/**
 * Single-workgroup exclusive prefix-sum of the project shader's
 * per-splat `coverage[]` into `emitOffset[]`. Also writes the total
 * pair count into `totalPairs[0]` so downstream kernels and the radix
 * sort can size their dispatches without a CPU round trip.
 *
 * Layout: 256 threads, each processes `SCAN_PER_THREAD` elements
 * serially (chosen so that 256 × SCAN_PER_THREAD ≥ chunkCap). Phase 1
 * computes a per-thread partial sum; phase 2 has thread 0 scan the 256
 * partials in shared memory (negligible vs the 800-element block work);
 * phase 3 each thread re-walks its block writing the exclusive prefix.
 *
 * @param scanPerThread - Per-thread element budget; must satisfy `256 * scanPerThread >= chunkCap`.
 * @returns WGSL source for the prefix-sum compute shader.
 */
declare const prefixSumWgsl: (scanPerThread: number) => string;
export { prefixSumWgsl };