-
Deb Mukherjee authored
Adds an experiment for lower precision transform blend in order to enable 8- and 10- bit blends to be more efficiently vectorizable. Also disable av1_convolve_y function temporarily since they are incompatible with 7 - round_0 - round_1 being negative. Also disable SIMD version of warp if round_1 > 0, because of incompatibility. Noise level difference in coding efficiency: +0.014% (lowres, 15 frames cpu-used = 1) Change-Id: Ic6b9233901ca35ab735b0a5cdb850f96d3cd70ab
c40cb70c