dav1d / libaom single-thread performance analysis
I ran perf reports on both dav1d and aomdec using 3 different video's. They ran with a single thread to determine which functions aomdec is still faster than dav1d.
- aomdec (d909c2c8d):
aomdec --noblit --skip-film-grain --summary - dav1d (7f7b5586):
-q --muxer null --filmgrain 0
This issue can be used to discus things that stand out.
Here are the files (31 = SSSE3, 63 = SSE4.1, 511 = AVX2):
- aomdec-chimera-31.txt
- aomdec-chimera-63.txt
- aomdec-chimera-511.txt
- aomdec-dualipa-31.txt
- aomdec-dualipa-63.txt
- aomdec-dualipa-511.txt
- aomdec-morocco-31.txt
- aomdec-morocco-63.txt
- aomdec-morocco-511.txt
- dav1d-chimera-avx2.txt
- dav1d-chimera-ssse3.txt
- dav1d-dualipa-avx2.txt
- dav1d-dualipa-ssse3.txt
- dav1d-morocco-avx2.txt
- dav1d-morocco-ssse3.txt
Edited by Ewout ter Hoeven