[Graphical Bug, Nightly] Texture Animations render incorrectly on non-20 fps

[Xandervile]

Noticed on the nightly build (downloaded 30/05/2026) - Textures when Match Refresh Rate is enabled that are animated tend to have some visual issue where they stutter. This is not occuring when frame rate is fixed at standard OoT frame rate, so may be an interpretation issue for framerates higher than 20.

EDIT: I'm number 6666, let's go

[A-Green-Spoon]

Narrowed this down when this was introduced to the LUS bump commit in #6601

Just a visual example:
https://github.com/user-attachments/assets/7a558f1c-c6fc-4e2a-8ca0-fea4051c61a1

[banteg]

also affects lava, not just water
https://github.com/user-attachments/assets/5810c003-7427-4e9b-86a6-cf6950e676bf

[banteg]

I did some local instrumentation on the Fire Temple lava case and I think this is not frame pacing or skipped interpolation. The interpolation replay cycle looked regular (0 -> 1 -> 2 -> 0 ...), but one replay phase derived a different texture window size from the interpolated tile coordinates.

Example from the bad phase:

replay=1
tile0=(0.000,6.333,124.000,130.333)
renderSize0=(32, 31)
texSize0=(32, 31)

The adjacent phases for the same lava surface were 32x32:

replay=0 tile0=(0.000,6.000,124.000,130.000) renderSize0=(32, 32)
replay=2 tile0=(0.000,6.667,124.000,130.667) renderSize0=(32, 32)

So the visual jitter seems to come from the renderer alternating between a 32x32 and 32x31 texture window. The intended height is still 32 texels, but the current conversion effectively truncates (lrt - ult + 4) / 4 after interpolation. With float values like 6.333 / 130.333, precision can land just below 32.0, so the cast becomes 31.

A minimal local fix that made the lava case look stable for me was to round this derived tile size instead of truncating it, in the places that compute rendered/imported tile dimensions from uls/ult/lrs/lrt in libultraship/src/fast/interpreter.cpp (ImportTextureRgba16, ImportTextureRgba32, ImportTextureCi4, ImportTextureCi8, and GfxSpTri1). Shape of the change:

static uint32_t GetTileSizeFromCoordinates(float low, float high) {
    float size = (high - low + 4.0f) / 4.0f;
    if (size <= 0.0f) {
        return 0;
    }
    return static_cast<uint32_t>(lroundf(size));
}

Then use that helper instead of directly casting the (high - low + 4) / 4 result to uint32_t. I did not need to change the interpolation scheduling or texture cache behavior for this specific symptom.