OpenGL combine Textures - performance

Question

I'm programming a 2D-Game in OpenGL and I have to output a level which consists of 20x15 fields.
So I'm currently outputting a texture for each field which is quite slow (300 textures/frame).
But due to the reason that the level never changes, I wondered if it's possible to combine the textures to a big, single texture before the game-loop starts.
Then I would have to output only one texture with 4 Texture Coordinates (0/0)(0/1)(1/1)(1/0) and 4 glVertex2f() which specifies the position in the Window.
This is my current Code for each of the 300 fields:
glColor3f(1,1,1);
glBindTexture(GL_TEXTURE_2D,textur);
glBegin(GL_QUADS);
glTexCoord2f(textArea.a.x,textArea.b.y);glVertex2f(display.a.x,display.a.y);
glTexCoord2f(textArea.a.x,textArea.a.y);glVertex2f(display.a.x,display.b.y);
glTexCoord2f(textArea.b.x,textArea.a.y);glVertex2f(display.b.x,display.b.y);
glTexCoord2f(textArea.b.x,textArea.b.y);glVertex2f(display.b.x,display.a.y);
glEnd();
Note that I have the images for all possible field-types in one .tga-File. So I'm choosing the right one with glTexCoord2f().
The image-File with all Tiles is loaded into
GLuint textur;
So I bind the same texture for every field.
My target is to decrease CPU-time. Display-Lists didn't work because there is so many data to load in the Graphics Card, that, in the end, display-Lists were even slower.
I also wasn't able to use VBOs because I don't use extensions like GLUT.
So my idea was to generate a single texture which should be quite easy and effective.
I hope you can give me feedback how I can combine textures and if this method would be the easiest one to increase performance
EDIT: that are the OpenGl-Functions I use in my program:
When I start the program, I initialize the window:
glfwInit();
if( !glfwOpenWindow(windowSize.x,windowSize.y, 0, 0, 0, 0, 0, 0, GLFW_WINDOW ) )
{ glfwTerminate();
return;
}
And that's all what the game-loop does with OpenG:
int main()
{
//INIT HERE (see code above)
glBlendFunc(GL_SRC_ALPHA,GL_ONE_MINUS_SRC_ALPHA);
glEnable(GL_BLEND);
glAlphaFunc(GL_GREATER,0.1f);
glEnable(GL_ALPHA_TEST);
long loopStart;//measure loopcycle-time
do{
height = height > 0 ? height : 1;
glViewport( 0, 0, width, height ); //set Origin
glClearColor( 0.0f, 0.0f, 0.0f, 0.0f ); //background-color
glClear(GL_COLOR_BUFFER_BIT);
glMatrixMode(GL_PROJECTION);
glLoadIdentity();
glOrtho(0,windowSize.x,0,windowSize.y,0,128); //2D-Mode
glMatrixMode(GL_MODELVIEW);
loopStart=clock();
//(...) OUTPUT HERE (code see above)
glfwSwapBuffers(); //erzeugte Grafikdaten ausgeben
printf("%4dms -> ",clock()-loopStart);
}while(...);
glDisable(GL_ALPHA_TEST);
glDisable(GL_TEXTURE_2D);
glfwTerminate();
}

Answer 1

I see you're using GLFW. You can add GLEW and GLM and then you should use OpenGL 3.x or higher.
Here is a FULL example, how you can easily draw 2000 Textured Quads (With Alphablending) or more with FPS of 200 or more on a lost budget Laptop. It has only one little Texture, but it will work also with an 4096x4096 Texture Atlas. You get one HUGE Performance-Hit, if the Subtexture Size in the big texture EXACTLY matches the size of your Quad you draw! You should use 50x50 Pixels also in the Big-Texture! The following Deme-Code here also UPDATES ALL 2000 Quads each frame and send them to the GPU. If you will not have to update them each frame and put the Scroll-Coordinates to the Shader..you will gain performance again.
If you need no blending...use Alpha-Tests..you will gain again more speed.
#define GLEW_STATIC
#include "glew.h"
#include "glfw.h"
#include "glm.hpp"
#include "glm/gtc/matrix_transform.hpp"
#include "glm/gtx/transform.hpp"
#include <sstream>
#include <fstream>
#include <vector>
#define BUFFER_OFFSET(i) ((char *)NULL + (i))
std::ofstream logger("Log\\Ausgabe.txt", (std::ios::out | std::ios::app));
class Vertex
{
public:
float x;
float y;
float z;
float tx;
float ty;
};
class Quad
{
public:
float x;
float y;
float width;
float height;
};
int getHighResTimeInMilliSeconds(bool bFirstRun);
GLuint buildShader();
void addQuadToLocalVerticeArray(Vertex * ptrVertexArrayLocal, Quad *quad, int *iQuadCounter);
int main()
{
logger << "Start" << std::endl;
if(!glfwInit())
exit(EXIT_FAILURE);
glfwOpenWindowHint(GLFW_OPENGL_VERSION_MAJOR,3);
glfwOpenWindowHint(GLFW_OPENGL_VERSION_MINOR,3);
glfwOpenWindowHint(GLFW_OPENGL_FORWARD_COMPAT, 1);
glfwOpenWindowHint(GLFW_OPENGL_PROFILE,GLFW_OPENGL_CORE_PROFILE);
if( !glfwOpenWindow(1366, 768,8,8,8,8,32,32,GLFW_FULLSCREEN) )
{
glfwTerminate();
exit( EXIT_FAILURE );
}
if (glewInit() != GLEW_OK)
exit( EXIT_FAILURE );
//Init
GLuint VertexArrayID;
GLuint vertexbuffer;
GLuint MatrixID;
GLuint TextureID;
GLuint Texture;
GLuint programID = buildShader();
//Texture in Video-Speicher erstellen
GLFWimage img;
int iResult = glfwReadImage("Graphics\\gfx.tga", &img, GLFW_NO_RESCALE_BIT);
glEnable(GL_TEXTURE_2D);
glGenTextures(1, &Texture);
glBindTexture(GL_TEXTURE_2D, Texture);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA,32,32, 0, GL_RGBA, GL_UNSIGNED_BYTE, img.Data);
glfwFreeImage(&img);
Vertex * ptrVertexArrayLocal = new Vertex[12000];
glGenVertexArrays(1, &VertexArrayID);
glBindVertexArray(VertexArrayID);
glGenBuffers(1, &vertexbuffer);
glBindBuffer(GL_ARRAY_BUFFER, VertexArrayID);
glBufferData(GL_ARRAY_BUFFER, sizeof(Vertex) * 12000, NULL, GL_DYNAMIC_DRAW);
glm::mat4 Projection = glm::ortho(0.0f, (float)1366,0.0f, (float)768, 0.0f, 100.0f);
glm::mat4 Model = glm::mat4(1.0f);
glm::mat4 MVP = Projection * Model;
glViewport( 0, 0, 1366, 768 );
MatrixID = glGetUniformLocation(programID, "MVP");
glEnable(GL_CULL_FACE);
glEnable (GL_BLEND);
glBlendFunc (GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
TextureID = glGetUniformLocation(programID, "myTextureSampler");
glUseProgram(programID);
glUniformMatrix4fv(MatrixID, 1, GL_FALSE, &MVP[0][0]);
glActiveTexture(GL_TEXTURE0);
glBindTexture(GL_TEXTURE_2D, Texture);
glUniform1i(TextureID, 0);
int iQuadVerticeCounter=0;
int iNumOfQuads = 2000;
Quad * ptrQuads = new Quad[iNumOfQuads];
//LOCAL VERTICES CHANGES EACH LOOP
for (int i=0; i<iNumOfQuads; i++)
{
ptrQuads[i].width = 32;
ptrQuads[i].height = 32;
ptrQuads[i].x = (float)(rand() % (1334));
ptrQuads[i].y = (float)(rand() % (736));
}
int iCurrentTime=0;
int iFPS=0;
int iFrames=0;
int iFrameCounterTimeStart=0;
int running = GL_TRUE;
bool bFirstRun=true;
while( running )
{
iCurrentTime = getHighResTimeInMilliSeconds(bFirstRun);
bFirstRun=false;
//UPDATE ALL QUADS EACH FRAME!
for (int i=0; i<iNumOfQuads; i++)
{
ptrQuads[i].width = 32;
ptrQuads[i].height = 32;
ptrQuads[i].x = ptrQuads[i].x;
ptrQuads[i].y = ptrQuads[i].y;
addQuadToLocalVerticeArray(ptrVertexArrayLocal, &ptrQuads[i], &iQuadVerticeCounter);
}
//DO THE RENDERING
glClear( GL_COLOR_BUFFER_BIT );
glBindBuffer(GL_ARRAY_BUFFER, VertexArrayID);
glBufferSubData(GL_ARRAY_BUFFER, 0,sizeof(Vertex) * iQuadVerticeCounter, ptrVertexArrayLocal);
glEnableVertexAttribArray(0);
glBindBuffer(GL_ARRAY_BUFFER, vertexbuffer);
glVertexAttribPointer(0,3,GL_FLOAT,GL_FALSE,sizeof(Vertex),BUFFER_OFFSET(0));
glEnableVertexAttribArray(1);
glBindBuffer(GL_ARRAY_BUFFER, vertexbuffer);
glVertexAttribPointer(1,2,GL_FLOAT,GL_FALSE,sizeof(Vertex),BUFFER_OFFSET(3*sizeof(GL_FLOAT)));
glDrawArrays(GL_TRIANGLES, 0, iQuadVerticeCounter);
glDisableVertexAttribArray(0);
glDisableVertexAttribArray(1);
iQuadVerticeCounter=0;
glfwSwapBuffers();
//END OF DOING THE RENDERING
running = !glfwGetKey( GLFW_KEY_ESC ) &&glfwGetWindowParam( GLFW_OPENED );
iFrames++;
if (iCurrentTime >= iFrameCounterTimeStart + 1000.0f)
{
iFPS = (int)((iCurrentTime - iFrameCounterTimeStart) / 1000.0f * iFrames);
iFrameCounterTimeStart = iCurrentTime;
iFrames = 0;
logger << "FPS: " << iFPS << std::endl;
}
}
glfwTerminate();
exit( EXIT_SUCCESS );
}
int getHighResTimeInMilliSeconds(bool bFirstRun)
{
if (bFirstRun)
glfwSetTime(0);
return (int)((float)glfwGetTime()*1000.0f);
}
GLuint buildShader()
{
//Hint: Shader in the TXT-File looks like this
/*std::stringstream ssVertexShader;
ssVertexShader << "#version 330 core"<< std::endl
<< "layout(location = 0) in vec3 vertexPosition_modelspace;"<< std::endl
<< "layout(location = 1) in vec2 vertexUV;"<< std::endl
<< "out vec2 UV;"<< std::endl
<< "uniform mat4 MVP;"<< std::endl
<< "void main(){"<< std::endl
<< "vec4 v = vec4(vertexPosition_modelspace,1);"<< std::endl
<< "gl_Position = MVP * v;"<< std::endl
<< "UV = vertexUV;"<< std::endl
<< "}"<< std::endl;*/
std::string strVertexShaderCode;
std::ifstream VertexShaderStream("Shader\\VertexShader.txt", std::ios::in);
if(VertexShaderStream.is_open())
{
std::string Line = "";
while(getline(VertexShaderStream, Line))
strVertexShaderCode += "\n" + Line;
VertexShaderStream.close();
}
//Hint: Shader in the TXT-File looks like this
/*std::stringstream ssFragmentShader;
ssFragmentShader << "#version 330 core\n"
"in vec2 UV;\n"
"out vec4 color;\n"
"uniform sampler2D myTextureSampler;\n"
"void main(){\n"
"color = texture( myTextureSampler, UV ).rgba;\n"
"}\n";*/
std::string strFragmentShaderCode;
std::ifstream FragmentShaderStream("Shader\\FragmentShader.txt", std::ios::in);
if(FragmentShaderStream.is_open())
{
std::string Line = "";
while(getline(FragmentShaderStream, Line))
strFragmentShaderCode += "\n" + Line;
FragmentShaderStream.close();
}
GLuint gluiVertexShaderId = glCreateShader(GL_VERTEX_SHADER);
char const * VertexSourcePointer = strVertexShaderCode.c_str();
glShaderSource(gluiVertexShaderId, 1, &VertexSourcePointer , NULL);
glCompileShader(gluiVertexShaderId);
GLint Result = GL_FALSE;
int InfoLogLength;
glGetShaderiv(gluiVertexShaderId, GL_COMPILE_STATUS, &Result);
glGetShaderiv(gluiVertexShaderId, GL_INFO_LOG_LENGTH, &InfoLogLength);
std::vector<char> VertexShaderErrorMessage(InfoLogLength);
glGetShaderInfoLog(gluiVertexShaderId, InfoLogLength, NULL, &VertexShaderErrorMessage[0]);
std::string strInfoLog = std::string(&VertexShaderErrorMessage[0]);
GLuint gluiFragmentShaderId = glCreateShader(GL_FRAGMENT_SHADER);
char const * FragmentSourcePointer = strFragmentShaderCode.c_str();
glShaderSource(gluiFragmentShaderId, 1, &FragmentSourcePointer , NULL);
glCompileShader(gluiFragmentShaderId);
Result = GL_FALSE;
glGetShaderiv(gluiFragmentShaderId, GL_COMPILE_STATUS, &Result);
glGetShaderiv(gluiFragmentShaderId, GL_INFO_LOG_LENGTH, &InfoLogLength);
std::vector<char> FragmentShaderErrorMessage(InfoLogLength);
glGetShaderInfoLog(gluiFragmentShaderId, InfoLogLength, NULL, &FragmentShaderErrorMessage[0]);
strInfoLog = std::string(&FragmentShaderErrorMessage[0]);
GLuint gluiProgramId = glCreateProgram();
glAttachShader(gluiProgramId, gluiVertexShaderId);
glAttachShader(gluiProgramId, gluiFragmentShaderId);
glLinkProgram(gluiProgramId);
Result = GL_FALSE;
glGetProgramiv(gluiProgramId, GL_LINK_STATUS, &Result);
glGetProgramiv(gluiProgramId, GL_INFO_LOG_LENGTH, &InfoLogLength);
std::vector<char> ProgramErrorMessage( std::max(InfoLogLength, int(1)) );
glGetProgramInfoLog(gluiProgramId, InfoLogLength, NULL, &ProgramErrorMessage[0]);
strInfoLog = std::string(&ProgramErrorMessage[0]);
glDeleteShader(gluiVertexShaderId);
glDeleteShader(gluiFragmentShaderId);
return gluiProgramId;
}
void addQuadToLocalVerticeArray(Vertex * ptrVertexArrayLocal, Quad *quad, int *ptrQuadVerticeCounter)
{
//Links oben
ptrVertexArrayLocal[*ptrQuadVerticeCounter].x = quad->x;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].y = quad->y;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].z = 0.0f;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].tx = 0.0f;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].ty = 1.0f;
++(*ptrQuadVerticeCounter);
//Links unten
ptrVertexArrayLocal[*ptrQuadVerticeCounter].x = quad->x;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].y = quad->y - quad->height;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].z = 0.0f;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].tx = 0.0f;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].ty = 0.0f;
++(*ptrQuadVerticeCounter);
//Rechts unten
ptrVertexArrayLocal[*ptrQuadVerticeCounter].x = quad->x + quad->width;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].y = quad->y - quad->height;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].z = 0.0f;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].tx = 1.0f;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].ty = 0.0f;
++(*ptrQuadVerticeCounter);
//Rechts unten
ptrVertexArrayLocal[*ptrQuadVerticeCounter].x = quad->x + quad->width;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].y = quad->y - quad->height;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].z = 0.0f;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].tx = 1.0f;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].ty = 0.0f;
++(*ptrQuadVerticeCounter);
//Rechts oben
ptrVertexArrayLocal[*ptrQuadVerticeCounter].x = quad->x + quad->width;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].y = quad->y;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].z = 0.0f;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].tx = 1.0f;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].ty = 1.0f;
++(*ptrQuadVerticeCounter);
//Links oben
ptrVertexArrayLocal[*ptrQuadVerticeCounter].x = quad->x;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].y = quad->y;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].z = 0.0f;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].tx = 0.0f;
ptrVertexArrayLocal[*ptrQuadVerticeCounter].ty = 1.0f;
++(*ptrQuadVerticeCounter);
}

Answer 2

I identified a huge time-killer now. The textures I was using were too large, and the resolution was very unefficient.
The main-texture which included the level sprites had a resolution of 2200x2200 Pixels. So the GPU increased the size to 4096x4096 and calculated it with a huge amount of data.
The image contains 10x10 different Level-Tiles which are outputed on the screen with a resolution of 50x50 pixels each.
So I saved the Tiles-File with a lower resolution (1020 x 1020 Pixels -> each tile=102x102px) and now I have a loop-cycle time of <=15ms.
This isn't perfect, but in comparison with my previous 30-60ms it was a huge progress.