memory (GB) | 2.2 | 2.5 | ||
./SvtHevcEncApp -i 7680x3840_P444.yuv -w 7680 -h 3840 -color-format 3 -b 0.265 -n 500 | Encoding Performance (FPS) | 16.56 | 18.25 | |
Memory (GB) | 26.5 | 26.9 |
Hi @Austin-Hu , I will re-enable out_alloc and increase stream output fifo to fix the hang issue. will close this one. Thank you for doing the profiling.
of FIFO objects.
Signed-off-by: Austin Hu austin.hu@intel.com
As the 3rd (work around) option, to fix #497 , #514 .