First post, by `Moe`
- Rank
- Oldbie
Hi!
During the work on the hq2x scaler, I had to learn that dosbox renders the screen line-by-line, probably emulating the internals of the VGA chip quite closely. Unfortunately, the hq2x scaler needs more memory than the other scalers, so I am seeing lots of L1 cache misses. In the standalone implementation of hq2x which I use as base, this is not as bad, since the loop runs tighter.
So my questions:
Is there a good reason for calling the scaler line-by-line instead of on a whole buffer at a time?
Is it possible that this is changed so the scaler gets a full screen at a time?
Or else, if there are serious technical issues, could a shortcut be made for the game-typical 320x200x8, just like my code does? (in hq2x, having a separate 320x200 implementation gained 2fps due to better compiler optimization)