官方网站下载免费软件大兴企业官网网站建设咨询

Cesium

首先想到的就是查看Cesium的代码看看他是怎么解决的对吧
这里有个最简代码片段，我们顺着往里找

* void main()* {*   vec4 p = czm_translateRelativeToEye(positionHigh, positionLow);*   gl_Position = czm_modelViewProjectionRelativeToEye * p;* }

vec4 czm_translateRelativeToEye(vec3 high, vec3 low)
{vec3 highDifference = high - czm_encodedCameraPositionMCHigh;// This check handles the case when NaN values have gotten into `highDifference`.// Such a thing could happen on devices running iOS.if (length(highDifference) == 0.0) {  highDifference = vec3(0);  }vec3 lowDifference = low - czm_encodedCameraPositionMCLow;return vec4(highDifference + lowDifference, 1.0);
}

这里可以看见他把camera的坐标也传进来了，然后high low分别和camera的high low减一下再合并到一起，就得到了相对相机的坐标系

也就是说，他不仅把坐标拆成了high low，还减去了相机坐标，也就是减去相机之后的坐标还拆，这确实是能支持很大的数据了

czm_modelViewProjectionRelativeToEye 这个值的创建就是普通的modelViewMatirx只是把translate清空了，如下代码

czm_modelViewRelativeToEye: new AutomaticUniform({size: 1,datatype: WebGLConstants.FLOAT_MAT4,getValue: function (uniformState) {return uniformState.modelViewRelativeToEye;},
}),modelViewRelativeToEye: {get: function () {cleanModelViewRelativeToEye(this);return this._modelViewRelativeToEye;},}
function cleanModelViewRelativeToEye(uniformState) {if (uniformState._modelViewRelativeToEyeDirty) {uniformState._modelViewRelativeToEyeDirty = false;const mv = uniformState.modelView;const mvRte = uniformState._modelViewRelativeToEye;mvRte[0] = mv[0];mvRte[1] = mv[1];mvRte[2] = mv[2];mvRte[3] = mv[3];mvRte[4] = mv[4];mvRte[5] = mv[5];mvRte[6] = mv[6];mvRte[7] = mv[7];mvRte[8] = mv[8];mvRte[9] = mv[9];mvRte[10] = mv[10];mvRte[11] = mv[11];mvRte[12] = 0.0;mvRte[13] = 0.0;mvRte[14] = 0.0;mvRte[15] = mv[15];}
}

modelViewProjectionRelativeToEye只是一并在CPU做完几个乘法

modelViewProjectionRelativeToEye: {get: function () {cleanModelViewProjectionRelativeToEye(this);return this._modelViewProjectionRelativeToEye;},
},
function cleanModelViewProjectionRelativeToEye(uniformState) {if (uniformState._modelViewProjectionRelativeToEyeDirty) {uniformState._modelViewProjectionRelativeToEyeDirty = false;Matrix4.multiply(uniformState._projection,uniformState.modelViewRelativeToEye,uniformState._modelViewProjectionRelativeToEye,);}
}

这个是high low的解释
https://help.agi.com/AGIComponents/html/BlogPrecisionsPrecisions.htm
这是Cesium里的实现

EncodedCartesian3.encode = function (value, result) {//>>includeStart('debug', pragmas.debug);Check.typeOf.number("value", value);//>>includeEnd('debug');if (!defined(result)) {result = {high: 0.0,low: 0.0,};}let doubleHigh;if (value >= 0.0) {doubleHigh = Math.floor(value / 65536.0) * 65536.0;result.high = doubleHigh;result.low = value - doubleHigh;} else {doubleHigh = Math.floor(-value / 65536.0) * 65536.0;result.high = -doubleHigh;result.low = value + doubleHigh;}return result;
};

这是CPP代码

void CDoubleToTwoFloats::Convert(double doubleValue,float& floatHigh, float& floatLow)
{if (doubleValue >= 0.0){double doubleHigh = floor(doubleValue / 65536.0) * 65536.0;floatHigh = (float)doubleHigh;floatLow = (float)(doubleValue - doubleHigh);}else{double doubleHigh = floor(-doubleValue / 65536.0) * 65536.0;floatHigh = (float)-doubleHigh;floatLow = (float)(doubleValue + doubleHigh);}
}

UE

https://dev.epicgames.com/documentation/en-us/unreal-engine/large-world-coordinates-rendering-in-unreal-engine-5
这里是UE的文档
之前就听说UE有这个支持，好奇他怎么搞的
这里是文档里最后一段话，看起来也是类似Cesium的解法
不过他还有一个空间，是纠正了旋转不只是减去offset，建议能做就做，那估计就是没做:)

FDFScalar(double Input)
{float High = (float)Input;float  Low = (float)(Input - High);
}

总结

这种做法每个坐标都需要存储两倍大小的坐标信息high low两个vector3，但是似乎如果要支持double的话，没有其他好法子