Files
iris/core/src/render/shader/prelude.wgsl
T
iris-aiandClaude Opus 5 1f9dc48b80 Carry a box in a move slot, not a translation
A slot now holds the box its contents are placed within, in the coordinates
of the slot it names, and `prelude.wgsl` composes the chain with `within`
instead of adding a delta. A translation is the special case where the box
has its parent's relative extent, so every caller passes
`UiRegion::FULL.offset(delta)` and nothing changes on screen yet: 42 tests
pass and `tabs` at 1920x1200 is byte-identical.

`Moves::resolve` takes the region to compose rather than returning a sum, so
the CPU walk is the same operation the shader performs.

Measured against the translate slot on the same binary with
`tests/chain_cost.rs`, 200k instances: +0.6% at depth 1, +0.5% at 2, +0.8% at
4, then +9.6% at 8 and +32.2% at 64. Free at the depth opt-in slots produce,
which is the next commit; the per-level cost was always the dependent load
rather than the arithmetic.

The identity is `UiRegion::FULL` rather than zero, which `MoveOffset`'s
comment says beside the `Zeroable` that `Pod` requires: a zeroed entry is a
box of no extent and collapses its subtree to a point.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-14 12:14:24 -04:00

151 lines
4.2 KiB
WebGPU Shading Language

// Prepended to every primitive's shader, which declares its own instance data
// as `var<storage> <name>: array<T>` at group 1 binding 0, and an `fs_main`
// shading one instance of it. What it samples, if anything, is bound at group
// 2: the texture at binding 0 and the sampler at binding 1.
@group(0) @binding(0)
var<uniform> window: WindowUniform;
@group(0) @binding(1)
var<storage> masks: array<Mask>;
@group(0) @binding(2)
var<storage> move_offsets: array<MoveOffset>;
struct WindowUniform {
dim: vec2<f32>,
};
struct Mask {
x: UiSpan,
y: UiSpan,
move_idx: u32,
}
struct MoveOffset {
x: UiSpan,
y: UiSpan,
parent: u32,
}
struct Region {
x: UiSpan,
y: UiSpan,
}
const MOVE_NONE: u32 = 4294967295u;
// Keep in step with `iris_core::CHAIN_LIMIT`. It bounds a malformed cycle
// rather than any real tree, and the CPU walk uses the same number so both
// resolve a deep one the same way.
const CHAIN_LIMIT: u32 = 64u;
fn scalar_within(s: UiScalar, p: UiSpan) -> UiScalar {
return UiScalar(
mix(p.start.rel, p.end.rel, s.rel),
s.abs + mix(p.start.abs, p.end.abs, s.rel),
);
}
fn span_within(s: UiSpan, p: UiSpan) -> UiSpan {
return UiSpan(scalar_within(s.start, p), scalar_within(s.end, p));
}
fn resolve_move(idx: u32, local: Region) -> Region {
var r = local;
var at = idx;
for (var step = 0u; step < CHAIN_LIMIT; step++) {
if at == MOVE_NONE {
break;
}
let entry = move_offsets[at];
r = Region(span_within(r.x, entry.x), span_within(r.y, entry.y));
at = entry.parent;
}
return r;
}
struct UiSpan {
start: UiScalar,
end: UiScalar,
}
struct UiScalar {
rel: f32,
abs: f32,
}
struct InstanceInput {
@location(0) x_start: vec2<f32>,
@location(1) x_end: vec2<f32>,
@location(2) y_start: vec2<f32>,
@location(3) y_end: vec2<f32>,
@location(4) mask_idx: u32,
@location(5) move_idx: u32,
}
struct VertexOutput {
@location(0) top_left: vec2<f32>,
@location(1) bot_right: vec2<f32>,
@location(2) uv: vec2<f32>,
@location(3) @interpolate(flat) mask_idx: u32,
@location(4) @interpolate(flat) idx: u32,
@builtin(position) clip_position: vec4<f32>,
};
@vertex
fn vs_main(
@builtin(vertex_index) vi: u32,
@builtin(instance_index) ii: u32,
in: InstanceInput,
) -> VertexOutput {
var out: VertexOutput;
let local = Region(
UiSpan(UiScalar(in.x_start.x, in.x_start.y), UiScalar(in.x_end.x, in.x_end.y)),
UiSpan(UiScalar(in.y_start.x, in.y_start.y), UiScalar(in.y_end.x, in.y_end.y)),
);
let r = resolve_move(in.move_idx, local);
let top_left_rel = vec2(r.x.start.rel, r.y.start.rel);
let top_left_abs = vec2(r.x.start.abs, r.y.start.abs);
let bot_right_rel = vec2(r.x.end.rel, r.y.end.rel);
let bot_right_abs = vec2(r.x.end.abs, r.y.end.abs);
let top_left = floor(top_left_rel * window.dim) + floor(top_left_abs);
let bot_right = floor(bot_right_rel * window.dim) + floor(bot_right_abs);
let size = bot_right - top_left;
let uv = vec2<f32>(
f32(vi % 2u),
f32(vi / 2u)
);
let pos = (top_left + uv * size) / window.dim * 2.0 - 1.0;
out.clip_position = vec4<f32>(pos.x, -pos.y, 0.0, 1.0);
out.uv = uv;
out.top_left = top_left;
out.bot_right = bot_right;
out.mask_idx = in.mask_idx;
out.idx = ii;
return out;
}
fn masked(in: VertexOutput, color: vec4<f32>) -> vec4<f32> {
if in.mask_idx == 4294967295u {
return color;
}
let mask = masks[in.mask_idx];
// Its own chain, not the drawn primitive's, so a stationary viewport
// clips content that moves inside it.
let m = resolve_move(mask.move_idx, Region(mask.x, mask.y));
let tl = vec2(m.x.start.rel, m.y.start.rel);
let tl_abs = vec2(m.x.start.abs, m.y.start.abs);
let br = vec2(m.x.end.rel, m.y.end.rel);
let br_abs = vec2(m.x.end.abs, m.y.end.abs);
let top_left = floor(tl * window.dim) + floor(tl_abs);
let bot_right = floor(br * window.dim) + floor(br_abs);
let pos = in.clip_position.xy;
if pos.x < top_left.x || pos.x > bot_right.x || pos.y < top_left.y || pos.y > bot_right.y {
return color * 0.0;
}
return color;
}