Line2: Add support for Alpha To Coverage

[gkjohnson]

Related issue: --

Description

Adds support for alphaToCoverage to Line2. It's a subtle effect but it enables anti aliasing on the rounded tips of the line segments which can be seen on the line ends and when the line is curving in the demo:

https://raw.githack.com/gkjohnson/three.js/atc-line2/examples/webgl_lines_fat.html

cc @WestLangley

[WestLangley]

Interesting idea: anti-aliasing of the line end-caps. It is very subtle.

I'll defer to @Mugen87 on this one.

[WestLangley]

Although with very fat (50px) straight lines I see artifacts.

[gkjohnson]

@WestLangley

It is very subtle.

Heh I've noticed it forever and it's always bugged me and now it can be addressed! It's definitely more noticeable on lower DPI displays and it also helps some jaggedness around curved corners on paths where the end points meet.

Although with very fat (50px) straight lines I see artifacts.

I'm not seeing this on the two machines I'm able to test with but I suspect it's because a derivative was being taken in a conditional the behavior of which probably varies depending on hardware. I've attempted to fix this by defining len2 before the conditional to ensure it's defined. Can you give it a try to see if that fixed it? Should I update the example so you can raise the line thickness to 50px?

A couple other fixes I just pushed:

• Adjust how the alpha coverage factor is computed.
• Set shader to use the derivatives extension so WebGL1 is supported.

[Mugen87]

I can see the artifacts, too. The only appear with AtC.

This glitch is visible with a 10px wide line and only at the start and end.

[gkjohnson]

I can see the artifacts, too. The only appear with AtC.

Heh uh oh 😅

@Mugen87

What hardware are you using? I'm hoping I can reproduce it somehow otherwise this won't be a fun one to track down.

[Mugen87]

It's an iMac (10.15.7) with a AMD Radeon R9 M295X 4 GB. Chrome version: 89.0.4389.82

[gkjohnson]

I was able to reproduce the issue on a mac laptop and have fixed them (in my case). It seems that on some hardware taking a derivative inside a conditional at all can cause undefined behavior even if the variable is defined outside the condition. I should have fixed it in the last commit -- let me know!

[Mugen87]

Looks better now!

[WestLangley]

Yep, looks better.

[mrdoob]

Thanks!