the recently added shaders support might provide you with the desiderd results

you should try out one of the latest SVN vanilla build
https://www.dosbox.com/wiki/SVN_Builds#Plain. … lean_SVN_builds

and set up your configuration as follows:
fullscreen=false
windowresolution=1280x960
output=opengl
scaler=none
glshader=sharp

Bilinear is basically averaging. Thinking about the positioning of the picture, it will most likely be offset by half a pixel compared to your method, so no original lines will remain, as every one will be a linear combination of two input lines. I have not played with the shaders myself, although this is on my to-do list, but I am afraid that your method is not achievable without additional patching. This is all just speculation, though. The pixel-perfect guys might be able to contribute here.

Ringding wrote on 2020-12-29, 12:24:

Bilinear is basically averaging. Thinking about the positioning of the picture, it will most likely be offset by half a pixel compared to your method, so no original lines will remain, as every one will be a linear combination of two input lines. I have not played with the shaders myself, although this is on my to-do list, but I am afraid that your method is not achievable without additional patching.

Adding (or subtracting) a quarter-pixel to the uv co-ordinates in the fragment shader would do it (assuming the output size is exactly double the input size).

I tried implementing this to see what it looked like, the answer: bloody terrible. Don't stare at it too long or you'll feel like you need glasses.

This is why a proper bilinear 2x resize uses alternating weights of 0.75/0.25 instead of 1.0/0.5.

jmarsh wrote on 2020-12-30, 12:32:

I tried implementing this to see what it looked like, the answer: bloody terrible. Don't stare at it too long or you'll feel like you need glasses.

This is why a proper bilinear 2x resize uses alternating weights of 0.75/0.25 instead of 1.0/0.5.

How is this terrible? It preserves all image data intact without any jaggies. Are we pretending our 640x480 games are suddenly HD and hence must have sharp edges, image accuracy be damned?

Do you know how I can get this result myself? That's what I am looking for.

PS: I agree text looks a bit blurry, but something like software-rendered Screamer Rally would look very good.

Make sure you are viewing the image at its original size (1280x960).

The problem is it adds a "hard" fuzzy edge to everything, which makes it look as if it's out of focus. A standard bilinear resize (which is what is actually used in the video industry - it's the default resampling algorithm on all video cards, after all) compensates for this.

jmarsh wrote on 2020-12-31, 01:34:

Make sure you are viewing the image at its original size (1280x960).

The problem is it adds a "hard" fuzzy edge to everything, which makes it look as if it's out of focus. A standard bilinear resize (which is what is actually used in the video industry - it's the default resampling algorithm on all video cards, after all) compensates for this.

Ok, I would still use it for certain types of games. Is there a way I can have it?