Browse Source

Shaders: extract reusable line drawing GLSL code to separate files.

Will need those in the Ui library, in a form that's quite restricted
compared to what the generic shader provides (2D only, uniform
buffers always, ...). Plus it doesn't make sense to make the Ui library
depend on Shaders, so it would be just these two files copied verbatim
between the two.
pull/651/merge
Vladimír Vondruš 1 year ago
parent
commit
59a60026f3
  1. 57
      src/Magnum/Shaders/Line.frag
  2. 76
      src/Magnum/Shaders/Line.in.frag
  3. 334
      src/Magnum/Shaders/Line.in.vert
  4. 314
      src/Magnum/Shaders/Line.vert
  5. 2
      src/Magnum/Shaders/LineGL.cpp
  6. 6
      src/Magnum/Shaders/resources-gl.conf
  7. 2
      src/Magnum/Shaders/resources-gles2.conf

57
src/Magnum/Shaders/Line.frag

@ -200,6 +200,15 @@ layout(location = OBJECT_ID_OUTPUT_ATTRIBUTE_LOCATION)
out highp uint fragmentObjectId;
#endif
/* Coming from Line.in.frag. That file is added after this one in order to have
#extension directives before any code. */
mediump float lineBlendFactor(
in highp vec2 centerDistanceSigned,
in highp float halfSegmentLength,
in highp float hasCap,
in mediump const float width,
in mediump const float smoothness);
void main() {
#ifdef UNIFORM_BUFFERS
#ifdef OBJECT_ID
@ -218,48 +227,12 @@ void main() {
mediump const float smoothness = materials[materialId].material_smoothness;
#endif
/* Calculate a distance from the original line endpoint (B). Assuming a cap
that's not a butt, actual quad vertices (2, 3 on the left diagram) would
be at a distance `width/2` in both X and Y (in the space of the line
segment, where X is in direction of the segment and Y is in direction to
the line edges):
----------2 --------2
| |
[0,0] B | [0,0] B
| |
----------3 --------3
For a butt cap, the endpoint B would be at the edge instead (right
diagram) -- to have handling consistent for all cap styles, add
`width/2` to the center distance in that case. For fragments on the left
of B the X distance would be negative, make it 0 in that case
instead. */
highp const vec2 centerDistance = abs(centerDistanceSigned);
highp vec2 endpointDistance = vec2(max(centerDistance.x
#ifdef CAP_STYLE_BUTT
+ width*0.5
#endif
- halfSegmentLength, 0.0), centerDistance.y);
/* If hasCap is negative, it means the nearest endpoint is a join, not a
cap. Thus no smoothing happens in the direction of a cap, i.e. same as
if we'd be at the center of the line. */
if(hasCap < 0.0) endpointDistance.x = 0.0;
/* Calculate a single distance factor out of the two-dimensional endpoint
distance. This will form the cap shape. */
#if defined(CAP_STYLE_BUTT) || defined(CAP_STYLE_SQUARE)
highp const float distance1D = max(endpointDistance.x, endpointDistance.y);
#elif defined(CAP_STYLE_ROUND)
highp const float distance1D = length(endpointDistance);
#elif defined(CAP_STYLE_TRIANGLE)
highp const float distance1D = endpointDistance.x + endpointDistance.y;
#else
#error
#endif
mediump const float factor = smoothstep(width*0.5 - smoothness, width*0.5 + smoothness, distance1D);
mediump const float factor = lineBlendFactor(
centerDistanceSigned,
halfSegmentLength,
hasCap,
width,
smoothness);
fragmentColor = mix(
#ifdef VERTEX_COLOR

76
src/Magnum/Shaders/Line.in.frag

@ -0,0 +1,76 @@
/*
This file is part of Magnum.
Copyright © 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018, 2019,
2020, 2021, 2022, 2023, 2024, 2025
Vladimír Vondruš <mosra@centrum.cz>
Permission is hereby granted, free of charge, to any person obtaining a
copy of this software and associated documentation files (the "Software"),
to deal in the Software without restriction, including without limitation
the rights to use, copy, modify, merge, publish, distribute, sublicense,
and/or sell copies of the Software, and to permit persons to whom the
Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included
in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
DEALINGS IN THE SOFTWARE.
*/
mediump float lineBlendFactor(
in highp vec2 centerDistanceSigned,
in highp float halfSegmentLength,
in highp float hasCap,
in mediump const float width,
in mediump const float smoothness)
{
/* Calculate a distance from the original line endpoint (B). Assuming a cap
that's not a butt, actual quad vertices (2, 3 on the left diagram) would
be at a distance `width/2` in both X and Y (in the space of the line
segment, where X is in direction of the segment and Y is in direction to
the line edges):
----------2 --------2
| |
[0,0] B | [0,0] B
| |
----------3 --------3
For a butt cap, the endpoint B would be at the edge instead (right
diagram) -- to have handling consistent for all cap styles, add
`width/2` to the center distance in that case. For fragments on the left
of B the X distance would be negative, make it 0 in that case
instead. */
highp const vec2 centerDistance = abs(centerDistanceSigned);
highp vec2 endpointDistance = vec2(max(centerDistance.x
#ifdef CAP_STYLE_BUTT
+ width*0.5
#endif
- halfSegmentLength, 0.0), centerDistance.y);
/* If hasCap is negative, it means the nearest endpoint is a join, not a
cap. Thus no smoothing happens in the direction of a cap, i.e. same as
if we'd be at the center of the line. */
if(hasCap < 0.0) endpointDistance.x = 0.0;
/* Calculate a single distance factor out of the two-dimensional endpoint
distance. This will form the cap shape. */
#if defined(CAP_STYLE_BUTT) || defined(CAP_STYLE_SQUARE)
highp const float distance1D = max(endpointDistance.x, endpointDistance.y);
#elif defined(CAP_STYLE_ROUND)
highp const float distance1D = length(endpointDistance);
#elif defined(CAP_STYLE_TRIANGLE)
highp const float distance1D = endpointDistance.x + endpointDistance.y;
#else
#error
#endif
return smoothstep(width*0.5 - smoothness, width*0.5 + smoothness, distance1D);
}

334
src/Magnum/Shaders/Line.in.vert

@ -0,0 +1,334 @@
/*
This file is part of Magnum.
Copyright © 2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018, 2019,
2020, 2021, 2022, 2023, 2024, 2025
Vladimír Vondruš <mosra@centrum.cz>
Permission is hereby granted, free of charge, to any person obtaining a
copy of this software and associated documentation files (the "Software"),
to deal in the Software without restriction, including without limitation
the rights to use, copy, modify, merge, publish, distribute, sublicense,
and/or sell copies of the Software, and to permit persons to whom the
Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included
in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
DEALINGS IN THE SOFTWARE.
*/
/* Point annotation, matching the LineVertexAnnotation enum bits */
#define ANNOTATION_UP_MASK 1u
#define ANNOTATION_JOIN_MASK 2u
#define ANNOTATION_BEGIN_MASK 4u
/* Same as Math::Vector2::perpendicular() */
vec2 perpendicular(vec2 a) {
return vec2(-a.y, a.x);
}
highp vec2 expandLineVertex(
in highp const vec2 transformedPosition,
in highp const vec2 transformedPreviousPosition,
in highp const vec2 transformedNextPosition,
in lowp const uint annotation,
in mediump const float width,
in mediump const float smoothness,
in highp const float miterLimit,
in lowp const vec2 viewportSize,
out highp vec2 centerDistanceSigned,
out highp float halfSegmentLength,
out highp float hasCap)
{
/* Decide about the line direction vector `d` and edge direction vector `e`
from the `pointMarkerComponent` input. Quad corners 0 and 1 come from
segment endpoint A, are marked with the ANNOTATION_BEGIN_MASK bit and so
their line direction is taken from `nextPosition`, quad corners 2 and 3
come from B and are *not* marked with ANNOTATION_BEGIN_MASK and so their
line direction is taken from `previousPosition`, with the direction
being always from point A to point B. The edge direction is then
perpendicular to the line direction, with points 0 and 2 marked with
ANNOTATION_UP_MASK using it directly, while points 1 and 3 don't have
ANNOTATION_UP_MASK and have to negate it:
^ ^
e e
| |
[UP, BEGIN] 0-d--> 2-d--> [UP]
A B
[BEGIN] 1-d--> 3-d--> []
| |
e e
v v
The ANNOTATION_CAP_MASK is then used below. */
highp const vec2 lineDirection = bool(annotation & ANNOTATION_BEGIN_MASK) ?
transformedNextPosition - transformedPosition :
transformedPosition - transformedPreviousPosition;
mediump const float edgeSign = bool(annotation & ANNOTATION_UP_MASK) ? 1.0 : -1.0;
mediump const float neighborSign = bool(annotation & ANNOTATION_BEGIN_MASK) ? -1.0 : 1.0;
/* Line direction and its length converted from the [-1, 1] unit square to
the screen space so we properly take aspect ratio into account. In the
end it undoes the transformation by multiplying by 2.0/viewportSize
again. */
highp const vec2 screenspaceLineDirection = lineDirection*viewportSize/2.0;
highp const float screenspaceLineDirectionLength = length(screenspaceLineDirection);
/* Normalized screenspace line and edge direction. In case of zero-sized
lines (i.e., points) the X axis is picked as line direction instead, and
thus Y axis for edge direction. */
highp const vec2 screenspaceLineDirectionNormalized = screenspaceLineDirectionLength == 0.0 ? vec2(1.0, 0.0) : screenspaceLineDirection/screenspaceLineDirectionLength;
highp const vec2 screenspaceEdgeDirectionNormalized = perpendicular(screenspaceLineDirectionNormalized);
/* Line width includes also twice the smoothness (because it's a radius
instead of a diameter, and is on both sides of the line), and is rounded
to whole pixels. So for the edge distance we need half of it. */
mediump const float edgeDistance = ceil(width + 2.0*smoothness)*0.5;
#ifdef CAP_STYLE_BUTT
mediump const float capDistance = ceil(2.0*smoothness)*0.5;
#elif defined(CAP_STYLE_SQUARE) || defined(CAP_STYLE_ROUND) || defined(CAP_STYLE_TRIANGLE)
mediump const float capDistance = edgeDistance;
#else
#error
#endif
/* Line segment half-length, passed to the fragment shader. Same for all
four points. */
halfSegmentLength = screenspaceLineDirectionLength*0.5;
/* Calculate the actual endpoint parameters depending on whether we're at a
line cap, line join bevel, line join miter etc.
- `screenspacePointDirection` contains screenspace direction from
`transformedPosition` to the actual point. After undoing the
screenspace projection the sum of the two is written to
gl_Position.
- `centerDistanceSigned` contains signed distance from the edge to
center, passed to the fragment shader. It's chosen in a way that
interpolates to zero in the quad center, and the area where
`all(abs(centerDistanceSigned) <= vec2(halfSegmentLength +
capDistance, edgeDistance))` is inside the line.
- `hasCap` contains `abs(centerDistanceSigned.x)` with a sign
positive if the point is a cap and negative if it isn't. Given
segment endpoints A and B (and quad points 0/1 and 2/3
corresponding to these), the following cases can happen:
- if both have a cap, it's a negative value in both, thus has a
constant negative value in the fragment shader
- if neither have a cap, it's a positive value in both, thus has
a constant positive value in the fragment shader
- if one has a cap and the other not, it's a negative value in
one and positive in the other, interpolating to zero in the
quad center
In the fragment shader, `abs(centerDistanceSigned)` and `sign(hasCap)`
is then used to perform cap rendering and antialiasing. For example,
with a standalone line segment that has square caps on both ends, the
value of `centerDistanceSigned` is like in the following diagram, with
`d` being `halfSegmentLength`, `w` being `edgeDistance`, `c` being
`capDistance`, and an extra margin for `smoothness` indicated by `s` and
the double border:
[-d-c-s,+w+s] [+d+c+s,+w+s]
0-----------------------------2
[-d-c,+w]------------------[+d+c,+w]
| | | | hasCap[0] = hasCap[1] = +d+c+s
[-d-c,0] [0,0] [+d+c,0]
| | | | hasCap[2] = hasCap[3] = +d+c+s
[-d-c,-w]------------------[+d+c,-w]
1-----------------------------3
[-d-c-s,-w-s] [+d+c+s,-w+s]
With a cap only on the left side, `centerDistanceSigned` would be like
this. Note the absence of a smoothness margin on the right side:
[-d-c-s,+w+s] [+d,+w+s]
0---------------------------2
[-d-c,+w]-------------------[+d,+w]
| | | hasCap[0] = hasCap[1] = +d+c+s
[-d-c,0] [0,0] [+d,0]
| | | hasCap[2] = hasCap[3] = -d
[-d-c,-w]-------------------[+d,-w]
1---------------------------3
[-d-c-s,-w-s] [+d,-w-s]
*/
centerDistanceSigned =
/* The the Y coordinate is same for all cases, X coordinate gets
further adjusted below */
vec2(halfSegmentLength*neighborSign, edgeDistance*edgeSign);
highp vec2 screenspacePointDirection;
/* Line join */
if(bool(annotation & ANNOTATION_JOIN_MASK)) {
/* Neighbor direction `nd`, needed to distinguish whether this is the
inner or outer join point. Calculated with basically an inverse of
the logic used to calculate `lineDirection`, with the neighbor
direction always pointing from the A/B endpoint to the other
neighbor line endpoint:
<--nd-0 [BEGIN] [END] 2-nd-->
A B
<--nd-1 [BEGIN] [END] 3-nd--> */
highp const vec2 neighborDirection = bool(annotation & ANNOTATION_BEGIN_MASK) ?
transformedPreviousPosition - transformedPosition :
transformedNextPosition - transformedPosition;
/* Screenspace neighbor direction and its length, calculated
equivalently to screenspace line direction above */
highp const vec2 screenspaceNeighborDirectionNormalized = normalize(neighborDirection*viewportSize/2.0);
/* If the edge direction vector `e` and the neighbor direction vector
`nd` point to the opposite direction (i.e., their dot product is
negative), this is an outer point of the line and a candidate for
a bevel.
^
e
|
-d->-2
|\
B | nd
| \
-----3 v
If a miter join is used instead of a bevel, the point is beveled
only if the line direction `d` and neighbor direction `nd` is
sharper than a limit (i.e., their dot product, or a cosine of their
angle, is between `[-1, -miterLimit]`). */
const bool outerBeveledPoint =
dot(screenspaceEdgeDirectionNormalized*edgeSign, screenspaceNeighborDirectionNormalized) < 0.0
#if defined(JOIN_STYLE_MITER)
&& dot(screenspaceLineDirectionNormalized*neighborSign, screenspaceNeighborDirectionNormalized) < -miterLimit
#elif !defined(JOIN_STYLE_BEVEL)
#error
#endif
;
/* Outer point of a beveled join -- although
https://www.w3.org/TR/svg-strokes/#LineJoin doesn't define *what
exactly* is a bevel, it's defined as "Cuts the outside edge off
where a circle the diameter of the stroke intersects the stroke." at
e.g. https://apike.ca/prog_svg_line_cap_join.html.
0--- ----2a
| |^\
| | e -_
| | |ρ \
A-- ----|--B-e->2b
| | | _-|
| | _- |
| | _- | |
1-- --3 | |
| | |
C
Which ultimately means the `2a` and `2b` quad endpoints are simply
the edge direction vector `e` away from point B, in one case with
the `e` calculated from the AB segment, and in the other from the BC
segment. */
if(outerBeveledPoint) {
screenspacePointDirection = screenspaceEdgeDirectionNormalized*edgeDistance*edgeSign;
/* centerDistanceSigned doesn't need any adjustment, hasCap is set
below for both */
/* Otherwise it's either an outer point of a miter join (basically
points 2a and 2b from above evaluated to the same position), or the
inner point, which is the same for bevel and mitter joins. Given
normalized direction `d` and neighbor direction `nd`,
`normalized(d + nd)` is the "average" direction of the two and `perpendicular(normalized(d + nd))` gives us the direction from B to
2 (or from 3 to B):
0--- --------+---2
| | α/ \
| w | / j \
| |/ \
A-- +_-----d->-B \
| -_ α/α\ \
| -_ / nd \
| d + nd /-_ v \
1---- ----3 -_ \
\ -+
\ \
C
With `2α` being the angle between `d` and `nd`, `α` appears in two
right triangles and the following holds, `w` being the edge distance
from above, and `j` having the length that's needed to scale
`perpendicular(normalized(d + nd))` to get point 2:
|d + nd| w 2 w
sin(α) = -------- = --- --> |j| = --------
2 |d| |j| |d + nd|
Then, vector j is the following, meaning we avoid the normalization
square root completely:
perp(d + nd) (2 w)perp(d + nd)
j = |j| ------------ = -----------------
|d + nd| dot(d + nd)
Point 3 is then just in the opposite direction; for the other side
it's done equivalently. */
} else {
highp const vec2 averageDirection = neighborSign*screenspaceLineDirectionNormalized + screenspaceNeighborDirectionNormalized;
screenspacePointDirection = (perpendicular(averageDirection)*(neighborSign*edgeSign*2.0*edgeDistance/dot(averageDirection, averageDirection)));
/* By projecting the point direction onto the line direction we
get a signed distance from the endpoint, adjust center distance
with that */
centerDistanceSigned.x += dot(screenspacePointDirection, screenspaceLineDirectionNormalized);
}
/* No cap here, store a negative value. TODO If
sign(centerDistanceSigned.x) is different from neighborSign, then
the sign here should be taken based on whether the other point is a
join -- add more bits to the vertex annotation? */
hasCap = -abs(centerDistanceSigned.x);
/* Line cap otherwise -- the quad corner 0/1/2/3 a sum of the signed cap
distance (`cdS`) and signed edge distance vectors (`eDS`), which are
formed by the line direction vector `d` and its perpendicular vector.
Neighbor direction (i.e., the other input from the one used to calculate
`lineDirection`) isn't used at all in this case.
cDS
0<---+----------
| ^
| | eDS
| |
| A--d-->
|
|
|
1---
The signed center distance a sum of half segment length and the cap
distance, multiplied by the cap sign (thus negative for points derived
from A and positive for B). */
} else {
screenspacePointDirection =
screenspaceLineDirectionNormalized*capDistance*neighborSign +
screenspaceEdgeDirectionNormalized*edgeDistance*edgeSign;
/* Add signed cap distance to the center distance */
centerDistanceSigned.x += capDistance*neighborSign;
/* Cap is here, store a positive value */
hasCap = abs(centerDistanceSigned.x);
}
/* Undo the screenspace projection */
return screenspacePointDirection*2.0/viewportSize;
}

314
src/Magnum/Shaders/Line.vert

@ -292,10 +292,20 @@ flat out highp uint interpolatedInstanceObjectId;
flat out highp uint drawId;
#endif
/* Same as Math::Vector2::perpendicular() */
vec2 perpendicular(vec2 a) {
return vec2(-a.y, a.x);
}
/* Coming from Line.in.vert. That file is added after this one in order to have
#extension directives before any code. */
highp vec2 expandLineVertex(
in highp const vec2 transformedPosition,
in highp const vec2 transformedPreviousPosition,
in highp const vec2 transformedNextPosition,
in lowp const uint annotation,
in mediump const float width,
in mediump const float smoothness,
in highp const float miterLimit,
in lowp const vec2 viewportSize,
out highp vec2 centerDistanceSigned,
out highp float halfSegmentLength,
out highp float hasCap);
void main() {
#ifdef UNIFORM_BUFFERS
@ -373,290 +383,18 @@ void main() {
#error
#endif
/* Decide about the line direction vector `d` and edge direction vector `e`
from the `pointMarkerComponent` input. Quad corners 0 and 1 come from
segment endpoint A, are marked with the ANNOTATION_BEGIN_MASK bit and so
their line direction is taken from `nextPosition`, quad corners 2 and 3
come from B and are *not* marked with ANNOTATION_BEGIN_MASK and so their
line direction is taken from `previousPosition`, with the direction
being always from point A to point B. The edge direction is then
perpendicular to the line direction, with points 0 and 2 marked with
ANNOTATION_UP_MASK using it directly, while points 1 and 3 don't have
ANNOTATION_UP_MASK and have to negate it:
^ ^
e e
| |
[UP, BEGIN] 0-d--> 2-d--> [UP]
A B
[BEGIN] 1-d--> 3-d--> []
| |
e e
v v
The ANNOTATION_CAP_MASK is then used below. */
highp const vec2 lineDirection = bool(annotation & ANNOTATION_BEGIN_MASK) ?
transformedNextPosition - transformedPosition :
transformedPosition - transformedPreviousPosition;
mediump const float edgeSign = bool(annotation & ANNOTATION_UP_MASK) ? 1.0 : -1.0;
mediump const float neighborSign = bool(annotation & ANNOTATION_BEGIN_MASK) ? -1.0 : 1.0;
/* Line direction and its length converted from the [-1, 1] unit square to
the screen space so we properly take aspect ratio into account. In the
end it undoes the transformation by multiplying by 2.0/viewportSize
again. */
highp const vec2 screenspaceLineDirection = lineDirection*viewportSize/2.0;
highp const float screenspaceLineDirectionLength = length(screenspaceLineDirection);
/* Normalized screenspace line and edge direction. In case of zero-sized
lines (i.e., points) the X axis is picked as line direction instead, and
thus Y axis for edge direction. */
highp const vec2 screenspaceLineDirectionNormalized = screenspaceLineDirectionLength == 0.0 ? vec2(1.0, 0.0) : screenspaceLineDirection/screenspaceLineDirectionLength;
highp const vec2 screenspaceEdgeDirectionNormalized = perpendicular(screenspaceLineDirectionNormalized);
/* Line width includes also twice the smoothness (because it's a radius
instead of a diameter, and is on both sides of the line), and is rounded
to whole pixels. So for the edge distance we need half of it. */
mediump const float edgeDistance = ceil(width + 2.0*smoothness)*0.5;
#ifdef CAP_STYLE_BUTT
mediump const float capDistance = ceil(2.0*smoothness)*0.5;
#elif defined(CAP_STYLE_SQUARE) || defined(CAP_STYLE_ROUND) || defined(CAP_STYLE_TRIANGLE)
mediump const float capDistance = edgeDistance;
#else
#error
#endif
/* Line segment half-length, passed to the fragment shader. Same for all
four points. */
halfSegmentLength = screenspaceLineDirectionLength*0.5;
/* Calculate the actual endpoint parameters depending on whether we're at a
line cap, line join bevel, line join miter etc.
- `screenspacePointDirection` contains screenspace direction from
`transformedPosition` to the actual point. After undoing the
screenspace projection the sum of the two is written to
gl_Position.
- `centerDistanceSigned` contains signed distance from the edge to
center, passed to the fragment shader. It's chosen in a way that
interpolates to zero in the quad center, and the area where
`all(abs(centerDistanceSigned) <= vec2(halfSegmentLength +
capDistance, edgeDistance))` is inside the line.
- `hasCap` contains `abs(centerDistanceSigned.x)` with a sign
positive if the point is a cap and negative if it isn't. Given
segment endpoints A and B (and quad points 0/1 and 2/3
corresponding to these), the following cases can happen:
- if both have a cap, it's a negative value in both, thus has a
constant negative value in the fragment shader
- if neither have a cap, it's a positive value in both, thus has
a constant positive value in the fragment shader
- if one has a cap and the other not, it's a negative value in
one and positive in the other, interpolating to zero in the
quad center
In the fragment shader, `abs(centerDistanceSigned)` and `sign(hasCap)`
is then used to perform cap rendering and antialiasing. For example,
with a standalone line segment that has square caps on both ends, the
value of `centerDistanceSigned` is like in the following diagram, with
`d` being `halfSegmentLength`, `w` being `edgeDistance`, `c` being
`capDistance`, and an extra margin for `smoothness` indicated by `s` and
the double border:
[-d-c-s,+w+s] [+d+c+s,+w+s]
0-----------------------------2
[-d-c,+w]------------------[+d+c,+w]
| | | | hasCap[0] = hasCap[1] = +d+c+s
[-d-c,0] [0,0] [+d+c,0]
| | | | hasCap[2] = hasCap[3] = +d+c+s
[-d-c,-w]------------------[+d+c,-w]
1-----------------------------3
[-d-c-s,-w-s] [+d+c+s,-w+s]
With a cap only on the left side, `centerDistanceSigned` would be like
this. Note the absence of a smoothness margin on the right side:
[-d-c-s,+w+s] [+d,+w+s]
0---------------------------2
[-d-c,+w]-------------------[+d,+w]
| | | hasCap[0] = hasCap[1] = +d+c+s
[-d-c,0] [0,0] [+d,0]
| | | hasCap[2] = hasCap[3] = -d
[-d-c,-w]-------------------[+d,-w]
1---------------------------3
[-d-c-s,-w-s] [+d,-w-s]
*/
centerDistanceSigned =
/* The the Y coordinate is same for all cases, X coordinate gets
further adjusted below */
vec2(halfSegmentLength*neighborSign, edgeDistance*edgeSign);
highp vec2 screenspacePointDirection;
/* Line join */
if(bool(annotation & ANNOTATION_JOIN_MASK)) {
/* Neighbor direction `nd`, needed to distinguish whether this is the
inner or outer join point. Calculated with basically an inverse of
the logic used to calculate `lineDirection`, with the neighbor
direction always pointing from the A/B endpoint to the other
neighbor line endpoint:
<--nd-0 [BEGIN] [END] 2-nd-->
A B
<--nd-1 [BEGIN] [END] 3-nd--> */
highp const vec2 neighborDirection = bool(annotation & ANNOTATION_BEGIN_MASK) ?
transformedPreviousPosition - transformedPosition :
transformedNextPosition - transformedPosition;
/* Screenspace neighbor direction and its length, calculated
equivalently to screenspace line direction above */
highp const vec2 screenspaceNeighborDirectionNormalized = normalize(neighborDirection*viewportSize/2.0);
/* If the edge direction vector `e` and the neighbor direction vector
`nd` point to the opposite direction (i.e., their dot product is
negative), this is an outer point of the line and a candidate for
a bevel.
^
e
|
-d->-2
|\
B | nd
| \
-----3 v
If a miter join is used instead of a bevel, the point is beveled
only if the line direction `d` and neighbor direction `nd` is
sharper than a limit (i.e., their dot product, or a cosine of their
angle, is between `[-1, -miterLimit]`). */
const bool outerBeveledPoint =
dot(screenspaceEdgeDirectionNormalized*edgeSign, screenspaceNeighborDirectionNormalized) < 0.0
#if defined(JOIN_STYLE_MITER)
&& dot(screenspaceLineDirectionNormalized*neighborSign, screenspaceNeighborDirectionNormalized) < -miterLimit
#elif !defined(JOIN_STYLE_BEVEL)
#error
#endif
;
/* Outer point of a beveled join -- although
https://www.w3.org/TR/svg-strokes/#LineJoin doesn't define *what
exactly* is a bevel, it's defined as "Cuts the outside edge off
where a circle the diameter of the stroke intersects the stroke." at
e.g. https://apike.ca/prog_svg_line_cap_join.html.
0--- ----2a
| |^\
| | e -_
| | |ρ \
A-- ----|--B-e->2b
| | | _-|
| | _- |
| | _- | |
1-- --3 | |
| | |
C
Which ultimately means the `2a` and `2b` quad endpoints are simply
the edge direction vector `e` away from point B, in one case with
the `e` calculated from the AB segment, and in the other from the BC
segment. */
if(outerBeveledPoint) {
screenspacePointDirection = screenspaceEdgeDirectionNormalized*edgeDistance*edgeSign;
/* centerDistanceSigned doesn't need any adjustment, hasCap is set
below for both */
/* Otherwise it's either an outer point of a miter join (basically
points 2a and 2b from above evaluated to the same position), or the
inner point, which is the same for bevel and mitter joins. Given
normalized direction `d` and neighbor direction `nd`,
`normalized(d + nd)` is the "average" direction of the two and `perpendicular(normalized(d + nd))` gives us the direction from B to
2 (or from 3 to B):
0--- --------+---2
| | α/ \
| w | / j \
| |/ \
A-- +_-----d->-B \
| -_ α/α\ \
| -_ / nd \
| d + nd /-_ v \
1---- ----3 -_ \
\ -+
\ \
C
With `2α` being the angle between `d` and `nd`, `α` appears in two
right triangles and the following holds, `w` being the edge distance
from above, and `j` having the length that's needed to scale
`perpendicular(normalized(d + nd))` to get point 2:
|d + nd| w 2 w
sin(α) = -------- = --- --> |j| = --------
2 |d| |j| |d + nd|
Then, vector j is the following, meaning we avoid the normalization
square root completely:
perp(d + nd) (2 w)perp(d + nd)
j = |j| ------------ = -----------------
|d + nd| dot(d + nd)
Point 3 is then just in the opposite direction; for the other side
it's done equivalently. */
} else {
highp const vec2 averageDirection = neighborSign*screenspaceLineDirectionNormalized + screenspaceNeighborDirectionNormalized;
screenspacePointDirection = (perpendicular(averageDirection)*(neighborSign*edgeSign*2.0*edgeDistance/dot(averageDirection, averageDirection)));
/* By projecting the point direction onto the line direction we
get a signed distance from the endpoint, adjust center distance
with that */
centerDistanceSigned.x += dot(screenspacePointDirection, screenspaceLineDirectionNormalized);
}
/* No cap here, store a negative value. TODO If
sign(centerDistanceSigned.x) is different from neighborSign, then
the sign here should be taken based on whether the other point is a
join -- add more bits to the vertex annotation? */
hasCap = -abs(centerDistanceSigned.x);
/* Line cap otherwise -- the quad corner 0/1/2/3 a sum of the signed cap
distance (`cdS`) and signed edge distance vectors (`eDS`), which are
formed by the line direction vector `d` and its perpendicular vector.
Neighbor direction (i.e., the other input from the one used to calculate
`lineDirection`) isn't used at all in this case.
cDS
0<---+----------
| ^
| | eDS
| |
| A--d-->
|
|
|
1---
The signed center distance a sum of half segment length and the cap
distance, multiplied by the cap sign (thus negative for points derived
from A and positive for B). */
} else {
screenspacePointDirection =
screenspaceLineDirectionNormalized*capDistance*neighborSign +
screenspaceEdgeDirectionNormalized*edgeDistance*edgeSign;
/* Add signed cap distance to the center distance */
centerDistanceSigned.x += capDistance*neighborSign;
/* Cap is here, store a positive value */
hasCap = abs(centerDistanceSigned.x);
}
/* Undo the screenspace projection */
highp const vec2 pointDirection = screenspacePointDirection*2.0/viewportSize;
highp const vec2 pointDirection = expandLineVertex(
transformedPosition,
transformedPreviousPosition,
transformedNextPosition,
annotation,
width,
smoothness,
miterLimit,
viewportSize,
centerDistanceSigned,
halfSegmentLength,
hasCap);
#ifdef TWO_DIMENSIONS
gl_Position = vec4(transformedPosition + pointDirection, 0.0, 1.0);

2
src/Magnum/Shaders/LineGL.cpp

@ -181,6 +181,7 @@ template<UnsignedInt dimensions> typename LineGL<dimensions>::CompileState LineG
}
vert.addSource(rs.getString("generic.glsl"_s))
.addSource(rs.getString("Line.vert"_s))
.addSource(rs.getString("Line.in.vert"_s))
.submitCompile();
GL::Shader frag{version, GL::Shader::Type::Fragment};
@ -212,6 +213,7 @@ template<UnsignedInt dimensions> typename LineGL<dimensions>::CompileState LineG
}
frag.addSource(rs.getString("generic.glsl"_s))
.addSource(rs.getString("Line.frag"_s))
.addSource(rs.getString("Line.in.frag"_s))
.submitCompile();
LineGL<dimensions> out{NoInit};

6
src/Magnum/Shaders/resources-gl.conf

@ -16,9 +16,15 @@ filename=generic.glsl
[file]
filename=Line.vert
[file]
filename=Line.in.vert
[file]
filename=Line.frag
[file]
filename=Line.in.frag
[file]
filename=MeshVisualizer.vert

2
src/Magnum/Shaders/resources-gles2.conf

@ -13,7 +13,7 @@ filename=FullScreenTriangle.glsl
[file]
filename=generic.glsl
# Line.vert and Line.frag missing
# Line{,.in}.vert and Line{,.in}.frag missing
[file]
filename=MeshVisualizer.vert

Loading…
Cancel
Save