OpenGL原子锁

先看一代没有用原子锁的代码及输出：

#include "vgl.h"
#include "vmath.h"
#include "LoadShaders.h"
#include "vbm.h"
#include <iostream>
using namespace std;

GLuint vbo;

GLuint ProgCompure;
GLuint tex;

void init()
{   
    glGenBuffers(1, &vbo);	
	glBindBuffer(GL_SHADER_STORAGE_BUFFER, vbo);
	glBufferData(GL_SHADER_STORAGE_BUFFER, sizeof(GLint) * 2, NULL, GL_DYNAMIC_COPY);
		
	glBindBuffer(GL_SHADER_STORAGE_BUFFER, vbo);
	GLint *tempP = (GLint *)glMapBuffer(GL_SHADER_STORAGE_BUFFER, GL_WRITE_ONLY);
	for (int i = 0; i < 2; ++i)
	{
		*(tempP++) = 0;
	}
	glUnmapBuffer(GL_SHADER_STORAGE_BUFFER);

    ShaderInfo csi[] = {
        { GL_COMPUTE_SHADER, "Computer.comp" },
        {GL_NONE, NULL}
    };
    ProgCompure = LoadShaders(csi);

}

void display()
{
	glUseProgram(ProgCompure);	
	glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, vbo);
	
	glDispatchCompute(20, 10, 10);

	glBindBuffer(GL_SHADER_STORAGE_BUFFER, vbo);
	GLint *tempP = (GLint *)glMapBuffer(GL_SHADER_STORAGE_BUFFER, GL_READ_ONLY);

	for (int j = 0; j < 2; ++j)
	{
		cout << *(tempP++) << " ";
	}		
	glUnmapBuffer(GL_SHADER_STORAGE_BUFFER);
	
}

void idle()
{
    static unsigned int Time = GetTickCount();
    unsigned int timeNow = GetTickCount();

    if (timeNow - Time > 15)
    {
        Time = timeNow;
        glutPostRedisplay();
    }
	else
	{
		Sleep(30);
	}
}

int main(int argc, char *argv[])
{
    glutInit(&argc, argv);
    glutInitDisplayMode(GLUT_RGBA | GLUT_DEPTH | GLUT_DOUBLE | GLUT_STENCIL);
    glutInitContextProfile(GLUT_CORE_PROFILE);
    glutInitContextVersion(4, 3);
    glutInitWindowSize(400, 400);
    glutCreateWindow("Test OpenGL");

    if (glewInit() != 0)
    {
        cout << "Can not initilizate GLEW" << endl;
        return 1;
    }

    init();
    glutIdleFunc(idle);
    glutDisplayFunc(display);
    glutMainLoop();
    return 0;
}

#version 430 core

layout(local_size_x = 4,local_size_y = 4,local_size_z = 4) in;

layout(binding = 0) buffer outBuf
{
    int datas[];
}Out;

void main()
{	
	Out.datas[1] += 1;
}

计算着色器是执行了20101044*4次，每执行一次，内存变量加1，输出如下，结果显然是错误的。

修改着色器代码，实现一个原子锁:

#version 430 core

layout(local_size_x = 2,local_size_y = 1,local_size_z = 1) in;

layout(binding = 0) buffer outBuf
{
    int datas[];
}Out;

void lock()
{
	int lock_available;
	do{
		//0与内存中的值比较，等于0把1写入内存，并返回原先内存中的值0
		//不等于0时返回内存中的值 
		//第一次执行时，内存中原先值为0，执行完毕值变成1，返回0，跳出循环
		//第二次执行时，内存中原先值为1，返回1，继续循环，等待值0
		lock_available = atomicCompSwap(Out.datas[0],0,1);
	}while(lock_available != 0);
}
void unlock()
{
	//内存中的值为1，把0写入
	atomicCompSwap(Out.datas[0],1,0);
}
void main()
{	
	lock();
	Out.datas[1] += 1;
	unlock();
}

遇到问题了。。。。。。只执行一次时没问题，执行两次就会死循环、崩溃。。。。加上volatile也不管用。。
没有找到解决办法，说是个bug，http://ati.cchtml.com/show_bug.cgi?id=579，但这都是好久之前的了。。。。