Skip to content

x86-64: Add msac_decode_bool_equi asm

Henrik Gramner requested to merge gramner/dav1d:msac_decode_equi_asm into master

Around 50% faster then the C code, but it varies a bit depending on which compiler is used.

C perf:

     1.32%  dav1d    libdav1d.so.1.1.0   [.] dav1d_msac_decode_bool_equi_c
     1.33%  dav1d    libdav1d.so.1.1.0   [.] dav1d_msac_decode_symbol_adapt4_sse2.renorm3

asm perf:

     0.45%  dav1d    libdav1d.so.1.1.0   [.] dav1d_msac_decode_bool_equi_sse2
     1.73%  dav1d    libdav1d.so.1.1.0   [.] dav1d_msac_decode_symbol_adapt4_sse2.renorm3

e.g. it goes from 1.32% of overall runtime to 0.85%.

Merge request reports