Performance and test improvements

README.md

@@ -12,7 +12,7 @@ and fairly fast (about 20 nanoseconds for a sample of 2d noise) and tree shakeab
 
 ## Demos
 
-- [Synth Wave Demo](https://29a.ch/sandbox/2022/simplex-noise-synthwave/) shown in the header 
+- [Synthwave Demo](https://29a.ch/sandbox/2022/simplex-noise-synthwave/) shown in the header 
 - Simple 2D plasma on [codepen.io](http://codepen.io/jwagner/pen/BNmpdm/?editors=001).
 - [3D voxel world generation](https://29a.ch/sandbox/2012/voxelworld/) example.
 - Film grain in [analog film emulator](https://29a.ch/film-emulator/).
@@ -90,11 +90,9 @@ So ~20 nanoseconds per call.
 
 ```
 $ node perf/index.js
-27745787.933336906
-init: 192,590 ops/sec ±1%
-noise2D: 57,928,891 ops/sec ±1%
-noise3D: 34,159,230 ops/sec ±0%
-noise4D: 24,589,786 ops/sec ±0%
+noise2D: 66,608,762 ops/sec ±0% 
+noise3D: 41,059,121 ops/sec ±0% 
+noise4D: 33,406,638 ops/sec ±0% 
 ```
 
 At least at a glance it also seems to be faster than 'fast-simplex-noise':
@@ -159,6 +157,8 @@ const simplex = {
   When combined with tree-shaking this helps with build sizes.
 - Removed the built in version of the alea PRNG to focus the library to do only one thing.
    If you want to continue to use it you'll have to install and import it separately.
+- Noise functions are a bit faster (~ 10-20%) due to using integers in some places
+- Noise values can be different from previous versions especially for inputs > 2^31
 - Test coverage is now at 100%
 
 ### 3.0.1

package.json

@@ -77,6 +77,6 @@
     "build": "./build.sh",
     "docs": "typedoc --excludePrivate --out public/docs simplex-noise.ts",
     "prepare": "npm run-script build",
-    "benchmark": "parcel build && node ./perf/benchmark.js"
+    "benchmark": "npm run build && cd perf && ./benchmark.sh"
   }
 }

perf/benchmark.sh

@@ -1,5 +1,6 @@
 #!/bin/sh
-sudo cpufreq-set -g performance
-# use core 0 + 4 (which is hyperthreading for core 0)
-sudo nice -n -10 sudo -u $USER taskset -c 0,4 `which node` --random_seed=1 --hash_seed=1 --expose-gc index.js
-sudo cpufreq-set -g powersave
+# using powersave yields more consistent results
+sudo cpupower frequency-set -g powersave
+# use core 0 + 16 (which is hyperthreading for core 0)
+sudo nice -n -10 sudo -u $USER taskset -c 0,16 `which node` --random_seed=1 --hash_seed=1 --expose-gc index.js
+sudo cpufreq-set -g schedutil

perf/index.js

@@ -1,13 +1,13 @@
 /* eslint-disable @typescript-eslint/no-var-requires */
 const Benchmark = require('benchmark');
 const Alea = require('alea');
-const {noiseFunction2D, noiseFunction3D, noiseFunction4D} = require('..');
+const {createNoise2D, createNoise3D, createNoise4D} = require('..');
 
 const invocationsPerRun = 8*8*8;
 const rnd = () => new Alea('seed');
-const noise2D = noiseFunction2D(rnd());
-const noise3D = noiseFunction3D(rnd());
-const noise4D = noiseFunction4D(rnd());
+const noise2D = createNoise2D(rnd());
+const noise3D = createNoise3D(rnd());
+const noise4D = createNoise4D(rnd());
 
 // prevent the compiler from optimizing away the calls
 let sideEffect = 0.0;
@@ -23,7 +23,7 @@ new Benchmark.Suite('simplex-noise')
       }
     }
     sideEffect += a;
-  })
+  }, {minTime: 30})
   .add('noise3D', function() {
     let a = 0.0;
     for (let x = 0; x < 8; x++) {
@@ -58,4 +58,4 @@ new Benchmark.Suite('simplex-noise')
       );
     });
   })
-  .run({delay: 10, minTime: 20, maxTime: 25});
+  .run({delay: 10, minTime: 50, maxTime: 120, minSamples: 10000});

simplex-noise.ts

@@ -34,7 +34,30 @@ const G3 = 1.0 / 6.0;
 const F4 = (Math.sqrt(5.0) - 1.0) / 4.0;
 const G4 = (5.0 - Math.sqrt(5.0)) / 20.0;
 
-const grad3 = new Float32Array([1, 1, 0,
+// I'm really not sure why this | 0 (basically a coercion to int)
+// is making this faster but I get ~5 million ops/sec more on the
+// benchmarks across the board or a ~10% speedup.
+const fastFloor = (x: number) => Math.floor(x) | 0;
+
+const grad2 = new Float64Array([1, 1,
+  -1, 1,
+  1, -1,
+
+  -1, -1,
+  1, 0,
+  -1, 0,
+
+  1, 0,
+  -1, 0,
+  0, 1,
+
+  0, -1,
+  0, 1,
+  0, -1]);
+
+// double seems to be faster than single or int's
+// probably because most operations are in double precision
+const grad3 = new Float64Array([1, 1, 0,
   -1, 1, 0,
   1, -1, 0,
 
@@ -50,7 +73,8 @@ const grad3 = new Float32Array([1, 1, 0,
   0, 1, -1,
   0, -1, -1]);
 
-const grad4 = new Float32Array([0, 1, 1, 1, 0, 1, 1, -1, 0, 1, -1, 1, 0, 1, -1, -1,
+// double is a bit quicker here as well
+const grad4 = new Float64Array([0, 1, 1, 1, 0, 1, 1, -1, 0, 1, -1, 1, 0, 1, -1, -1,
   0, -1, 1, 1, 0, -1, 1, -1, 0, -1, -1, 1, 0, -1, -1, -1,
   1, 0, 1, 1, 1, 0, 1, -1, 1, 0, -1, 1, 1, 0, -1, -1,
   -1, 0, 1, 1, -1, 0, 1, -1, -1, 0, -1, 1, -1, 0, -1, -1,
@@ -79,15 +103,18 @@ export type NoiseFunction2D = (x: number, y: number) => number;
  */
 export function createNoise2D(random: RandomFn = Math.random): NoiseFunction2D {
   const perm = buildPermutationTable(random);
-  const permMod12 = perm.map(v => v % 12);
+  // precalculating this yields a little ~3% performance improvement.
+  const permGrad2x = new Float64Array(perm).map(v => grad2[(v % 12) * 2]);
+  const permGrad2y = new Float64Array(perm).map(v => grad2[(v % 12) * 2 + 1]);
   return function noise2D(x: number, y: number): number {
+    // if(!isFinite(x) || !isFinite(y)) return 0;
     let n0 = 0; // Noise contributions from the three corners
     let n1 = 0;
     let n2 = 0;
     // Skew the input space to determine which simplex cell we're in
     const s = (x + y) * F2; // Hairy factor for 2D
-    const i = Math.floor(x + s);
-    const j = Math.floor(y + s);
+    const i = fastFloor(x + s);
+    const j = fastFloor(y + s);
     const t = (i + j) * G2;
     const X0 = i - t; // Unskew the cell origin back to (x,y) space
     const Y0 = j - t;
@@ -117,21 +144,30 @@ export function createNoise2D(random: RandomFn = Math.random): NoiseFunction2D {
     // Calculate the contribution from the three corners
     let t0 = 0.5 - x0 * x0 - y0 * y0;
     if (t0 >= 0) {
-      const gi0 = permMod12[ii + perm[jj]] * 3;
+      const gi0 = ii + perm[jj];
+      const g0x = permGrad2x[gi0];
+      const g0y = permGrad2y[gi0];
       t0 *= t0;
-      n0 = t0 * t0 * (grad3[gi0] * x0 + grad3[gi0 + 1] * y0); // (x,y) of grad3 used for 2D gradient
+      // n0 = t0 * t0 * (grad2[gi0] * x0 + grad2[gi0 + 1] * y0); // (x,y) of grad3 used for 2D gradient
+      n0 = t0 * t0 * (g0x * x0 + g0y * y0);
     }
     let t1 = 0.5 - x1 * x1 - y1 * y1;
     if (t1 >= 0) {
-      const gi1 = permMod12[ii + i1 + perm[jj + j1]] * 3;
+      const gi1 = ii + i1 + perm[jj + j1];
+      const g1x = permGrad2x[gi1];
+      const g1y = permGrad2y[gi1];
       t1 *= t1;
-      n1 = t1 * t1 * (grad3[gi1] * x1 + grad3[gi1 + 1] * y1);
+      // n1 = t1 * t1 * (grad2[gi1] * x1 + grad2[gi1 + 1] * y1);
+      n1 = t1 * t1 * (g1x * x1 + g1y * y1);
     }
     let t2 = 0.5 - x2 * x2 - y2 * y2;
     if (t2 >= 0) {
-      const gi2 = permMod12[ii + 1 + perm[jj + 1]] * 3;
+      const gi2 = ii + 1 + perm[jj + 1];
+      const g2x = permGrad2x[gi2];
+      const g2y = permGrad2y[gi2];
       t2 *= t2;
-      n2 = t2 * t2 * (grad3[gi2] * x2 + grad3[gi2 + 1] * y2);
+      // n2 = t2 * t2 * (grad2[gi2] * x2 + grad2[gi2 + 1] * y2);
+      n2 = t2 * t2 * (g2x * x2 + g2y * y2);
     }
     // Add contributions from each corner to get the final noise value.
     // The result is scaled to return values in the interval [-1,1].
@@ -155,15 +191,13 @@ export type NoiseFunction3D = (x: number, y: number, z: number) => number;
  */
 export function createNoise3D(random: RandomFn = Math.random): NoiseFunction3D {
   const perm = buildPermutationTable(random);
-  const permMod12 = perm.map(v => v % 12);
-
   return function noise3D(x: number, y: number, z: number): number {
     let n0, n1, n2, n3; // Noise contributions from the four corners
     // Skew the input space to determine which simplex cell we're in
     const s = (x + y + z) * F3; // Very nice and simple skew factor for 3D
-    const i = Math.floor(x + s);
-    const j = Math.floor(y + s);
-    const k = Math.floor(z + s);
+    const i = fastFloor(x + s);
+    const j = fastFloor(y + s);
+    const k = fastFloor(z + s);
     const t = (i + j + k) * G3;
     const X0 = i - t; // Unskew the cell origin back to (x,y,z) space
     const Y0 = j - t;
@@ -248,28 +282,28 @@ export function createNoise3D(random: RandomFn = Math.random): NoiseFunction3D {
     let t0 = 0.6 - x0 * x0 - y0 * y0 - z0 * z0;
     if (t0 < 0) n0 = 0.0;
     else {
-      const gi0 = permMod12[ii + perm[jj + perm[kk]]] * 3;
+      const gi0 = (perm[ii + perm[jj + perm[kk]]] % 12) * 3;
       t0 *= t0;
       n0 = t0 * t0 * (grad3[gi0] * x0 + grad3[gi0 + 1] * y0 + grad3[gi0 + 2] * z0);
     }
     let t1 = 0.6 - x1 * x1 - y1 * y1 - z1 * z1;
     if (t1 < 0) n1 = 0.0;
     else {
-      const gi1 = permMod12[ii + i1 + perm[jj + j1 + perm[kk + k1]]] * 3;
+      const gi1 = (perm[ii + i1 + perm[jj + j1 + perm[kk + k1]]] % 12) * 3;
       t1 *= t1;
       n1 = t1 * t1 * (grad3[gi1] * x1 + grad3[gi1 + 1] * y1 + grad3[gi1 + 2] * z1);
     }
     let t2 = 0.6 - x2 * x2 - y2 * y2 - z2 * z2;
     if (t2 < 0) n2 = 0.0;
     else {
-      const gi2 = permMod12[ii + i2 + perm[jj + j2 + perm[kk + k2]]] * 3;
+      const gi2 = (perm[ii + i2 + perm[jj + j2 + perm[kk + k2]]] % 12) * 3;
       t2 *= t2;
       n2 = t2 * t2 * (grad3[gi2] * x2 + grad3[gi2 + 1] * y2 + grad3[gi2 + 2] * z2);
     }
     let t3 = 0.6 - x3 * x3 - y3 * y3 - z3 * z3;
     if (t3 < 0) n3 = 0.0;
     else {
-      const gi3 = permMod12[ii + 1 + perm[jj + 1 + perm[kk + 1]]] * 3;
+      const gi3 = (perm[ii + 1 + perm[jj + 1 + perm[kk + 1]]] % 12) * 3;
       t3 *= t3;
       n3 = t3 * t3 * (grad3[gi3] * x3 + grad3[gi3 + 1] * y3 + grad3[gi3 + 2] * z3);
     }
@@ -300,10 +334,10 @@ export function createNoise4D(random: RandomFn = Math.random) {
     let n0, n1, n2, n3, n4; // Noise contributions from the five corners
     // Skew the (x,y,z,w) space to determine which cell of 24 simplices we're in
     const s = (x + y + z + w) * F4; // Factor for 4D skewing
-    const i = Math.floor(x + s);
-    const j = Math.floor(y + s);
-    const k = Math.floor(z + s);
-    const l = Math.floor(w + s);
+    const i = fastFloor(x + s);
+    const j = fastFloor(y + s);
+    const k = fastFloor(z + s);
+    const l = fastFloor(w + s);
     const t = (i + j + k + l) * G4; // Factor for 4D unskewing
     const X0 = i - t; // Unskew the cell origin back to (x,y,z,w) space
     const Y0 = j - t;

test/matches-snapshot.ts

@@ -68,4 +68,10 @@ export function sampleFunctionToImageData(f: SampleFunction, width: number, heig
     }
   }
   return imageData;
+}
+
+// same as sampleFunctionToImageData but x and y go from -1 .. 1 instead of 0 .. width
+// output is scaled from -1 .. 1 to 0 .. 255
+export function sampleFunctionToImageDataOne(f: SampleFunction, width: number, height: number): ImageDataLike {
+  return sampleFunctionToImageData((x, y) => f(x / width * 2 - 1, y / height * 2 - 1) * 128 + 127, width, height);
 }

test/simplex-noise-test.ts

@@ -4,7 +4,7 @@ import alea from 'alea';
 import { assert } from 'chai';
 
 
-import { assertMatchesImage, sampleFunctionToImageData } from './matches-snapshot';
+import { assertMatchesImage, sampleFunctionToImageDataOne } from './matches-snapshot';
 
 function getRandom(seed = 'seed') {
   return alea(seed);
@@ -91,9 +91,20 @@ describe('createNoise2D', () => {
     it('should return similar values for similar inputs', function () {
       assert(Math.abs(noise2D(0.1, 0.2) - noise2D(0.101, 0.202)) < 0.1);
     });
-    it('should match snapshot', function () {
-      const actual = sampleFunctionToImageData((x, y) => noise2D(x / 10, y / 10) * 128 + 128, 64, 64);
-      assertMatchesImage(actual, 'noise2D.png');
+    it('should match snapshot for small inputs', function () {
+      const size = 64;
+      const actual = sampleFunctionToImageDataOne((x, y) => noise2D(x * 2, y * 2), size, size);
+      assertMatchesImage(actual, 'noise2Dsmall.png');
+    });
+    it('should match snapshot for large inputs', function () {
+      const size = 64;
+      const actual = sampleFunctionToImageDataOne((x, y) => noise2D(x * 1000, y * 1000), size, size);
+      assertMatchesImage(actual, 'noise2Dlarge.png');
+    });
+    it('should match snapshot for gigantic inputs', function () {
+      const size = 64;
+      const actual = sampleFunctionToImageDataOne((x, y) => noise2D(x * 5e9, y * 5e9), size, size);
+      assertMatchesImage(actual, 'noise2Dgiga.png');
     });
   });
 });
@@ -131,9 +142,15 @@ describe('createNoise3D', () => {
     it('should return similar values for similar inputs', function () {
       assert(Math.abs(noise3D(0.1, 0.2, 0.3) - noise3D(0.101, 0.202, 0.303)) < 0.1);
     });
-    it('should match snapshot', function () {
-      const actual = sampleFunctionToImageData((x, y) => noise3D(x / 10, y / 10, (x + y) / 2) * 128 + 128, 64, 64);
-      assertMatchesImage(actual, 'noise3D.png');
+    it('should match snapshot for small inputs', function () {
+      const size = 64;
+      const actual = sampleFunctionToImageDataOne((x, y) => noise3D(x * 2, y * 2, (x + y)), size, size);
+      assertMatchesImage(actual, 'noise3Dsmall.png');
+    });
+    it('should match snapshot for large inputs', function () {
+      const size = 64;
+      const actual = sampleFunctionToImageDataOne((x, y) => noise3D(x * 1000, y * 1000, (x + y) * 500), size, size);
+      assertMatchesImage(actual, 'noise3Dlarge.png');
     });
   });
 });
@@ -171,9 +188,15 @@ describe('createNoise4D', () => {
     it('should return similar values for similar inputs', function () {
       assert(Math.abs(noise4D(0.1, 0.2, 0.3, 0.4) - noise4D(0.101, 0.202, 0.303, 0.404)) < 0.1);
     });
-    it('should match snapshot', function () {
-      const actual = sampleFunctionToImageData((x, y) => noise4D(x / 10, y / 10, x / 4, y / 3) * 128 + 128, 64, 64);
-      assertMatchesImage(actual, 'noise4D.png');
+    it('should match snapshot for small inputs', function () {
+      const size = 64;
+      const actual = sampleFunctionToImageDataOne((x, y) => noise4D(x * 2, y * 2, x + y, x - y), size, size);
+      assertMatchesImage(actual, 'noise4Dsmall.png');
+    });
+    it('should match snapshot for large inputs', function () {
+      const size = 64;
+      const actual = sampleFunctionToImageDataOne((x, y) => noise4D(x * 1000, y * 1000, (x + y) * 500, (x - y) * 500), size, size);
+      assertMatchesImage(actual, 'noise4Dlarge.png');
     });
   });
 });