OpenGL 之 Compute Shader(通用计算并行加速)
平常我们使用的Shader有顶点着色器、几何着色器、片段着色器,这几个都是为光栅化图形渲染服务的,OpenGL 4.3之后新出了一个Compute Shader,用于通用计算并行加速,现在对其进行介绍。
介绍Compute Shader之前需要先介绍一下ImageTexture:
普通的Texture在GLSL中只能进行读取(sampler采样获取数据),写入则必须在Fragment Shader中写入帧缓冲绑定的附件Texture当前像素中,不能随意指定位置写入,并且不能同时读写同一张纹理(我试过不行,有博客同样说不行,应该是不行吧)。
1、生成Texture
void WKS::ImageTexture::setupTexture() { glGenTextures(1, &this->textureID); glBindTexture(GL_TEXTURE_2D, this->textureID); glTexStorage2D(GL_TEXTURE_2D, 1, GL_RGBA32F, width, height); // turn off filtering and wrap modes glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP); glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP); glBindTexture(GL_TEXTURE_2D, 0); }
注意,要是用 glTexStorage2D()生成固定大小纹理,不能使用glTexImage2D()
2、生成ImageTexture
glBindImageTexture(0, this->inputTexture, 0, GL_FALSE, 0, GL_READ_ONLY, GL_RGBA32F);
inputTexture对应1、中生成的Texture纹理ID。第一个参数是ImageTexture绑定点,与texture纹理绑定点应该不重合。
3、GLSL中声明
layout (rgba32f, binding = 0) uniform image2D input_image;
补充:ImageTexture底层是Texture,那么在Host上可以进行访问
a、初始化,传入数据
void WKS::ImageTexture::Transfer2Texture(float* data) { glBindTexture(GL_TEXTURE_2D, this->textureID); glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, width, height, GL_RGBA, GL_FLOAT, data); }
b、读取数据
float* WKS::Texture::GetTextureData(GLuint width, GLuint height, GLuint channels, GLuint texID) { float* data = new float[width * height * channels]; glBindTexture(GL_TEXTURE_2D, texID); if(channels==1) glGetTexImage(GL_TEXTURE_2D, 0, GL_RED, GL_FLOAT, data); if(channels==3) glGetTexImage(GL_TEXTURE_2D, 0, GL_RGB, GL_FLOAT, data); if (channels == 4) glGetTexImage(GL_TEXTURE_2D, 0, GL_RGBA, GL_FLOAT, data); glBindTexture(GL_TEXTURE_2D, 0); return data; }
现在来介绍Compute Shader:
#version 430 core layout (local_size_x=16, local_size_y=16) in; uniform float v[4]; layout (rgba32f, binding = 0) uniform image2D input_image; layout (rgba32f, binding = 1) uniform image2D output_image; shared vec4 mat_shared[16][16]; void main(void) { ivec2 pos=ivec2(gl_GlobalInvocationID.xy); mat_shared[pos.x][pos.y]=imageLoad(input_image,pos); barrier(); vec4 data=mat_shared[pos.x][pos.y]; data.r=v[0]+data.r; data.g=v[1]+data.g; data.b=v[2]+data.b; data.a=v[3]+data.a; imageStore(output_image,pos.xy,data); }
计算由一个一个计算单元完成,layout (local_size_x=16, local_size_y=16) in; 是表示本地工作组的由16*16的计算单元组成,本地工作组可以共享Shadered变量。 多个本地工作组构成全局工作组,由:
glDispatchCompute(1, 1, 1);
启动计算,参数表示全局工作组的维度(以本地工作组为单位),(1,1,1)表示只有一个本地工作组。
注意:Compute Shader 只有一个阶段(渲染一般是vertex+fragment 2个阶段),编译类型选择GL_COMPUTE_SHADER
Shader(const char* computePath) :programId(0) { std::vector<ShaderFile> fileVec; fileVec.push_back(ShaderFile(GL_COMPUTE_SHADER, computePath)); loadFromFile(fileVec); }
示例:
对一个4*4的vec4矩阵的所有元素加上vec4(0, 0.1,0.2,0.3)
初始化:
void SceneRendering::setupAddData() { int num = 4 * 4 * 4; this->inputData = new float[num]; for (int i = 0; i < num; i++) inputData[i] = i; for (int i = 0; i < 4; i++) v[i] = i*0.1f; shader_add = new Shader("./Shader/add.comp"); WKS::ImageTexture* texturePtr = new WKS::ImageTexture(4, 4); this->inputTexture = texturePtr->GetTextureID(); this->outputTexture = (new WKS::ImageTexture(4, 4))->GetTextureID(); texturePtr->Transfer2Texture(inputData); }
调用Compute Shader:
void SceneRendering::performCompute() { this->shader_add->use(); this->shader_add->setVecN("v", 4, v); glBindImageTexture(0, this->inputTexture, 0, GL_FALSE, 0, GL_READ_ONLY, GL_RGBA32F); glBindImageTexture(1, this->outputTexture, 0, GL_FALSE, 0, GL_WRITE_ONLY, GL_RGBA32F); glDispatchCompute(1, 1, 1); glMemoryBarrier(GL_SHADER_IMAGE_ACCESS_BARRIER_BIT); glFinish(); }
主函数调用,结果输出:
glClearColor(0.5f, 0.5f, 0.5f, 1.0f); glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT); // 我们现在不使用模板缓冲//Compute Shader this->performCompute(); float* data = WKS::Texture::GetTextureData(4, 4, 4, this->outputTexture); int index = 0; for (int i = 0; i < 4; i++) { for (int j = 0; j < 4; j++) { std::cout << "(" <<data[index]<<","<<data[index+1]<<","<<data[index+2]<<","<<data[index+3]<< ")" << " "; index += 4; } std::cout << std::endl; } std::cout<< std::endl; free(data);
图片: