Basic GPU stroke rendering with support for caps and joins #408

armansito · 2023-11-08T18:06:17Z

This PR implements support for GPU-side stroke expansion with initial support for caps and joins. This is further progress towards implementing the design outlined in #303

The flatten stage implements rudimentary stroke offsetting based on displacing the endpoints of lines that result from flattening the baseline curve along their (transformed) normal vector. This results in a crude subdivision estimate for the offset curves but this is initially acceptable as it unblocks follow-up stroke work. @dragostis suggested a solution that sounds promising, which we could use until Euler Spiral based offsetting arrives.
The scheme described in Stroke rework #303 involving a "stroke cap marker" segment to detect when to render a start vs end cap on the GPU has been implemented with a slight tweak. For an open path, the start position and tangent require two points to be encoded in the marker which cannot be done with a lineTo without first inserting a moveTo. Since inserting a moveTo would terminate the subpath prematurely and because we want to preserve the invariant that only one segment has the SUBPATH_END_BIT set, the marker for an open path gets encoded as a quadTo instead. The marker for a closed path is encoded as a lineTo since the final control point of the preceding segment (i.e. the close) should already match the start control point.

Since we can distinguish between an open and close path based on whether the marker is encoded as a quadTo or a lineTo, adding a "close vs open" bit to the encoding wasn't necessary.
Currently only butt caps and bevel joins are supported. Support for other stroke styles will be added as a follow-up.
PathEncoder has some new handling for zero-length segments to avoid encoding a cap marker with a degenerate start tangent. If the lengths of all possible tangent vectors are within an epsilon threshold (currently 1e-12) then the segment is dropped. This will get extended to all segments in a follow-up PR to avoid encoding zero-length segments for strokes in general.
Added additional stroke test cases for closed paths, including a single curve with a co-incident start and end control point.

Implemented a stroking scheme in the flatten stage in which every flattened line segment gets offset along the curve normal by the line width. The line vertices are offset _after_ subdivision and does not try to accurately estimate subdivisions for the offset curve. Cusps are not detected and are prone to failure. Regions with high curvature (especially around self-intersections) are handled by simply boosting the subdivision count using the curve offset which is not correct but avoids awfully missing samples when the tolerance is set too low. This is intended as a placeholder but is the first step towards implementing correct strokes without optimizing for performance.

- Implemented the stroke cap marker segment encoding scheme outlined in #303. The proposed scheme has been slightly modified such that instead of an open/closed bit, the segment is encoded as a lineto or quadto to distinguish between the different states. Encoding the marker segment as a quad to for open segments also handles a limitation with only using linetos: for an open path, the marker must encode both the start tangent and the coordinates of the first control point to correctly position a start cap. Encoding these as 4 f32s requires an implicit moveto before the lineto which must terminate the preceding subpath. Encoding this as a quadto preserves the invariant that every stroked subpath only contains one segment with the SUBPATH_END_BIT set to 1, which now serves as the stroke cap marker. For a closed path, a lineto is sufficient since the preceding coordinate (from the lineto encoded for a "closepath" verb) must already equal the subpath's first control point. - Shitty stroking continues to work by detecting and ignoring the stroke cap marker.

The flatten stage now makes use of the stroke cap marker encoding scheme to draw caps and joins at the right locations. At this stage only the butt cap and bevel join styles are supported.

Make sure that a path start segment has a tangent that has a non-zero length.

- Add versions of the cubics without co-incident start and end points - Reposition the closed paths so they fit within the view

Make it so that tangent vectors are NOT normalized and normals are normalized. Vector helpers that derive normals from tangents now always return normalized vectors. This keeps things internally consistent.

raphlinus

I've done a quick pass, may come back later with a deeper look. Overall it's what I expected to see. Thanks! I'll go on to 410 next.

raphlinus · 2023-11-11T18:27:03Z

crates/encoding/src/path.rs

+    /// segment tells the GPU stroker whether to draw a cap or a join based on the topology of the
+    /// path:
+    ///
+    /// 1. This marker segment is encoded as a `quad-to` for an open path and a `line-to` for a


Is it worth mentioning the fact that quad-to is just a convention to make the number of encoded coordinate pairs come out right? Could be as simple as saying (2 additional points) and (1 additional point) after the verb.

raphlinus · 2023-11-11T18:28:36Z

crates/encoding/src/path.rs

+            const EPS: f32 = 1e-12;
+            if (x - x0).abs() < EPS && (y - y0).abs() < EPS {
+                // Drop the segment if its length is zero
+                // TODO: do this for all not segments, not just start?


Can take the question mark off :)

raphlinus · 2023-11-11T18:28:49Z

crates/encoding/src/path.rs

+                (x2, y2)
+            } else {
+                // Drop the segment if its length is zero
+                // TODO: do this for all not segments, not just start?


raphlinus · 2023-11-11T18:29:19Z

crates/encoding/src/path.rs

+                (x3, y3)
+            } else {
+                // Drop the segment if its length is zero
+                // TODO: do this for all not segments, not just start?


Same. I'm also wondering if there might be an opportunity to factor for DRY.

Done. I initially thought of using kurbo::BezPath::tangents here (which does the same thing) but instead ended up writing a simpler helper for just the start tangent. Let me know if you'd rather use the kurbo utility here.

raphlinus · 2023-11-11T18:32:27Z

shader/flatten.wgsl

+
+        // HACK: this increase subdivision count as function of the stroke width for shitty strokes.
+        var tol = sqrt(REM_ACCURACY);
+        if cubic.flags == 1u {


Name the flag rather than having a magic constant?

Done. With the way the code is evolving, I may want to make some changes to the Cubic struct (such as removing the stroke field). That may be an opportunity to more strictly re-define the flags field.

raphlinus · 2023-11-11T18:38:00Z

shader/flatten.wgsl

-                    p3 = read_f32_point(tm.pathseg_offset + 6u);
-                }
+struct CubicPoints {
+    p0: vec2f,


Just noticing that we're not being consistent whether to use vec2f or vec2. I'm generally in favor of being more concise.

vec2f is a predeclared alias for vec2<f32> in type declarations. vec2 seems to work as a constructor expression where the element type gets inferred from the parameters. I've been using vec2f in new type declarations and I tried to match vec2(...) in constructor expressions since that's what the existing code did. I'd be happy to standardize on vec2f though.

raphlinus · 2023-11-11T18:44:16Z

shader/shared/pathtag.wgsl

@@ -21,6 +21,7 @@ let PATH_TAG_TRANSFORM = 0x20u;
 let PATH_TAG_PATH = 0x10u;
 let PATH_TAG_STYLE = 0x40u;
 #endif
+let PATH_TAG_SUBPATH_END_BIT = 4u;


We're not using _BIT convention for other flags, but that's a minor style nit. Also a self-reminder, at some point I'd like to pack this more tightly so a single tag word can have path, transform, and style set along with the other content.

* Fix the phrasing in a comment regarding subdivision count * Use the vec4f and vec2f predeclared aliases for vec4<f32> and vec2<f32> in `flatten`. I've been consistent with this style for new code in this shader and I may make the other shaders use the alises in a follow up since it's more concise.

armansito added 9 commits November 8, 2023 09:55

Add closed paths to the stroke styles test scene

f72a066

Update caps and joins to use the new stroke cap marker encoding

73dd128

The flatten stage now makes use of the stroke cap marker encoding scheme to draw caps and joins at the right locations. At this stage only the butt cap and bevel join styles are supported.

[encoding] Eliminate zero-length start segments

13810b8

Make sure that a path start segment has a tangent that has a non-zero length.

Update closed stroke tests

deba7c8

- Add versions of the cubics without co-incident start and end points - Reposition the closed paths so they fit within the view

Minor cleanup in flatten.wgsl

83359c5

Keep GPU stroking off by default

1f5a93d

Minor documentation and cleanup to zero-length tangent detection

24681fa

armansito requested review from raphlinus and dfrg November 8, 2023 18:07

armansito mentioned this pull request Nov 9, 2023

Stroke rework #303

Closed

[flatten] Settle on a consistent vector normalization scheme

3df413a

Make it so that tangent vectors are NOT normalized and normals are normalized. Vector helpers that derive normals from tangents now always return normalized vectors. This keeps things internally consistent.

armansito force-pushed the gpu-shitty-strokes branch from d653b27 to 3df413a Compare November 10, 2023 19:12

armansito mentioned this pull request Nov 10, 2023

Refactor transform handling in flatten & support miter join #410

Merged

raphlinus approved these changes Nov 11, 2023

View reviewed changes

Address review comments

eded0a1

armansito force-pushed the gpu-shitty-strokes branch from 0bbbe73 to eded0a1 Compare November 13, 2023 23:14

armansito merged commit e6c5bcf into main Nov 14, 2023
4 checks passed

armansito deleted the gpu-shitty-strokes branch November 14, 2023 00:29

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Basic GPU stroke rendering with support for caps and joins #408

Basic GPU stroke rendering with support for caps and joins #408

armansito commented Nov 8, 2023 •

edited

Loading

raphlinus left a comment

raphlinus Nov 11, 2023

armansito Nov 13, 2023

raphlinus Nov 11, 2023

raphlinus Nov 11, 2023

raphlinus Nov 11, 2023

armansito Nov 13, 2023

raphlinus Nov 11, 2023

armansito Nov 13, 2023

raphlinus Nov 11, 2023

armansito Nov 13, 2023 •

edited

Loading

raphlinus Nov 11, 2023

armansito Nov 13, 2023

Basic GPU stroke rendering with support for caps and joins #408

Basic GPU stroke rendering with support for caps and joins #408

Conversation

armansito commented Nov 8, 2023 • edited Loading

raphlinus left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

armansito Nov 13, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

armansito commented Nov 8, 2023 •

edited

Loading

armansito Nov 13, 2023 •

edited

Loading