New test scene: hugepath

[armansito]

This test draws lines with very large floating point y-coordinates (spanning 12345678901234567890.0 and -9.8765432109876543210e+19 respectively). This causes an overflow in bump estimation and results in rendering issues in Vello.

Bump Estimation

This test causes a panic in debug builds with the bump_estimate feature enabled:

thread 'main' panicked at /Users/armansito/work/vello/crates/encoding/src/estimate.rs:157:26:
attempt to multiply with overflow

This happens because the line generates an exorbitant amount of tile crossings which causes the segment count to overflow. This can be avoided by taking the viewport size and culling into account during estimation since a majority of the shape lies outside the viewport. See #541.

Rendering Issues

1. The frame rate (on my M1 Pro) doesn't go higher than 6 fps for this scene.
2. The expected rendering result is similar to the following (barring colors):

With vello, the top 3 strokes (winding from positive to negative) are missing under the default transform:

Under rotation, (particularly >90°) causes artifacts that look like this:

[DJMcNab]

I think I see the opposite problem (?) when pressing the C key in with_winit, where the strokes for the frame times are scaled to 1/u64::MAX, which seems to lead to terrible performance (~200ms frame times)

[DJMcNab]

Thanks, this looks like it brings in a real case where we have issues

I don't know what the path towards fixing this would be

[armansito]

Thanks, this looks like it brings in a real case where we have issues

I don't know what the path towards fixing this would be

There are some ideas, all mainly point towards implementing some type of viewport culling. For this particular test, we probably want to clamp the coordinates down to a reasonable range such that the ULP is sufficiently small for math.

[DJMcNab]

The only blocker is the clippy failure now

[waywardmonkeys]

@armansito Is this something that we want to land? Trying to clear up some PRs before doing a Vello 0.2 release...

[DJMcNab]

This only impacts the examples, so is entirely orthogonal to the release imo

But we should get this out of limbo.