This works out a bit more favourably than VP8's due to:
- additional multiplications that can be vectored,
- hardware-supported fixed-point rounding mode.

vp7_luma_dc_wht_c:       3.2
vp7_luma_dc_wht_rvv_i64: 2.0