Approximate cos with simd

gistfile1.dart

import "dart:typed_data";
import "dart:math";

class AMath {

  final Float32x4 a = new Float32x4.splat(0.9999932946);
  final Float32x4 b = new Float32x4.splat(-0.4999124376);
  final Float32x4 c = new Float32x4.splat(0.0414877472);
  final Float32x4 d = new Float32x4.splat(-0.0012712095);

  Float32x4 apcos(Float32x4 x) {
    Float32x4 x2 = x * x;
    return a + x2 * (b + x2 * (c + x2 * d));
  }

}

double test(Float64List rnd, int iter) {
  int rs = rnd.length;
  double total = 0.0;
  for (int j = 0; j < iter; j++) {
    for (int i = 0; i < rs; i++) {
      total += cos(rnd[i]);
    }
  }
  return total;
}

Float32x4 testp(AMath m, Float32x4List rndp, int iter) {
  int rs = rndp.length;
  Float32x4 total = new Float32x4.zero();
  for (int j = 0; j < iter; j++) {
    for (int i = 0; i < rs; i++) {
      total += m.apcos(rndp[i]);
    }
  }
  return total;
}

void main() {
  int rs = 1 << 20;
  int iter = 100;
  Float64List rnd = new Float64List(rs);
  Random r = new Random(1);
  // Create a random array with values 0-Pi/4
  for (int i = 0; i < rs; i++) {
    rnd[i] = r.nextDouble() * PI / 4;
  }
  // Create packed version of the same array
  Float32x4List rndp = new Float32x4List(rs >> 2);
  int j = 0;
  for (int i = 0; i < rs; i += 4) {
    rndp[j++] = new Float32x4(rnd[i], rnd[i + 1], rnd[i + 2], rnd[i + 3]);
  }

  // math.cos
  // Warm up
  test(rnd, 1);
  Stopwatch sw = new Stopwatch();
  sw.start();
  // Run the test
  test(rnd, iter);
  var e = sw.elapsedMilliseconds;
  print("math.cos time: $e");

  // Approximate simd cos
  // Warmup
  AMath m = new AMath();
  testp(m, rndp, 1);
  sw.reset();
  sw.start();
  // Run the test
  testp(m, rndp, iter);
  e = sw.elapsedMilliseconds;
  print("Approx cos time: $e");

  double maxErr = 0.0;
  for (int i = 0; i < rndp.length; i++) {
    Float32x4 ap = m.apcos(rndp[i]);
    j = i * 4;
    Float32x4 mv = new Float32x4(cos(rnd[j]), cos(rnd[j + 1]), cos(rnd[j + 2]),
        cos(rnd[j + 3]));
    Float32x4 d = (ap - mv).abs();
    maxErr = max(maxErr, max(max(d.x, d.y), max(d.z, d.w)));
  }
  print("Max err: $maxErr");
}