Post

I often see recomps advertise 'native performance'. esp with wave of the slop AI recomps. Verbatim ASM -> C translation is actually quite poor from my experience. Around 25-50% of native, depending on arch. CPU register flags+stack emulation is expensive! Lot of headroom.

2 reposts3 likes

I've seen a function for distance squared go from 7 instructions to 40 instructions in one game, for instance. Due to all that overhead.