discussion: compatibility of presets & implementation of shaders

[auralarch]

I'm just curious…

Has anyone experimented with doing side-by-side comparisons of presets?

One 'side' running the Milkdrop plugin in the final release of Winamp; the other 'side' running the sdl app on a Mac. Use the same audio source for both, ensure that an identical set of presets is installed on both (including any textures or other required extras), set both apps to select presets in alphabetical order, configure as many of the other settings as possible to the same options…

This seems like an interesting & fun way to check for anomalies. Compare performance; see if anything that may not be working or implemented in the same way; discover how much of various presets is tightly controlled by the music and the parameters of the preset, and how much is left to chance.

[milkdropper]

It looks like https://github.com/mvsoft74/BeatDrop might be a good tool to test with. It seems to be completely based on the opensource Milkdrop2 and does loopback capture as well.

[milkdropper]

I grabbed two presets at random (I swear, no bias) and compared them side by side with BeatDrop.

The results are drastically different.

11.milk Video: https://streamable.com/w0gif1 This type of preset is always in motion. They look different.

EoS - glowsticks v2 03 music shifter edit b (Jelly V2).milk Video: https://streamable.com/6g7bbz This preset relies entirely on music. To get a really honest comparison I paused my music, reset both (hard transitions to it in both apps) then resumed the music while recording. They look different.

[revmischa]

Very interesting! I think it'd be a great project to look into the milkdrop source and compare https://sourceforge.net/projects/milkdrop2/files/milkdrop2_v2.25c_OPEN_SOURCED_20130514_orig_code.zip/download

[kblaschke]

There's definitely a huge amount of differences between Milkdrop and projectM. I've already fixed some issues in my fork, it's not merged yet, but from what I've seen there are numerous issues (EDIT: items with :heavy_check_mark: are fixed and merged into in current master):

Issues affecting Milkdrop 1.x features:

• Waveform rendering is way too thick, like 10 pixels wide on 4K screens. Milkdrop only draws a single-pixel line, or a 2x2 grid for "thick" lines, but not more. (:heavy_check_mark:)
• Waveform type numbers aren't properly used by projectM, specifically the indices are off by 1. Type 0 (the circular wave) may actually cause crashes (or render garbage or the static waveform vertices from the previous preset) as the waveform enum in projectM starts with index 1. (:heavy_check_mark:)
• Some shapecode/waveform parameters aren't even supported. (:heavy_check_mark:)
• Instanced shapes are not rendered at all, only a single shape is displayed. (:heavy_check_mark:)
• Some equation variables/states aren't properly propagated or reset.
• Aspect variables are not properly assigned to the waveform/per-pixel equations. (:heavy_check_mark:)

Shader issues:

• Noise textures are created wrongly. Instead of having different noise on each of the four channels, the noise is the same in the RGB channels and full white on the alpha channel. (:heavy_check_mark:)
• Some texture samplers use the wrong wrap mode (clamp to edge instead of repeat/wrap), causing weird visual artifacts as the texture's border pixels get stretched out to the screen edges. (:heavy_check_mark:)
• More shader issues with yet unknown reasons. the shader transpiler we use isn't really complete and might not work with all possible shaders found in presets. Will require a thorough investigation for each preset.

[kblaschke]

With the above linked PR, most of the shader issues we had in the pst should be fixed. There will be a few remaining problems left, some which we probably can't solve due to differences between HLSL and GLSL. Only a few presets should be affected, so fixing the preset files might be a good approach (won't work for user-provided presets though).