Waveform issues with "High details" enabled

[daschuer]

Bug Description

RGB high details waveform are showing an offset:
High details, the beat marker is not on the visual beat:

No High details, the beat marker is at the expected position:

Tested with QT 6.2

Version

2.6~alpha

OS

Ubuntu Focal

[daschuer]

With another track, I see only the upper half. The waveform disappears entirely when I disable "Stereo coloration".

[Swiftb0y]

To clarify, the missing bottom half only applies to the "High Details" version?

[Swiftb0y]

Both of these refer to #13471 (comment) and #13471 (comment) respectively, right?

[daschuer]

To clarify, the missing bottom half only applies to the "High Details" version?

Yes.

Non of these issues are in 2.5 with Qt5
I am just building Qt6

[daschuer]

The 2.5 + Qt 6 build is also not affected.

[ywwg]

as I suspected, the stem support is causing the problem ( @acolombier ) commit 2911c38

[acolombier]

I believe I know how to fix it, I will prepare a fix for it soon.

[ywwg]

it looks like the waveform is still stretched by 2x, reopening

[acolombier]

Could you please attach some screenshots or reproducing steps?

[ywwg]

looks like it's not 2x, but some small fraction
Screencast from 2024-07-17 13-42-06.webm

[ywwg]

seeking to the end of the waveform is sufficient for seeing the issue

[acolombier]

Wondering if there could be some issue due to memory alignment. I should have some time this weekend, if anyone wants to give it a go in the meantime, please feel free to!

[Swiftb0y]

Not trying to be "I told you so", but I think this is a convincing argument for why abusing a texture for uploading a blob to the GPU is a bad idea...

[ywwg]

if we could get the same level of detail with another method I'd be all for it

[ywwg]

ah it's not an offset, there's just a point in the waveform at which suddenly it starts showing the wrong data. In this video you can see the left half of the waveform is correct, and the right half is wrong.

Screencast.from.2024-07-19.11-10-37.webm

[ywwg]

it looks like it wraps around to the beginning of the waveform -- I'm guessing there's a miscalculation of the total length of the data?

[ywwg]

Screencast.from.2024-07-19.11-20-13.webm

[ywwg]

error always starts at 2:28.60

[ywwg]

are we overflowing int32

[ywwg]

hmmmm it all goes wrong at visualindex==65535 :)

[ywwg]

ok well I guess this is technically a different problem than the 2x issue last time... but it seems ok to reuse the bug