@gnsx/three 0.185.11 → 0.185.13

package/examples/jsm/tsl/display/GTAONode.js

@@ -1,5 +1,5 @@
 import { DataTexture, RenderTarget, RepeatWrapping, Vector2, Vector3, TempNode, QuadMesh, NodeMaterial, RendererUtils, RedFormat } from '@gnsx/three/webgpu';
-import { reference, logarithmicDepthToViewZ, viewZToPerspectiveDepth, getNormalFromDepth, getScreenPosition, getViewPosition, nodeObject, Fn, float, NodeUpdateType, uv, uniform, Loop, vec2, vec3, vec4, int, dot, max, pow, abs, If, textureSize, sin, cos, PI, texture, passTexture, mat3, add, normalize, mul, cross, div, mix, acos, clamp } from '@gnsx/three/tsl';
+import { reference, logarithmicDepthToViewZ, viewZToPerspectiveDepth, getNormalFromDepth, getScreenPosition, getViewPosition, nodeObject, Fn, float, NodeUpdateType, uv, uniform, Loop, vec2, vec3, vec4, int, dot, max, min, pow, abs, If, textureSize, sin, cos, PI, texture, passTexture, mat3, add, normalize, mul, cross, div, mix, acos, clamp } from '@gnsx/three/tsl';
 
 const _quadMesh = /*@__PURE__*/ new QuadMesh();
 const _size = /*@__PURE__*/ new Vector2();
@@ -70,6 +70,18 @@ class GTAONode extends TempNode {
 		 */
 		this.normalNode = normalNode;
 
+		// WITH_GENESYS
+		/**
+		 * Optional node that marks pixels where GTAO should run. This is sampled only
+		 * for the center pixel, allowing neighbor depth taps to use the raw depth
+		 * texture without repeating an auxiliary validity texture fetch.
+		 *
+		 * @type {?Node<float>}
+		 * @default null
+		 */
+		this.validityNode = null;
+		// !WITH_GENESYS
+
 		/**
 		 * The resolution scale. By default the effect is rendered in full resolution
 		 * for best quality but a value of `0.5` should be sufficient for most scenes.
@@ -106,6 +118,18 @@ class GTAONode extends TempNode {
 		 */
 		this.radius = uniform( 0.25 );
 
+		// WITH_GENESYS
+		/**
+		 * Maximum screen-space radius in pixels. Clamps the projected world-space
+		 * radius when the camera is close to surfaces so sample UVs stay local and
+		 * avoid texture-cache thrashing (Activision GTAO guidance).
+		 *
+		 * @type {UniformNode<float>}
+		 * @default 64
+		 */
+		this.maxScreenRadius = uniform( 64 );
+		// !WITH_GENESYS
+
 		/**
 		 * The resolution of the effect. Can be scaled via
 		 * `resolutionScale`.
@@ -343,12 +367,30 @@ class GTAONode extends TempNode {
 
 			const depth = sampleDepth( uvNode ).toVar();
 
+			// WITH_GENESYS
+			if ( this.validityNode !== null ) {
+
+				this.validityNode.sample( uvNode ).lessThan( 0.5 ).discard();
+
+			}
+			// !WITH_GENESYS
+
 			depth.greaterThanEqual( 1.0 ).discard();
 
 			const viewPosition = getViewPosition( uvNode, depth, this._cameraProjectionMatrixInverse ).toVar();
 			const viewNormal = sampleNormal( uvNode ).toVar();
 
-			const radiusToUse = this.radius;
+			// WITH_GENESYS
+			// Clamp world radius so its screen footprint never exceeds maxScreenRadius
+			// pixels. Without this, near-camera surfaces project a huge UV stride and
+			// the depth/normal taps thrash the texture cache.
+			const probeScreen = getScreenPosition( viewPosition.add( vec3( this.radius, 0.0, 0.0 ) ), this._cameraProjectionMatrix );
+			const centerScreen = getScreenPosition( viewPosition, this._cameraProjectionMatrix );
+			const screenRadiusPx = abs( probeScreen.x.sub( centerScreen.x ) ).mul( this.resolution.x );
+			const radiusScale = min( float( 1.0 ), this.maxScreenRadius.div( max( screenRadiusPx, float( 1e-4 ) ) ) );
+			const radiusToUse = this.radius.mul( radiusScale );
+			// !WITH_GENESYS
+			// const radiusToUse = this.radius;
 
 			const noiseResolution = textureSize( this._noiseNode, 0 );
 			let noiseUv = vec2( uvNode.x, uvNode.y.oneMinus() );
@@ -360,8 +402,16 @@ class GTAONode extends TempNode {
 			const bitangent = vec3( tangent.y.mul( - 1.0 ), tangent.x, 0.0 );
 			const kernelMatrix = mat3( tangent, bitangent, vec3( 0.0, 0.0, 1.0 ) );
 
-			const DIRECTIONS = this.samples.lessThan( 30 ).select( 3, 5 ).toVar();
-			const STEPS = add( this.samples, DIRECTIONS.sub( 1 ) ).div( DIRECTIONS ).toVar();
+			// WITH_GENESYS
+			// A clamped radius covers less screen space, so proportionally reduce the
+			// sample budget as well. Keeping all samples only changes tap locality and
+			// does not address the near-camera shader cost.
+			const samplesToUse = min( this.samples, max( float( 6 ), this.samples.mul( radiusScale ) ) ).toVar();
+			const DIRECTIONS = samplesToUse.lessThan( 30 ).select( 3, 5 ).toVar();
+			const STEPS = add( samplesToUse, DIRECTIONS.sub( 1 ) ).div( DIRECTIONS ).toVar();
+			// !WITH_GENESYS
+			// const DIRECTIONS = this.samples.lessThan( 30 ).select( 3, 5 ).toVar();
+			// const STEPS = add( this.samples, DIRECTIONS.sub( 1 ) ).div( DIRECTIONS ).toVar();
 
 			const ao = float( 0 ).toVar();
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gnsx/three",
-  "version": "0.185.11",
+  "version": "0.185.13",
   "description": "JavaScript 3D library",
   "type": "module",
   "main": "./build/three.cjs",

package/src/profiler/ProfilerService.js

@@ -10,6 +10,7 @@ import { warnOnce } from '../utils.js';
  *   // play for a few seconds
  *   __gnsx_profiler.report()         // sorted console.table (aggregated stats)
  *   __gnsx_profiler.downloadTrace()  // download gnsx-trace.json for Speedscope / Perfetto
+ *   __gnsx_profiler.downloadTrace('gnsx-trace.json', 0.05)  // omit spans < 0.05 ms
  *   __gnsx_profiler.reset()          // clear samples and trace
  *   __gnsx_profiler.disable()
  *
@@ -1216,11 +1217,16 @@ class ProfilerServiceClass {
 	}
 
 	/**
+	 * @param {number} [minDurationMs=0] Omit complete spans shorter than this duration (ms). Capture is unaffected.
 	 * @return {import('./ProfilerService.js').ChromeTrace}
 	 */
-	exportChromeTrace() {
+	exportChromeTrace( minDurationMs = 0 ) {
 
-		const slices = [ ...this.traceEvents ];
+		const thresholdMs = Number.isFinite( minDurationMs ) && minDurationMs > 0 ? minDurationMs : 0;
+		const minDurUs = thresholdMs * 1000;
+		const slices = minDurUs > 0
+			? this.traceEvents.filter( e => ( e.dur ?? 0 ) >= minDurUs )
+			: [ ...this.traceEvents ];
 		slices.sort( ( a, b ) => {
 
 			if ( a.ts !== b.ts ) return a.ts - b.ts;
@@ -1298,8 +1304,9 @@ class ProfilerServiceClass {
 
 	/**
 	 * @param {string} [filename='gnsx-trace.json']
+	 * @param {number} [minDurationMs=0] Omit complete spans shorter than this duration (ms).
 	 */
-	downloadTrace( filename = 'gnsx-trace.json' ) {
+	downloadTrace( filename = 'gnsx-trace.json', minDurationMs = 0 ) {
 
 		if ( typeof document === 'undefined' ) {
 
@@ -1322,7 +1329,10 @@ class ProfilerServiceClass {
 
 		}
 
-		const json = JSON.stringify( this.exportChromeTrace() );
+		const thresholdMs = Number.isFinite( minDurationMs ) && minDurationMs > 0 ? minDurationMs : 0;
+		const trace = this.exportChromeTrace( thresholdMs );
+		const eventCount = trace.traceEvents.filter( e => e.ph === 'X' ).length;
+		const json = JSON.stringify( trace );
 		const blob = new Blob( [ json ], { type: 'application/json' } );
 		const url = URL.createObjectURL( blob );
 		const a = document.createElement( 'a' );
@@ -1330,7 +1340,10 @@ class ProfilerServiceClass {
 		a.download = filename;
 		a.click();
 		URL.revokeObjectURL( url );
-		console.log( `[ProfilerService] Trace downloaded: ${filename} (${this.traceEvents.length} events)` );
+		const thresholdNote = thresholdMs > 0
+			? ` (filtered ${this.traceEvents.length - eventCount} spans < ${thresholdMs} ms)`
+			: '';
+		console.log( `[ProfilerService] Trace downloaded: ${filename} (${eventCount} events${thresholdNote})` );
 
 	}
 

package/src/renderers/common/Renderer.js

@@ -1378,20 +1378,26 @@ class Renderer {
 	 */
 	render( scene, camera ) {
 
-		// WITH_GENESYS
-		ProfilerService.begin( 'Renderer.render' );
-		// !WITH_GENESYS
-
 		if ( this._initialized === false ) {
 
 			throw new Error( 'THREE.Renderer: .render() called before the backend is initialized. Use "await renderer.init();" before rendering.' );
 
 		}
 
-		this._renderScene( scene, camera );
-
 		// WITH_GENESYS
-		ProfilerService.end( 'Renderer.render' );
+		const label = `Renderer.render (${scene.name || scene.type})`;
+		ProfilerService.begin( label );
+		const handle = ProfilerService.beginGpu( label, this );
+		try {
+
+			this._renderScene( scene, camera );
+
+		} finally {
+
+			ProfilerService.endGpu( handle );
+			ProfilerService.end( label );
+
+		}
 		// !WITH_GENESYS
 
 	}
@@ -3436,7 +3442,7 @@ class Renderer {
 	_renderObjects( renderList, camera, scene, lightsNode, passId = null ) {
 
 		// WITH_GENESYS
-		const label = `${this._currentRenderObjectFunction?.name || 'renderObject'} (${renderList.length})`;
+		const label = `_renderObjects (${renderList.length})`;
 		ProfilerService.begin( label );
 		// !WITH_GENESYS
 
@@ -3671,6 +3677,11 @@ class Renderer {
 	 */
 	renderObject( object, scene, camera, geometry, material, group, lightsNode, clippingContext = null, passId = null ) {
 
+		// WITH_GENESYS
+		const label = `renderObject (${object.name || object.type} - ${material.name || material.type})`;
+		ProfilerService.begin( label );
+		// !WITH_GENESYS
+
 		let materialOverride = false;
 		let materialColorNode;
 		let materialDepthNode;
@@ -3782,6 +3793,10 @@ class Renderer {
 
 		object.onAfterRender( this, scene, camera, geometry, material, group );
 
+		// WITH_GENESYS
+		ProfilerService.end( label );
+		// !WITH_GENESYS
+
 	}
 
 	/**