Add 2D axis-aligned bounding box type #259
Conversation
Like graphene_box_t, but for 2D areas. It's more efficient than rectangles, at the cost of legibility.
| { | ||
| /*< private >*/ | ||
| GRAPHENE_PRIVATE_FIELD (graphene_vec2_t, min); | ||
| GRAPHENE_PRIVATE_FIELD (graphene_vec2_t, max); |
There was a problem hiding this comment.
Should this be a single graphene_vec4_t because the vec2's are just vec4's in disguise afaik?
The Rect type in GSK's Vulkan shader uses a vec4 itself, and that worked really well.
There was a problem hiding this comment.
Partial lane reads are always slower than full lane ones; if we store the min/max in the same SIMD type like this:
| min_x | min_y | max_x | max_y |
then it should be possible to optimise the access using:
graphene_simd4f_t min = graphene_simd4f_merge_low (box->value, graphene_sim4f_zero ());
graphene_simd4f_t max = graphene_simd4f_merge_high (box->value, graphene_simd4f_zero ());| graphene_point_t *max); | ||
| GRAPHENE_AVAILABLE_IN_1_12 | ||
| void graphene_box2d_get_vertices (const graphene_box2d_t *box, | ||
| graphene_vec2_t vertices[]); |
There was a problem hiding this comment.
A graphene_box2d_to_float (box, float[4]) would come in handy here for uploading to GLSL.
| GRAPHENE_AVAILABLE_IN_1_12 | ||
| void graphene_box2d_expand_scalar (const graphene_box2d_t *box, | ||
| float scalar, | ||
| graphene_box2d_t *res); |
There was a problem hiding this comment.
a linear transform function / fused multiply-add ala scale * box + offset is the one thing I've needed in the GSK clip rects to handle affine transforms. And it works as a fallback for all the other transforms (scales, offsets, etc), too.
| GRAPHENE_AVAILABLE_IN_1_12 | ||
| bool graphene_box2d_intersection (const graphene_box2d_t *a, | ||
| const graphene_box2d_t *b, | ||
| graphene_box2d_t *res); |
There was a problem hiding this comment.
intersection checking is the one absolutely most critical operation we are doing when rendering in GTK.
In fact, in certain workloads (think a GtkListbox in a scrolledwindow, like gtk4-demo --run=listbox) intersecting alone can take >50% of time spend in gsk_renderer_render() (according to sysprof).
So having an (inline?) function that just does an intersection check and returns true/false would be very neat. Something like bool graphene_box2d_intersects (a, b)
| GRAPHENE_AVAILABLE_IN_1_12 | ||
| graphene_box2d_t * graphene_box2d_init_from_vec2 (graphene_box2d_t *box, | ||
| const graphene_vec2_t *min, | ||
| const graphene_vec2_t *max); |
There was a problem hiding this comment.
Transitioning GTK code from using graphene_rect_t would benefit from graphene_box2d_init_from_rect() and graphene_rect_init_from_box2d().
| graphene_point_t *min); | ||
| GRAPHENE_AVAILABLE_IN_1_12 | ||
| void graphene_box2d_get_max (const graphene_box2d_t *box, | ||
| graphene_point_t *max); |
There was a problem hiding this comment.
What's your opinion on graphene_box2d_get_minmax(box, &min, &max) instead/on top of these two?
There was a problem hiding this comment.
A fast path copy operation for both vertices would probably be okay, in the same way as sincos(); I mainly kept the same API as graphene_box_t, but I can see it being useful there, too.
Works like
graphene_box_t, but in two dimensions.