Cocos2dx的Shader架构
Cocos2dx的Shader由GLProgram、GLProgramState、GLProgramCache、GLProgramStateCache组成。
GLProgram是Cocos2dx对Program的封装,一般提供一些静态方法,被多个节点复用。而GLProgramState是对GLProgram的封装,提供更加便捷的操作接口,内部记录了Uniform和Attribute,可以看作一个动态的对象。GLProgram和GLProgramState可以对比理解类和对象的区别。
它们分别对应的xxCache类顾名思义用来做缓冲的,如Cocos2dx中所有内置Shader的名字,就在GLProgramCache的loadDefaultGLProgram()方法中,将所有内置的Shader都创建了出来并缓存。
Cocos2dx内置Shader规则
下面我们先来分析下源码,看看Cocos2dx对Shader是怎么封装的,它与一般OpenGL编写的Shader脚本不同。
首先是GLProgram的compileShader()方法中,编译Shader前使用了precision关键字来设定数值的精度。总共有低精度lowp、中精度mediump、高精度highp可以选择,可以为浮点数或整数指定精度,精度越高效果越好,但消耗也越大。另外,顶点着色器支持的精度要比片段着色器高,片段着色器的highp精度的支持对于显卡而言,是可选的。
然后,它还会在Shader脚本前定义一系列Uniform变量。这些都是Cocos2dx内置的Uniform,在Shader中可以直接使用。下面是源码剪切:
bool GLProgram::compileShader(GLuint* shader, GLenum type, const GLchar* source, const std::string& convertedDefines)
{
GLint status;
if (!source)
{
return false;
}
const GLchar *sources[] = {
#if CC_TARGET_PLATFORM == CC_PLATFORM_WINRT
(type == GL_VERTEX_SHADER ? "precision mediump float;\n precision mediump int;\n" : "precision mediump float;\n precision mediump int;\n"),
#elif (CC_TARGET_PLATFORM != CC_PLATFORM_WIN32 && CC_TARGET_PLATFORM != CC_PLATFORM_LINUX && CC_TARGET_PLATFORM != CC_PLATFORM_MAC)
(type == GL_VERTEX_SHADER ? "precision highp float;\n precision highp int;\n" : "precision mediump float;\n precision mediump int;\n"),
#endif
COCOS2D_SHADER_UNIFORMS,
convertedDefines.c_str(),
source};
*shader = glCreateShader(type);
glShaderSource(*shader, sizeof(sources)/sizeof(*sources), sources, nullptr);
glCompileShader(*shader);
glGetShaderiv(*shader, GL_COMPILE_STATUS, &status);
if (! status)
{
GLsizei length;
glGetShaderiv(*shader, GL_SHADER_SOURCE_LENGTH, &length);
GLchar* src = (GLchar *)malloc(sizeof(GLchar) * length);
glGetShaderSource(*shader, length, nullptr, src);
CCLOG("cocos2d: ERROR: Failed to compile shader:\n%s", src);
if (type == GL_VERTEX_SHADER)
{
CCLOG("cocos2d: %s", getVertexShaderLog().c_str());
}
else
{
CCLOG("cocos2d: %s", getFragmentShaderLog().c_str());
}
free(src);
return false;
}
return (status == GL_TRUE);
}
static const char * COCOS2D_SHADER_UNIFORMS =
"uniform mat4 CC_PMatrix;\n"
"uniform mat4 CC_MVMatrix;\n"
"uniform mat4 CC_MVPMatrix;\n"
"uniform mat3 CC_NormalMatrix;\n"
"uniform vec4 CC_Time;\n"
"uniform vec4 CC_SinTime;\n"
"uniform vec4 CC_CosTime;\n"
"uniform vec4 CC_Random01;\n"
"uniform sampler2D CC_Texture0;\n"
"uniform sampler2D CC_Texture1;\n"
"uniform sampler2D CC_Texture2;\n"
"uniform sampler2D CC_Texture3;\n"
"//CC INCLUDES END\n\n";
在初始化GLProgram时,Cocos2dx会自动根据Shader是否使用了相应的Uniform变量来初始化_flags成员变量相对应的属性。在编译Shader时,GLSL的编译器会自动将没有使用到的Uniform移除(用glGetUniformLocation()方法判断,返回-1就是没有使用)。
void GLProgram::updateUniforms()
{
_builtInUniforms[UNIFORM_AMBIENT_COLOR] = glGetUniformLocation(_program, UNIFORM_NAME_AMBIENT_COLOR);
_builtInUniforms[UNIFORM_P_MATRIX] = glGetUniformLocation(_program, UNIFORM_NAME_P_MATRIX);
_builtInUniforms[UNIFORM_MV_MATRIX] = glGetUniformLocation(_program, UNIFORM_NAME_MV_MATRIX);
_builtInUniforms[UNIFORM_MVP_MATRIX] = glGetUniformLocation(_program, UNIFORM_NAME_MVP_MATRIX);
_builtInUniforms[UNIFORM_NORMAL_MATRIX] = glGetUniformLocation(_program, UNIFORM_NAME_NORMAL_MATRIX);
_builtInUniforms[UNIFORM_TIME] = glGetUniformLocation(_program, UNIFORM_NAME_TIME);
_builtInUniforms[UNIFORM_SIN_TIME] = glGetUniformLocation(_program, UNIFORM_NAME_SIN_TIME);
_builtInUniforms[UNIFORM_COS_TIME] = glGetUniformLocation(_program, UNIFORM_NAME_COS_TIME);
_builtInUniforms[UNIFORM_RANDOM01] = glGetUniformLocation(_program, UNIFORM_NAME_RANDOM01);
_builtInUniforms[UNIFORM_SAMPLER0] = glGetUniformLocation(_program, UNIFORM_NAME_SAMPLER0);
_builtInUniforms[UNIFORM_SAMPLER1] = glGetUniformLocation(_program, UNIFORM_NAME_SAMPLER1);
_builtInUniforms[UNIFORM_SAMPLER2] = glGetUniformLocation(_program, UNIFORM_NAME_SAMPLER2);
_builtInUniforms[UNIFORM_SAMPLER3] = glGetUniformLocation(_program, UNIFORM_NAME_SAMPLER3);
_flags.usesP = _builtInUniforms[UNIFORM_P_MATRIX] != -1;
_flags.usesMV = _builtInUniforms[UNIFORM_MV_MATRIX] != -1;
_flags.usesMVP = _builtInUniforms[UNIFORM_MVP_MATRIX] != -1;
_flags.usesNormal = _builtInUniforms[UNIFORM_NORMAL_MATRIX] != -1;
_flags.usesTime = (
_builtInUniforms[UNIFORM_TIME] != -1 ||
_builtInUniforms[UNIFORM_SIN_TIME] != -1 ||
_builtInUniforms[UNIFORM_COS_TIME] != -1
);
_flags.usesRandom = _builtInUniforms[UNIFORM_RANDOM01] != -1;
this->use();
// Since sample most probably won't change, set it to 0,1,2,3 now.
if(_builtInUniforms[UNIFORM_SAMPLER0] != -1)
setUniformLocationWith1i(_builtInUniforms[UNIFORM_SAMPLER0], 0);
if(_builtInUniforms[UNIFORM_SAMPLER1] != -1)
setUniformLocationWith1i(_builtInUniforms[UNIFORM_SAMPLER1], 1);
if(_builtInUniforms[UNIFORM_SAMPLER2] != -1)
setUniformLocationWith1i(_builtInUniforms[UNIFORM_SAMPLER2], 2);
if(_builtInUniforms[UNIFORM_SAMPLER3] != -1)
setUniformLocationWith1i(_builtInUniforms[UNIFORM_SAMPLER3], 3);
}
上面我们分析了Cocos2dx的Shader会内置很多Uniform变量,也知道了这些Uniform变量是否使用了。那么下一个问题,如果使用了,这些Uniform是什么时候被设置的呢?它们的值又是多少呢?
如果是纹理Uniform,在一开始就会被设置为固定的值,上面源码中也可以看到,同时可以使用4个纹理,在渲染的时候,Cocos2dx会将纹理绑定到指定的纹理采样器中,我们在Shader脚本中可以访问这些采样器,而采样器对应的纹理,由Cocos2dx调用OpenGL的glBindTexture()等方法绑定,因此这个Uniform的值不需要我们设置。
而其它的如矩阵、时间、法线、随机数等Uniform,是在运行的时候动态设置的,这些属性是实时变化的,在程序运行时,Cocos2dx会每帧调用setUniformsForBuiltins()自动设置它们。
void GLProgram::setUniformsForBuiltins(const Mat4 &matrixMV)
{
const auto& matrixP = _director->getMatrix(MATRIX_STACK_TYPE::MATRIX_STACK_PROJECTION);
if (_flags.usesP)
setUniformLocationWithMatrix4fv(_builtInUniforms[UNIFORM_P_MATRIX], matrixP.m, 1);
if (_flags.usesMV)
setUniformLocationWithMatrix4fv(_builtInUniforms[UNIFORM_MV_MATRIX], matrixMV.m, 1);
if (_flags.usesMVP)
{
Mat4 matrixMVP = matrixP * matrixMV;
setUniformLocationWithMatrix4fv(_builtInUniforms[UNIFORM_MVP_MATRIX], matrixMVP.m, 1);
}
if (_flags.usesNormal)
{
Mat4 mvInverse = matrixMV;
mvInverse.m[12] = mvInverse.m[13] = mvInverse.m[14] = 0.0f;
mvInverse.inverse();
mvInverse.transpose();
GLfloat normalMat[9];
normalMat[0] = mvInverse.m[0];normalMat[1] = mvInverse.m[1];normalMat[2] = mvInverse.m[2];
normalMat[3] = mvInverse.m[4];normalMat[4] = mvInverse.m[5];normalMat[5] = mvInverse.m[6];
normalMat[6] = mvInverse.m[8];normalMat[7] = mvInverse.m[9];normalMat[8] = mvInverse.m[10];
setUniformLocationWithMatrix3fv(_builtInUniforms[UNIFORM_NORMAL_MATRIX], normalMat, 1);
}
if (_flags.usesTime) {
// This doesn't give the most accurate global time value. // Cocos2D doesn't store a high precision time value, so this will have to do.
// Getting Mach time per frame per shader using time could be extremely expensive.
float time = _director->getTotalFrames() * _director->getAnimationInterval();
setUniformLocationWith4f(_builtInUniforms[GLProgram::UNIFORM_TIME], time/10.0, time, time*2, time*4);
setUniformLocationWith4f(_builtInUniforms[GLProgram::UNIFORM_SIN_TIME], time/8.0, time/4.0, time/2.0, sinf(time));
setUniformLocationWith4f(_builtInUniforms[GLProgram::UNIFORM_COS_TIME], time/8.0, time/4.0, time/2.0, cosf(time));
}
if (_flags.usesRandom)
setUniformLocationWith4f(_builtInUniforms[GLProgram::UNIFORM_RANDOM01], CCRANDOM_0_1(), CCRANDOM_0_1(), CCRANDOM_0_1(), CCRANDOM_0_1());
}
当然,除了以上内置的Uniform外,Cocos2dx还内置了一些Attribute,但这些变量是需要在Shader脚本中手动输入的,主要是颜色、坐标、纹理坐标、法线等常用属性。它们被定义在CCGLProgram.cpp中,在Renderer进行渲染时,会调用glVertexAttribPointer()设置它们。
// Attribute names
const char* GLProgram::ATTRIBUTE_NAME_COLOR = "a_color";
const char* GLProgram::ATTRIBUTE_NAME_POSITION = "a_position";
const char* GLProgram::ATTRIBUTE_NAME_TEX_COORD = "a_texCoord";
const char* GLProgram::ATTRIBUTE_NAME_TEX_COORD1 = "a_texCoord1";
const char* GLProgram::ATTRIBUTE_NAME_TEX_COORD2 = "a_texCoord2";
const char* GLProgram::ATTRIBUTE_NAME_TEX_COORD3 = "a_texCoord3";
const char* GLProgram::ATTRIBUTE_NAME_NORMAL = "a_normal";
const char* GLProgram::ATTRIBUTE_NAME_BLEND_WEIGHT = "a_blendWeight";
const char* GLProgram::ATTRIBUTE_NAME_BLEND_INDEX = "a_blendIndex";
const char* GLProgram::ATTRIBUTE_NAME_TANGENT = "a_tangent";
const char* GLProgram::ATTRIBUTE_NAME_BINORMAL = "a_binormal";
整个Cocos2dx的Shader不同的规则大体就是这样了,主要是内置了一系列的变量,由引擎自动设置值,我们可以在自己Shader中直接使用。
Cocos2dx中使用Shader示例
我们就编写一个常用的灰化Shader来演示下Cocos2dx中使用Shader的流程。
灰化原理:灰白图片的特征就是每个像素的RGB三个分量的值均相等,值越高颜色越白,反之颜色越黑。常用的算法有3种,平均值、最大值和加权平均值。
首先,我们得分别编写顶点Shader和片段Shader:
//顶点Shader
attribute vec4 a_position; //位置(内置)
attribute vec2 a_texCoord; //纹理坐标(内置)
varying vec2 v_texCoord; //自定义易变变量(写入值)
void main()
{
gl_Position = CC_PMatrix * a_position;
v_texCoord = a_texCoord;
}
//片段着色器
//使用加权平均值算RGB分量,即给每个分量指定一个比例,指定比例没有写死,通过一个Uniform变量来设置
varying vec2 v_texCoord; //接收顶点着色器传递的值(接收值)
uniform vec4 u_grayParam; //指定rgb颜色分量的比例
void main()
{
vec4 texColor = texture2D(CC_Texture0, v_texCoord);
texColor.rgb = texColor.r * u_grayParam.r + texColor.g * u_grayParam.g + texColor.b * u_grayParam.b;
gl_FragColor = texColor;
}
分别命名为gray.vert和gray.frag保存到磁盘。
好,脚本写完了,下面来看看怎么使用:
#include "renderer/CCGLProgramStateCache.h"
//置灰接口
void grayNode(Node* node)
{
//对应xxCache的运用,先成缓存中找,不需要每次都重新创建
GLProgram* program = GLProgramCache::getInstance()->getGLProgram("MyGrayShader");
if(nullptr == program)
{
//本地磁盘加载shader创建program,还一种字符串的形式效率要高createWithByteArrays()
program = GLProgram::createWithFilenames("gray.vert", "gray.frag");
//将GLProgram对象缓存到Cache中
GLProgramCache::getInstance()->addGLProgram(program, "MyGrayShader");
}
//通过program创建programState,用它来设置上面自定义的Uniform变量的值
GLProgramState* programState = GLProgramState::getOrCreateWithGLProgram(program);
programState->setUniformVec4("u_grayParam", Vec4(0.2f,0.3f,0.5f,1.0f));
//调用node节点的setGLProgramState,以后此节点渲染就会使用这个Shader了
node->setGLProgramState(programState);
}
//使用
auto sprite = Sprite::create(xx.png);
grayNode(sprite);
至此,Cocos2dx中使用Shader流程就讲完了~ 最后说下,怎么把置灰的图片再还原回去呢?OpenGL是个状态机的运行机制,所以只需要把设置的Shader还原回去重新设置就行了。这里Sprite对象默认的Shader是GLProgram::SHADER_NAME_POSITION_TEXTRUE_COLOR_NO_MVP,可以通过GLProgramState::getOrCreateWithGLProgramName()传入上面字符串常量来获取它的GLProgramState.