mirror of
https://github.com/gpujs/gpu.js.git
synced 2025-12-08 20:35:56 +00:00
258 lines
7.1 KiB
JavaScript
258 lines
7.1 KiB
JavaScript
const { assert, skip, test, module: describe, only } = require('qunit');
|
|
const { GPU } = require('../../src');
|
|
|
|
describe('issue # 233');
|
|
|
|
//TODO: Write for 2D and 3D and textures
|
|
//TODO: Write for pipeline as well
|
|
function kernelMapSinglePrecision(mode) {
|
|
const lst = [1, 2, 3, 4, 5, 6, 7];
|
|
const gpu = new GPU({ mode });
|
|
const kernels = gpu.createKernelMap({
|
|
stepA: function (x) {
|
|
return x * x;
|
|
},
|
|
stepB: function (x) {
|
|
return x + 1;
|
|
}
|
|
}, function (lst) {
|
|
const val = lst[this.thread.x];
|
|
|
|
stepA(val);
|
|
stepB(val);
|
|
|
|
return val;
|
|
}, {
|
|
precision: 'single',
|
|
output: [lst.length]
|
|
});
|
|
|
|
const result = kernels(lst);
|
|
const unwrap = gpu.createKernel(function(x) {
|
|
return x[this.thread.x];
|
|
}, {
|
|
output: [lst.length],
|
|
precision: 'single',
|
|
optimizeFloatMemory: true,
|
|
});
|
|
const stepAResult = unwrap(result.stepA);
|
|
const stepBResult = unwrap(result.stepB);
|
|
|
|
assert.deepEqual(Array.from(stepAResult), lst.map((x) => x * x));
|
|
assert.deepEqual(Array.from(stepBResult), lst.map((x) => x + 1));
|
|
assert.deepEqual(Array.from(result.result), lst);
|
|
gpu.destroy();
|
|
}
|
|
|
|
(GPU.isSinglePrecisionSupported && GPU.isKernelMapSupported ? test : skip)('Issue #233 - kernel map with single precision auto', () => {
|
|
kernelMapSinglePrecision();
|
|
});
|
|
|
|
(GPU.isSinglePrecisionSupported && GPU.isKernelMapSupported ? test : skip)('Issue #233 - kernel map with single precision gpu', () => {
|
|
kernelMapSinglePrecision('gpu');
|
|
});
|
|
|
|
(GPU.isSinglePrecisionSupported && GPU.isWebGLSupported ? test : skip)('Issue #233 - kernel map with single precision webgl', () => {
|
|
kernelMapSinglePrecision('webgl');
|
|
});
|
|
|
|
(GPU.isWebGL2Supported ? test : skip)('Issue #233 - kernel map with single precision webgl2', () => {
|
|
kernelMapSinglePrecision('webgl2');
|
|
});
|
|
|
|
(GPU.isHeadlessGLSupported && GPU.isKernelMapSupported ? test : skip)('Issue #233 - kernel map with single precision headlessgl', () => {
|
|
kernelMapSinglePrecision('headlessgl');
|
|
});
|
|
|
|
test('Issue #233 - kernel map with single precision cpu', () => {
|
|
kernelMapSinglePrecision('cpu');
|
|
});
|
|
|
|
|
|
function kernelMapSinglePrecision2D(mode) {
|
|
const lst = [
|
|
[1,2,3],
|
|
[4,5,6],
|
|
[7,8,9]
|
|
];
|
|
const stepAExpected = [
|
|
[1,4,9],
|
|
[16,25,36],
|
|
[49,64,81],
|
|
];
|
|
const stepBExpected = [
|
|
[2,3,4],
|
|
[5,6,7],
|
|
[8,9,10]
|
|
];
|
|
const gpu = new GPU({ mode });
|
|
const kernels = gpu.createKernelMap({
|
|
stepA: function (x) {
|
|
return x * x;
|
|
},
|
|
stepB: function (x) {
|
|
return x + 1;
|
|
}
|
|
}, function (lst) {
|
|
const val = lst[this.thread.y][this.thread.x];
|
|
|
|
stepA(val);
|
|
stepB(val);
|
|
|
|
return val;
|
|
}, {
|
|
precision: 'single',
|
|
output: [3, 3]
|
|
});
|
|
|
|
const result = kernels(lst);
|
|
assert.deepEqual(result.stepA.map(v => Array.from(v)), stepAExpected);
|
|
assert.deepEqual(result.stepB.map(v => Array.from(v)), stepBExpected);
|
|
assert.deepEqual(result.result.map(v => Array.from(v)), lst);
|
|
const memoryOptimize = gpu.createKernel(function(x) {
|
|
return x[this.thread.y][this.thread.x];
|
|
}, {
|
|
output: [3, 3],
|
|
precision: 'single',
|
|
optimizeFloatMemory: true,
|
|
});
|
|
const stepAOptimized = memoryOptimize(result.stepA);
|
|
const stepBOptimized = memoryOptimize(result.stepB);
|
|
const resultOptimized = memoryOptimize(result.result);
|
|
|
|
assert.deepEqual(stepAOptimized.map(v => Array.from(v)), stepAExpected);
|
|
assert.deepEqual(stepBOptimized.map(v => Array.from(v)), stepBExpected);
|
|
assert.deepEqual(resultOptimized.map(v => Array.from(v)), lst);
|
|
gpu.destroy();
|
|
}
|
|
|
|
(GPU.isSinglePrecisionSupported && GPU.isKernelMapSupported ? test : skip)('Issue #233 - kernel map with single precision 2d auto', () => {
|
|
kernelMapSinglePrecision2D();
|
|
});
|
|
|
|
(GPU.isSinglePrecisionSupported && GPU.isKernelMapSupported ? test : skip)('Issue #233 - kernel map with single precision 2d gpu', () => {
|
|
kernelMapSinglePrecision2D('gpu');
|
|
});
|
|
|
|
(GPU.isSinglePrecisionSupported && GPU.isWebGLSupported ? test : skip)('Issue #233 - kernel map with single precision 2d webgl', () => {
|
|
kernelMapSinglePrecision2D('webgl');
|
|
});
|
|
|
|
(GPU.isWebGL2Supported ? test : skip)('Issue #233 - kernel map with single precision 2d webgl2', () => {
|
|
kernelMapSinglePrecision2D('webgl2');
|
|
});
|
|
|
|
(GPU.isHeadlessGLSupported && GPU.isKernelMapSupported ? test : skip)('Issue #233 - kernel map with single precision 2d headlessgl', () => {
|
|
kernelMapSinglePrecision2D('headlessgl');
|
|
});
|
|
|
|
test('Issue #233 - kernel map with single precision 2d cpu', () => {
|
|
kernelMapSinglePrecision2D('cpu');
|
|
});
|
|
|
|
function kernelMapSinglePrecision3D(mode) {
|
|
const lst = [
|
|
[
|
|
[1,2,3],
|
|
[4,5,6],
|
|
[7,8,9]
|
|
],
|
|
[
|
|
[10,11,12],
|
|
[13,14,15],
|
|
[16,17,18]
|
|
]
|
|
];
|
|
const stepAExpected = [
|
|
[
|
|
[1,4,9],
|
|
[16,25,36],
|
|
[49,64,81],
|
|
],
|
|
[
|
|
[100,121,144],
|
|
[169,196,225],
|
|
[256,289,324],
|
|
]
|
|
];
|
|
const stepBExpected = [
|
|
[
|
|
[2,3,4],
|
|
[5,6,7],
|
|
[8,9,10]
|
|
],
|
|
[
|
|
[11,12,13],
|
|
[14,15,16],
|
|
[17,18,19]
|
|
]
|
|
];
|
|
const gpu = new GPU({ mode });
|
|
const kernels = gpu.createKernelMap({
|
|
stepA: function (x) {
|
|
return x * x;
|
|
},
|
|
stepB: function (x) {
|
|
return x + 1;
|
|
}
|
|
}, function (lst) {
|
|
const val = lst[this.thread.z][this.thread.y][this.thread.x];
|
|
|
|
stepA(val);
|
|
stepB(val);
|
|
|
|
return val;
|
|
}, {
|
|
precision: 'single',
|
|
output: [3, 3, 2]
|
|
});
|
|
|
|
const result = kernels(lst);
|
|
assert.deepEqual(arrayFromCube(result.stepA), stepAExpected);
|
|
assert.deepEqual(arrayFromCube(result.stepB), stepBExpected);
|
|
assert.deepEqual(arrayFromCube(result.result), lst);
|
|
const memoryOptimize = gpu.createKernel(function(x) {
|
|
return x[this.thread.z][this.thread.y][this.thread.x];
|
|
}, {
|
|
output: [3, 3, 2],
|
|
precision: 'single',
|
|
optimizeFloatMemory: true,
|
|
});
|
|
const stepAOptimized = memoryOptimize(result.stepA);
|
|
const stepBOptimized = memoryOptimize(result.stepB);
|
|
const resultOptimized = memoryOptimize(result.result);
|
|
|
|
assert.deepEqual(arrayFromCube(stepAOptimized), stepAExpected);
|
|
assert.deepEqual(arrayFromCube(stepBOptimized), stepBExpected);
|
|
assert.deepEqual(arrayFromCube(resultOptimized), lst);
|
|
|
|
function arrayFromCube(cube) {
|
|
return cube.map(matrix => matrix.map(row => Array.from(row)));
|
|
}
|
|
gpu.destroy();
|
|
}
|
|
|
|
(GPU.isSinglePrecisionSupported && GPU.isKernelMapSupported ? test : skip)('Issue #233 - kernel map with single precision 3d auto', () => {
|
|
kernelMapSinglePrecision3D();
|
|
});
|
|
|
|
(GPU.isSinglePrecisionSupported && GPU.isKernelMapSupported ? test : skip)('Issue #233 - kernel map with single precision 3d gpu', () => {
|
|
kernelMapSinglePrecision3D('gpu');
|
|
});
|
|
|
|
(GPU.isSinglePrecisionSupported && GPU.isWebGLSupported ? test : skip)('Issue #233 - kernel map with single precision 3d webgl', () => {
|
|
kernelMapSinglePrecision3D('webgl');
|
|
});
|
|
|
|
(GPU.isWebGL2Supported ? test : skip)('Issue #233 - kernel map with single precision 3d webgl2', () => {
|
|
kernelMapSinglePrecision3D('webgl2');
|
|
});
|
|
|
|
(GPU.isHeadlessGLSupported && GPU.isKernelMapSupported ? test : skip)('Issue #233 - kernel map with single precision 3d headlessgl', () => {
|
|
kernelMapSinglePrecision3D('headlessgl');
|
|
});
|
|
|
|
test('Issue #233 - kernel map with single precision 3d cpu', () => {
|
|
kernelMapSinglePrecision3D('cpu');
|
|
});
|