'use strict';
// Reproducible CPU/no-DOM benchmark. Prints timings/counts, never pixels/points.
const fs = require('node:fs'), path = require('node:path'), vm = require('node:vm'), zlib = require('node:zlib');
const root = path.resolve(__dirname, '..'), cache = {};
const context = vm.createContext({ console: { log() {}, warn() {}, error() {} }, setTimeout, clearTimeout });
function load(file) {
  file = path.resolve(file); if (cache[file]) return cache[file].exports;
  const module = { exports: {} }; cache[file] = module;
  const run = vm.runInContext('(function(exports,require,module){' + fs.readFileSync(file, 'utf8') + '\n})', context, { filename: file });
  run(module.exports, name => { if (!name.startsWith('.')) throw new Error('Unexpected runtime dependency'); return load(path.resolve(path.dirname(file), name + '.js')); }, module);
  return module.exports;
}
const fixture = JSON.parse(fs.readFileSync(path.join(root, 'test/fixtures/native-face-c3.json'), 'utf8'));
const source = new Uint8Array(zlib.gunzipSync(Buffer.from(fixture.gzipBase64, 'base64')));
const portrait = { width: 540, height: 960, data: new Uint8Array(540 * 960 * 4) };
for (let at = 3; at < portrait.data.length; at += 4) portrait.data[at] = 255;
for (let y = 0; y < 405; y++) for (let x = 0; x < 540; x++) {
  const at = (Math.floor(y * fixture.height / 405) * fixture.width + Math.floor(x * fixture.width / 540)) * 4;
  portrait.data.set(source.subarray(at, at + 4), ((y + 277) * 540 + x) * 4);
}
const repeatArg = process.argv.find(value => value.startsWith('--repeats='));
const repeats = repeatArg ? Math.max(1, Math.min(10, Number(repeatArg.split('=')[1]) || 2)) : 2;
const cpuOptimization = !process.argv.includes('--baseline-cpu');
const maxSideArg = process.argv.find(value => value.startsWith('--max-side='));
const maxSide = maxSideArg ? Number(maxSideArg.split('=')[1]) : 320;
const sizesArg = process.argv.find(value => value.startsWith('--input-sizes='));
const sizes = sizesArg ? sizesArg.split('=')[1].split(',').map(Number) : [256, 192, 160, 128];
const api = load(path.join(root, 'utils/native-face.js'));
(async () => {
  if (process.argv.includes('--guided')) {
    const seed = api.create({ inputSize: 256, landmarkModel: 'full' });
    const originalPortrait = await seed.detect(portrait);
    const reference = await seed.detect({ width: fixture.width, height: fixture.height, data: source });
    seed.dispose();
    const box = reference.detections[0].detection.box, scale = 960 * 0.45 / box.height;
    const centerX = box.x + box.width / 2, centerY = box.y + box.height / 2;
    for (let y = 0; y < 960; y++) for (let x = 0; x < 540; x++) {
      const sx = Math.floor((x - 270) / scale + centerX), sy = Math.floor((y - 480) / scale + centerY), to = (y * 540 + x) * 4;
      portrait.data.fill(0, to, to + 3); portrait.data[to + 3] = 255;
      if (sx >= 0 && sx < fixture.width && sy >= 0 && sy < fixture.height) portrait.data.set(source.subarray((sy * fixture.width + sx) * 4, (sy * fixture.width + sx) * 4 + 4), to);
    }
    process.stdout.write(JSON.stringify({ setup: 'guided-portrait', originalPortraitFaceHeightRatio: originalPortrait.detections[0].detection.box.height / 960, requestedFaceHeightRatio: 0.45 }) + '\n');
  }
  for (const landmarkModel of ['full', 'tiny']) for (const inputSize of sizes) {
    const engine = api.create({ defaultPixelFormat: 'RGBA', maxSide, inputSize, landmarkModel, cpuOptimization });
    const initialized = Date.now(); await engine.init(); const initMs = Date.now() - initialized;
    const samples = [];
    for (let index = 0; index < repeats; index++) {
      const result = await engine.detect(portrait);
      samples.push({ ...result.timings, faces: result.faceCount, landmarks: result.detections[0] ? result.detections[0].landmarks.positions.length : 0,
        faceHeightRatio: result.detections[0] ? result.detections[0].detection.box.height / result.height : 0 });
    }
    const mean = {};
    for (const key of ['snapshotMs', 'modelMs', 'tensorMs', 'detectorMs', 'landmarkMs', 'totalMs']) mean[key] = Math.round(samples.reduce((sum, item) => sum + item[key], 0) / samples.length);
    const memory = engine.stats(); engine.dispose();
    process.stdout.write(JSON.stringify({ landmarkModel, inputSize, maxSide, cpuOptimization, frame: '540x960', initMs, mean,
      allSamplesDetected: samples.every(item => item.faces > 0 && item.landmarks === 68), samples,
      modelBytes: memory.numBytes, modelTensors: memory.numTensors, afterDisposeTensors: engine.stats().numTensors }) + '\n');
  }
})().catch(error => { process.stderr.write(error.message + '\n'); process.exitCode = 1; });