rv40: make rv40_adaptive_loop_filter() always_inline This function is called from the rv40_[hv]_loop_filter() wrappers with some arguments constant, so it really needs to be inlined. 3% faster overall on Cortex-A8.