[Impeller] rrect blur has weird artifacts sometimes

[gaaclarke]

reproduction

TEST_P(DlGoldenTest, LargeDownscaleRrect) {
  impeller::Point content_scale = GetContentScale();
  auto draw = [&](DlCanvas* canvas, const std::vector<sk_sp<DlImage>>& images) {
    canvas->Scale(content_scale.x, content_scale.y);
    DlPaint paint;
    paint.setColor(DlColor::kYellow());
    paint.setMaskFilter(
        DlBlurMaskFilter::Make(DlBlurStyle::kNormal, /*sigma=*/10));
    canvas->DrawCircle(SkPoint::Make(0, 0), 500, paint);
  };

  DisplayListBuilder builder;
  draw(&builder, /*images=*/{});

  ASSERT_TRUE(OpenPlaygroundHere(builder.Build()));
}

observed output

[jonahwilliams]

@gaaclarke what is the scaling factor here? Does this look weird at regular scales, or is this a "galaxy" scale issue?

[gaaclarke]

@gaaclarke what is the scaling factor here? Does this look weird at regular scales, or is this a "galaxy" scale issue?

The scaling factor isn't big, it would just be the 2x from content_scale.x.

[gaaclarke]

@flar have you seen this before? Any theories?

[flar]

I remember having seen banding like that. When I looked at the math of the fast blurs I believed that it was due to the way the minimal number of bands being averaged were chosen.

I don't think the squircle solutions suffer from this effect because they provide a more continuous answer.

[gaaclarke]

fixed in flutter/engine#55604

[github-actions]

This thread has been automatically locked since there has not been any recent activity after it was closed. If you are still experiencing a similar issue, please open a new bug, including the output of flutter doctor -v and a minimal reproduction of the issue.