add new feature combineKernels

and fix a return on cpu when in graphic mode
gpujs · May 17, 2017 · 323a4e1 · 323a4e1
1 parent acc3651
commit 323a4e1
Show file tree

Hide file tree

Showing 5 changed files with 99 additions and 0 deletions.
diff --git a/src/backend/cpu/kernel.js b/src/backend/cpu/kernel.js
@@ -95,6 +95,7 @@ module.exports = class CPUKernel extends KernelBase {
 			if (this.graphical) {
 				this._imageData.data.set(this._colorData);
 				this._canvasCtx.putImageData(this._imageData, 0, 0);
+				return;
 			}
 
 			if (this.dimensions.length === 1) {

diff --git a/src/gpu.js b/src/gpu.js
@@ -81,6 +81,51 @@ module.exports = class GPU {
 		return this._kernelSynchronousExecutor = this._runner.buildKernel(fn, settings || {});
 	}
 
+	combineKernels() {
+		const lastKernel = arguments[arguments.length - 2];
+		const combinedKernel = arguments[arguments.length - 1];
+		if (this.mode === 'cpu') return combinedKernel;
+
+		const canvas = arguments[0].canvas;
+		let webGl =  arguments[0].webGl;
+
+		for (let i = 0; i < arguments.length - 1; i++) {
+			arguments[i]
+				.setCanvas(canvas)
+				.setWebGl(webGl)
+				.setOutputToTexture(true);
+		}
+
+		return function() {
+			combinedKernel.apply(null, arguments);
+			const texSize = lastKernel.texSize;
+			const gl = lastKernel.webGl;
+			const threadDim = lastKernel.threadDim;
+			let result;
+			if (lastKernel.floatOutput) {
+				result = new Float32Array(texSize[0] * texSize[1] * 4);
+				gl.readPixels(0, 0, texSize[0], texSize[1], gl.RGBA, gl.FLOAT, result);
+			} else {
+				const bytes = new Uint8Array(texSize[0] * texSize[1] * 4);
+				gl.readPixels(0, 0, texSize[0], texSize[1], gl.RGBA, gl.UNSIGNED_BYTE, bytes);
+				result = Float32Array.prototype.slice.call(new Float32Array(bytes.buffer));
+			}
+
+			result = result.subarray(0, threadDim[0] * threadDim[1] * threadDim[2]);
+
+			if (lastKernel.dimensions.length === 1) {
+				return result;
+			} else if (lastKernel.dimensions.length === 2) {
+				return utils.splitArray(result, lastKernel.dimensions[0]);
+			} else if (lastKernel.dimensions.length === 3) {
+				const cube = utils.splitArray(result, lastKernel.dimensions[0] * lastKernel.dimensions[1]);
+				return cube.map(function(x) {
+					return utils.splitArray(x, lastKernel.dimensions[0]);
+				});
+			}
+		};
+	}
+
 	get mode() {
 		return this._runner.mode;
 	}

diff --git a/test/html/features/combine-kernels.html b/test/html/features/combine-kernels.html
@@ -0,0 +1,17 @@
+<!DOCTYPE html>
+<html>
+<head>
+	<meta charset="utf-8">
+	<title>GPU.JS : Combine Kernels</title>
+	<link rel="stylesheet" href="../../../node_modules/qunitjs/qunit/qunit.css">
+
+	<!-- gpu.js scripts -->
+	<script src="../../../bin/gpu.js"></script>
+</head>
+<body>
+	<div id="qunit"></div>
+	<div id="qunit-fixture"></div>
+	<script src="../../../node_modules/qunitjs/qunit/qunit.js"></script>
+	<script src="../../src/features/combine-kernels.js"></script>
+</body>
+</html>
diff --git a/test/html/test-all.html b/test/html/test-all.html
@@ -24,6 +24,7 @@
 	<script src="../src/features/add-custom-function.js"></script>
 	<script src="../src/features/nested-function.js"></script>
 	<script src="../src/features/get-canvas.js"></script>
+  <script src="../src/features/combine-kernels.js"></script>
 
 	<!-- internal -->
 	<script src="../src/internal/utils.js"></script>

diff --git a/test/src/features/combine-kernels.js b/test/src/features/combine-kernels.js
@@ -0,0 +1,35 @@
+function combineKernels(mode) {
+	var gpu = new GPU({ mode: mode });
+
+	var kernel1 = gpu.createKernel(function(a, b) {
+		return a[this.thread.x] + b[this.thread.x];
+	}, { dimensions: [5] });
+
+	var kernel2 = gpu.createKernel(function(c, d) {
+		return c[this.thread.x] * d[this.thread.x];
+	}, { dimensions: [5] });
+
+	return gpu.combineKernels(kernel1, kernel2, function(array1, array2, array3) {
+		return kernel2(kernel1(array1, array2), array3);
+	});
+}
+
+QUnit.test( "combineKernels (auto)", function() {
+	var superKernel = combineKernels(null);
+	var result = QUnit.extend([], superKernel([1,2,3,4,5], [1,2,3,4,5], [1,2,3,4,5]));
+	QUnit.assert.deepEqual(result, [2, 8, 18, 32, 50]);
+});
+
+QUnit.test( "combineKernels (WebGL)", function() {
+	var superKernel = combineKernels('webgl');
+	var result = QUnit.extend([], superKernel([1,2,3,4,5], [1,2,3,4,5], [1,2,3,4,5]));
+	QUnit.assert.deepEqual(result, [2, 8, 18, 32, 50]);
+});
+
+QUnit.test( "combineKernels (CPU)", function() {
+	var superKernel = combineKernels('cpu');
+	var result = QUnit.extend([], superKernel([1,2,3,4,5], [1,2,3,4,5], [1,2,3,4,5]));
+	QUnit.assert.deepEqual(result, [2, 8, 18, 32, 50]);
+});
+
+console.log(combineKernels('gpu')([1,2,3,4,5], [1,2,3,4,5], [1,2,3,4,5]));