rayzee 9.4.2 → 9.5.0

package/src/Processor/TreeletOptimizer.js

@@ -1,574 +0,0 @@
-// No external dependencies — uses inline float fields for bounds (no Vector3)
-
-/**
- * Treelet optimization for BVH trees (Bittner et al. 2013).
- *
- * For each small subtree ("treelet") of N leaves, enumerates all
- * Catalan(N-1) distinct binary tree topologies, tries all leaf
- * permutations (or greedy assignment for N > 5), and replaces the
- * subtree with the arrangement that minimises SAH cost.
- */
-export class TreeletOptimizer {
-
-	constructor( traversalCost, intersectionCost ) {
-
-		this.traversalCost = traversalCost;
-		this.intersectionCost = intersectionCost;
-		this.maxTreeletLeaves = 7;
-		this.minImprovement = 0.02; // relative improvement threshold
-
-		// Precompute topologies for leaf counts 3..maxTreeletLeaves
-		this.topologyCache = new Map();
-		for ( let n = 3; n <= this.maxTreeletLeaves; n ++ ) {
-
-			this.topologyCache.set( n, this.generateTopologies( n ) );
-
-		}
-
-		// Stats
-		this.stats = {
-			treeletsProcessed: 0,
-			treeletsImproved: 0,
-			totalSAHImprovement: 0,
-			averageSAHImprovement: 0,
-			optimizationTime: 0
-		};
-
-	}
-
-	// ------------------------------------------------------------------
-	// Topology generation — nested binary trees via Catalan decomposition
-	// ------------------------------------------------------------------
-
-	/**
-	 * Generate all distinct binary tree topologies for `n` leaves.
-	 * Leaves are represented as integers 0..n-1 (slot indices).
-	 * Internal nodes are 2-element arrays [left, right].
-	 *
-	 * Examples for n=3 (Catalan(2)=2):
-	 *   [[0,1],2]   and   [0,[1,2]]
-	 */
-	generateTopologies( n ) {
-
-		if ( n === 1 ) return [ 0 ];
-		if ( n === 2 ) return [[ 0, 1 ]];
-
-		const results = [];
-
-		// Split n leaves into left (k) and right (n-k) groups.
-		// Base cases return arrays of topologies:
-		//   n=1 → [0]        (one topology: leaf 0)
-		//   n=2 → [[0,1]]    (one topology: pair [0,1])
-		// So `for (const t of topos)` yields individual topologies in all cases.
-		for ( let k = 1; k < n; k ++ ) {
-
-			const leftTopos = this.generateTopologies( k );
-			const rightTopos = this.generateTopologies( n - k );
-
-			for ( const lt of leftTopos ) {
-
-				for ( const rt of rightTopos ) {
-
-					results.push( [ lt, this.offsetTopology( rt, k ) ] );
-
-				}
-
-			}
-
-		}
-
-		return results;
-
-	}
-
-	/**
-	 * Offset all leaf indices in a topology by `offset`.
-	 */
-	offsetTopology( topo, offset ) {
-
-		if ( typeof topo === 'number' ) return topo + offset;
-		return [ this.offsetTopology( topo[ 0 ], offset ), this.offsetTopology( topo[ 1 ], offset ) ];
-
-	}
-
-	// ------------------------------------------------------------------
-	// Main entry point
-	// ------------------------------------------------------------------
-
-	optimizeBVH( bvhRoot ) {
-
-		const startTime = performance.now();
-		const maxTime = 30000; // 30s safety timeout
-
-		this.stats = {
-			treeletsProcessed: 0,
-			treeletsImproved: 0,
-			totalSAHImprovement: 0,
-			averageSAHImprovement: 0,
-			optimizationTime: 0
-		};
-
-		// Identify all treelet roots (bottom-up, non-overlapping)
-		const treeletRoots = this.identifyTreeletRoots( bvhRoot );
-
-		for ( let i = 0; i < treeletRoots.length; i ++ ) {
-
-			if ( performance.now() - startTime > maxTime ) {
-
-				console.warn( `TreeletOptimizer: timeout after ${i}/${treeletRoots.length} treelets` );
-				break;
-
-			}
-
-			this.optimizeTreelet( treeletRoots[ i ] );
-
-		}
-
-		this.stats.optimizationTime = performance.now() - startTime;
-		this.stats.averageSAHImprovement = this.stats.treeletsProcessed > 0
-			? this.stats.totalSAHImprovement / this.stats.treeletsProcessed
-			: 0;
-
-		return bvhRoot;
-
-	}
-
-	// ------------------------------------------------------------------
-	// Treelet identification — bottom-up, non-overlapping
-	// ------------------------------------------------------------------
-
-	identifyTreeletRoots( bvhRoot ) {
-
-		const roots = [];
-		const processed = new Set();
-
-		// Post-order traversal: children before parents so we process
-		// bottom-up and skip subtrees already covered by a deeper treelet.
-		const stack = [ { node: bvhRoot, visited: false } ];
-
-		while ( stack.length > 0 ) {
-
-			const top = stack[ stack.length - 1 ];
-
-			if ( top.visited ) {
-
-				stack.pop();
-				const node = top.node;
-
-				// Skip leaves
-				if ( node.triangleCount > 0 ) continue;
-				// Skip if already inside a processed treelet
-				if ( processed.has( node ) ) continue;
-
-				const leafCount = this.countLeaves( node );
-				if ( leafCount >= 3 && leafCount <= this.maxTreeletLeaves ) {
-
-					roots.push( node );
-					this.markSubtree( node, processed );
-
-				}
-
-			} else {
-
-				top.visited = true;
-				const node = top.node;
-				if ( node.triangleCount > 0 ) continue;
-				if ( node.rightChild ) stack.push( { node: node.rightChild, visited: false } );
-				if ( node.leftChild ) stack.push( { node: node.leftChild, visited: false } );
-
-			}
-
-		}
-
-		return roots;
-
-	}
-
-	countLeaves( node ) {
-
-		if ( ! node ) return 0;
-		if ( node.triangleCount > 0 ) return 1;
-		return this.countLeaves( node.leftChild ) + this.countLeaves( node.rightChild );
-
-	}
-
-	markSubtree( node, set ) {
-
-		if ( ! node ) return;
-		set.add( node );
-		if ( node.triangleCount > 0 ) return;
-		this.markSubtree( node.leftChild, set );
-		this.markSubtree( node.rightChild, set );
-
-	}
-
-	// ------------------------------------------------------------------
-	// Single treelet optimization
-	// ------------------------------------------------------------------
-
-	optimizeTreelet( treeletRoot ) {
-
-		// Extract leaves
-		const leaves = [];
-		this.extractLeaves( treeletRoot, leaves );
-		const n = leaves.length;
-
-		if ( n < 3 || n > this.maxTreeletLeaves ) return;
-
-		this.stats.treeletsProcessed ++;
-
-		const originalCost = this.evaluateSubtreeSAH( treeletRoot );
-		const topologies = this.topologyCache.get( n );
-		if ( ! topologies || topologies.length === 0 ) return;
-
-		let bestCost = originalCost;
-		let bestTopo = null;
-		let bestPerm = null;
-
-		if ( n <= 5 ) {
-
-			// Full permutation search — feasible for n<=5 (max 120 perms × 14 topos)
-			const perms = this.generatePermutations( n );
-
-			for ( const topo of topologies ) {
-
-				for ( const perm of perms ) {
-
-					const cost = this.evaluateTopology( topo, leaves, perm );
-					if ( cost < bestCost ) {
-
-						bestCost = cost;
-						bestTopo = topo;
-						bestPerm = perm;
-
-					}
-
-				}
-
-			}
-
-		} else {
-
-			// Greedy assignment for n>5: for each topology, use identity perm
-			// plus a few SAH-guided swaps
-			const identityPerm = Array.from( { length: n }, ( _, i ) => i );
-
-			for ( const topo of topologies ) {
-
-				// Try identity permutation
-				const cost = this.evaluateTopology( topo, leaves, identityPerm );
-				if ( cost < bestCost ) {
-
-					bestCost = cost;
-					bestTopo = topo;
-					bestPerm = identityPerm;
-
-				}
-
-				// Try greedy swap improvements
-				const result = this.greedySwapOptimize( topo, leaves, identityPerm, cost );
-				if ( result.cost < bestCost ) {
-
-					bestCost = result.cost;
-					bestTopo = topo;
-					bestPerm = result.perm;
-
-				}
-
-			}
-
-		}
-
-		// Apply if relative improvement exceeds threshold
-		const relativeImprovement = ( originalCost - bestCost ) / originalCost;
-		if ( bestTopo && relativeImprovement > this.minImprovement ) {
-
-			this.reconstructTreelet( treeletRoot, bestTopo, leaves, bestPerm );
-			this.stats.treeletsImproved ++;
-			this.stats.totalSAHImprovement += relativeImprovement;
-
-		}
-
-	}
-
-	extractLeaves( node, leaves ) {
-
-		if ( ! node ) return;
-
-		if ( node.triangleCount > 0 ) {
-
-			// BVHNode uses inline floats (minX/maxX etc.)
-			leaves.push( {
-				minX: node.minX, minY: node.minY, minZ: node.minZ,
-				maxX: node.maxX, maxY: node.maxY, maxZ: node.maxZ,
-				triangleOffset: node.triangleOffset,
-				triangleCount: node.triangleCount
-			} );
-			return;
-
-		}
-
-		this.extractLeaves( node.leftChild, leaves );
-		this.extractLeaves( node.rightChild, leaves );
-
-	}
-
-	// ------------------------------------------------------------------
-	// SAH evaluation
-	// ------------------------------------------------------------------
-
-	evaluateSubtreeSAH( node ) {
-
-		if ( ! node ) return 0;
-
-		if ( node.triangleCount > 0 ) {
-
-			return this.surfaceAreaFlat( node.minX, node.minY, node.minZ, node.maxX, node.maxY, node.maxZ ) * node.triangleCount * this.intersectionCost;
-
-		}
-
-		const leftCost = this.evaluateSubtreeSAH( node.leftChild );
-		const rightCost = this.evaluateSubtreeSAH( node.rightChild );
-		const sa = this.surfaceAreaFlat( node.minX, node.minY, node.minZ, node.maxX, node.maxY, node.maxZ );
-		return sa * this.traversalCost + leftCost + rightCost;
-
-	}
-
-	/**
-	 * Evaluate SAH cost for a topology with a given leaf permutation.
-	 * Returns { cost, boundsMin, boundsMax } for internal use, or just cost number.
-	 */
-	evaluateTopology( topo, leaves, perm ) {
-
-		const result = this.evalTopoRecursive( topo, leaves, perm );
-		return result.cost;
-
-	}
-
-	evalTopoRecursive( topo, leaves, perm ) {
-
-		if ( typeof topo === 'number' ) {
-
-			// Leaf slot — look up actual leaf via permutation
-			const leaf = leaves[ perm[ topo ] ];
-			return {
-				cost: this.surfaceAreaFlat( leaf.minX, leaf.minY, leaf.minZ, leaf.maxX, leaf.maxY, leaf.maxZ ) * leaf.triangleCount * this.intersectionCost,
-				minX: leaf.minX, minY: leaf.minY, minZ: leaf.minZ,
-				maxX: leaf.maxX, maxY: leaf.maxY, maxZ: leaf.maxZ
-			};
-
-		}
-
-		const left = this.evalTopoRecursive( topo[ 0 ], leaves, perm );
-		const right = this.evalTopoRecursive( topo[ 1 ], leaves, perm );
-
-		const mnX = Math.min( left.minX, right.minX );
-		const mnY = Math.min( left.minY, right.minY );
-		const mnZ = Math.min( left.minZ, right.minZ );
-		const mxX = Math.max( left.maxX, right.maxX );
-		const mxY = Math.max( left.maxY, right.maxY );
-		const mxZ = Math.max( left.maxZ, right.maxZ );
-
-		const sa = this.surfaceAreaFlat( mnX, mnY, mnZ, mxX, mxY, mxZ );
-		return {
-			cost: sa * this.traversalCost + left.cost + right.cost,
-			minX: mnX, minY: mnY, minZ: mnZ,
-			maxX: mxX, maxY: mxY, maxZ: mxZ
-		};
-
-	}
-
-	surfaceAreaFlat( minX, minY, minZ, maxX, maxY, maxZ ) {
-
-		const dx = maxX - minX;
-		const dy = maxY - minY;
-		const dz = maxZ - minZ;
-		return 2 * ( dx * dy + dy * dz + dz * dx );
-
-	}
-
-	// ------------------------------------------------------------------
-	// Permutation helpers
-	// ------------------------------------------------------------------
-
-	generatePermutations( n ) {
-
-		const result = [];
-		const arr = Array.from( { length: n }, ( _, i ) => i );
-
-		const permute = ( start ) => {
-
-			if ( start === n ) {
-
-				result.push( [ ...arr ] );
-				return;
-
-			}
-
-			for ( let i = start; i < n; i ++ ) {
-
-				[ arr[ start ], arr[ i ] ] = [ arr[ i ], arr[ start ] ];
-				permute( start + 1 );
-				[ arr[ start ], arr[ i ] ] = [ arr[ i ], arr[ start ] ];
-
-			}
-
-		};
-
-		permute( 0 );
-		return result;
-
-	}
-
-	/**
-	 * Greedy pairwise swap optimization for large treelet sizes.
-	 * Starting from an initial permutation, try all pairwise swaps
-	 * and accept any that improve cost. Repeat until no improvement.
-	 */
-	greedySwapOptimize( topo, leaves, initialPerm, initialCost ) {
-
-		const perm = [ ...initialPerm ];
-		let cost = initialCost;
-		let improved = true;
-
-		while ( improved ) {
-
-			improved = false;
-
-			for ( let i = 0; i < perm.length - 1; i ++ ) {
-
-				for ( let j = i + 1; j < perm.length; j ++ ) {
-
-					// Swap
-					[ perm[ i ], perm[ j ] ] = [ perm[ j ], perm[ i ] ];
-					const newCost = this.evaluateTopology( topo, leaves, perm );
-
-					if ( newCost < cost ) {
-
-						cost = newCost;
-						improved = true;
-
-					} else {
-
-						// Swap back
-						[ perm[ i ], perm[ j ] ] = [ perm[ j ], perm[ i ] ];
-
-					}
-
-				}
-
-			}
-
-		}
-
-		return { perm, cost };
-
-	}
-
-	// ------------------------------------------------------------------
-	// Reconstruction — builds proper BVHNode instances
-	// ------------------------------------------------------------------
-
-	reconstructTreelet( treeletRoot, topo, leaves, perm ) {
-
-		const built = this.buildSubtree( topo, leaves, perm );
-
-		// Copy built structure into the existing treelet root node (inline floats)
-		treeletRoot.minX = built.minX; treeletRoot.minY = built.minY; treeletRoot.minZ = built.minZ;
-		treeletRoot.maxX = built.maxX; treeletRoot.maxY = built.maxY; treeletRoot.maxZ = built.maxZ;
-		treeletRoot.leftChild = built.leftChild;
-		treeletRoot.rightChild = built.rightChild;
-		treeletRoot.triangleOffset = built.triangleOffset;
-		treeletRoot.triangleCount = built.triangleCount;
-
-	}
-
-	/**
-	 * Recursively build a BVHNode subtree from a topology + leaf permutation.
-	 */
-	buildSubtree( topo, leaves, perm ) {
-
-		if ( typeof topo === 'number' ) {
-
-			// Leaf — create a lightweight node with inline bounds
-			const leaf = leaves[ perm[ topo ] ];
-			const node = new TreeletBVHNode();
-			node.minX = leaf.minX; node.minY = leaf.minY; node.minZ = leaf.minZ;
-			node.maxX = leaf.maxX; node.maxY = leaf.maxY; node.maxZ = leaf.maxZ;
-			node.triangleOffset = leaf.triangleOffset;
-			node.triangleCount = leaf.triangleCount;
-			return node;
-
-		}
-
-		const left = this.buildSubtree( topo[ 0 ], leaves, perm );
-		const right = this.buildSubtree( topo[ 1 ], leaves, perm );
-
-		const node = new TreeletBVHNode();
-		node.leftChild = left;
-		node.rightChild = right;
-		node.minX = Math.min( left.minX, right.minX );
-		node.minY = Math.min( left.minY, right.minY );
-		node.minZ = Math.min( left.minZ, right.minZ );
-		node.maxX = Math.max( left.maxX, right.maxX );
-		node.maxY = Math.max( left.maxY, right.maxY );
-		node.maxZ = Math.max( left.maxZ, right.maxZ );
-
-		return node;
-
-	}
-
-	// ------------------------------------------------------------------
-	// Configuration
-	// ------------------------------------------------------------------
-
-	setTreeletSize( size ) {
-
-		this.maxTreeletLeaves = Math.max( 3, Math.min( 7, size ) );
-
-		// Regenerate topology cache if needed
-		for ( let n = 3; n <= this.maxTreeletLeaves; n ++ ) {
-
-			if ( ! this.topologyCache.has( n ) ) {
-
-				this.topologyCache.set( n, this.generateTopologies( n ) );
-
-			}
-
-		}
-
-	}
-
-	setMinImprovement( threshold ) {
-
-		this.minImprovement = Math.max( 0.001, threshold );
-
-	}
-
-	getStatistics() {
-
-		return { ...this.stats };
-
-	}
-
-}
-
-/**
- * Lightweight BVH node used during reconstruction.
- * Duck-type compatible with the main BVHNode class —
- * uses inline float fields (minX/maxX etc.).
- */
-class TreeletBVHNode {
-
-	constructor() {
-
-		this.minX = 0; this.minY = 0; this.minZ = 0;
-		this.maxX = 0; this.maxY = 0; this.maxZ = 0;
-		this.leftChild = null;
-		this.rightChild = null;
-		this.triangleOffset = 0;
-		this.triangleCount = 0;
-
-	}
-
-}