计算WebGL中的uniforms变量使用数
在使用Three.js为人体模型加载皮肤材料时,启用了skinning:true的参数。
有时候会导致GL编译错误,提示“too many uniforms”。下面的文章有助于理解错误原因和检测uniforms的使用情况。
For a recent consulting project I was attempting to render some fairly complex skeletal animations in WebGL on Firefox and Chrome. I quickly ran into a situation where the animation was rendering on Linux and Mac computers, but not on Windows. All the test machines had up-to-date graphics drivers, but the Windows machines threw a “too many uniforms” error when attempting to link the shader programs. So what does this error mean, and how do we fix it? In this post I’ll share the results of my research that allowed me to get everything working.
Uniforms
So what is a uniform? According to the OpenGL wiki, a uniform is “a global GLSL variable declared with the ‘uniform’ storage qualifier.” To be a little more specific: your shader executes on the GPU, which is physically distinct from the rest of the computer, separated by a bus. This means there’s a very limited amount of data that can be shared quickly between the GPU and your web browser. A uniform variable is what lets you declare “this is data that is exposed externally for transport over the bus”, and there’s a specific API on the JavaScript/WebGL side that lets you get and set this data. For example, you might declare the following variables at the start of a GLSL vertex shader:
uniform mat4 uMVMatrix; uniform mat4 uPMatrix; uniform vec3 someVector; varying vec4 vColor;
That shader has, you guessed it, three uniforms. Later, in your JavaScript program code, you might access, transform, and return one of the uniforms like this:
// get a copy of our vec3 uniform var vec3Uni = gl.getUniformLocation(shaderProgram, "someVector"); // increment the x component of the vector by 10 vec3Uni[0] += 10; // write it back to the GPU gl.uniform3fv(shaderProgram, vec3Uni);
If you’ve ever used web workers, this data flow should be familiar to you. This is for good reason: we’re simply doing parallel computing here, but instead of the shader code running on a spawned worker thread, it’s running on the GPU.
How Many is Too Many?
So, we know what a uniform is: it’s data that’s transferred back and forth between the GPU. Because the GPU is physically distinct from the rest of the computer, we are severely limited in the amount of data that we can quickly push back and forth over the bus between the CPU and the GPU. So we need to watch our uniform usage very closely, and we need to understand that different combinations of graphics cards and graphics drivers allow for different uniform caps.
When it seems like I’m hitting a performance limit, the first thing I do is benchmark my system. Fortunately, webgl-bench is a great website that can benchmark your WebGL performance in seconds, giving you a lot of statistics that can be valuable for debugging. For this problem in particular, we’re interested in the value returned forMAX_VERTEX_UNIFORM_VECTORS. As you might imagine, having more active uniforms than the maximum is what causes a “too many uniforms” error.
Running the benchmark on Mac, Linux, and Windows machines highlighted something odd: while most of the Mac and Linux machines capped out at 1024 uniforms, the Windows computers supported only 256 vertex uniforms, even if they had amazing graphics cards and the latest drivers.
This is due to a peculiarity in how Chrome and Firefox render WebGL. In Mac and Linux, the browsers render WebGL directly, using native OpenGL drivers. The problem on Windows is that the native OpenGL drivers are notoriously unstable, causing sometimes massive performance hits. To avoid these problems, the Windows versions of both browsers default to using ANGLE, which translates OpenGL ES 2.0 API calls to DirectX 9 or DirectX 11 API calls.
(Frank Olivier of the IE GPU team informs me that IE11 WebGL supports more than 256 vertex uniforms because it’s already using the DX11 runtime.)
The problem is that the ANGLE configuration currently used by both browsers uses DirectX 9. The DX9 configuration imposes an artificial cap of 256 vertex uniforms in use at any given time, no matter how good the graphics card is. Users can manually disable ANGLE (more info here) but it’s an onerous process only meant for developers; you can’t expect an end user to do it.
Fortunately, an upcoming Chrome release (version 34 or possibly earlier) is going to default to using DirectX 11, which should increase the cap to 1024 maximum vertex uniforms. You can follow the tracking issue for this feature here. Firefox is actively looking into doing the same, but they don’t have a timetable yet.
Counting the Uniforms
So now I needed to figure out how many uniforms I was using with my complex model. Obviously I was using more than 256, but exactly how many more?
For simple shaders you can manually count the number of uniforms by visually inspecting your script. But for most combinations of shaders and dynamic scenes, it’s likely that the number of uniforms in use will scale with the complexity and the number of the models being rendered. We need to dynamically count the active uniforms.
After a bunch of research, I wrote a function does this for you. You pass it a shader program and it returns an object containing useful data.
Here’s the code for the function:
function getProgramInfo(gl, program) { var result = { attributes: [], uniforms: [], attributeCount: 0, uniformCount: 0 }, activeUniforms = gl.getProgramParameter(program, gl.ACTIVE_UNIFORMS), activeAttributes = gl.getProgramParameter(program, gl.ACTIVE_ATTRIBUTES); // Taken from the WebGl spec: // http://www.khronos.org/registry/webgl/specs/latest/1.0/#5.14 var enums = { 0x8B50: 'FLOAT_VEC2', 0x8B51: 'FLOAT_VEC3', 0x8B52: 'FLOAT_VEC4', 0x8B53: 'INT_VEC2', 0x8B54: 'INT_VEC3', 0x8B55: 'INT_VEC4', 0x8B56: 'BOOL', 0x8B57: 'BOOL_VEC2', 0x8B58: 'BOOL_VEC3', 0x8B59: 'BOOL_VEC4', 0x8B5A: 'FLOAT_MAT2', 0x8B5B: 'FLOAT_MAT3', 0x8B5C: 'FLOAT_MAT4', 0x8B5E: 'SAMPLER_2D', 0x8B60: 'SAMPLER_CUBE', 0x1400: 'BYTE', 0x1401: 'UNSIGNED_BYTE', 0x1402: 'SHORT', 0x1403: 'UNSIGNED_SHORT', 0x1404: 'INT', 0x1405: 'UNSIGNED_INT', 0x1406: 'FLOAT' }; // Loop through active uniforms for (var i=0; i < activeUniforms; i++) { var uniform = gl.getActiveUniform(program, i); uniform.typeName = enums[uniform.type]; result.uniforms.push(uniform); result.uniformCount += uniform.size; } // Loop through active attributes for (var i=0; i < activeAttributes; i++) { var attribute = gl.getActiveAttrib(program, i); attribute.typeName = enums[attribute.type]; result.attributes.push(attribute); result.attributeCount += attribute.size; } return result; }
It returns an object containing the following data:
uniformCount – the total number of currently active uniforms in this program
uniforms – an array containing the name, size, type, and typeName of the each active uniform. In particular, typeName can be useful since otherwise you’d have to look up the integer type value and cross reference it to the enum declarations in the webGL spec
attributeCount, attributes – same as above, but for attributes (attributes are out of scope for this article, but are similarly capped and nice to keep count of)
Here’s some sample output using console.table (more info on that function here) to show the information from the returned object’s uniforms array. This is data from my original, complex model:
Looking at this I was able to immediately discern that we had 288 active vertex uniforms, and most of them were in the joint matrix (aka the complexity of our model’s skeleton). This helped me tell our artist that we needed to reduce the number of joints in the model by about 20% for it to work on Windows machines.
原文链接:
https://bocoup.com/weblog/counting-uniforms-in-webgl
- 相关文章
CentOS6 Apache2.2多站点HTTPS配置
可以使用letsencrypt(certbot)免费证书服务。支持多系统、多站点和多目录,支持wildcard(通配符域名),90天生效,可用定时任务自动更新。需要注意一点的是apache2.4以下版本需要在默认的ssl配置中添加如下的指令:NameVirtualHost
2019年NodeJS框架Koa和Express选型比较
Koa和Express都是NodeJS的主流应用开发框架。
Express是一个完整的nodejs应用框架。Koa是由Express团队开发的,但是它有不同的关注点。Koa致力于核心中间件...OpenGL/WebGL顶点坐标变换过程简介
世界坐标是按照笛卡尔坐标系定义出来的绝对坐标系,下面的各种坐标系都建立在世界坐标的基础上。对象坐标系对象被应用于任何...
增强现实引擎ARToolKit工作原理简介
ARToolkit是一个基于CV(计算机视觉)和Marker(标识)的开源增强现实引擎。其具备如下功能特性:A. 鲁棒跟踪,包括基于标记的跟踪与基于特征的跟踪;
常用光照类型基本概念工作原理及其计算公式
在三维场景中,原理上物体的渲染效果取决于光照与物体表面的相互作用,对于渲染程序而言,可以通过把一些数学公式应用于像素着色来实现,从而模拟出真实生活中的...
CSS3原生变量(Native Variables)新特性简介
对Web开发者来说,一个盼望已久的特性是CSS终于支持原生变量了!
变量是程序语言中用来解决代码重复和进行表达式计算的关键概念(想想数学方程式中的x)。...Web界面编程状态变化和JS开发框架(React/Angular/Ember)
UI编程中的一个关键课题就是界面组件化(可复用)以及组件状态管理。稍早一些的windows程序员可能接触过MFC,其界面编程中有一个DDX(DoDataExchange)的机制,...
使用HTML5 Canvas实现的界面元素截屏功能
HTTP1.1协议现状、问题和解决方案
HTTP的现状最早的HTTP协议非常简单,只能用来传送文本,方法也只有GET,后来逐步发展到1.1,能够支持多种MIME格式数据(如文本、文件),支持GET,POST,HEAD,OPTI...
Three.js入门教程4 - 创建粒子系统动画
嗨,又见面了。这么说我们已经开始学习Three.js了,如果你还没有看过之前三篇教程,建议你先读完。如果你已经读完前面的教程了,你可能会想做一些关于粒子的东西。让我们直面这个话题吧,每个人都爱粒子效果。不管你是否知道,你可以很轻易地创建它们。
Three.js入门教程2 - 着色器(上)
CSS3图片混合(Blend)效果及其参考计算公式一览表
在Photoshop软件中,混合是将两个图层的色彩值进行合成,从而创造出大量的效果。在这些效果的背后实际是一些简单的数学公式在起作用。下面所介绍的公式仅适用于R...
jQuery Ribbles - 基于WebGL的水面涟漪动效插件
使用jQuery
Three.js 3D打印数据模型文件(.STL)加载
3D打印是当下和未来10年产品技术主流方向之一,影响深远。对于电子商务类的3D打印网站,一个主要功能是把商品以3D的方式呈现出来,也就是3D数据可视化技术。HTML...
如何使用CSS3实现一个3D商品标签
使用3D缎带形状的标签是常见的一个设计模式,用在商品折扣、文章标题或网站推荐信息上,来突出显示重点内容,吸引用户视觉焦点。实现的方法有2种,一种是使用背...
更多...