Rotation/orientation

[star69rem]

Is there documentation about the view matrix? I'm assuming that you have to pass it settings there to get images back that are rotated 90, 180, 270, etc.

I wanted to just rotate it myself, but I can't, because I end up having an input with a width that's less than the output that I want, and so I'm getting vertical scanlines because it probably thinks the image is coming in rotated.

[chyyran]

As of 82ad516 Vulkan and DX11 run internal passes with an identity matrix so hopefully that should mimic RetroArch's behaviour.

The default MVP for use in all runtimes on the final pass is

pub static DEFAULT_MVP: &[f32; 16] = &[
    2f32, 0.0, 0.0, 0.0,
    0.0, 2.0, 0.0, 0.0,
    0.0, 0.0, 0.0, 0.0,
    -1.0, -1.0, 0.0, 1.0,
];

If you apply a rotation on that matrix it should work, Just pass a pointer to the float[16] in *_frame.

It remains to be seen if I want to provide a 'rotation' API option instead in radians or if just providing the final pass MVP is sufficient.

[star69rem]

A game's internal resolution can be 692x224, and if I correct the aspect ratio for a 90 degree rotation at 3840x1600, I want a 1600x666 resolution image. For some reason, though, when I do this, and then I rotate the output myself, I get vertical scanlines.

It's really inconvenient to have to deal with another set of rotation code when I already have code to do it that works with my non shader path.

[chyyran]

The default MVP assumes that the input is unrotated, so it’ll apply horizontal scanlines to that. After you rotate it then the scanlines being vertical makes sense. You need to provide an MVP such that vertical scanlines are applied.

I would think that just rotating the final MVP would be sufficient but if not I can expose the internal identity matrix as a parameter as well.

[star69rem]

The default MVP assumes that the input is unrotated, so it’ll apply horizontal scanlines to that. After you rotate it then the scanlines being vertical makes sense. You need to provide an MVP such that vertical scanlines are applied.

I would think that just rotating the final MVP would be sufficient but if not I can expose the internal identity matrix as a parameter as well.

That's why I'm having problems. I'm passing in an unrotated image, expecting to get one back with horizontal scanlines, and then rotating myself, but what's happening is that I'm passing in an unrotated image and getting back one with vertical scanlines.

[chyyran]

Have you tried passing in a rotated MVP?

[star69rem]

Have you tried passing in a rotated MVP?

No, because I want horizontal scanlines.

[chyyran]

If you pass in the rotated MVP you'll get vertical scanlines on the unrotated image, then when you rotate it they'll turn out horizontal.

[star69rem]

That doesn't make sense to me. I don't want vertical scanlines. I want horizontal scanlines and to rotate the whole image.

[star69rem]

Please close this issue. What happened is that I had a super wide ultrawide image, and it just so happens when you have an image that thin rotated, just due to lack of resolution it looked like the scanlines were vertical when they weren't. When I tried rotating 4:3 aspect ratio modes, it looked okay. Sorry!