Skip to content

A fast 2D and 3D game engine — TypeScript SDK, C++/WebAssembly core, visual editor. Ship one project to web, desktop, WeChat MiniGames, playable ads, and native Android / iOS.

feat(render): the per-object record leaves the vertex attributes

esengine93a3fec4f039
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 91d761966..500cf75ab 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -94,10 +94,31 @@ published separately; it ships inside the editor.
longer than the cap carries no irradiance by construction, so its tail reads
the zeroes and falls back to the environment, which is its answer anyway.
- Per-instance, not per-object-record, because the record is out of room: the 16
- vertex attribute slots are a fixed vocabulary — 0-7 the mesh's own channels
- (serialised in `.esmesh`), 8-15 the per-object transform — with nothing spare.
- Moving that record to a texture is what would free them, and is its own change.
+ Per-instance, not per-object-record, because the record was out of room at the
+ time. The entry below is the change that moved it.
+
+- **A mesh has somewhere to put its own vertex attributes.** All 16 vertex
+ attribute slots had a semantic owner — 0-7 the mesh's own channels (serialised
+ in `.esmesh`), 8-15 the engine's per-object record — so a mesh that wanted to
+ carry anything else had nowhere to carry it, and neither did the record when it
+ wanted to grow.
+
+ The record is a texture now, read by index rather than by whichever instance
+ the hardware happens to be on, and **slots 8-15 are free**. A draw says where
+ its own run of records begins (`gl_InstanceID` restarts every draw while the
+ frame's records are one run), and the whole layout collapses to one fixed
+ eight-texel shape: the stride the layout, the allocation and the packer each
+ used to derive for themselves is gone, and so is every variant of it.
+
+ Full floats, because what a record carries is a world-space translation and a
+ half loses whole units of one a thousand away — so `RGBA32F` is a format the
+ engine has now, on both backends. On WebGPU it is unfilterable, which the
+ bind-group layout has to say: a unit that declares no sampler cannot be
+ filtered, so that is the answer read out of the shader itself rather than a
+ second table to keep.
+
+ Every one of the 174 pixel scenes draws what it drew before, on both backends,
+ and 200 copies of a crate still cost 1-2 draw calls.
- **Automation can read what a frame cost.** A probe could ask what a game was
doing and not what it was paying: `Stats` is installed by StatsPlugin, which a
diff --git a/examples/effects-gallery/assets/materials/dissolve.esshader b/examples/effects-gallery/assets/materials/dissolve.esshader
index f1011a579..92d87fcfb 100644
--- a/examples/effects-gallery/assets/materials/dissolve.esshader
+++ b/examples/effects-gallery/assets/materials/dissolve.esshader
@@ -40,7 +40,7 @@ void main() {
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 0a687499da49fe50
-// assembled-hash: 60636c76d066e1af
+// assembled-hash: 3b7c3069a5e2ac75
struct gl_PerVertex {
@builtin(position) gl_Position: vec4<f32>,
gl_PointSize: f32,
@@ -64,6 +64,10 @@ struct MaterialConstants {
u_noiseScale: f32,
}
+struct InstanceConstants {
+ u_instanceBase: vec4<u32>,
+}
+
struct VertexOutput {
@location(0) member: vec4<f32>,
@builtin(position) gl_Position: vec4<f32>,
@@ -82,19 +86,25 @@ var<private> a_texCoord_1: vec2<f32>;
var<uniform> unnamed_2: TimeConstants;
@group(0) @binding(1)
var<uniform> unnamed_3: MaterialConstants;
+@group(0) @binding(4)
+var<uniform> unnamed_4: InstanceConstants;
+@group(1) @binding(23)
+var u_instanceData_est: texture_2d<f32>;
+@group(1) @binding(31)
+var u_instanceData_ess: sampler;
fn main_1() {
var world: vec4<f32>;
- let _e12 = a_position_1;
- world = vec4<f32>(_e12.x, _e12.y, _e12.z, 1f);
- let _e17 = a_color_1;
- v_color = _e17;
- let _e19 = unnamed_1.u_projection;
- let _e20 = world;
- unnamed.gl_Position = (_e19 * _e20);
- let _e23 = a_texCoord_1;
- v_texCoord = _e23;
+ let _e15 = a_position_1;
+ world = vec4<f32>(_e15.x, _e15.y, _e15.z, 1f);
+ let _e20 = a_color_1;
+ v_color = _e20;
+ let _e22 = unnamed_1.u_projection;
+ let _e23 = world;
+ unnamed.gl_Position = (_e22 * _e23);
+ let _e26 = a_texCoord_1;
+ v_texCoord = _e26;
return;
}
@@ -115,7 +125,7 @@ fn vs_main(@location(0) a_position: vec3<f32>, @location(1) a_color: vec4<f32>,
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 0a687499da49fe50
-// assembled-hash: 60636c76d066e1af
+// assembled-hash: 3b7c3069a5e2ac75
struct MaterialConstants {
u_progress: f32,
u_edgeColor: vec4<f32>,
diff --git a/examples/effects-gallery/assets/materials/hit-flash.esshader b/examples/effects-gallery/assets/materials/hit-flash.esshader
index 4d434aa56..439ff4d28 100644
--- a/examples/effects-gallery/assets/materials/hit-flash.esshader
+++ b/examples/effects-gallery/assets/materials/hit-flash.esshader
@@ -25,7 +25,7 @@ void main() {
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 674537b45fb865fb
-// assembled-hash: 72ca189990dafa80
+// assembled-hash: 51a5fa5bc7072ae0
struct gl_PerVertex {
@builtin(position) gl_Position: vec4<f32>,
gl_PointSize: f32,
@@ -47,6 +47,10 @@ struct MaterialConstants {
u_flashColor: vec4<f32>,
}
+struct InstanceConstants {
+ u_instanceBase: vec4<u32>,
+}
+
struct VertexOutput {
@location(0) member: vec4<f32>,
@builtin(position) gl_Position: vec4<f32>,
@@ -65,19 +69,25 @@ var<private> a_texCoord_1: vec2<f32>;
var<uniform> unnamed_2: TimeConstants;
@group(0) @binding(1)
var<uniform> unnamed_3: MaterialConstants;
+@group(0) @binding(4)
+var<uniform> unnamed_4: InstanceConstants;
+@group(1) @binding(23)
+var u_instanceData_est: texture_2d<f32>;
+@group(1) @binding(31)
+var u_instanceData_ess: sampler;
fn main_1() {
var world: vec4<f32>;
- let _e12 = a_position_1;
- world = vec4<f32>(_e12.x, _e12.y, _e12.z, 1f);
- let _e17 = a_color_1;
- v_color = _e17;
- let _e19 = unnamed_1.u_projection;
- let _e20 = world;
- unnamed.gl_Position = (_e19 * _e20);
- let _e23 = a_texCoord_1;
- v_texCoord = _e23;
+ let _e15 = a_position_1;
+ world = vec4<f32>(_e15.x, _e15.y, _e15.z, 1f);
+ let _e20 = a_color_1;
+ v_color = _e20;
+ let _e22 = unnamed_1.u_projection;
+ let _e23 = world;
+ unnamed.gl_Position = (_e22 * _e23);
+ let _e26 = a_texCoord_1;
+ v_texCoord = _e26;
return;
}
@@ -98,7 +108,7 @@ fn vs_main(@location(0) a_position: vec3<f32>, @location(1) a_color: vec4<f32>,
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 674537b45fb865fb
-// assembled-hash: 72ca189990dafa80
+// assembled-hash: 51a5fa5bc7072ae0
struct MaterialConstants {
u_flash: f32,
u_flashColor: vec4<f32>,
diff --git a/examples/effects-gallery/assets/materials/outline.esshader b/examples/effects-gallery/assets/materials/outline.esshader
index 8f5e71d1a..f3dd29fcd 100644
--- a/examples/effects-gallery/assets/materials/outline.esshader
+++ b/examples/effects-gallery/assets/materials/outline.esshader
@@ -34,7 +34,7 @@ void main() {
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 7ed4f4686c6bfd8d
-// assembled-hash: e924eb284d528261
+// assembled-hash: 659232dd4e620239
struct gl_PerVertex {
@builtin(position) gl_Position: vec4<f32>,
gl_PointSize: f32,
@@ -56,6 +56,10 @@ struct MaterialConstants {
u_outlineWidth: f32,
}
+struct InstanceConstants {
+ u_instanceBase: vec4<u32>,
+}
+
struct VertexOutput {
@location(0) member: vec4<f32>,
@builtin(position) gl_Position: vec4<f32>,
@@ -74,19 +78,25 @@ var<private> a_texCoord_1: vec2<f32>;
var<uniform> unnamed_2: TimeConstants;
@group(0) @binding(1)
var<uniform> unnamed_3: MaterialConstants;
+@group(0) @binding(4)
+var<uniform> unnamed_4: InstanceConstants;
+@group(1) @binding(23)
+var u_instanceData_est: texture_2d<f32>;
+@group(1) @binding(31)
+var u_instanceData_ess: sampler;
fn main_1() {
var world: vec4<f32>;
- let _e12 = a_position_1;
- world = vec4<f32>(_e12.x, _e12.y, _e12.z, 1f);
- let _e17 = a_color_1;
- v_color = _e17;
- let _e19 = unnamed_1.u_projection;
- let _e20 = world;
- unnamed.gl_Position = (_e19 * _e20);
- let _e23 = a_texCoord_1;
- v_texCoord = _e23;
+ let _e15 = a_position_1;
+ world = vec4<f32>(_e15.x, _e15.y, _e15.z, 1f);
+ let _e20 = a_color_1;
+ v_color = _e20;
+ let _e22 = unnamed_1.u_projection;
+ let _e23 = world;
+ unnamed.gl_Position = (_e22 * _e23);
+ let _e26 = a_texCoord_1;
+ v_texCoord = _e26;
return;
}
@@ -107,7 +117,7 @@ fn vs_main(@location(0) a_position: vec3<f32>, @location(1) a_color: vec4<f32>,
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 7ed4f4686c6bfd8d
-// assembled-hash: e924eb284d528261
+// assembled-hash: 659232dd4e620239
struct MaterialConstants {
u_outlineColor: vec4<f32>,
u_outlineWidth: f32,
diff --git a/examples/effects-gallery/assets/materials/pixelate.esshader b/examples/effects-gallery/assets/materials/pixelate.esshader
index 8a5021c14..1bee84001 100644
--- a/examples/effects-gallery/assets/materials/pixelate.esshader
+++ b/examples/effects-gallery/assets/materials/pixelate.esshader
@@ -23,7 +23,7 @@ void main() {
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 0c2fb4b31599ec7a
-// assembled-hash: 8d6d37a7023e822f
+// assembled-hash: aa5efc32d552a8db
struct gl_PerVertex {
@builtin(position) gl_Position: vec4<f32>,
gl_PointSize: f32,
@@ -44,6 +44,10 @@ struct MaterialConstants {
u_pixels: f32,
}
+struct InstanceConstants {
+ u_instanceBase: vec4<u32>,
+}
+
struct VertexOutput {
@location(0) member: vec4<f32>,
@builtin(position) gl_Position: vec4<f32>,
@@ -62,19 +66,25 @@ var<private> a_texCoord_1: vec2<f32>;
var<uniform> unnamed_2: TimeConstants;
@group(0) @binding(1)
var<uniform> unnamed_3: MaterialConstants;
+@group(0) @binding(4)
+var<uniform> unnamed_4: InstanceConstants;
+@group(1) @binding(23)
+var u_instanceData_est: texture_2d<f32>;
+@group(1) @binding(31)
+var u_instanceData_ess: sampler;
fn main_1() {
var world: vec4<f32>;
- let _e12 = a_position_1;
- world = vec4<f32>(_e12.x, _e12.y, _e12.z, 1f);
- let _e17 = a_color_1;
- v_color = _e17;
- let _e19 = unnamed_1.u_projection;
- let _e20 = world;
- unnamed.gl_Position = (_e19 * _e20);
- let _e23 = a_texCoord_1;
- v_texCoord = _e23;
+ let _e15 = a_position_1;
+ world = vec4<f32>(_e15.x, _e15.y, _e15.z, 1f);
+ let _e20 = a_color_1;
+ v_color = _e20;
+ let _e22 = unnamed_1.u_projection;
+ let _e23 = world;
+ unnamed.gl_Position = (_e22 * _e23);
+ let _e26 = a_texCoord_1;
+ v_texCoord = _e26;
return;
}
@@ -95,7 +105,7 @@ fn vs_main(@location(0) a_position: vec3<f32>, @location(1) a_color: vec4<f32>,
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 0c2fb4b31599ec7a
-// assembled-hash: 8d6d37a7023e822f
+// assembled-hash: aa5efc32d552a8db
struct MaterialConstants {
u_pixels: f32,
}
diff --git a/examples/effects-gallery/assets/materials/uv-scroll.esshader b/examples/effects-gallery/assets/materials/uv-scroll.esshader
index 1789680f9..aabefe441 100644
--- a/examples/effects-gallery/assets/materials/uv-scroll.esshader
+++ b/examples/effects-gallery/assets/materials/uv-scroll.esshader
@@ -24,7 +24,7 @@ void main() {
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 301c5b2b73d966f3
-// assembled-hash: d2039f3013a5d165
+// assembled-hash: 31e405ee96a6823e
struct gl_PerVertex {
@builtin(position) gl_Position: vec4<f32>,
gl_PointSize: f32,
@@ -45,6 +45,10 @@ struct MaterialConstants {
u_scrollSpeed: vec2<f32>,
}
+struct InstanceConstants {
+ u_instanceBase: vec4<u32>,
+}
+
struct VertexOutput {
@location(0) member: vec4<f32>,
@builtin(position) gl_Position: vec4<f32>,
@@ -63,19 +67,25 @@ var<private> a_texCoord_1: vec2<f32>;
var<uniform> unnamed_2: TimeConstants;
@group(0) @binding(1)
var<uniform> unnamed_3: MaterialConstants;
+@group(0) @binding(4)
+var<uniform> unnamed_4: InstanceConstants;
+@group(1) @binding(23)
+var u_instanceData_est: texture_2d<f32>;
+@group(1) @binding(31)
+var u_instanceData_ess: sampler;
fn main_1() {
var world: vec4<f32>;
- let _e12 = a_position_1;
- world = vec4<f32>(_e12.x, _e12.y, _e12.z, 1f);
- let _e17 = a_color_1;
- v_color = _e17;
- let _e19 = unnamed_1.u_projection;
- let _e20 = world;
- unnamed.gl_Position = (_e19 * _e20);
- let _e23 = a_texCoord_1;
- v_texCoord = _e23;
+ let _e15 = a_position_1;
+ world = vec4<f32>(_e15.x, _e15.y, _e15.z, 1f);
+ let _e20 = a_color_1;
+ v_color = _e20;
+ let _e22 = unnamed_1.u_projection;
+ let _e23 = world;
+ unnamed.gl_Position = (_e22 * _e23);
+ let _e26 = a_texCoord_1;
+ v_texCoord = _e26;
return;
}
@@ -96,7 +106,7 @@ fn vs_main(@location(0) a_position: vec3<f32>, @location(1) a_color: vec4<f32>,
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 301c5b2b73d966f3
-// assembled-hash: d2039f3013a5d165
+// assembled-hash: 31e405ee96a6823e
struct TimeConstants {
u_time: vec4<f32>,
u_viewport: vec4<f32>,
diff --git a/fixtures/scenes/depth-layers/opaque.esshader b/fixtures/scenes/depth-layers/opaque.esshader
index df8e73d64..08673f0ad 100644
--- a/fixtures/scenes/depth-layers/opaque.esshader
+++ b/fixtures/scenes/depth-layers/opaque.esshader
@@ -24,7 +24,7 @@ void main() {
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 930787555114ce9c
-// assembled-hash: 4fbd4cec4b7a3eab
+// assembled-hash: cab87529431e525a
struct gl_PerVertex {
@builtin(position) gl_Position: vec4<f32>,
gl_PointSize: f32,
@@ -41,6 +41,10 @@ struct TimeConstants {
u_viewport: vec4<f32>,
}
+struct InstanceConstants {
+ u_instanceBase: vec4<u32>,
+}
+
struct VertexOutput {
@location(0) member: vec4<f32>,
@builtin(position) gl_Position: vec4<f32>,
@@ -57,19 +61,25 @@ var<private> v_texCoord: vec2<f32>;
var<private> a_texCoord_1: vec2<f32>;
@group(0) @binding(3)
var<uniform> unnamed_2: TimeConstants;
+@group(0) @binding(4)
+var<uniform> unnamed_3: InstanceConstants;
+@group(1) @binding(23)
+var u_instanceData_est: texture_2d<f32>;
+@group(1) @binding(31)
+var u_instanceData_ess: sampler;
fn main_1() {
var world: vec4<f32>;
- let _e11 = a_position_1;
- world = vec4<f32>(_e11.x, _e11.y, _e11.z, 1f);
- let _e16 = a_color_1;
- v_color = _e16;
- let _e18 = unnamed_1.u_projection;
- let _e19 = world;
- unnamed.gl_Position = (_e18 * _e19);
- let _e22 = a_texCoord_1;
- v_texCoord = _e22;
+ let _e14 = a_position_1;
+ world = vec4<f32>(_e14.x, _e14.y, _e14.z, 1f);
+ let _e19 = a_color_1;
+ v_color = _e19;
+ let _e21 = unnamed_1.u_projection;
+ let _e22 = world;
+ unnamed.gl_Position = (_e21 * _e22);
+ let _e25 = a_texCoord_1;
+ v_texCoord = _e25;
return;
}
@@ -90,7 +100,7 @@ fn vs_main(@location(0) a_position: vec3<f32>, @location(1) a_color: vec4<f32>,
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 930787555114ce9c
-// assembled-hash: 4fbd4cec4b7a3eab
+// assembled-hash: cab87529431e525a
struct FrameConstants {
u_projection: mat4x4<f32>,
u_camera: vec4<f32>,
diff --git a/fixtures/scenes/mat-builtin-lit/lit.esshader b/fixtures/scenes/mat-builtin-lit/lit.esshader
index 4d467d302..91f7ef0cd 100644
--- a/fixtures/scenes/mat-builtin-lit/lit.esshader
+++ b/fixtures/scenes/mat-builtin-lit/lit.esshader
@@ -27,7 +27,7 @@ void main() {
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 5330079245724fa2
-// assembled-hash: 5d2188654785ea29
+// assembled-hash: cc964016cf869c0b
struct gl_PerVertex {
@builtin(position) gl_Position: vec4<f32>,
gl_PointSize: f32,
@@ -48,6 +48,10 @@ struct MaterialConstants {
u_tint: vec4<f32>,
}
+struct InstanceConstants {
+ u_instanceBase: vec4<u32>,
+}
+
struct VertexOutput {
@location(1) member: vec4<f32>,
@builtin(position) gl_Position: vec4<f32>,
@@ -75,23 +79,29 @@ var<uniform> unnamed_3: MaterialConstants;
var u_normalMap_est: texture_2d<f32>;
@group(1) @binding(24)
var u_normalMap_ess: sampler;
+@group(0) @binding(4)
+var<uniform> unnamed_4: InstanceConstants;
+@group(1) @binding(23)
+var u_instanceData_est: texture_2d<f32>;
+@group(1) @binding(31)
+var u_instanceData_ess: sampler;
fn main_1() {
var world: vec4<f32>;
- let _e17 = a_position_1;
- world = vec4<f32>(_e17.x, _e17.y, _e17.z, 1f);
- let _e22 = a_color_1;
- v_color = _e22;
- let _e24 = unnamed_1.u_projection;
- let _e25 = world;
- unnamed.gl_Position = (_e24 * _e25);
- let _e28 = a_texCoord_1;
- v_texCoord = _e28;
- let _e29 = world;
- v_worldPos = _e29.xy;
- let _e31 = gl_InstanceIndex_1;
- v_probeSlot = f32(_e31);
+ let _e20 = a_position_1;
+ world = vec4<f32>(_e20.x, _e20.y, _e20.z, 1f);
+ let _e25 = a_color_1;
+ v_color = _e25;
+ let _e27 = unnamed_1.u_projection;
+ let _e28 = world;
+ unnamed.gl_Position = (_e27 * _e28);
+ let _e31 = a_texCoord_1;
+ v_texCoord = _e31;
+ let _e32 = world;
+ v_worldPos = _e32.xy;
+ let _e34 = gl_InstanceIndex_1;
+ v_probeSlot = f32(_e34);
return;
}
@@ -115,7 +125,7 @@ fn vs_main(@location(0) a_position: vec3<f32>, @location(1) a_color: vec4<f32>,
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 5330079245724fa2
-// assembled-hash: 5d2188654785ea29
+// assembled-hash: cc964016cf869c0b
struct FrameConstants {
u_projection: mat4x4<f32>,
u_camera: vec4<f32>,
diff --git a/fixtures/scenes/mat-dissolve/dissolve.esshader b/fixtures/scenes/mat-dissolve/dissolve.esshader
index f1011a579..92d87fcfb 100644
--- a/fixtures/scenes/mat-dissolve/dissolve.esshader
+++ b/fixtures/scenes/mat-dissolve/dissolve.esshader
@@ -40,7 +40,7 @@ void main() {
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 0a687499da49fe50
-// assembled-hash: 60636c76d066e1af
+// assembled-hash: 3b7c3069a5e2ac75
struct gl_PerVertex {
@builtin(position) gl_Position: vec4<f32>,
gl_PointSize: f32,
@@ -64,6 +64,10 @@ struct MaterialConstants {
u_noiseScale: f32,
}
+struct InstanceConstants {
+ u_instanceBase: vec4<u32>,
+}
+
struct VertexOutput {
@location(0) member: vec4<f32>,
@builtin(position) gl_Position: vec4<f32>,
@@ -82,19 +86,25 @@ var<private> a_texCoord_1: vec2<f32>;
var<uniform> unnamed_2: TimeConstants;
@group(0) @binding(1)
var<uniform> unnamed_3: MaterialConstants;
+@group(0) @binding(4)
+var<uniform> unnamed_4: InstanceConstants;
+@group(1) @binding(23)
+var u_instanceData_est: texture_2d<f32>;
+@group(1) @binding(31)
+var u_instanceData_ess: sampler;
fn main_1() {
var world: vec4<f32>;
- let _e12 = a_position_1;
- world = vec4<f32>(_e12.x, _e12.y, _e12.z, 1f);
- let _e17 = a_color_1;
- v_color = _e17;
- let _e19 = unnamed_1.u_projection;
- let _e20 = world;
- unnamed.gl_Position = (_e19 * _e20);
- let _e23 = a_texCoord_1;
- v_texCoord = _e23;
+ let _e15 = a_position_1;
+ world = vec4<f32>(_e15.x, _e15.y, _e15.z, 1f);
+ let _e20 = a_color_1;
+ v_color = _e20;
+ let _e22 = unnamed_1.u_projection;
+ let _e23 = world;
+ unnamed.gl_Position = (_e22 * _e23);
+ let _e26 = a_texCoord_1;
+ v_texCoord = _e26;
return;
}
@@ -115,7 +125,7 @@ fn vs_main(@location(0) a_position: vec3<f32>, @location(1) a_color: vec4<f32>,
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 0a687499da49fe50
-// assembled-hash: 60636c76d066e1af
+// assembled-hash: 3b7c3069a5e2ac75
struct MaterialConstants {
u_progress: f32,
u_edgeColor: vec4<f32>,
diff --git a/fixtures/scenes/mat-graph/graph-tint.esshader b/fixtures/scenes/mat-graph/graph-tint.esshader
index 8ab803576..8e2868dae 100644
--- a/fixtures/scenes/mat-graph/graph-tint.esshader
+++ b/fixtures/scenes/mat-graph/graph-tint.esshader
@@ -39,7 +39,7 @@ void main() {
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 5382212ba4021113
-// assembled-hash: 948218796d1bc56c
+// assembled-hash: f490343500205763
struct gl_PerVertex {
@builtin(position) gl_Position: vec4<f32>,
gl_PointSize: f32,
@@ -113,7 +113,7 @@ fn vs_main(@location(0) a_position: vec3<f32>, @location(1) a_color: vec4<f32>,
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 5382212ba4021113
-// assembled-hash: 948218796d1bc56c
+// assembled-hash: f490343500205763
struct MaterialConstants {
u_tint: vec4<f32>,
}
diff --git a/fixtures/scenes/mat-sw/switched.esshader b/fixtures/scenes/mat-sw/switched.esshader
index 315603270..009d82cd5 100644
--- a/fixtures/scenes/mat-sw/switched.esshader
+++ b/fixtures/scenes/mat-sw/switched.esshader
@@ -38,7 +38,7 @@ void main() {
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 0ba3d3644597ce83
-// assembled-hash: eb67150bbfaf17cb
+// assembled-hash: da96256e5deec553
#ifdef USE_GREEN
struct gl_PerVertex {
@builtin(position) gl_Position: vec4<f32>,
@@ -152,7 +152,7 @@ fn vs_main(@location(0) a_position: vec3<f32>, @location(1) a_color: vec4<f32>,
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 0ba3d3644597ce83
-// assembled-hash: eb67150bbfaf17cb
+// assembled-hash: da96256e5deec553
#ifdef USE_GREEN
struct FrameConstants {
u_projection: mat4x4<f32>,
diff --git a/fixtures/scenes/mat-tex/masked.esshader b/fixtures/scenes/mat-tex/masked.esshader
index 1adff28c1..282219971 100644
--- a/fixtures/scenes/mat-tex/masked.esshader
+++ b/fixtures/scenes/mat-tex/masked.esshader
@@ -38,7 +38,7 @@ void main() {
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 11b26db499fb2fc6
-// assembled-hash: 995c87fac03d97a3
+// assembled-hash: d081d31e97177956
struct gl_PerVertex {
@builtin(position) gl_Position: vec4<f32>,
gl_PointSize: f32,
@@ -106,7 +106,7 @@ fn vs_main(@location(0) a_position: vec3<f32>, @location(1) a_color: vec4<f32>,
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 11b26db499fb2fc6
-// assembled-hash: 995c87fac03d97a3
+// assembled-hash: d081d31e97177956
struct FrameConstants {
u_projection: mat4x4<f32>,
u_camera: vec4<f32>,
diff --git a/fixtures/scenes/mat-tint/tint.esshader b/fixtures/scenes/mat-tint/tint.esshader
index 692585474..5edbda82f 100644
--- a/fixtures/scenes/mat-tint/tint.esshader
+++ b/fixtures/scenes/mat-tint/tint.esshader
@@ -78,7 +78,7 @@ struct VSOut {
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 7be4a77f9921c078
-// assembled-hash: 153358b0bdd2ce6a
+// assembled-hash: 2907a720efa553dc
struct gl_PerVertex {
@builtin(position) gl_Position: vec4<f32>,
gl_PointSize: f32,
@@ -148,7 +148,7 @@ fn vs_main(@location(0) a_position: vec3<f32>, @location(1) a_color: vec4<f32>,
// GENERATED by tools/gen-shader-twins.mjs (glslang + naga) - do not edit.
// Regenerate after changing the GLSL stages above.
// source-hash: 7be4a77f9921c078
-// assembled-hash: 153358b0bdd2ce6a
+// assembled-hash: 2907a720efa553dc
struct MaterialConstants {
u_tint: vec4<f32>,
}
diff --git a/src/esengine/bindings/WebSDKEntry.cpp b/src/esengine/bindings/WebSDKEntry.cpp
index 1289541cb..2617e9f8f 100644
--- a/src/esengine/bindings/WebSDKEntry.cpp
+++ b/src/esengine/bindings/WebSDKEntry.cpp
@@ -125,7 +125,7 @@ emscripten::val esshader_cookInfo(const std::string& source, const std::string&
out.set("hasWgslFragment", parsed.wgslStages.count(resource::ShaderStage::Fragment) != 0);
out.set("hasSwitches", !parsed.switches.empty());
// The switch/feature NAMES, so a cook can enumerate the permutations it has
- // to emit a twin for — WGSL resolves them at assembly time (preprocessWGSL),
+ // to emit a twin for — WGSL resolves them at assembly time (preprocessConditionals),
// which needs every branch present in the body.
emscripten::val switches = emscripten::val::array();
u32 switchCount = 0;
diff --git a/src/esengine/data/shaders/mesh.esshader b/src/esengine/data/shaders/mesh.esshader
index 91cdcdd01..ab0c38fc4 100644
--- a/src/esengine/data/shaders/mesh.esshader
+++ b/src/esengine/data/shaders/mesh.esshader
@@ -59,31 +59,10 @@ layout(location = 6) in vec4 a_weights;
layout(location = 7) in vec2 a_texCoord1;
#endif
-// 8.. and not 3..: a channel's semantic IS its location, so adding normals to a
-// mesh must not move the per-object record (see MESH_INSTANCE_FIRST_LOCATION).
-#ifndef SKINNED
-// Three ROWS of an affine 4x4. The fourth is (0,0,0,1) on every transform a
-// Transform can state, so carrying it spends an attribute slot and 16 bytes an
-// instance on a constant.
-layout(location = 8) in vec4 a_model0;
-layout(location = 9) in vec4 a_model1;
-layout(location = 10) in vec4 a_model2;
-#ifdef MESH_LIGHTMAP
-// Appended to the record, so the tint and the normal matrix keep the offsets
-// they had. Never on a skinned draw: bones move a character, and a character is
-// what a bake cannot hold still long enough to light.
-layout(location = 11) in vec4 a_lightmapRect;
-#endif
-#endif
-layout(location = 12) in vec4 a_instTint;
-#if defined(MESH_NORMALS) && !defined(SKINNED)
-// The normal matrix (inverse transpose of the model's 3x3), per object: under a
-// non-uniform scale the model matrix skews a normal, and inverting one per
-// vertex is the alternative.
-layout(location = 13) in vec3 a_nrm0;
-layout(location = 14) in vec3 a_nrm1;
-layout(location = 15) in vec3 a_nrm2;
-#endif
+// The per-object record comes from the frame's texture, not from attributes:
+// esInstanceModel / esInstanceTint / esInstanceNormalMatrix /
+// esInstanceLightmapRect are injected by the assembler for any vertex stage that
+// names one, which is what leaves every slot above the mesh channels free.
#ifdef SKINNED
// This draw's pose, rewritten immediately before it (SkinConstants, binding 5).
@@ -169,18 +148,16 @@ void main() {
+ a_weights.w * u_bones[a_joints.w];
vec4 world = skin * vec4(local, 1.0);
#else
- mat4 model = mat4(vec4(a_model0.x, a_model1.x, a_model2.x, 0.0),
- vec4(a_model0.y, a_model1.y, a_model2.y, 0.0),
- vec4(a_model0.z, a_model1.z, a_model2.z, 0.0),
- vec4(a_model0.w, a_model1.w, a_model2.w, 1.0));
+ mat4 model = esInstanceModel();
vec4 world = model * vec4(local, 1.0);
#endif
gl_Position = u_projection * world;
v_texCoord = a_texCoord;
- v_color = a_color * a_instTint;
+ v_color = a_color * esInstanceTint();
#ifdef MESH_LIGHTMAP
- v_lightmap = vec3(a_texCoord1 * a_lightmapRect.xy + a_lightmapRect.zw,
- a_lightmapRect.x > 0.0 ? 1.0 : 0.0);
+ highp vec4 lmRect = esInstanceLightmapRect();
+ v_lightmap = vec3(a_texCoord1 * lmRect.xy + lmRect.zw,
+ lmRect.x > 0.0 ? 1.0 : 0.0);
#endif
#ifdef SHADOW_DEPTH
v_shadowClip = gl_Position;
@@ -192,7 +169,7 @@ void main() {
#if defined(MESH_NORMALS) && defined(SKINNED)
v_worldNormal = mat3(skin) * localNormal;
#elif defined(MESH_NORMALS)
- v_worldNormal = mat3(a_nrm0, a_nrm1, a_nrm2) * localNormal;
+ v_worldNormal = esInstanceNormalMatrix() * localNormal;
#else
// A surface with no normal channel faces the viewer, exactly as a sprite does.
v_worldNormal = vec3(0.0, 0.0, 1.0);
@@ -301,27 +278,12 @@ struct VSIn {
#ifdef SKINNED
@location(5) a_joints : vec4u,
@location(6) a_weights : vec4f,
-#else
- @location(8) a_model0 : vec4f,
- @location(9) a_model1 : vec4f,
- @location(10) a_model2 : vec4f,
-#ifdef MESH_LIGHTMAP
- @location(11) a_lightmapRect : vec4f,
-#endif
#endif
#ifdef MESH_LIGHTMAP
@location(7) a_texCoord1 : vec2f,
#endif
- @location(12) a_instTint : vec4f,
// What the GLSL twin reads as gl_VertexID: a morph delta is addressed by it.
@builtin(vertex_index) vertexIndex : u32,
-#ifdef MESH_NORMALS
-#ifndef SKINNED
- @location(13) a_nrm0 : vec3f,
- @location(14) a_nrm1 : vec3f,
- @location(15) a_nrm2 : vec3f,
-#endif
-#endif
};
struct VSOut {
@builtin(position) pos : vec4f,
@@ -344,6 +306,7 @@ struct VSOut {
};
@vertex fn vs_main(v : VSIn, @builtin(instance_index) inst : u32) -> VSOut {
+ g_instanceId = i32(inst);
var local = v.a_position;
#ifdef MESH_NORMALS
var localNormal = v.a_normal;
@@ -370,20 +333,17 @@ struct VSOut {
+ v.a_weights.w * skin.bones[v.a_joints.w];
let world = pose * vec4f(local, 1.0);
#else
- let model = mat4x4f(vec4f(v.a_model0.x, v.a_model1.x, v.a_model2.x, 0.0),
- vec4f(v.a_model0.y, v.a_model1.y, v.a_model2.y, 0.0),
- vec4f(v.a_model0.z, v.a_model1.z, v.a_model2.z, 0.0),
- vec4f(v.a_model0.w, v.a_model1.w, v.a_model2.w, 1.0));
- let world = model * vec4f(local, 1.0);
+ let world = esInstanceModel() * vec4f(local, 1.0);
#endif
var out : VSOut;
out.pos = frame.projection * world;
out.v_texCoord = v.a_texCoord;
- out.v_color = v.a_color * v.a_instTint;
+ out.v_color = v.a_color * esInstanceTint();
#ifdef MESH_LIGHTMAP
- out.v_lightmap = vec3f(v.a_texCoord1 * v.a_lightmapRect.xy + v.a_lightmapRect.zw,
- select(0.0, 1.0, v.a_lightmapRect.x > 0.0));
+ let lmRect = esInstanceLightmapRect();
+ out.v_lightmap = vec3f(v.a_texCoord1 * lmRect.xy + lmRect.zw,
+ select(0.0, 1.0, lmRect.x > 0.0));
#endif
#ifdef SHADOW_DEPTH
out.v_shadowClip = out.pos;
@@ -395,7 +355,7 @@ struct VSOut {
#ifdef SKINNED
out.v_worldNormal = mat3x3f(pose[0].xyz, pose[1].xyz, pose[2].xyz) * localNormal;
#else
- out.v_worldNormal = mat3x3f(v.a_nrm0, v.a_nrm1, v.a_nrm2) * localNormal;
+ out.v_worldNormal = esInstanceNormalMatrix() * localNormal;
#endif
#else
// A surface with no normal channel faces the viewer, exactly as a sprite does.
diff --git a/src/esengine/renderer/draw/DrawList.cpp b/src/esengine/renderer/draw/DrawList.cpp
index 3281160cc..9f9ba96b4 100644
--- a/src/esengine/renderer/draw/DrawList.cpp
+++ b/src/esengine/renderer/draw/DrawList.cpp
@@ -3,6 +3,7 @@
#include "./DrawList.hpp"
#include "./BatchVertex.hpp"
#include "../store/MaterialStore.hpp"
+#include "../store/InstanceConstants.hpp"
#include "../store/SkinConstants.hpp"
#include "../../core/FrameProfiler.hpp"
@@ -195,11 +196,15 @@ void DrawList::finalize(TransientBufferPool& pool) {
void DrawList::execute(GfxDevice& device, TransientBufferPool& buffers,
MaterialStore& materials, u32 white_texture_id,
- FrameCapture* capture, PerDrawBlocks* skin_blocks,
- PerDrawBlocks* morph_blocks, PerDrawBlocks* probe_blocks) {
+ FrameCapture* capture, const PerDrawBlockSet& blocks) {
PipelineDesc lastDesc{};
PipelineHandle lastHandle = PipelineHandle::Invalid;
+ // Once for the pass, not once per draw: every resident mesh reads the same
+ // texture. Rebound each pass because growing it mints a new handle.
+ if (buffers.instanceTexture() != TextureHandle::Invalid) {
+ device.bindTexture(MESH_INSTANCE_TEXTURE_UNIT, buffers.instanceTexture());
+ }
for (u32 i = 0; i < merged_draw_calls_; ++i) {
const auto& cmd = commands_[i];
@@ -243,28 +248,28 @@ void DrawList::execute(GfxDevice& device, TransientBufferPool& buffers,
// The pose, for a draw that has one — in a buffer of this draw's own. A
// block shared between two draws of one pass is read by both with what
// the last wrote, which is a room of characters in one pose.
- if (cmd.skin_count > 0 && skin_blocks) {
+ if (cmd.skin_count > 0 && blocks.skin) {
device.setUniformBuffer(
SKIN_CONSTANTS_BINDING,
- skin_blocks->write(skin_matrices_.data() + cmd.skin_offset,
+ blocks.skin->write(skin_matrices_.data() + cmd.skin_offset,
cmd.skin_count * static_cast<u32>(sizeof(glm::mat4))));
}
// The shapes, the same way — and a draw with none binds the zeroed block
// rather than erasing the last one's, so "unshaped" is an object instead
// of an erasure someone has to remember to perform.
- if (morph_blocks) {
+ if (blocks.morph) {
device.setUniformBuffer(MORPH_CONSTANTS_BINDING,
cmd.morph_index > 0
- ? morph_blocks->write(&morph_shapes_[cmd.morph_index - 1],
+ ? blocks.morph->write(&morph_shapes_[cmd.morph_index - 1],
sizeof(MorphConstants))
- : morph_blocks->zero());
+ : blocks.morph->zero());
}
// WHERE this draw's instances read their irradiance, not what it is: the
// coefficients are in the frame's texture, one run per instance, so a
// merged draw holding objects in many places still says one thing here.
- if (probe_blocks) {
+ if (blocks.probe) {
// Every instance of this run, in the order it draws them, capped at
// what the block holds — a longer run carries no irradiance at all,
// so its tail reads zeroes and takes the environment.
@@ -280,7 +285,7 @@ void DrawList::execute(GfxDevice& device, TransientBufferPool& buffers,
}
device.setUniformBuffer(
PROBE_CONSTANTS_BINDING,
- probe_blocks->write(probe_run_.data(),
+ blocks.probe->write(probe_run_.data(),
static_cast<u32>(probe_run_.size() * sizeof(glm::vec4))));
}
@@ -312,12 +317,15 @@ void DrawList::execute(GfxDevice& device, TransientBufferPool& buffers,
}
if (cmd.hasPersistentGeometry()) {
- // The mesh's buffers for geometry, the frame's pool for transforms.
- // Only the second was written this frame: an unchanged mesh costs no
- // upload, and drawing it twice costs one more transform.
+ // gl_InstanceID restarts at zero every draw while the frame's
+ // records are one continuous run, so a draw has to say where its own
+ // run begins in them.
+ if (blocks.instance) {
+ const InstanceConstants at{cmd.vertex_byte_offset / MESH_INSTANCE_STRIDE, {}};
+ device.setUniformBuffer(INSTANCE_CONSTANTS_BINDING,
+ blocks.instance->write(&at, sizeof(at)));
+ }
device.setVertexBuffer(0, cmd.vertex_buffer, 0);
- device.setVertexBuffer(1, buffers.vertexBuffer(LayoutId::MeshInstance),
- cmd.vertex_byte_offset);
device.setIndexBuffer(cmd.index_buffer);
device.drawElementsInstanced(
cmd.index_count, GfxDataType::UnsignedInt,
diff --git a/src/esengine/renderer/draw/DrawList.hpp b/src/esengine/renderer/draw/DrawList.hpp
index d17964ed6..5b7861c38 100644
--- a/src/esengine/renderer/draw/DrawList.hpp
+++ b/src/esengine/renderer/draw/DrawList.hpp
@@ -52,9 +52,7 @@ public:
void execute(GfxDevice& device, TransientBufferPool& buffers,
MaterialStore& materials, u32 white_texture_id = 0,
FrameCapture* capture = nullptr,
- PerDrawBlocks* skin_blocks = nullptr,
- PerDrawBlocks* morph_blocks = nullptr,
- PerDrawBlocks* probe_blocks = nullptr);
+ const PerDrawBlockSet& blocks = {});
/** Append one draw's bone matrices; returns where they start. Frame-scoped,
* like the vertex pool: a command references an offset, not a copy. */
diff --git a/src/esengine/renderer/frame/RenderContext.cpp b/src/esengine/renderer/frame/RenderContext.cpp
index 1796ae50e..95c4924d2 100644
--- a/src/esengine/renderer/frame/RenderContext.cpp
+++ b/src/esengine/renderer/frame/RenderContext.cpp
@@ -17,6 +17,7 @@
#include "../draw/DrawParams.hpp"
#include "../store/SkinConstants.hpp"
#include "../store/MorphConstants.hpp"
+#include "../store/InstanceConstants.hpp"
#include "../store/ProbeConstants.hpp"
#include "../../core/Log.hpp"
@@ -155,10 +156,12 @@ void RenderContext::initFrameUbo() {
// says about every object in it, and GL wants the whole declared block bound.
probeBlocks_.init(device_,
static_cast<u32>(sizeof(ProbeConstants)) * PROBE_MAX_INSTANCES);
+ instanceBlocks_.init(device_, static_cast<u32>(sizeof(InstanceConstants)));
// What a draw with no shapes and no volume reads. A pose has no zero: geometry
// without one is drawn by a shader that does not declare the block.
device_.setUniformBuffer(MORPH_CONSTANTS_BINDING, morphBlocks_.zero());
device_.setUniformBuffer(PROBE_CONSTANTS_BINDING, probeBlocks_.zero());
+ device_.setUniformBuffer(INSTANCE_CONSTANTS_BINDING, instanceBlocks_.zero());
ES_LOG_DEBUG("FrameConstants UBO created (handle: {})", static_cast<u32>(frameUbo_));
}
diff --git a/src/esengine/renderer/frame/RenderContext.hpp b/src/esengine/renderer/frame/RenderContext.hpp
index e5ed1f60f..f2f689ed6 100644
--- a/src/esengine/renderer/frame/RenderContext.hpp
+++ b/src/esengine/renderer/frame/RenderContext.hpp
@@ -216,20 +216,26 @@ private:
BufferHandle frameUbo_ = BufferHandle::Invalid;
BufferHandle timeUbo_ = BufferHandle::Invalid;
BufferHandle drawParamsFallback_ = BufferHandle::Invalid;
- /// A bone-matrix block per skinned draw, a shape-weight block per morphed one,
- /// and an indirect-light block per draw standing in a volume. One buffer EACH
- /// and not one shared: see PerDrawBlocks.
+ /// A block each per skinned draw, per morphed one, per draw standing in a
+ /// volume, and per draw reading the record texture. One buffer EACH and not
+ /// one shared: see PerDrawBlocks.
PerDrawBlocks skinBlocks_;
PerDrawBlocks morphBlocks_;
PerDrawBlocks probeBlocks_;
+ PerDrawBlocks instanceBlocks_;
public:
- /** @brief The pool a skinned draw takes its pose buffer from. */
- PerDrawBlocks& skinBlocks() { return skinBlocks_; }
- /** @brief The pool a morphed draw takes its shape buffer from. */
- PerDrawBlocks& morphBlocks() { return morphBlocks_; }
- /** @brief The pool a draw inside a probe volume takes its light buffer from. */
- PerDrawBlocks& probeBlocks() { return probeBlocks_; }
+ /** @brief The blocks the draw loop rewrites before each draw. */
+ PerDrawBlockSet perDrawBlocks() {
+ return {&skinBlocks_, &morphBlocks_, &probeBlocks_, &instanceBlocks_};
+ }
+ /** @brief Hand every one of them back at the top of a frame. */
+ void beginPerDrawBlocks() {
+ skinBlocks_.beginFrame();
+ morphBlocks_.beginFrame();
+ probeBlocks_.beginFrame();
+ instanceBlocks_.beginFrame();
+ }
private:
f32 lastElapsed_ = 0.0f;
diff --git a/src/esengine/renderer/frame/RenderFrame.cpp b/src/esengine/renderer/frame/RenderFrame.cpp
index 5abc90994..f11a8a8ac 100644
--- a/src/esengine/renderer/frame/RenderFrame.cpp
+++ b/src/esengine/renderer/frame/RenderFrame.cpp
@@ -288,8 +288,7 @@ void RenderFrame::drawScene() {
{
ES_PROFILE_SCOPE("render.submit");
draw_list_.execute(device_, pool_, context_.materials(), context_.getWhiteTextureId(),
- &frame_capture_, &context_.skinBlocks(), &context_.morphBlocks(),
- &context_.probeBlocks());
+ &frame_capture_, context_.perDrawBlocks());
}
// Handed over to whatever the graph runs next. Blend/depth/colour-mask come
@@ -333,9 +332,7 @@ void RenderFrame::beginFrame() {
// The per-draw blocks go back at the FRAME boundary and not at each pass's:
// a frame reaches the device as several passes with one submission behind
// them, so a buffer reused between two of them is read by both.
- context_.skinBlocks().beginFrame();
- context_.morphBlocks().beginFrame();
- context_.probeBlocks().beginFrame();
+ context_.beginPerDrawBlocks();
}
void RenderFrame::applySceneDepthNeed() {
@@ -685,8 +682,7 @@ void RenderFrame::replayToDrawCall(i32 stopAtDrawCall) {
context_.updateCameraConstants(view_projection_);
context_.lights().uploadAndBind();
draw_list_.execute(device_, pool_, context_.materials(), context_.getWhiteTextureId(),
- &frame_capture_, &context_.skinBlocks(), &context_.morphBlocks(),
- &context_.probeBlocks());
+ &frame_capture_, context_.perDrawBlocks());
// Leave scissor disabled for whatever renders next; invalidate so the next
// setPipeline re-applies its full state (stencil included).
@@ -759,8 +755,7 @@ void RenderFrame::renderSurface(ecs::Registry& registry, const glm::mat4& viewPr
context_.updateCameraConstants(viewProjection);
context_.lights().uploadAndBind();
draw_list_.execute(device_, pool_, context_.materials(), context_.getWhiteTextureId(),
- &frame_capture_, &context_.skinBlocks(), &context_.morphBlocks(),
- &context_.probeBlocks());
+ &frame_capture_, context_.perDrawBlocks());
frame_capture_.endCapture();
rt->unbind();
@@ -1748,8 +1743,7 @@ void RenderFrame::executeShadowPass(ecs::Registry& registry) {
// nothing — the whole pass came back empty for exactly this.
context_.lights().uploadAndBind();
list.execute(device_, shadow_pool_, context_.materials(),
- context_.getWhiteTextureId(), nullptr, &context_.skinBlocks(),
- &context_.morphBlocks(), &context_.probeBlocks());
+ context_.getWhiteTextureId(), nullptr, context_.perDrawBlocks());
++shadowTiles;
shadowDraws += list.mergedDrawCallCount();
}
diff --git a/src/esengine/renderer/frame/RenderFrameOverlay.cpp b/src/esengine/renderer/frame/RenderFrameOverlay.cpp
index 2ec913946..3b1e8dd7d 100644
--- a/src/esengine/renderer/frame/RenderFrameOverlay.cpp
+++ b/src/esengine/renderer/frame/RenderFrameOverlay.cpp
@@ -80,8 +80,7 @@ void RenderFrame::drawScreenOverlay() {
{
ES_PROFILE_SCOPE("render.overlay.submit");
draw_list_.execute(device_, pool_, context_.materials(), context_.getWhiteTextureId(),
- &frame_capture_, &context_.skinBlocks(), &context_.morphBlocks(),
- &context_.probeBlocks());
+ &frame_capture_, context_.perDrawBlocks());
}
device_.invalidatePipelineCache();
device_.setScissorTest(false);
diff --git a/src/esengine/renderer/plugins/MeshPlugin.cpp b/src/esengine/renderer/plugins/MeshPlugin.cpp
index 646097e1a..aa1443a0e 100644
--- a/src/esengine/renderer/plugins/MeshPlugin.cpp
+++ b/src/esengine/renderer/plugins/MeshPlugin.cpp
@@ -663,10 +663,13 @@ void MeshPlugin::collect(RenderCollectContext& collect_ctx) {
// the record is read through that layout, so the fact that built
// it must pack it. (It already excludes a skinned mesh.)
const bool lightmapped = resident->hasLightmapUV;
- const u32 stride = meshInstanceStride(skinned, resident->hasNormals, lightmapped);
- u32 instOffset = buffers.allocVertices(LayoutId::MeshInstance, stride);
+ u32 instOffset = buffers.allocVertices(LayoutId::MeshInstance,
+ MESH_INSTANCE_STRIDE);
auto* dst = buffers.vertexData(LayoutId::MeshInstance) + instOffset;
- u32 tintRGBA = packColor(mesh.color);
+ // The record is one fixed shape, so every texel this variant
+ // does not write still has to say something. Zero is the answer
+ // each of them reads as absent.
+ std::memset(dst, 0, MESH_INSTANCE_STRIDE);
// The shapes, whether or not bones move it: the two deform the
// same vertices, and a morph target is what the mesh is BEFORE a
// pose is applied to it.
@@ -675,8 +678,8 @@ void MeshPlugin::collect(RenderCollectContext& collect_ctx) {
key.morphIndex = draw_list.addMorphShapes(shapes);
key.morphTextureId = static_cast<u32>(resident->morphTexture);
}
+ std::memcpy(dst + MESH_INSTANCE_TEXEL_TINT * 16, &mesh.color, 16);
if (skinned) {
- std::memcpy(dst, &tintRGBA, 4);
key.skinOffset = draw_list.addSkinMatrices(pose_scratch_.data(), poseSize);
key.skinCount = poseSize;
} else {
@@ -687,15 +690,14 @@ void MeshPlugin::collect(RenderCollectContext& collect_ctx) {
// the one an affine transform never varies.
for (u32 row = 0; row < 3; ++row) {
const glm::vec4 r{model[0][row], model[1][row], model[2][row], model[3][row]};
- std::memcpy(dst + row * 16, &r, 16);
+ std::memcpy(dst + (MESH_INSTANCE_TEXEL_MODEL + row) * 16, &r, 16);
}
- std::memcpy(dst + 48, &tintRGBA, 4);
if (resident->hasNormals) {
// Written per object rather than derived per vertex: this is
// the transform a normal takes under a non-uniform scale.
const glm::mat3 nrm = glm::transpose(glm::inverse(glm::mat3(model)));
for (u32 row = 0; row < 3; ++row) {
- std::memcpy(dst + 52 + row * 12, &nrm[row][0], 12);
+ std::memcpy(dst + (MESH_INSTANCE_TEXEL_NORMAL + row) * 16, &nrm[row][0], 12);
}
}
if (lightmapped) {
@@ -709,7 +711,7 @@ void MeshPlugin::collect(RenderCollectContext& collect_ctx) {
key.lightmapTextureId = atlas->getId();
}
}
- std::memcpy(dst + meshInstanceLightmapOffset(resident->hasNormals), &rect, 16);
+ std::memcpy(dst + MESH_INSTANCE_TEXEL_LIGHTMAP * 16, &rect, 16);
}
}
@@ -743,7 +745,7 @@ void MeshPlugin::collect(RenderCollectContext& collect_ctx) {
key.shaderId = materialProgram != 0 ? materialProgram : residentShader;
key.layoutId = LayoutId::MeshInstance;
key.instanceCount = 1;
- key.instanceStride = stride;
+ key.instanceStride = MESH_INSTANCE_STRIDE;
key.vertexBuffer = resident->vertexBuffer;
key.indexBuffer = resident->indexBuffer;
key.vertexLayout = resident->layout;
diff --git a/src/esengine/renderer/rhi/GLDevice.cpp b/src/esengine/renderer/rhi/GLDevice.cpp
index f033a4096..7ddae4ea8 100644
--- a/src/esengine/renderer/rhi/GLDevice.cpp
+++ b/src/esengine/renderer/rhi/GLDevice.cpp
@@ -63,6 +63,9 @@
#ifndef GL_RGBA16F
#define GL_RGBA16F 0x881A
#endif
+#ifndef GL_RGBA32F
+ #define GL_RGBA32F 0x8814
+#endif
#ifndef GL_HALF_FLOAT
#define GL_HALF_FLOAT 0x140B
#endif
@@ -190,6 +193,7 @@ GLPixelFormatInfo toGLPixelFormat(GfxPixelFormat fmt) {
case GfxPixelFormat::RGBA8: return { GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE };
case GfxPixelFormat::SRGB8_ALPHA8: return { GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_BYTE };
case GfxPixelFormat::RGBA16F: return { GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT };
+ case GfxPixelFormat::RGBA32F: return { GL_RGBA32F, GL_RGBA, GL_FLOAT };
case GfxPixelFormat::DepthComponent24: return { GL_DEPTH_COMPONENT24, GL_DEPTH_COMPONENT, GL_UNSIGNED_INT };
case GfxPixelFormat::Depth24Stencil8: return { GL_DEPTH24_STENCIL8, GL_DEPTH_STENCIL, GL_UNSIGNED_INT_24_8 };
default: return { GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE };
diff --git a/src/esengine/renderer/rhi/GfxEnums.hpp b/src/esengine/renderer/rhi/GfxEnums.hpp
index 97fd893e5..223ad1bb0 100644
--- a/src/esengine/renderer/rhi/GfxEnums.hpp
+++ b/src/esengine/renderer/rhi/GfxEnums.hpp
@@ -93,22 +93,35 @@ enum class LayoutId : u8 {
static constexpr u32 LAYOUT_COUNT = 5;
-/// One per-object record for a resident mesh: three ROWS of an affine model
-/// matrix, then a tint. The matrix's fourth row is (0,0,0,1) on every transform
-/// a Transform can state, so it is neither sent nor given an attribute slot.
-static constexpr u32 MESH_INSTANCE_STRIDE = 52;
-/// The same plus a normal matrix (three vec3 rows), for geometry with normals.
-/// Carried per object because a non-uniform scale makes the model matrix the
-/// wrong transform for a normal, and inverting one per vertex is the alternative.
-static constexpr u32 MESH_INSTANCE_STRIDE_LIT = 88;
-/// A SKINNED object's record: the tint, and nothing else. No model matrix — a
-/// skinned mesh's own transform is ignored (glTF says so) because its joints are
-/// already placed in the world — and the pose itself is a uniform block.
-static constexpr u32 MESH_INSTANCE_STRIDE_SKINNED = 4;
-/// What a lightmapped object APPENDS: the atlas rectangle its second UV set is
-/// read through (xy scale, zw offset). Per object, because the same mesh placed
-/// twice is lit twice and occupies two patches of one atlas.
-static constexpr u32 MESH_INSTANCE_LIGHTMAP_BYTES = 16;
+/// One per-object record for a resident mesh, in texels of the frame's RGBA32F
+/// instance texture. FIXED at one shape for every variant, so no stride has to be
+/// derived by the layout, the allocation and the packer separately.
+/// @see MESH_INSTANCE_TEXEL_MODEL and the offsets below.
+static constexpr u32 MESH_INSTANCE_TEXELS = 8;
+/// The model matrix's three ROWS, translation in .w. Its fourth row is (0,0,0,1)
+/// on every transform a Transform can state, so it is not sent.
+static constexpr u32 MESH_INSTANCE_TEXEL_MODEL = 0;
+static constexpr u32 MESH_INSTANCE_TEXEL_TINT = 3;
+/// The normal matrix's three rows in .xyz. Per object because a non-uniform scale
+/// makes the model matrix the wrong transform for a normal.
+static constexpr u32 MESH_INSTANCE_TEXEL_NORMAL = 4;
+/// The bake's atlas rectangle (xy scale, zw offset). Per object, because the same
+/// mesh placed twice is lit twice and occupies two patches of one atlas.
+static constexpr u32 MESH_INSTANCE_TEXEL_LIGHTMAP = 7;
+/// Bytes one record occupies. RGBA32F: a world-space translation is what this
+/// carries, and half precision loses whole units of it a thousand away.
+static constexpr u32 MESH_INSTANCE_STRIDE = MESH_INSTANCE_TEXELS * 16;
+/// Texels per row of the instance texture. A multiple of the record, so no
+/// record straddles a row; 2048 is the smallest MAX_TEXTURE_SIZE WebGL2 grants.
+static constexpr u32 MESH_INSTANCE_TEXTURE_WIDTH = 2048;
+/// Records the texture holds. Beyond it the frame's tail draws unplaced, which
+/// is visible; a texture the device refuses would draw nothing at all.
+static constexpr u32 MESH_INSTANCE_MAX_RECORDS =
+ MESH_INSTANCE_TEXTURE_WIDTH * MESH_INSTANCE_TEXTURE_WIDTH / MESH_INSTANCE_TEXELS;
+/// The unit the frame's record texture binds on. Not one of the eight a draw's
+/// own stream can claim, because it is FRAME state; and the TOP of what remains,
+/// since a material's textures are handed units upward from unit 8.
+static constexpr u32 MESH_INSTANCE_TEXTURE_UNIT = 15;
/// Bone matrices one skinned draw may carry. 64 mat4 is 4KB, inside the 16KB a
/// WebGL2 uniform block is guaranteed; a mesh wanting more is drawn static.
static constexpr u32 MESH_MAX_BONES = 64;
@@ -123,34 +136,10 @@ static constexpr u32 MORPH_TEXTURE_WIDTH = 2048;
/// Beyond it the geometry draws unmorphed — a texture the device refuses is a
/// mesh that does not draw at all.
static constexpr u32 MORPH_MAX_TEXELS = MORPH_TEXTURE_WIDTH * MORPH_TEXTURE_WIDTH;
-/// The attributes that record occupies (4 matrix rows + the tint + the atlas
-/// rectangle) — the budget check, so the widest record is the one measured.
-static constexpr u32 MESH_INSTANCE_ATTRIBUTES = 6;
/// Programs the resident-mesh shader has: one per combination of {normals, lit,
/// normal-mapped, skinned, depth-only, env-mapped, lightmapped}. Seven bits, not
/// one: the geometry, the draw, the frame and the pass each answer a different.
static constexpr u32 MESH_VARIANTS = 128;
-/// Where those attributes start. FIXED, not "after the mesh's channels": a mesh
-/// that gains normals would otherwise move them, and every mesh shader with it.
-static constexpr u32 MESH_INSTANCE_FIRST_LOCATION = 8;
-
-/**
- * @brief How many bytes one object's record is, for the shape it is drawn in.
- * @details One author for a number three places need — the layout, the
- * allocation and the packer — so a part added to the record cannot
- * reach two of them and miss the third.
- */
-constexpr u32 meshInstanceStride(bool skinned, bool hasNormals, bool lightmapped) {
- if (skinned) return MESH_INSTANCE_STRIDE_SKINNED;
- return (hasNormals ? MESH_INSTANCE_STRIDE_LIT : MESH_INSTANCE_STRIDE)
- + (lightmapped ? MESH_INSTANCE_LIGHTMAP_BYTES : 0);
-}
-
-/** @brief Where the atlas rectangle sits: appended, so nothing before it moves. */
-constexpr u32 meshInstanceLightmapOffset(bool hasNormals) {
- return hasNormals ? MESH_INSTANCE_STRIDE_LIT : MESH_INSTANCE_STRIDE;
-}
-
/**
* @brief What a mesh vertex channel MEANS, and the attribute location it binds to.
* @details One vocabulary for the file format, the layout and the shaders: a
@@ -167,8 +156,7 @@ enum class MeshChannel : u8 {
Joints = 5,
Weights = 6,
/// A second UV set — what a bake is read through, laid out so no two of a
- /// model's surfaces overlap. Location 7 is the last before the per-object
- /// record at MESH_INSTANCE_FIRST_LOCATION, which is why that is fixed.
+ /// model's surfaces overlap.
TexCoord1 = 7,
};
@@ -284,6 +272,10 @@ enum class GfxPixelFormat : u8 {
/// Half-float HDR target; gate render-target use on
/// GfxDevice::supportsFloatTargets (EXT_color_buffer_float on WebGL2).
RGBA16F,
+ /// Full-float storage, for values a half cannot hold — a world-space
+ /// translation is one. Fetched by texel index only: filtering full floats is
+ /// an optional device feature.
+ RGBA32F,
DepthComponent24,
Depth24Stencil8,
};
@@ -296,6 +288,7 @@ inline u32 gfxBytesPerPixel(GfxPixelFormat fmt) {
switch (fmt) {
case GfxPixelFormat::RGB8: return 3;
case GfxPixelFormat::RGBA16F: return 8;
+ case GfxPixelFormat::RGBA32F: return 16;
default: return 4;
}
}
diff --git a/src/esengine/renderer/rhi/PerDrawBlocks.hpp b/src/esengine/renderer/rhi/PerDrawBlocks.hpp
index 1c165f7a7..11a02ffe2 100644
--- a/src/esengine/renderer/rhi/PerDrawBlocks.hpp
+++ b/src/esengine/renderer/rhi/PerDrawBlocks.hpp
@@ -65,4 +65,18 @@ private:
BufferHandle zero_ = BufferHandle::Invalid;
};
+/**
+ * @brief The uniform blocks a draw rewrites immediately before it.
+ *
+ * @details One parameter and not four: every caller passes the same set from the
+ * same owner, so a block added to the vocabulary reaches the draw loop
+ * without every call site having to learn its name.
+ */
+struct PerDrawBlockSet {
+ PerDrawBlocks* skin = nullptr;
+ PerDrawBlocks* morph = nullptr;
+ PerDrawBlocks* probe = nullptr;
+ PerDrawBlocks* instance = nullptr;
+};
+
} // namespace esengine
diff --git a/src/esengine/renderer/rhi/Shader.cpp b/src/esengine/renderer/rhi/Shader.cpp
index a44c40eb5..ae511dfbd 100644
--- a/src/esengine/renderer/rhi/Shader.cpp
+++ b/src/esengine/renderer/rhi/Shader.cpp
@@ -20,6 +20,7 @@
#include "../store/MaterialConstants.hpp"
#include "../store/SkinConstants.hpp"
#include "../store/LightConstants.hpp"
+#include "../store/InstanceConstants.hpp"
#include "../store/MorphConstants.hpp"
#include "../store/ProbeConstants.hpp"
#include "../../core/Log.hpp"
@@ -269,6 +270,13 @@ bool Shader::compile(const std::string& vertexSrc, const std::string& fragmentSr
device_->uniformBlockBinding(program_, probeBlock, PROBE_CONSTANTS_BINDING);
}
+ // And where this draw's objects start in the frame's record texture, which
+ // DrawList rewrites per draw for the same reason.
+ u32 instanceBlock = device_->getUniformBlockIndex(program_, INSTANCE_CONSTANTS_BLOCK);
+ if (instanceBlock != GFX_INVALID_UNIFORM_BLOCK) {
+ device_->uniformBlockBinding(program_, instanceBlock, INSTANCE_CONSTANTS_BINDING);
+ }
+
// Same for the per-draw params block (rewriteLooseUniforms generates it for
// shaders whose loose uniforms were lifted); commitParams binds the UBO.
u32 drawParamsBlock = device_->getUniformBlockIndex(program_, DRAW_PARAMS_BLOCK);
@@ -276,34 +284,26 @@ bool Shader::compile(const std::string& vertexSrc, const std::string& fragmentSr
device_->uniformBlockBinding(program_, drawParamsBlock, DRAW_PARAMS_BINDING);
}
- // Engine-injected samplers sit on fixed units, pinned here for the same reason the
- // blocks above are: the header reaches every Lit shader, so a compile site that
- // forgot the unit would sample slot 0 silently. GLSL ES 300 has no layout(binding).
- if (hasUniform(SHADOW_MAP_SAMPLER) || hasUniform(ENV_MAP_SAMPLER)
- || hasUniform(SHADOW_2D_SAMPLER) || hasUniform(SHAPE_2D_SAMPLER)
- || hasUniform(MORPH_DELTA_SAMPLER) || hasUniform(LIGHTMAP_SAMPLER)) {
- bind();
- if (hasUniform(SHADOW_MAP_SAMPLER)) {
- setUniform(SHADOW_MAP_SAMPLER, static_cast<i32>(SHADOW_MAP_TEXTURE_UNIT));
- }
- if (hasUniform(ENV_MAP_SAMPLER)) {
- setUniform(ENV_MAP_SAMPLER, static_cast<i32>(ENV_MAP_TEXTURE_UNIT));
- }
- if (hasUniform(SHADOW_2D_SAMPLER)) {
- setUniform(SHADOW_2D_SAMPLER, static_cast<i32>(SHADOW_2D_TEXTURE_UNIT));
- }
- if (hasUniform(SHAPE_2D_SAMPLER)) {
- setUniform(SHAPE_2D_SAMPLER, static_cast<i32>(SHAPE_2D_TEXTURE_UNIT));
- }
- // The one sampler here read by the VERTEX stage.
- if (hasUniform(MORPH_DELTA_SAMPLER)) {
- setUniform(MORPH_DELTA_SAMPLER, static_cast<i32>(MORPH_DELTA_TEXTURE_UNIT));
- }
- if (hasUniform(LIGHTMAP_SAMPLER)) {
- setUniform(LIGHTMAP_SAMPLER, static_cast<i32>(LIGHTMAP_TEXTURE_UNIT));
- }
- unbind();
+ // Engine-injected samplers sit on fixed units, pinned here because GLSL ES 300
+ // has no layout(binding) and a compile site that forgot one samples slot 0
+ // silently. The last two are read by the VERTEX stage.
+ struct PinnedSampler { const char* name; u32 unit; };
+ static const PinnedSampler kPinned[] = {
+ {SHADOW_MAP_SAMPLER, SHADOW_MAP_TEXTURE_UNIT},
+ {ENV_MAP_SAMPLER, ENV_MAP_TEXTURE_UNIT},
+ {SHADOW_2D_SAMPLER, SHADOW_2D_TEXTURE_UNIT},
+ {SHAPE_2D_SAMPLER, SHAPE_2D_TEXTURE_UNIT},
+ {LIGHTMAP_SAMPLER, LIGHTMAP_TEXTURE_UNIT},
+ {MORPH_DELTA_SAMPLER, MORPH_DELTA_TEXTURE_UNIT},
+ {INSTANCE_DATA_SAMPLER, MESH_INSTANCE_TEXTURE_UNIT},
+ };
+ bool pinned = false;
+ for (const PinnedSampler& p : kPinned) {
+ if (!hasUniform(p.name)) continue;
+ if (!pinned) { bind(); pinned = true; }
+ setUniform(p.name, static_cast<i32>(p.unit));
}
+ if (pinned) unbind();
ES_LOG_DEBUG("Shader compiled successfully (program handle: {}, active uniforms: {})",
static_cast<u32>(program_), activeUniforms_.size());
diff --git a/src/esengine/renderer/rhi/ShaderEmbeds.generated.hpp b/src/esengine/renderer/rhi/ShaderEmbeds.generated.hpp
index ba1893284..2e253631e 100644
--- a/src/esengine/renderer/rhi/ShaderEmbeds.generated.hpp
+++ b/src/esengine/renderer/rhi/ShaderEmbeds.generated.hpp
@@ -412,30 +412,9 @@ layout(location = 7) in vec2 a_texCoord1;
-#ifndef SKINNED
-layout(location = 8) in vec4 a_model0;
-layout(location = 9) in vec4 a_model1;
-layout(location = 10) in vec4 a_model2;
-#ifdef MESH_LIGHTMAP
-
-
-
-layout(location = 11) in vec4 a_lightmapRect;
-#endif
-#endif
-layout(location = 12) in vec4 a_instTint;
-#if defined(MESH_NORMALS) && !defined(SKINNED)
-
-
-
-layout(location = 13) in vec3 a_nrm0;
-layout(location = 14) in vec3 a_nrm1;
-layout(location = 15) in vec3 a_nrm2;
-#endif
-
#ifdef SKINNED
layout(std140) uniform SkinConstants {
@@ -520,18 +499,16 @@ void main() {
+ a_weights.w * u_bones[a_joints.w];
vec4 world = skin * vec4(local, 1.0);
#else
- mat4 model = mat4(vec4(a_model0.x, a_model1.x, a_model2.x, 0.0),
- vec4(a_model0.y, a_model1.y, a_model2.y, 0.0),
- vec4(a_model0.z, a_model1.z, a_model2.z, 0.0),
- vec4(a_model0.w, a_model1.w, a_model2.w, 1.0));
+ mat4 model = esInstanceModel();
vec4 world = model * vec4(local, 1.0);
#endif
gl_Position = u_projection * world;
v_texCoord = a_texCoord;
- v_color = a_color * a_instTint;
+ v_color = a_color * esInstanceTint();
#ifdef MESH_LIGHTMAP
- v_lightmap = vec3(a_texCoord1 * a_lightmapRect.xy + a_lightmapRect.zw,
- a_lightmapRect.x > 0.0 ? 1.0 : 0.0);
+ highp vec4 lmRect = esInstanceLightmapRect();
+ v_lightmap = vec3(a_texCoord1 * lmRect.xy + lmRect.zw,
+ lmRect.x > 0.0 ? 1.0 : 0.0);
#endif
#ifdef SHADOW_DEPTH
v_shadowClip = gl_Position;
@@ -543,7 +520,7 @@ void main() {
#if defined(MESH_NORMALS) && defined(SKINNED)
v_worldNormal = mat3(skin) * localNormal;
#elif defined(MESH_NORMALS)
- v_worldNormal = mat3(a_nrm0, a_nrm1, a_nrm2) * localNormal;
+ v_worldNormal = esInstanceNormalMatrix() * localNormal;
#else
v_worldNormal = vec3(0.0, 0.0, 1.0);
@@ -652,27 +629,12 @@ struct VSIn {
#ifdef SKINNED
@location(5) a_joints : vec4u,
@location(6) a_weights : vec4f,
-#else
- @location(8) a_model0 : vec4f,
- @location(9) a_model1 : vec4f,
- @location(10) a_model2 : vec4f,
-#ifdef MESH_LIGHTMAP
- @location(11) a_lightmapRect : vec4f,
-#endif
#endif
#ifdef MESH_LIGHTMAP
@location(7) a_texCoord1 : vec2f,
#endif
- @location(12) a_instTint : vec4f,
@builtin(vertex_index) vertexIndex : u32,
-#ifdef MESH_NORMALS
-#ifndef SKINNED
- @location(13) a_nrm0 : vec3f,
- @location(14) a_nrm1 : vec3f,
- @location(15) a_nrm2 : vec3f,
-#endif
-#endif
};
struct VSOut {
@builtin(position) pos : vec4f,
@@ -695,6 +657,7 @@ struct VSOut {
};
@vertex fn vs_main(v : VSIn, @builtin(instance_index) inst : u32) -> VSOut {
+ g_instanceId = i32(inst);
var local = v.a_position;
#ifdef MESH_NORMALS
var localNormal = v.a_normal;
@@ -721,20 +684,17 @@ struct VSOut {
+ v.a_weights.w * skin.bones[v.a_joints.w];
let world = pose * vec4f(local, 1.0);
#else
- let model = mat4x4f(vec4f(v.a_model0.x, v.a_model1.x, v.a_model2.x, 0.0),
- vec4f(v.a_model0.y, v.a_model1.y, v.a_model2.y, 0.0),
- vec4f(v.a_model0.z, v.a_model1.z, v.a_model2.z, 0.0),
- vec4f(v.a_model0.w, v.a_model1.w, v.a_model2.w, 1.0));
- let world = model * vec4f(local, 1.0);
+ let world = esInstanceModel() * vec4f(local, 1.0);
#endif
var out : VSOut;
out.pos = frame.projection * world;
out.v_texCoord = v.a_texCoord;
- out.v_color = v.a_color * v.a_instTint;
+ out.v_color = v.a_color * esInstanceTint();
#ifdef MESH_LIGHTMAP
- out.v_lightmap = vec3f(v.a_texCoord1 * v.a_lightmapRect.xy + v.a_lightmapRect.zw,
- select(0.0, 1.0, v.a_lightmapRect.x > 0.0));
+ let lmRect = esInstanceLightmapRect();
+ out.v_lightmap = vec3f(v.a_texCoord1 * lmRect.xy + lmRect.zw,
+ select(0.0, 1.0, lmRect.x > 0.0));
#endif
#ifdef SHADOW_DEPTH
out.v_shadowClip = out.pos;
@@ -746,7 +706,7 @@ struct VSOut {
#ifdef SKINNED
out.v_worldNormal = mat3x3f(pose[0].xyz, pose[1].xyz, pose[2].xyz) * localNormal;
#else
- out.v_worldNormal = mat3x3f(v.a_nrm0, v.a_nrm1, v.a_nrm2) * localNormal;
+ out.v_worldNormal = esInstanceNormalMatrix() * localNormal;
#endif
#else
diff --git a/src/esengine/renderer/rhi/Texture.cpp b/src/esengine/renderer/rhi/Texture.cpp
index 3c0038cb3..fc017a7ed 100644
--- a/src/esengine/renderer/rhi/Texture.cpp
+++ b/src/esengine/renderer/rhi/Texture.cpp
@@ -30,6 +30,7 @@ GfxPixelFormat toGfxPixelFormat(TextureFormat format) {
case TextureFormat::RGBA8: return GfxPixelFormat::RGBA8;
case TextureFormat::SRGB8A8: return GfxPixelFormat::SRGB8_ALPHA8;
case TextureFormat::RGBA16F: return GfxPixelFormat::RGBA16F;
+ case TextureFormat::RGBA32F: return GfxPixelFormat::RGBA32F;
case TextureFormat::Depth24: return GfxPixelFormat::DepthComponent24;
default: return GfxPixelFormat::RGBA8;
}
diff --git a/src/esengine/renderer/rhi/Texture.hpp b/src/esengine/renderer/rhi/Texture.hpp
index 5fde4ea0d..cda6927c8 100644
--- a/src/esengine/renderer/rhi/Texture.hpp
+++ b/src/esengine/renderer/rhi/Texture.hpp
@@ -47,6 +47,7 @@ enum class TextureFormat {
RGBA8, ///< 4 channels, 8 bits each (32 bpp)
SRGB8A8, ///< sRGB-encoded RGBA8: hardware-linearized on sample (32 bpp)
RGBA16F, ///< Half-float HDR (64 bpp); render use gated on supportsFloatTargets
+ RGBA32F, ///< Full-float storage (128 bpp); fetched by texel, never filtered
Depth24 ///< Depth buffer format (24 bits)
};
diff --git a/src/esengine/renderer/rhi/TransientBufferPool.cpp b/src/esengine/renderer/rhi/TransientBufferPool.cpp
index 1389e7cc8..cacaee9fb 100644
--- a/src/esengine/renderer/rhi/TransientBufferPool.cpp
+++ b/src/esengine/renderer/rhi/TransientBufferPool.cpp
@@ -40,6 +40,8 @@ void TransientBufferPool::recreateGpuResources() {
s.ebo_capacity = 0;
s.vertex_write_pos = 0;
s.index_write_pos = 0;
+ s.texture = TextureHandle::Invalid;
+ s.texture_rows = 0;
}
initialized_ = false;
init(initial_vertex_bytes_, initial_index_count_);
@@ -56,6 +58,11 @@ void TransientBufferPool::shutdown() {
if (s.vbo != BufferHandle::Invalid) { device_.deleteBuffer(s.vbo); s.vbo = BufferHandle::Invalid; }
if (s.ebo != BufferHandle::Invalid) { device_.deleteBuffer(s.ebo); s.ebo = BufferHandle::Invalid; }
if (s.quad_vbo != BufferHandle::Invalid) { device_.deleteBuffer(s.quad_vbo); s.quad_vbo = BufferHandle::Invalid; }
+ if (s.texture != TextureHandle::Invalid) {
+ device_.deleteTexture(s.texture);
+ s.texture = TextureHandle::Invalid;
+ }
+ s.texture_rows = 0;
s.vertex_staging.clear();
s.index_staging.clear();
s.vertex_write_pos = 0;
@@ -81,6 +88,47 @@ const TransientBufferPool::Stream& TransientBufferPool::stream(LayoutId layout)
return streams_[static_cast<u32>(layout)];
}
+/**
+ * @brief The frame's records onto the GPU, as whole rows.
+ *
+ * @details Whole rows because a partial one asks the upload to describe a
+ * rectangle the staging is not laid out as. The record divides the row
+ * evenly, so a row boundary never falls inside one.
+ */
+void TransientBufferPool::uploadInstanceRows(Stream& s, u32& grows, u32& writes) {
+ const u32 rowBytes = instanceRowBytes();
+ const u32 rows = (s.vertex_write_pos + rowBytes - 1) / rowBytes;
+ if (static_cast<u64>(s.vertex_staging.size()) < static_cast<u64>(rows) * rowBytes) {
+ s.vertex_staging.resize(static_cast<usize>(rows) * rowBytes);
+ }
+ if (rows > s.texture_rows) {
+ // Doubling, the way the staging grows: a workload creeping upward
+ // otherwise reallocates a texture every frame.
+ const u32 want = std::max(rows, s.texture_rows * 2);
+ if (s.texture != TextureHandle::Invalid) device_.deleteTexture(s.texture);
+ TextureDesc desc;
+ desc.width = MESH_INSTANCE_TEXTURE_WIDTH;
+ desc.height = want;
+ desc.format = GfxPixelFormat::RGBA32F;
+ // Fetched by index, never filtered: a value between two records belongs
+ // to no object, and full floats are not filterable everywhere anyway.
+ desc.minFilter = TextureFilter::Nearest;
+ desc.magFilter = TextureFilter::Nearest;
+ desc.wrapS = TextureWrap::ClampToEdge;
+ desc.wrapT = TextureWrap::ClampToEdge;
+ desc.mipmaps = false;
+ s.texture = device_.createTexture(desc, s.vertex_staging.data());
+ s.texture_rows = s.texture == TextureHandle::Invalid ? 0 : want;
+ ++grows;
+ return;
+ }
+ // No flip: this is a record store addressed by index, not an image, so row 0
+ // has to stay the row the writer wrote first.
+ device_.updateTexture(s.texture, 0, 0, MESH_INSTANCE_TEXTURE_WIDTH, rows,
+ s.vertex_staging.data(), /*flipY=*/false);
+ ++writes;
+}
+
u32 TransientBufferPool::allocVertices(LayoutId layout, u32 byteSize) {
Stream& s = stream(layout);
u32 offset = s.vertex_write_pos;
@@ -131,6 +179,13 @@ void TransientBufferPool::upload() {
// Growth goes through resizeBuffer, which keeps the handle stable, so per-draw
// buffer bindings and the backend's cached vertex state stay valid.
for (auto& s : streams_) {
+ if (&s == &stream(LayoutId::MeshInstance)) {
+ if (s.vertex_write_pos == 0) continue;
+ ++streams;
+ vertexBytes += s.vertex_write_pos;
+ uploadInstanceRows(s, grows, writes);
+ continue;
+ }
if (s.vbo == BufferHandle::Invalid) continue;
if (s.vertex_write_pos == 0 && s.index_write_pos == 0) continue;
++streams;
@@ -217,13 +272,12 @@ void TransientBufferPool::setupStream(LayoutId layout) {
Stream& s = stream(layout);
if (layout == LayoutId::MeshInstance) {
- // Only a stream. The geometry is the mesh's and so is the layout — this
- // holds the per-object transforms written for the frame, which is the one
- // part of drawing a resident mesh that is not resident.
- s.vertex_staging.resize(initial_vertex_bytes_);
- s.vbo_capacity = initial_vertex_bytes_;
- s.vbo = device_.createBuffer({GfxBufferUsage::Vertex, s.vbo_capacity, /*dynamic=*/true}, nullptr);
+ // Only a stream, and not a vertex one: the frame's per-object records go
+ // in a texture, so a shader can reach any record by index rather than only
+ // the one the hardware happens to be on.
s.vertex_stride = MESH_INSTANCE_STRIDE;
+ s.vertex_staging.resize(instanceRowBytes());
+ s.texture_rows = 0;
return;
}
diff --git a/src/esengine/renderer/rhi/TransientBufferPool.hpp b/src/esengine/renderer/rhi/TransientBufferPool.hpp
index 852f35199..ffdd8ea6f 100644
--- a/src/esengine/renderer/rhi/TransientBufferPool.hpp
+++ b/src/esengine/renderer/rhi/TransientBufferPool.hpp
@@ -75,12 +75,20 @@ public:
BufferHandle vertexBuffer(LayoutId layout) const;
BufferHandle indexBuffer(LayoutId layout) const;
+ /** @brief The frame's per-object records. Invalid until a frame writes some. */
+ TextureHandle instanceTexture() const { return stream(LayoutId::MeshInstance).texture; }
+
private:
struct Stream {
BufferHandle vbo = BufferHandle::Invalid; // for ParticleInstance: the per-instance (streamed) buffer
BufferHandle ebo = BufferHandle::Invalid;
VertexLayoutHandle layout = VertexLayoutHandle::Invalid;
BufferHandle quad_vbo = BufferHandle::Invalid; // ParticleInstance only: static unit-quad geometry (slot 0)
+ /// MeshInstance only: the frame's records live in a texture, because a
+ /// vertex attribute can only be read at the instance the hardware is on
+ /// and a record has to be reachable by index.
+ TextureHandle texture = TextureHandle::Invalid;
+ u32 texture_rows = 0; ///< Rows the texture currently holds.
std::vector<u8> vertex_staging;
std::vector<u32> index_staging; // 32-bit indices: a single Batch stream can exceed 65535 vertices
u32 vertex_stride = 0;
@@ -91,6 +99,8 @@ private:
};
void setupStream(LayoutId layout);
+ void uploadInstanceRows(Stream& s, u32& grows, u32& writes);
+ static constexpr u32 instanceRowBytes() { return MESH_INSTANCE_TEXTURE_WIDTH * 16; }
void growVertexStaging(Stream& s, u32 requiredBytes);
void growIndexStaging(Stream& s, u32 requiredCount);
diff --git a/src/esengine/renderer/store/InstanceConstants.hpp b/src/esengine/renderer/store/InstanceConstants.hpp
new file mode 100644
index 000000000..75291b29d
--- /dev/null
+++ b/src/esengine/renderer/store/InstanceConstants.hpp
@@ -0,0 +1,44 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright (c) 2024-present ESEngine Team
+/**
+ * @file InstanceConstants.hpp
+ * @brief WHERE one draw's objects start in the frame's per-object record texture.
+ *
+ * @details Per DRAW and rewritten immediately before it, the way @ref ProbeConstants
+ * and @ref MorphConstants are. `gl_InstanceID` restarts at zero every draw
+ * while the frame's records are one continuous run, so a draw has to say
+ * where its own run begins; a merge makes that run longer, never split.
+ *
+ * A uniform and not a vertex attribute: an attribute would cost the record
+ * the very slot this whole carrier exists to give back.
+ */
+#pragma once
+
+#include "../../core/Types.hpp"
+
+namespace esengine {
+
+/** @brief Indexed UBO binding point for the per-draw InstanceConstants block.
+ * Nine and not four: slot 4 is DRAW_PARAMS_BINDING, which every shader whose
+ * loose uniforms were lifted declares — two blocks on one binding is a pipeline
+ * the device refuses without naming either. */
+inline constexpr u32 INSTANCE_CONSTANTS_BINDING = 8;
+
+/** @brief GLSL block name; must match the injected mesh header + Shader::compile lookup. */
+inline constexpr const char* INSTANCE_CONSTANTS_BLOCK = "InstanceConstants";
+
+/** @brief GLSL name of the record texture; pinned to its unit by Shader::compile. */
+inline constexpr const char* INSTANCE_DATA_SAMPLER = "u_instanceData";
+
+/** @brief The first record of one draw's run. */
+struct InstanceConstants {
+ u32 base = 0;
+ /// std140 rounds a block up to a vec4 anyway; naming the rest keeps the C++
+ /// size equal to what the shader's block occupies rather than nearly equal.
+ u32 pad[3]{};
+};
+
+static_assert(sizeof(InstanceConstants) == 16,
+ "InstanceConstants must be one vec4 to match std140");
+
+} // namespace esengine
diff --git a/src/esengine/renderer/webgpu/WebGPUDevice.cpp b/src/esengine/renderer/webgpu/WebGPUDevice.cpp
index 41d104b47..f88ddae1f 100644
--- a/src/esengine/renderer/webgpu/WebGPUDevice.cpp
+++ b/src/esengine/renderer/webgpu/WebGPUDevice.cpp
@@ -1475,9 +1475,13 @@ WGPUBindGroupLayout WebGPUDevice::groupLayoutFor(u32 group, u32 mask, u32 depthM
WGPUBindGroupLayoutEntry e{};
e.binding = tb;
e.visibility = stages(tb);
- e.texture.sampleType = (depthMask & (1u << tb))
- ? WGPUTextureSampleType_Depth
- : WGPUTextureSampleType_Float;
+ // A unit with no sampler cannot be textureSample'd, so it is
+ // unfilterable by construction — which is what lets rgba32float
+ // bind here. Filterable formats satisfy it too.
+ e.texture.sampleType =
+ (depthMask & (1u << tb)) ? WGPUTextureSampleType_Depth
+ : (mask & (1u << sb)) ? WGPUTextureSampleType_Float
+ : WGPUTextureSampleType_UnfilterableFloat;
e.texture.viewDimension = WGPUTextureViewDimension_2D;
entries[count++] = e;
}
diff --git a/src/esengine/renderer/webgpu/WebGPUDevice.hpp b/src/esengine/renderer/webgpu/WebGPUDevice.hpp
index 0da515ca5..523485677 100644
--- a/src/esengine/renderer/webgpu/WebGPUDevice.hpp
+++ b/src/esengine/renderer/webgpu/WebGPUDevice.hpp
@@ -521,7 +521,10 @@ private:
u32 current_pipeline_ = 0;
u32 bound_index_buffer_ = 0;
i32 stencil_ref_ = 0; ///< Last user-set reference (re-applied after internal quads).
- static constexpr u32 kUniformSlots = 8;
+ /// Nine: the eight the engine had, plus where a draw's objects start in the
+ /// frame's record texture. WebGPU grants 12 uniform buffers a stage and
+ /// WebGL2 grants 12 blocks a stage, so nine is inside both.
+ static constexpr u32 kUniformSlots = 9;
u32 uniform_slots_[kUniformSlots] = {}; ///< BufferHandle id per UBO binding slot.
/// Engine units 0..7 (batch multi-texture) + material-param units 8..15
/// (== webgpu::kGroup1TextureUnits, static_asserted in the .cpp).
diff --git a/src/esengine/renderer/webgpu/WebGPUMappings.hpp b/src/esengine/renderer/webgpu/WebGPUMappings.hpp
index 542ea5426..c185b26c4 100644
--- a/src/esengine/renderer/webgpu/WebGPUMappings.hpp
+++ b/src/esengine/renderer/webgpu/WebGPUMappings.hpp
@@ -160,6 +160,7 @@ inline WGPUTextureFormat toWGPUTextureFormat(GfxPixelFormat fmt) {
case GfxPixelFormat::RGBA8: return WGPUTextureFormat_RGBA8Unorm;
case GfxPixelFormat::SRGB8_ALPHA8: return WGPUTextureFormat_RGBA8UnormSrgb;
case GfxPixelFormat::RGBA16F: return WGPUTextureFormat_RGBA16Float;
+ case GfxPixelFormat::RGBA32F: return WGPUTextureFormat_RGBA32Float;
case GfxPixelFormat::DepthComponent24: return WGPUTextureFormat_Depth24Plus;
case GfxPixelFormat::Depth24Stencil8: return WGPUTextureFormat_Depth24PlusStencil8;
default: return WGPUTextureFormat_RGBA8Unorm;
@@ -168,7 +169,9 @@ inline WGPUTextureFormat toWGPUTextureFormat(GfxPixelFormat fmt) {
/** Bytes per pixel of an uncompressed upload in @p fmt (RGB8 expands to RGBA8). */
inline u32 wgpuBytesPerPixel(GfxPixelFormat fmt) {
- return fmt == GfxPixelFormat::RGBA16F ? 8u : 4u;
+ if (fmt == GfxPixelFormat::RGBA16F) return 8u;
+ if (fmt == GfxPixelFormat::RGBA32F) return 16u;
+ return 4u;
}
inline WGPUTextureFormat toWGPUCompressedFormat(GfxCompressedFormat fmt) {
diff --git a/src/esengine/resource/ResourceManager.cpp b/src/esengine/resource/ResourceManager.cpp
index 718f505d5..45edc004a 100644
--- a/src/esengine/resource/ResourceManager.cpp
+++ b/src/esengine/resource/ResourceManager.cpp
@@ -502,9 +502,9 @@ bool ResourceManager::realizeMesh(Mesh& mesh, ConstSpan<u8> vertexBytes, ConstSp
ConstSpan<f32> inverseBind, const MeshMorphSource& morph) {
if (!device_ || vertexBytes.empty() || indices.empty() || channels.empty()) return false;
- // The mesh describes its own vertices; the per-object transform is the
- // engine's and is appended here, so no caller has to know how a transform
- // reaches the shader — the reason a mesh is drawn without touching its bytes.
+ // The mesh describes its own vertices and nothing else: the per-object
+ // transform reaches the shader through the frame's record texture, which is
+ // what leaves every slot above the mesh channels free.
VertexLayoutDesc layout;
// A channel no shader reads stays out of the layout (meshShaderReads): it
// would spend an attribute slot and a fetch per vertex. The stride is
@@ -512,7 +512,7 @@ bool ResourceManager::realizeMesh(Mesh& mesh, ConstSpan<u8> vertexBytes, ConstSp
u32 bound = 0;
for (const GfxVertexAttribute& c : channels) {
if (!meshShaderReads(static_cast<MeshChannel>(c.location))) continue;
- if (bound + MESH_INSTANCE_ATTRIBUTES >= MAX_VERTEX_ATTRIBUTES) {
+ if (bound >= MAX_VERTEX_ATTRIBUTES) {
ES_LOG_ERROR("realizeMesh: {} bound channels exceeds the layout budget", bound + 1);
return false;
}
@@ -535,31 +535,7 @@ bool ResourceManager::realizeMesh(Mesh& mesh, ConstSpan<u8> vertexBytes, ConstSp
const bool lightmapped = lightmapUV && !skinned;
layout.strides[0] = vertexStride;
- layout.strides[1] = meshInstanceStride(skinned, hasNormals, lightmapped);
- layout.instanceStep[1] = true;
- u32 next = bound;
- // Only where the shader will read them: a layout may not declare an attribute
- // its shader does not consume, which WebGPU rejects. A skinned record thus
- // carries neither model nor normal matrix — its bones are world-space.
- if (!skinned) {
- for (u32 row = 0; row < 3; ++row) {
- layout.attributes[next++] = {MESH_INSTANCE_FIRST_LOCATION + row, 4, GfxDataType::Float,
- false, row * 16u, 1};
- }
- }
- layout.attributes[next++] = {MESH_INSTANCE_FIRST_LOCATION + 4, 4, GfxDataType::UnsignedByte,
- true, skinned ? 0u : 48u, 1};
- if (hasNormals && !skinned) {
- for (u32 row = 0; row < 3; ++row) {
- layout.attributes[next++] = {MESH_INSTANCE_FIRST_LOCATION + 5 + row, 3,
- GfxDataType::Float, false, 52 + row * 12u, 1};
- }
- }
- if (lightmapped) {
- layout.attributes[next++] = {MESH_INSTANCE_FIRST_LOCATION + 3, 4, GfxDataType::Float,
- false, meshInstanceLightmapOffset(hasNormals), 1};
- }
- layout.attributeCount = next;
+ layout.attributeCount = bound;
// The old records go only once the new ones stand: a rematerialization that
// fails half-way must leave the mesh as it found it, not stripped of the
diff --git a/src/esengine/resource/ShaderParser.cpp b/src/esengine/resource/ShaderParser.cpp
index cb646bf2e..01b6dd042 100644
--- a/src/esengine/resource/ShaderParser.cpp
+++ b/src/esengine/resource/ShaderParser.cpp
@@ -13,7 +13,9 @@
#include "ShaderParser.hpp"
#include "../core/Log.hpp"
+#include "../renderer/store/InstanceConstants.hpp"
#include "../renderer/store/MaterialConstants.hpp"
+#include "../renderer/rhi/GfxEnums.hpp"
#include "../renderer/rhi/WgslBindings.hpp"
#include <sstream>
@@ -30,6 +32,79 @@ namespace esengine::resource {
namespace {
+/**
+ * @brief How a vertex stage reaches its own object's record.
+ *
+ * @details Built from the layout constants rather than written out, so the texel
+ * a shader reads and the texel the packer writes cannot drift apart.
+ * `esInstance*` is the name a stage reaches for; a stage that names one
+ * gets this header, the way a stage that names `tN` gets that unit.
+ */
+/** @brief The same record, for a WGSL vertex stage. @see instanceRecordGLSL */
+std::string instanceRecordWGSL() {
+ const std::string texels = std::to_string(MESH_INSTANCE_TEXELS);
+ const std::string width = std::to_string(MESH_INSTANCE_TEXTURE_WIDTH);
+ const std::string unit = std::to_string(MESH_INSTANCE_TEXTURE_UNIT);
+ return
+ "struct InstanceConstants { base : vec4u };\n"
+ "@group(0) @binding(" + std::to_string(INSTANCE_CONSTANTS_BINDING)
+ + ") var<uniform> inst : InstanceConstants;\n"
+ // vs_main's instance_index, put where the helpers can reach it — WGSL has
+ // no global for it the way GLSL has gl_InstanceID.
+ "var<private> g_instanceId : i32 = 0;\n"
+ "fn esInstanceTexel(texel : i32) -> vec4f {\n"
+ " let at = (i32(inst.base.x) + g_instanceId) * " + texels + " + texel;\n"
+ " return textureLoad(t" + unit + ", vec2i(at % " + width + ", at / " + width + "), 0);\n"
+ "}\n"
+ "fn esInstanceModel() -> mat4x4f {\n"
+ " let r0 = esInstanceTexel(" + std::to_string(MESH_INSTANCE_TEXEL_MODEL) + ");\n"
+ " let r1 = esInstanceTexel(" + std::to_string(MESH_INSTANCE_TEXEL_MODEL + 1) + ");\n"
+ " let r2 = esInstanceTexel(" + std::to_string(MESH_INSTANCE_TEXEL_MODEL + 2) + ");\n"
+ " return mat4x4f(vec4f(r0.x, r1.x, r2.x, 0.0), vec4f(r0.y, r1.y, r2.y, 0.0),\n"
+ " vec4f(r0.z, r1.z, r2.z, 0.0), vec4f(r0.w, r1.w, r2.w, 1.0));\n"
+ "}\n"
+ "fn esInstanceNormalMatrix() -> mat3x3f {\n"
+ " return mat3x3f(esInstanceTexel(" + std::to_string(MESH_INSTANCE_TEXEL_NORMAL) + ").xyz,\n"
+ " esInstanceTexel(" + std::to_string(MESH_INSTANCE_TEXEL_NORMAL + 1) + ").xyz,\n"
+ " esInstanceTexel(" + std::to_string(MESH_INSTANCE_TEXEL_NORMAL + 2) + ").xyz);\n"
+ "}\n"
+ "fn esInstanceTint() -> vec4f { return esInstanceTexel("
+ + std::to_string(MESH_INSTANCE_TEXEL_TINT) + "); }\n"
+ "fn esInstanceLightmapRect() -> vec4f { return esInstanceTexel("
+ + std::to_string(MESH_INSTANCE_TEXEL_LIGHTMAP) + "); }\n";
+}
+
+std::string instanceRecordGLSL() {
+ const std::string texels = std::to_string(MESH_INSTANCE_TEXELS);
+ const std::string width = std::to_string(MESH_INSTANCE_TEXTURE_WIDTH);
+ return
+ "layout(std140) uniform InstanceConstants {\n"
+ " highp uvec4 u_instanceBase;\n"
+ "};\n"
+ "uniform highp sampler2D u_instanceData;\n"
+ "highp vec4 esInstanceTexel(int texel) {\n"
+ " int at = (int(u_instanceBase.x) + ES_INSTANCE_ID) * " + texels + " + texel;\n"
+ " return texelFetch(u_instanceData, ivec2(at % " + width + ", at / " + width + "), 0);\n"
+ "}\n"
+ "highp mat4 esInstanceModel() {\n"
+ " highp vec4 r0 = esInstanceTexel(" + std::to_string(MESH_INSTANCE_TEXEL_MODEL) + ");\n"
+ " highp vec4 r1 = esInstanceTexel(" + std::to_string(MESH_INSTANCE_TEXEL_MODEL + 1) + ");\n"
+ " highp vec4 r2 = esInstanceTexel(" + std::to_string(MESH_INSTANCE_TEXEL_MODEL + 2) + ");\n"
+ " return mat4(vec4(r0.x, r1.x, r2.x, 0.0), vec4(r0.y, r1.y, r2.y, 0.0),\n"
+ " vec4(r0.z, r1.z, r2.z, 0.0), vec4(r0.w, r1.w, r2.w, 1.0));\n"
+ "}\n"
+ "highp mat3 esInstanceNormalMatrix() {\n"
+ " return mat3(esInstanceTexel(" + std::to_string(MESH_INSTANCE_TEXEL_NORMAL) + ").xyz,\n"
+ " esInstanceTexel(" + std::to_string(MESH_INSTANCE_TEXEL_NORMAL + 1) + ").xyz,\n"
+ " esInstanceTexel(" + std::to_string(MESH_INSTANCE_TEXEL_NORMAL + 2) + ").xyz);\n"
+ "}\n"
+ "highp vec4 esInstanceTint() { return esInstanceTexel("
+ + std::to_string(MESH_INSTANCE_TEXEL_TINT) + "); }\n"
+ "highp vec4 esInstanceLightmapRect() { return esInstanceTexel("
+ + std::to_string(MESH_INSTANCE_TEXEL_LIGHTMAP) + "); }\n";
+}
+
+
constexpr u32 kMaxIncludeDepth = 16;
std::string ltrim(const std::string& s) {
@@ -178,32 +253,17 @@ std::string canonicalVertexStageWGSL(bool lit) {
"#ifdef SKINNED\n"
" @location(5) a_joints : vec4u,\n"
" @location(6) a_weights : vec4f,\n"
- " @location(12) a_instTint : vec4f,\n"
- "#else\n"
- "#ifdef MESH\n"
- " @location(8) a_model0 : vec4f,\n"
- " @location(9) a_model1 : vec4f,\n"
- " @location(10) a_model2 : vec4f,\n"
- "#ifdef MESH_LIGHTMAP\n"
- " @location(11) a_lightmapRect : vec4f,\n"
- "#endif\n"
- " @location(12) a_instTint : vec4f,\n"
- "#endif\n"
- "#ifdef MESH_NORMALS\n"
- " @location(13) a_nrm0 : vec3f,\n"
- " @location(14) a_nrm1 : vec3f,\n"
- " @location(15) a_nrm2 : vec3f,\n"
- "#endif\n"
- "#endif\n"
"#endif\n"
"#ifdef MESH\n"
// What the GLSL twin reads as gl_VertexID.
" @builtin(vertex_index) vertexIndex : u32,\n"
"#endif\n"
+ "#endif\n"
"};\n"
"\n"
- "@vertex fn vs_main(v : VSIn) -> VSOut {\n"
+ "@vertex fn vs_main(v : VSIn, @builtin(instance_index) instanceId : u32) -> VSOut {\n"
" var out : VSOut;\n"
+ " g_instanceId = i32(instanceId);\n"
"#ifdef MESH\n"
" var local = v.a_position;\n"
"#ifdef MESH_NORMALS\n"
@@ -247,12 +307,10 @@ std::string canonicalVertexStageWGSL(bool lit) {
" + v.a_weights.z * skin.bones[v.a_joints.z]\n"
" + v.a_weights.w * skin.bones[v.a_joints.w];\n"
" let world = pose * vec4f(local, 1.0);\n"
- " out.v_color = v.a_color * v.a_instTint;\n"
+ " out.v_color = v.a_color * esInstanceTint();\n"
"#elif defined(MESH)\n"
- " let world = vec4f(dot(v.a_model0, vec4f(local, 1.0)),\n"
- " dot(v.a_model1, vec4f(local, 1.0)),\n"
- " dot(v.a_model2, vec4f(local, 1.0)), 1.0);\n"
- " out.v_color = v.a_color * v.a_instTint;\n"
+ " let world = esInstanceModel() * vec4f(local, 1.0);\n"
+ " out.v_color = v.a_color * esInstanceTint();\n"
"#else\n"
" let world = vec4f(v.a_position, 1.0);\n"
" out.v_color = v.a_color;\n"
@@ -268,16 +326,20 @@ std::string canonicalVertexStageWGSL(bool lit) {
"#ifdef SKINNED\n"
" out.v_worldNormal = mat3x3f(pose[0].xyz, pose[1].xyz, pose[2].xyz) * localNormal;\n"
"#else\n"
- " out.v_worldNormal = mat3x3f(v.a_nrm0, v.a_nrm1, v.a_nrm2) * localNormal;\n"
+ " out.v_worldNormal = esInstanceNormalMatrix() * localNormal;\n"
"#endif\n"
" out.v_worldXYZ = world.xyz;\n"
"#endif\n"
"#ifdef MESH_LIGHTMAP\n"
- " out.v_lightmap = vec3f(v.a_texCoord1 * v.a_lightmapRect.xy + v.a_lightmapRect.zw,\n"
- " select(0.0, 1.0, v.a_lightmapRect.x > 0.0));\n"
+ " let lmRect = esInstanceLightmapRect();\n"
+ " out.v_lightmap = vec3f(v.a_texCoord1 * lmRect.xy + lmRect.zw,\n"
+ " select(0.0, 1.0, lmRect.x > 0.0));\n"
"#endif\n";
if (lit) {
src += " out.v_worldPos = world.xy;\n";
+ // Declared by wgslVSOut under the same condition, so filled here —
+ // a slot left at zero makes every instance read the first probe.
+ src += " out.v_probeSlot = f32(instanceId);\n";
}
src +=
" return out;\n"
@@ -360,26 +422,10 @@ std::string canonicalVertexStage(bool lit) {
"#ifdef SKINNED\n"
"layout(location = 5) in uvec4 a_joints;\n"
"layout(location = 6) in vec4 a_weights;\n"
- "layout(location = 12) in vec4 a_instTint;\n"
// This draw's pose, rewritten immediately before it (SkinConstants, binding 5).
"layout(std140) uniform SkinConstants {\n"
" mat4 u_bones[64];\n"
"};\n"
- "#else\n"
- "#ifdef MESH\n"
- "layout(location = 8) in vec4 a_model0;\n"
- "layout(location = 9) in vec4 a_model1;\n"
- "layout(location = 10) in vec4 a_model2;\n"
- "#ifdef MESH_LIGHTMAP\n"
- "layout(location = 11) in vec4 a_lightmapRect;\n"
- "#endif\n"
- "layout(location = 12) in vec4 a_instTint;\n"
- "#endif\n"
- "#ifdef MESH_NORMALS\n"
- "layout(location = 13) in vec3 a_nrm0;\n"
- "layout(location = 14) in vec3 a_nrm1;\n"
- "layout(location = 15) in vec3 a_nrm2;\n"
- "#endif\n"
"#endif\n"
"#endif\n"
"#ifdef MESH\n"
@@ -460,12 +506,10 @@ std::string canonicalVertexStage(bool lit) {
" + a_weights.z * u_bones[a_joints.z]\n"
" + a_weights.w * u_bones[a_joints.w];\n"
" vec4 world = skin * vec4(local, 1.0);\n"
- " v_color = a_color * a_instTint;\n"
+ " v_color = a_color * esInstanceTint();\n"
"#elif defined(MESH)\n"
- " vec4 world = vec4(dot(a_model0, vec4(local, 1.0)),\n"
- " dot(a_model1, vec4(local, 1.0)),\n"
- " dot(a_model2, vec4(local, 1.0)), 1.0);\n"
- " v_color = a_color * a_instTint;\n"
+ " vec4 world = esInstanceModel() * vec4(local, 1.0);\n"
+ " v_color = a_color * esInstanceTint();\n"
"#else\n"
" vec4 world = vec4(a_position, 1.0);\n"
" v_color = a_color;\n"
@@ -481,13 +525,14 @@ std::string canonicalVertexStage(bool lit) {
"#ifdef SKINNED\n"
" v_worldNormal = mat3(skin) * localNormal;\n"
"#else\n"
- " v_worldNormal = mat3(a_nrm0, a_nrm1, a_nrm2) * localNormal;\n"
+ " v_worldNormal = esInstanceNormalMatrix() * localNormal;\n"
"#endif\n"
" v_worldXYZ = world.xyz;\n"
"#endif\n"
"#ifdef MESH_LIGHTMAP\n"
- " v_lightmap = vec3(a_texCoord1 * a_lightmapRect.xy + a_lightmapRect.zw,\n"
- " a_lightmapRect.x > 0.0 ? 1.0 : 0.0);\n"
+ " highp vec4 lmRect = esInstanceLightmapRect();\n"
+ " v_lightmap = vec3(a_texCoord1 * lmRect.xy + lmRect.zw,\n"
+ " lmRect.x > 0.0 ? 1.0 : 0.0);\n"
"#endif\n";
if (lit) {
src += " v_worldPos = world.xy;\n";
@@ -1426,7 +1471,7 @@ std::string trimWs(const std::string& s) {
// #ifdef/#ifndef/#else/#elif defined(NAME)/#endif over the assembled text,
// with the feature set as the defined names. GLSL keeps real #defines for the
// driver, so both targets see identical variant logic in the authored bodies.
-std::string preprocessWGSL(const std::string& source, const std::vector<std::string>& features) {
+std::string preprocessConditionals(const std::string& source, const std::vector<std::string>& features) {
const std::unordered_set<std::string> defined(features.begin(), features.end());
struct Frame {
bool parentActive; ///< Whether the enclosing region emits lines.
@@ -1509,7 +1554,7 @@ ShaderParser::AssembledStage assembleWGSLStage(const ParsedShader& parsed,
// injecting the shared headers would double-declare them.
auto fullIt = parsed.wgslStageFull.find(stage);
if (fullIt != parsed.wgslStageFull.end() && fullIt->second) {
- result.source = preprocessWGSL(bodyIt->second, features);
+ result.source = preprocessConditionals(bodyIt->second, features);
result.headerLineCount = 0;
return result;
}
@@ -1546,13 +1591,19 @@ ShaderParser::AssembledStage assembleWGSLStage(const ParsedShader& parsed,
if (lit) inject(kLitHeaderWGSL);
inject(kColorHelpersWGSL);
}
+ // The records, for a stage that names them — the same question, and the same
+ // answer, as the GLSL side. Before the texture completion below, so the unit
+ // this reaches for is one of the ones it sees.
+ if (stage == ShaderStage::Vertex && bodyIt->second.find("esInstance") != std::string::npos) {
+ inject(instanceRecordWGSL());
+ }
// Last, so it sees every declaration made above as well as the body's own.
inject(wgslReachedTextureDecls(assembled.str() + bodyIt->second,
parsed.domain == "PostProcess"));
assembled << bodyIt->second;
- result.source = preprocessWGSL(assembled.str(), features);
+ result.source = preprocessConditionals(assembled.str(), features);
result.headerLineCount = headerLines;
return result;
}
@@ -1688,18 +1739,30 @@ ShaderParser::AssembledStage ShaderParser::assembleStageEx(const ParsedShader& p
// carry explicit highp for the same reason MaterialConstants does: a fragment shader has no
// default float precision until its `precision` line, which follows this injected header.
- // Declared by every Lit vertex because every Lit fragment takes it in. One
- // that never fills it reads slot 0 (the fragment clamps), not garbage.
- if (stage == ShaderStage::Vertex && parsed.domain == "Lit") {
- // The twin generator compiles this stage under Vulkan semantics, where the
- // instance number is spelled differently and gl_InstanceID does not exist.
+ if (stage == ShaderStage::Vertex) {
+ // Every vertex stage, not only the Lit ones: an unlit mesh reads the same
+ // record. Spelled twice because the twin generator compiles this under
+ // Vulkan semantics, where gl_InstanceID does not exist.
assembled << "#ifdef VULKAN\n"
"#define ES_INSTANCE_ID gl_InstanceIndex\n"
"#else\n"
"#define ES_INSTANCE_ID gl_InstanceID\n"
- "#endif\n"
- "flat out highp float v_probeSlot;\n";
- headerLines += 6;
+ "#endif\n";
+ headerLines += 5;
+ // The records themselves, for a stage that names them. Reached-not-declared
+ // is how wgslReachedTextureDecls decides the same question: one author for
+ // the declaration, rather than a copy in every shader that reads a record.
+ if (stageIt->second.find("esInstance") != std::string::npos) {
+ const std::string decl = instanceRecordGLSL();
+ assembled << decl;
+ headerLines += static_cast<u32>(std::count(decl.begin(), decl.end(), '\n'));
+ }
+ }
+ // Declared by every Lit vertex because every Lit fragment takes it in. One
+ // that never fills it reads slot 0 (the fragment clamps), not garbage.
+ if (stage == ShaderStage::Vertex && parsed.domain == "Lit") {
+ assembled << "flat out highp float v_probeSlot;\n";
+ headerLines += 1;
}
// The light-array size and packing here MUST match renderer/LightConstants.hpp (MAX_LIGHTS,
@@ -2394,6 +2457,18 @@ void ShaderParser::computeMaterialLayout(ParsedShader& shader) {
for (auto& p : shader.properties) {
if (!p.fromParam) continue; // legacy properties-block entries are reflection-only
if (p.type == ShaderPropertyType::Texture) {
+ // The top unit is the frame's per-object record, on every mesh shader
+ // there is. Handing it out here would bind a material's texture over
+ // the records and draw every object at the origin.
+ if (textureUnit >= MESH_INSTANCE_TEXTURE_UNIT) {
+ ES_LOG_ERROR("shader {}: texture param \"{}\" would take unit {}, which is the"
+ " per-object record; a material reads at most {} textures",
+ shader.name, p.name, textureUnit,
+ MESH_INSTANCE_TEXTURE_UNIT - MATERIAL_TEXTURE_UNIT_BASE);
+ p.textureUnit = -1;
+ p.std140Offset = -1;
+ continue;
+ }
p.textureUnit = static_cast<i32>(textureUnit++);
p.std140Offset = -1;
continue;
diff --git a/tests/renderer/test_shader_device.cpp b/tests/renderer/test_shader_device.cpp
index 5d89d44c8..ffb63f9c2 100644
--- a/tests/renderer/test_shader_device.cpp
+++ b/tests/renderer/test_shader_device.cpp
@@ -7,6 +7,7 @@
#include "MockGfxDevice.hpp"
#include "esengine/renderer/rhi/Shader.hpp"
#include "esengine/renderer/store/LightConstants.hpp"
+#include "esengine/renderer/rhi/GfxEnums.hpp"
#include "esengine/renderer/store/MorphConstants.hpp"
#include <cstdio>
@@ -31,16 +32,16 @@ int main() {
CHECK(shader->getProgramId() == 1, "programId is the device-returned id");
CHECK(d.getActiveUniformsCalls == 1, "reflection routes through device.getActiveUniforms");
- // compile() pins the engine-injected samplers to their units, so the counts
- // below start from what that seeding left behind. Six: the shadow map, the
- // reflection, the two 2D masks, the morph deltas and the baked atlas.
- CHECK(d.setUniform1iCalls == 6
- && d.lastUniform1iVal == static_cast<i32>(LIGHTMAP_TEXTURE_UNIT),
+ // compile() pins every engine-injected sampler to its unit, so the counts
+ // below start from what that seeding left behind — seven of them, ending
+ // on the per-object records.
+ CHECK(d.setUniform1iCalls == 7
+ && d.lastUniform1iVal == static_cast<i32>(MESH_INSTANCE_TEXTURE_UNIT),
"compile pins the injected samplers to their texture units");
CHECK(d.useProgramCalls == 2, "seeding them binds and unbinds the program once");
shader->setUniform("u_tex", 3);
- CHECK(d.setUniform1iCalls == 7, "setUniform(name,int) routes through device.setUniform1i");
+ CHECK(d.setUniform1iCalls == 8, "setUniform(name,int) routes through device.setUniform1i");
CHECK(d.lastUniform1iVal == 3, "uniform value forwarded");
shader->setUniform("u_color", glm::vec4(1, 0, 0, 1));
diff --git a/tools/check-mesh-instance-record.mjs b/tools/check-mesh-instance-record.mjs
index 2eef97eae..dc34e2ca8 100644
--- a/tools/check-mesh-instance-record.mjs
+++ b/tools/check-mesh-instance-record.mjs
@@ -1,22 +1,24 @@
// SPDX-License-Identifier: Apache-2.0
// SPDX-FileCopyrightText: Copyright (c) 2024-present ESEngine Team
/**
- * @file check-mesh-instance-record.mjs — the per-object record says one thing.
+ * @file check-mesh-instance-record.mjs — the per-object record stays out of the slots.
*
- * The engine appends a per-object record to every resident mesh's layout, and
- * what it contains is written down FOUR times: mesh.esshader declares it in GLSL
- * and again in WGSL, and ShaderParser injects the same pair into a material's
- * mesh variant. ResourceManager builds the layout those four must match. Nothing
- * held them together, and dropping a row from the model matrix meant editing all
- * five by hand — one missed edit reads whatever the driver left in the slot.
+ * The record a resident mesh is drawn with lives in the frame's texture, read
+ * through the `esInstance*` helpers the assembler injects. That is what leaves
+ * every attribute slot above the mesh channels free, and it is a property with
+ * no compiler behind it: a shader that declares an instance attribute again
+ * compiles, draws, and quietly takes the room back.
*
- * Two claims, because they fail differently:
+ * Three claims, because they fail differently:
*
- * the four declarations disagree
- * a material draws its mesh through a layout built for the other one; the
- * frame is wrong and no error names the cause
- * the declarations and MESH_INSTANCE_STRIDE disagree
- * the record is read past its end, or a part of it is never sent
+ * an instance attribute is declared above the mesh channels
+ * the slots are spent again, and the next mesh channel or custom attribute
+ * has nowhere to go — nothing reports it until one is added
+ * a shader names an `esInstance*` helper the header does not define
+ * GLSL says so at compile; WGSL reports an invalid pipeline that names no
+ * shader, which is how a whole variant goes missing with no message
+ * the record's texels do not fit the record
+ * a part is written past the end of one object and read as the next one's
*
* Run: node tools/check-mesh-instance-record.mjs
*/
@@ -30,109 +32,92 @@ const read = (rel) => readFileSync(path.join(ROOT, rel), 'utf8');
const SHADER = 'src/esengine/data/shaders/mesh.esshader';
const PARSER = 'src/esengine/resource/ShaderParser.cpp';
const ENUMS = 'src/esengine/renderer/rhi/GfxEnums.hpp';
+/** The first slot a mesh's own channels may claim — the room this check keeps free. */
+const FIRST_FREE_SLOT = 8;
const enums = read(ENUMS);
const constant = (name) => {
const found = new RegExp(`${name}\\s*=\\s*(\\d+)`).exec(enums);
- if (!found) throw new Error(`${ENUMS} no longer declares ${name}`);
- return Number(found[1]);
+ if (found) return Number(found[1]);
+ throw new Error(`${ENUMS} no longer declares ${name}`);
};
-const FIRST = constant('MESH_INSTANCE_FIRST_LOCATION');
-const STRIDE = constant('MESH_INSTANCE_STRIDE');
-const STRIDE_LIT = constant('MESH_INSTANCE_STRIDE_LIT');
-const LIGHTMAP_BYTES = constant('MESH_INSTANCE_LIGHTMAP_BYTES');
+const TEXELS = constant('MESH_INSTANCE_TEXELS');
+const WIDTH = constant('MESH_INSTANCE_TEXTURE_WIDTH');
+const PARTS = {
+ MESH_INSTANCE_TEXEL_MODEL: 3,
+ MESH_INSTANCE_TEXEL_TINT: 1,
+ MESH_INSTANCE_TEXEL_NORMAL: 3,
+ MESH_INSTANCE_TEXEL_LIGHTMAP: 1,
+};
+
+const problems = [];
-/** Every instance-range attribute a source declares: location → name, per language. */
-function declarations(text) {
- const glsl = new Map();
- const wgsl = new Map();
+// Claim 1: nothing takes the slots back. Both languages, both sources — a
+// declaration in either one is a layout the other cannot serve anyway.
+for (const file of [SHADER, PARSER]) {
+ const text = read(file);
for (const [, loc, name] of text.matchAll(/layout\(location\s*=\s*(\d+)\)\s*in\s+\w+\s+(\w+)/g)) {
- if (Number(loc) >= FIRST) glsl.set(Number(loc), name);
+ if (Number(loc) >= FIRST_FREE_SLOT) {
+ problems.push(`${file} declares GLSL attribute "${name}" at location ${loc};`
+ + ` ${FIRST_FREE_SLOT} and up are free for a mesh's own channels`);
+ }
}
- for (const [, loc, name] of text.matchAll(/@location\((\d+)\)\s*(\w+)\s*:/g)) {
- if (Number(loc) >= FIRST) wgsl.set(Number(loc), name);
+ for (const [, loc, name] of text.matchAll(/@location\((\d+)\)\s+(a_\w+)\s*:/g)) {
+ if (Number(loc) >= FIRST_FREE_SLOT) {
+ problems.push(`${file} declares WGSL attribute "${name}" at location ${loc};`
+ + ` ${FIRST_FREE_SLOT} and up are free for a mesh's own channels`);
+ }
}
- return { glsl, wgsl };
}
-const problems = [];
-const sources = { [SHADER]: declarations(read(SHADER)), [PARSER]: declarations(read(PARSER)) };
-
-// All four must name the same thing at the same location. Compared as text so a
-// renamed attribute is a finding too: a name is how the twin's reader matches up.
-const spellings = new Map();
-for (const [file, langs] of Object.entries(sources)) {
- for (const [lang, found] of Object.entries(langs)) {
- for (const [loc, name] of found) {
- const seen = spellings.get(loc);
- if (!seen) { spellings.set(loc, { name, where: `${file} (${lang})` }); continue; }
- if (seen.name !== name) {
- problems.push(`location ${loc} is "${seen.name}" in ${seen.where}`
- + ` and "${name}" in ${file} (${lang}) — one of them reads the other's bytes`);
- }
+// Claim 2: every helper a shader reaches for is one the header defines. The
+// header is built in PARSER, so its own text answers both halves.
+const parser = read(PARSER);
+const defined = new Set([
+ ...[...parser.matchAll(/highp \w+ (esInstance\w+)\(/g)].map((m) => m[1]),
+ ...[...parser.matchAll(/fn (esInstance\w+)\(/g)].map((m) => m[1]),
+]);
+for (const file of [SHADER, PARSER]) {
+ for (const [, name] of read(file).matchAll(/\b(esInstance\w+)\s*\(/g)) {
+ if (!defined.has(name)) {
+ problems.push(`${file} names ${name}(), which the injected header does not define`);
}
}
}
-
-// Each declaration set must cover the same locations: a part declared in one
-// language and not its twin is a backend that draws the record differently.
-const every = [...spellings.keys()].sort((a, b) => a - b);
-for (const [file, langs] of Object.entries(sources)) {
- for (const [lang, found] of Object.entries(langs)) {
- const missing = every.filter((loc) => !found.has(loc));
- if (missing.length) {
- problems.push(`${file} (${lang}) declares no attribute at location(s) `
- + `${missing.join(', ')}, which the others do`);
- }
+for (const half of ['highp \\w+ (esInstance\\w+)\\(', 'fn (esInstance\\w+)\\(']) {
+ const found = new Set([...parser.matchAll(new RegExp(half, 'g'))].map((m) => m[1]));
+ const missing = [...defined].filter((n) => !found.has(n));
+ if (missing.length) {
+ problems.push(`${PARSER} defines ${missing.join(', ')} in one language and not the other`);
}
}
-/**
- * What the declarations add up to, against what the packer sends. Measured by
- * WHERE each part is rather than by how many there are: normals and a bake are
- * each optional, so a count alone cannot say which record a total is of.
- */
-const LIGHTMAP = FIRST + 3;
-const TINT = FIRST + 4;
-const NORMAL_FIRST = FIRST + 5;
-const modelRows = every.filter((loc) => loc < LIGHTMAP).length;
-const normalRows = every.filter((loc) => loc >= NORMAL_FIRST).length;
-const base = modelRows * 16 + 4;
-const lit = base + normalRows * 12;
-if (!spellings.has(TINT)) {
- problems.push(`no attribute at location ${TINT}, where the record's tint is`);
-}
-if (base !== STRIDE) {
- problems.push(`${modelRows} model row(s) plus the tint is ${base} bytes, but`
- + ` MESH_INSTANCE_STRIDE is ${STRIDE} — the record is packed to a different shape`
- + ' than it is declared');
-}
-if (lit !== STRIDE_LIT) {
- problems.push(`with ${normalRows} normal row(s) that is ${lit} bytes, but`
- + ` MESH_INSTANCE_STRIDE_LIT is ${STRIDE_LIT}`);
-}
-// The bake's rectangle is APPENDED, so its size is all the packer needs to agree
-// on: everything before it keeps the offset it had, which is what lets one
-// record serve an object with a bake and one without.
-if (spellings.has(LIGHTMAP) && LIGHTMAP_BYTES !== 16) {
- problems.push(`location ${LIGHTMAP} declares a vec4 atlas rectangle, which is 16 bytes,`
- + ` but MESH_INSTANCE_LIGHTMAP_BYTES is ${LIGHTMAP_BYTES}`);
+// Claim 3: the parts fit, do not overlap, and a record never straddles a row.
+const occupied = new Map();
+for (const [name, span] of Object.entries(PARTS)) {
+ const at = constant(name);
+ for (let i = at; i < at + span; ++i) {
+ if (i >= TEXELS) {
+ problems.push(`${name} reaches texel ${i}, past the ${TEXELS} a record holds`);
+ }
+ const seen = occupied.get(i);
+ if (seen) problems.push(`texel ${i} is both ${seen} and ${name}`);
+ occupied.set(i, name);
+ }
}
-if (!spellings.has(LIGHTMAP) && LIGHTMAP_BYTES !== 0) {
- problems.push(`MESH_INSTANCE_LIGHTMAP_BYTES is ${LIGHTMAP_BYTES}, but no attribute at`
- + ` location ${LIGHTMAP} carries an atlas rectangle — the bytes are sent and never read`);
+if (WIDTH % TEXELS !== 0) {
+ problems.push(`MESH_INSTANCE_TEXTURE_WIDTH is ${WIDTH}, not a multiple of the`
+ + ` ${TEXELS}-texel record — a record would straddle a row`);
}
if (problems.length) {
- console.error('check-mesh-instance-record: the per-object record is described'
- + ' more than one way.\n');
+ console.error('check-mesh-instance-record: the per-object record is not where it belongs.\n');
for (const p of problems) console.error(` ${p}`);
- console.error(`\nIt is declared in ${SHADER} (both languages) and injected by ${PARSER}`
- + ` (both languages); ${ENUMS} says how many bytes it is.`);
+ console.error(`\nThe record is read through the helpers ${PARSER} injects and written by`
+ + ` MeshPlugin; ${ENUMS} says what texel each part is at.`);
process.exit(1);
}
-const bake = spellings.has(LIGHTMAP) ? ` (+${LIGHTMAP_BYTES} where a bake is read)` : '';
-console.log(`check-mesh-instance-record: ${every.length} part(s) of the per-object record,`
- + ` spelled the same in all four declarations and adding to ${STRIDE}/${STRIDE_LIT}`
- + ` bytes${bake}.`);
+console.log(`check-mesh-instance-record: ${occupied.size} of ${TEXELS} texel(s) spoken for,`
+ + ` read through ${defined.size} helper(s) defined in both languages,`
+ + ` and no attribute slot from ${FIRST_FREE_SLOT} up is spent on them.`);
diff --git a/tools/check-shader-conditionals.mjs b/tools/check-shader-conditionals.mjs
index 170a211a0..e238e1598 100644
--- a/tools/check-shader-conditionals.mjs
+++ b/tools/check-shader-conditionals.mjs
@@ -5,7 +5,7 @@
* engine can actually resolve.
*
* A GLSL stage is preprocessed by the DRIVER, which has a full C preprocessor.
- * A WGSL twin is expanded by the engine (ShaderParser::preprocessWGSL), which
+ * A WGSL twin is expanded by the engine (ShaderParser::preprocessConditionals), which
* understands `#ifdef`, `#ifndef`, `#elif defined(X)`, `#else` and `#endif` —
* and nothing else. A `#if defined(A) && !defined(B)` there is not an error: the
* line is emitted verbatim into the WGSL, which then fails to compile, and what
@@ -13,6 +13,10 @@
*
* The two halves of a file therefore do NOT have the same powers, which is the
* kind of asymmetry nothing else would state.
+ *
+ * And a second claim, on the same expander: every `#ifdef` closes. An unclosed
+ * one swallows the whole rest of the module for the variants that take the dead
+ * branch — silently, because this expander does not report one.
*/
import { readFileSync, existsSync } from 'node:fs';
import { corpusRoots, censusFindings, sourceFiles } from './lib/sourceCensus.mjs';
@@ -45,13 +49,78 @@ for (const file of files) {
}
}
+/**
+ * The conditional directives of a text, in order, as `{ line, kind }`.
+ */
+function directives(text, firstLine) {
+ const out = [];
+ text.split('\n').forEach((raw, i) => {
+ const lead = raw.trim();
+ if (!lead.startsWith('#if') && !lead.startsWith('#else')
+ && !lead.startsWith('#elif') && !lead.startsWith('#endif')) return;
+ out.push({ line: firstLine + i, kind: lead.split(/[\s(]/)[0] });
+ });
+ return out;
+}
+
+/** What @p text leaves open, as a list of the lines that opened it. */
+function unclosed(text, firstLine) {
+ const open = [];
+ for (const d of directives(text, firstLine)) {
+ if (d.kind === '#ifdef' || d.kind === '#ifndef' || d.kind === '#if') open.push(d.line);
+ else if (d.kind === '#endif') open.pop();
+ }
+ return open;
+}
+
+for (const file of files) {
+ const text = readFileSync(path.join(ROOT, file), 'utf8');
+ for (const block of text.matchAll(/^#pragma\s+\w+\s+wgsl\s*$([\s\S]*?)^#pragma\s+end\s*$/gm)) {
+ const before = text.slice(0, block.index).split('\n').length;
+ for (const line of unclosed(block[1], before + 1)) {
+ problems.push(`${file}:${line} opens a conditional this block never closes`);
+ }
+ }
+}
+
+// The stages the engine BUILDS are shader sources too, and they run through the
+// same expander. Their conditionals live in C++ string literals, so the text a
+// variant sees is not one anybody reads whole.
+const BUILDER = 'src/esengine/resource/ShaderParser.cpp';
+let builders = 0;
+{
+ const cpp = readFileSync(path.join(ROOT, BUILDER), 'utf8');
+ // Literal contents, by the function that holds them. A hand-rolled scan
+ // because a comment on the way may carry a quote of its own.
+ for (const fn of cpp.matchAll(/^std::string (\w+)\([^)]*\)\s*\{$([\s\S]*?)^\}$/gm)) {
+ let text = '';
+ const body = fn[2];
+ for (let i = 0; i < body.length; ++i) {
+ if (body[i] === '/' && body[i + 1] === '/') { while (i < body.length && body[i] !== '\n') ++i; continue; }
+ if (body[i] !== '"') continue;
+ for (++i; i < body.length && body[i] !== '"'; ++i) {
+ if (body[i] !== '\\') { text += body[i]; continue; }
+ text += body[++i] === 'n' ? '\n' : body[i];
+ }
+ }
+ if (!/#if(def|ndef)?\b/.test(text)) continue;
+ ++builders;
+ if (unclosed(text, 0).length) {
+ problems.push(`${BUILDER} ${fn[1]}() leaves `
+ + `${unclosed(text, 0).length} conditional(s) open — every variant that takes`
+ + ' the dead branch loses the rest of the stage');
+ }
+ }
+}
+
if (problems.length) {
console.error('check-shader-conditionals: a WGSL block uses a conditional the engine'
+ " does not resolve — it reaches the compiler verbatim.\n");
for (const p of problems) console.error(` ${p}`);
- console.error('\nUse nested #ifdef / #ifndef, or #elif defined(X).');
+ console.error('\nUse nested #ifdef / #ifndef, or #elif defined(X), and close every one.');
process.exit(1);
}
console.log(`check-shader-conditionals: ${wgslBlocks} WGSL block(s) across ${files.length}`
- + ' shader(s) — every conditional is one the engine expands.');
+ + ` shader(s) and ${builders} built stage(s) — every conditional is one the engine`
+ + ' expands, and every one of them closes.');
diff --git a/tools/gen-shader-twins.mjs b/tools/gen-shader-twins.mjs
index 33e220a7b..ea5057901 100644
--- a/tools/gen-shader-twins.mjs
+++ b/tools/gen-shader-twins.mjs
@@ -56,6 +56,10 @@ const BLOCK_BINDINGS = {
MaterialConstants: 1,
LightConstants: 2,
TimeConstants: 3,
+ // Where a draw's objects start in the frame's record texture, which the
+ // injected header declares for every vertex stage that reads a record — see
+ // InstanceConstants.hpp.
+ InstanceConstants: 8,
// The indirect light where a draw stands, which the injected Lit header
// declares for every Lit fragment stage — see ProbeConstants.hpp.
ProbeConstants: 7,
@@ -77,8 +81,11 @@ const ENGINE_SAMPLERS = [
{ name: 'u_envMap', unit: 3 },
{ name: 'u_shadow2D', unit: 7 },
{ name: 'u_lightShape2D', unit: 6 },
- // The one the VERTEX stage reads, pinned in MorphConstants.hpp.
+ // The two the VERTEX stage reads, pinned in MorphConstants.hpp and
+ // InstanceConstants.hpp. The record sits on the TOP unit, above every unit a
+ // material's own textures are handed (see GfxEnums MESH_INSTANCE_TEXTURE_UNIT).
{ name: 'u_morphDeltas', unit: 4 },
+ { name: 'u_instanceData', unit: 15 },
];
// Group-1 unit map — mirrors WebGPUMappings textureBindingForUnit/samplerBindingForUnit.
@@ -375,7 +382,7 @@ const permutationKey = (on) => [...on].sort().join(',');
/**
* One body carrying every permutation, selected by the same `#ifdef` the GLSL
- * side gets from real defines — `preprocessWGSL` resolves it at assembly, so
+ * side gets from real defines — `preprocessConditionals` resolves it at assembly, so
* both targets see identical variant logic.
*/
export function composePermutations(toggles, bodies, remaining = toggles, on = []) {
diff --git a/tools/releaseNotes.mjs b/tools/releaseNotes.mjs
index 8ec96ff1e..9e9fb27d2 100644
--- a/tools/releaseNotes.mjs
+++ b/tools/releaseNotes.mjs
@@ -28,6 +28,9 @@ export const NOTES_FLOOR = 'v0.65.0';
* the reader recognises anyway.
*/
export const NOTED = {
+ 'feat(render): the per-object record leaves the vertex attributes':
+ { note: 'A mesh has somewhere to put its own vertex attributes.' },
+
'feat(render): indirect light per instance, so a probe stops refusing the merge':
{ note: 'A scene can hold a thousand of something again.' },