Develop Reference

How to draw Open GL Elements more efficienty

I've posted my drawing method which is called each frame.
I change the vertices each frame to move the object (which is basically a sprite/textured quad).
As you can see I was initially creating an array each frame, but I have changed this now and I create the array initially and just update it every frame, however I'm wondering if I can do anything more to improve the efficiency? (Although I'm getting about 90fps the sprite does not move smoothly all the time, every now and then it just pauses for a split second). I can't see garbage collector running but I'm guessing it's due to allocation).
As I add more sprites/quads the jerkiness gets worse, but event at 100+ quads, although the smoothness has all but gone, my frame rate is still around 60fps so I can't understand what is slowing this down?
I've also added a screencap from Allocation Tracker
Any help would be appreciated.
public void drawTest(float x, float y, float[] mvpMatrix){
//Convert Co-ordinates
//Left
xPlotLeft = (-MyGLRenderer.ratio)+((x)*MyGLRenderer.coordStepAmountWidth);
//Top
yPlotTop = +1-((y)*MyGLRenderer.coordStepAmountHeight);
//Right
xPlotRight = xPlotLeft+((quadWidth)*MyGLRenderer.coordStepAmountWidth);
//Bottom
yPlotBottom = yPlotTop-((quadHeight)*MyGLRenderer.coordStepAmountHeight);
// Following has been changed as per below. I am now declaring the array initially and just updating it every frame.
// float[] vertices = {
//Top Left
// xPlotLeft,yPlotTop,0, 0,0,
//Top Right
// xPlotRight,yPlotTop,0, 1,0,
//Bottom Left
// xPlotLeft,yPlotBottom,0, 0,1,
//Bottom Right
// xPlotRight,yPlotBottom,0, 1,1
// };
vertices[0]=xPlotLeft;
vertices[1]=yPlotTop;
vertices[2]=0;
vertices[3]=0;
vertices[4]=0;
vertices[5]=xPlotRight;
vertices[6]=yPlotTop;
vertices[7]=0;
vertices[8]=1;
vertices[9]=0;
vertices[10]=xPlotLeft;
vertices[11]=yPlotBottom;
vertices[12]=0;
vertices[13]=0;
vertices[14]=1;
vertices[15]=xPlotRight;
vertices[16]=yPlotBottom;
vertices[17]=0;
vertices[18]=1;
vertices[19]=1;
vertexBuf = ByteBuffer.allocateDirect(vertices.length * 4).order(ByteOrder.nativeOrder()).asFloatBuffer();
vertexBuf.put(vertices).position(0);
//GLES20.glActiveTexture(GLES20.GL_TEXTURE0);
//Bind texture
GLES20.glBindTexture(GLES20.GL_TEXTURE_2D, texID);
//Use program
GLES20.glUseProgram(iProgId);
// Combine the rotation matrix with the projection and camera view
Matrix.multiplyMM(mvpMatrix2, 0, mvpMatrix, 0, mRotationMatrix, 0);
// get handle to shape's transformation matrix
mMVPMatrixHandle = GLES20.glGetUniformLocation(iProgId, "uMVPMatrix");
// Apply the projection and view transformation
GLES20.glUniformMatrix4fv(mMVPMatrixHandle, 1, false, mvpMatrix2, 0);
//Set starting position for vertices (0 for position)
vertexBuf.position(0);
//Specify attributes for vertex
GLES20.glVertexAttribPointer(iPosition, 3, GLES20.GL_FLOAT, false, 5 * 4, vertexBuf);
//Enable attribute for position
GLES20.glEnableVertexAttribArray(iPosition);
//Set starting position for vertices (3 for texture)
vertexBuf.position(3);
//Specify attributes for vertex
GLES20.glVertexAttribPointer(iTexCoords, 2, GLES20.GL_FLOAT, false, 5 * 4, vertexBuf);
//Enable attribute for texture
GLES20.glEnableVertexAttribArray(iTexCoords);
//Enable Alpha blending and set blending function
GLES20.glEnable(GLES20.GL_BLEND);
GLES20.glBlendFunc(GLES20.GL_ONE, GLES20.GL_ONE_MINUS_SRC_ALPHA);
//Draw
GLES20.glDrawArrays(GLES20.GL_TRIANGLE_STRIP, 0, 4);
//Disable Alpha blending
GLES20.glDisable(GLES20.GL_BLEND);
}

ByteBuffer.allocateDirect() allocates a new buffer in memory every frame, you can create an initial buffer and overwrite the contents instead. Just use rewind() or position(0) before put().
To improve matters further, use a VBO (vertex buffer object, there are many tutorials online, and several questions on SO on this topic) and glBufferSubData to update the buffer.

Low FPS with OpenGL on Android

I contact because, I try to use openGL with android, in order to make a 2D game :)
Here is my way of working:
I have a class GlRender
public class GlRenderer implements Renderer
In this class, on onDrawFrame I do
GameRender() and GameDisplay()
And on gameDisplay() I have:
gl.glClear(GL10.GL_COLOR_BUFFER_BIT | GL10.GL_DEPTH_BUFFER_BIT);
// Reset the Modelview Matrix
gl.glMatrixMode(GL10.GL_PROJECTION); //Select The Modelview Matrix
gl.glLoadIdentity(); //Reset The Modelview Matrix
// Point to our buffers
gl.glEnableClientState(GL10.GL_VERTEX_ARRAY);
gl.glEnableClientState(GL10.GL_TEXTURE_COORD_ARRAY);
// Set the face rotation
gl.glFrontFace(GL10.GL_CW);
for(Sprites...)
{
sprite.draw(gl, att.getX(), att.getY());
}
//Disable the client state before leaving
gl.glDisableClientState(GL10.GL_VERTEX_ARRAY);
gl.glDisableClientState(GL10.GL_TEXTURE_COORD_ARRAY);
And in the draw method of sprite I have:
_vertices[0] = x;
_vertices[1] = y;
_vertices[3] = x;
_vertices[4] = y + height;
_vertices[6] = x + width;
_vertices[7] = y;
_vertices[9] = x + width;
_vertices[10] = y + height;
if(vertexBuffer != null)
{
vertexBuffer.clear();
}
// fill the vertexBuffer with the vertices
vertexBuffer.put(_vertices);
// set the cursor position to the beginning of the buffer
vertexBuffer.position(0);
// bind the previously generated texture
gl.glBindTexture(GL10.GL_TEXTURE_2D, textures[0]);
// Point to our vertex buffer
gl.glVertexPointer(3, GL10.GL_FLOAT, 0, vertexBuffer.mByteBuffer);
gl.glTexCoordPointer(2, GL10.GL_FLOAT, 0, textureBuffer.mByteBuffer);
// Draw the vertices as triangle strip
gl.glDrawArrays(GL10.GL_TRIANGLE_STRIP, 0, _vertices.length / 3);
My problem is that I have a low frame rate, even at 30 FPS I loose some frame sometimes with only 1 sprite (but it is the same with 50)
Am I doing something wrong? How can I improve FPS?

In general, you should not be changing your vertex buffer for every sprite drawn. And by "in general", I pretty much mean "never," unless you're making a particle system. And even then, you would use proper streaming techniques, not write a quad at a time.
For each sprite, you have a pre-built quad. To render it, you use shader uniforms to transform the sprite from a neutral position to the actual position you want to see it on screen.

OpenGL ES 1.1 strange lighting problems

I am examining an interesting problem I'm facing with OpenGL lighting on Android. I'm working on a 3D Viewer where you can add and manipulate 3D objects. You can also set a light with different attributes. The problem I was facing with my Viewer was that the highlight on the 3D objects from the light (it is a point light) behaved strangely. If the light source was in the exact same point as the camera, the highlight would move in the opposite direction you would expect. (So if you move the object to the left, the highlight moves to the leftedge of the object as well, instead of the right, which is what I was expecting.)
So to further narrow the problem down I've created a small sample application that only renders a square and then I rotate that square around the camera position (the origin), which is also where the light is placed. This should result in all squares facing the camera directly, so that they would be completely highlighted. The result though looked like that:
Can it be that these artifacts appear because of the distortion you get on the border due to the projection?
In the first image the distance between the sphere and the camera is about 20 units and the size of the sphere is about 2. If I move the light closer to the object the highlight looks a lot better, in the way I'm expecting it.
In the second image the radius in which the squares are located is 25 units.
I'm using OpenGL ES 1.1 (since I was struggling to get it to work with shaders in ES 2.0) on Android 3.1
Here is some of the code I'm using:
public void onDrawFrame(GL10 gl) {
// Setting the camera
GLU.gluLookAt(gl, 0, 0, 0, 0f, 0f, -1f, 0f, 1.0f, 0.0f);
gl.glMatrixMode(GL10.GL_MODELVIEW);
gl.glLoadIdentity();
for (int i = 0; i < 72; i++) {
gl.glPushMatrix();
gl.glRotatef(5f * i, 0, 1, 0);
gl.glTranslatef(0, 0, -25);
draw(gl);
gl.glPopMatrix();
}
}
public void draw(GL10 gl) {
setMaterial(gl);
gl.glEnable(GL10.GL_NORMALIZE);
gl.glEnableClientState(GL10.GL_NORMAL_ARRAY);
gl.glEnableClientState(GL10.GL_VERTEX_ARRAY);
gl.glFrontFace(GL10.GL_CCW);
// Enable the vertex and normal state
gl.glVertexPointer(3, GL10.GL_FLOAT, 0, mVertexBuffer);
gl.glNormalPointer(GL10.GL_FLOAT, 0, mNormalBuffer);
gl.glDrawElements(GL10.GL_TRIANGLES, mIndexBuffer.capacity(), GL10.GL_UNSIGNED_SHORT, mIndexBuffer);
gl.glDisableClientState(GL10.GL_VERTEX_ARRAY);
gl.glDisableClientState(GL10.GL_NORMAL_ARRAY);
}
// Setting the light
private void drawLights(GL10 gl) {
// Point Light
float[] position = { 0, 0, 0, 1 };
float[] diffuse = { .6f, .6f, .6f, 1f };
float[] specular = { 1, 1, 1, 1 };
float[] ambient = { .2f, .2f, .2f, 1 };
gl.glEnable(GL10.GL_LIGHTING);
gl.glEnable(GL10.GL_LIGHT0);
gl.glMatrixMode(GL10.GL_MODELVIEW);
gl.glLoadIdentity();
gl.glLightfv(GL10.GL_LIGHT0, GL_POSITION, position, 0);
gl.glLightfv(GL10.GL_LIGHT0, GL_DIFFUSE, diffuse, 0);
gl.glLightfv(GL10.GL_LIGHT0, GL_AMBIENT, ambient, 0);
gl.glLightfv(GL10.GL_LIGHT0, GL_SPECULAR, specular, 0);
}
private void setMaterial(GL10 gl) {
float shininess = 30;
float[] ambient = { 0, 0, .3f, 1 };
float[] diffuse = { 0, 0, .7f, 1 };
float[] specular = { 1, 1, 1, 1 };
gl.glMaterialfv(GL_FRONT_AND_BACK, GL_DIFFUSE, diffuse, 0);
gl.glMaterialfv(GL_FRONT_AND_BACK, GL_AMBIENT, ambient, 0);
gl.glMaterialfv(GL_FRONT_AND_BACK, GL_SPECULAR, specular, 0);
gl.glMaterialf(GL_FRONT_AND_BACK, GL_SHININESS, shininess);
}
I'm setting the light in the beginning, when the activity is started (in onSurfaceCreated) and the material everytime I draw a square.

The effect in your second example (with the squares) is rather due to the default non-local viewer that OpenGL uses. By default the eye-space view vector (the vector from vertex to camera, used for the specular highlight computation) is just taken to be the (0, 0, 1)-vector, instead of the normalized vertex position. This approximation is only correct if the vertex is in the middle of the screen, but gets more and more incorrect the farther you move to the boundary of the srceen.
To change this and let OpenGL use the real vector from the vertex to the camera, just use the glLightModel function, especially
glLightModeli(GL_LIGHT_MODEL_LOCAL_VIEWER, GL_TRUE);
I'm not sure if this is also the cause for your first problem (with the sphere), but maybe, just try it.
EDIT: It seems you cannot use GL_LIGHT_MODEL_LOCAL_VIEWER in OpenGL ES. In this case there is no way around this problem, except switching to OpenGL ES 2.0 and doing all lighting computations yourself, of course.

Your light is probably moving when you're moving your object.
Take a look at this answer http://www.opengl.org/resources/faq/technical/lights.htm#ligh0050

OpenGLES10: Apply Projection & Camera View

I need a little help with this:
android developers, Tutorials: OpenGLES10.
a link
It all works fine for the first Triangle, until I put in the code for Projection & Camera View. This should rezise OpenGLES Square view to match Phone's screen, so object stay in propotions.
As a Newbie watching, the code looks fine and i have cheked with referencefiles, that there's not missing a parameter or something like that. But now i'm lost..! Can't see what's wrong.
If Projection and Camera code are applied, there is no triangle, but the app. is runing and the View with backgroundcolor are shown.
Here is my code:
package notme.helloopengles10;
import java.nio.ByteBuffer;
import java.nio.ByteOrder;
import java.nio.FloatBuffer;
import javax.microedition.khronos.egl.EGLConfig;
import javax.microedition.khronos.opengles.GL10;
import android.opengl.GLSurfaceView;
import android.opengl.GLU;
public class HelloOpenGLES10Renderer implements GLSurfaceView.Renderer {
// Set the background frame color
public void onSurfaceCreated(GL10 gl, EGLConfig config) {
gl.glClearColor(0.5f, 0.5f, 0.5f, 1.0f);
// initialize the triangle vertex array
initShapes();
//enable use of vertex arrays
gl.glEnableClientState(GL10.GL_VERTEX_ARRAY);
}
public void onDrawFrame(GL10 gl) {
// Redraw background color
gl.glClear(GL10.GL_COLOR_BUFFER_BIT | GL10.GL_DEPTH_BUFFER_BIT);
/* // set GL_MODELVIEW transformation mode (If outline from here to after GLU.gluLookAt() - it works when also outlines further down i code!
gl.glMatrixMode(GL10.GL_MODELVIEW);
gl.glLoadIdentity(); // reset Matrix to its default state
// when using GL_MODELVIEW, you must set the view point
GLU.gluLookAt(gl, 0, 0, -5, 0f, 0f, 0f, 0f, 1.0f, 0.0f); */
//Draw Triangel
gl.glColor4f(0.63671875f, 0.76953125f, 0.22265625f, 0.0f);
gl.glVertexPointer(3, GL10.GL_FLOAT, 0, triangleVB);
gl.glDrawArrays(GL10.GL_TRIANGLES, 0, 3);
}
// Redraw on orientation changes // adjust for screen size ratio
public void onSurfaceChanged(GL10 gl, int width, int height) {
gl.glViewport(0, 0, width, height);
// Make adjustments for screen ratio
/*(If outline from here to after gl.Frumstumf() - it works!
float ratio = (float) width / height;
gl.glMatrixMode(GL10.GL_PROJECTION); // set matrix to projection mode
gl.glLoadIdentity(); // reset the matrix to its default state
gl.glFrustumf(-ratio, ratio, -1, 1, 3, 7); // apply the projection */
}
/*
* Draw a shape, a triangle. first add new member variable to contain
* the vertices of a triangle
*/
private FloatBuffer triangleVB;
//Create a method, initShaoe(), which populate the members variable
private void initShapes(){
//create a array
float triangleCoords[] = {
// X, Y, Z
-0.5f, -0.25f, 0,
0.5f, -0.25f, 0,
0.0f, 0,559016994f, 0
};
// initialize vertex Buffer for triangle
ByteBuffer vbb= ByteBuffer.allocateDirect(
//(# of coordinates values * 4 bytes per float)
triangleCoords.length * 4 );
vbb.order(ByteOrder.nativeOrder()); // use device hardware's native byte order
triangleVB = vbb.asFloatBuffer(); //create floating point buffer from the ByteBuffer
triangleVB.put(triangleCoords); // add coordinates to the FloatBuffer
triangleVB.position(0); // set the buffer to read the first coordinate
}
} // end
I hope some one can tell me, where things go wrong?
DevTool: Eclipse.

I had the same problem with this tutorial and it got solved when I changed the order of multiplying in the vertex shader code in the Triangle class. So instead of having uMVPMatrix * vPosition, replace it with vPosition * uMVPMatrix. I guess the reason for this is because vPosition is a row vector.

The code looks resonable (if you uncomment the parts that are commented out at the moment). Your matrix modification code is quite correct and all transformations are applied to the correct matrices.
But at the moment you are looking from the point (0,0,-5) to the point (0,0,0) and therefore along the +z axis. But since the default OpenGL view looks along the -z axis, you actually rotate the view 180 degrees around the y-axis. Whereas this is absolutely no problem, you now see the back-side of the triangle. So can it be, that you have back-face culling enabled and this back-side is just optimized away? Just try disabling back-face culling by calling glDisable(GL_CULL_FACE) or change the -5 in the gluLookAt call to a 5, so that you look along the -z axis.
You can also try to use gluPerspective(45, ratio, 3, 7) instead of the glFrustum call, but your arguments to glFrustum look quite reasonable. Of course, keep in mind that both calls create a perspective view, with farther objects getting smaller, like in reality. If you actually want a parallel/orthographic view (where size on screen is independent on depth) you should replace the glFrustum with a glOrtho, though the parameters can stay the same.

Your call to gluLookAt trashes your modelview matrix. You should call this function with the projection matrix active.
http://www.opengl.org/sdk/docs/man/xhtml/gluLookAt.xml
This code shows the triangle for me:
public void onDrawFrame(GL10 gl) {
// Redraw background color
gl.glClear(GL10.GL_COLOR_BUFFER_BIT | GL10.GL_DEPTH_BUFFER_BIT);
gl.glMatrixMode(GL10.GL_PROJECTION);
gl.glLoadIdentity();
// when using GL_MODELVIEW, you must set the view point
GLU.gluLookAt(gl, 0, 0, -5, 0f, 0f, 0f, 0f, 1.0f, 0.0f);
// set GL_MODELVIEW transformation mode (If outline from here to after GLU.gluLookAt() - it works when also outlines further down i code!
gl.glMatrixMode(GL10.GL_MODELVIEW);
gl.glLoadIdentity(); // reset Matrix to its default state
//Draw Triangel
gl.glColor4f(0.63671875f, 0.76953125f, 0.22265625f, 0.0f);
gl.glVertexPointer(3, GL10.GL_FLOAT, 0, triangleVB);
gl.glDrawArrays(GL10.GL_TRIANGLES, 0, 3);
}

Increasing drawing performance with a hexagon grid

I'm working on my very first openGL game, inspired by the game "Greed Corp" on the playstation network. It's a turn based strategy game that is based on a hex grid. Each hexagon tile has it's own height and texture.
I'm currently drawing a hexagon based on some examples and tutorials I've read. Here's my hextile class:
public class HexTile
{
private float height;
private int[] textures = new int[1];
private float vertices[] = { 0.0f, 0.0f, 0.0f, //center
0.0f, 1.0f, 0.0f, // top
-1.0f, 0.5f, 0.0f, // left top
-1.0f, -0.5f, 0.0f, // left bottom
0.0f, -1.0f, 0.0f, // bottom
1.0f, -0.5f, 0.0f, // right bottom
1.0f, 0.5f, 0.0f, // right top
};
private short[] indices = { 0, 1, 2, 3, 4, 5, 6, 1};
//private float texture[] = { };
private FloatBuffer vertexBuffer;
private ShortBuffer indexBuffer;
//private FloatBuffer textureBuffer;
public HexTile()
{
ByteBuffer vbb = ByteBuffer.allocateDirect(vertices.length * 4);
vbb.order(ByteOrder.nativeOrder());
vertexBuffer = vbb.asFloatBuffer();
vertexBuffer.put(vertices);
vertexBuffer.position(0);
ByteBuffer ibb = ByteBuffer.allocateDirect(indices.length * 2);
ibb.order(ByteOrder.nativeOrder());
indexBuffer = ibb.asShortBuffer();
indexBuffer.put(indices);
indexBuffer.position(0);
/*ByteBuffer tbb = ByteBuffer.allocateDirect(texture.length * 4);
tbb.order(ByteOrder.nativeOrder());
textureBuffer = tbb.asFloatBuffer();
textureBuffer.put(texture);
textureBuffer.position(0);*/
}
public void setHeight(float h)
{
height = h;
}
public float getHeight()
{
return height;
}
public void draw(GL10 gl)
{
gl.glBindTexture(GL10.GL_TEXTURE_2D, textures[0]);
gl.glVertexPointer(3, GL10.GL_FLOAT, 0, vertexBuffer);
//gl.glTexCoordPointer(2, GL10.GL_FLOAT, 0, textureBuffer);
gl.glDrawElements(GL10.GL_TRIANGLE_FAN, indices.length, GL10.GL_UNSIGNED_SHORT, indexBuffer);
}
public void loadGLTexture(GL10 gl, Context context)
{
textures[0] = -1;
Bitmap bitmap = BitmapFactory.decodeResource(context.getResources(), R.drawable.hex);
while(textures[0] <= 0)
gl.glGenTextures(1, textures, 0);
gl.glBindTexture(GL10.GL_TEXTURE_2D, textures[0]);
gl.glTexParameterf(GL10.GL_TEXTURE_2D, GL10.GL_TEXTURE_MIN_FILTER, GL10.GL_LINEAR);
gl.glTexParameterf(GL10.GL_TEXTURE_2D, GL10.GL_TEXTURE_MAG_FILTER, GL10.GL_LINEAR);
GLUtils.texImage2D(GL10.GL_TEXTURE_2D, 0, bitmap, 0);
bitmap.recycle();
}
}
Every frame I'm looping through all visible tiles to draw them, that would be 11 * 9 tiles MAX. This however drops my framerate to 38, and that is without even drawing textures on them, just the flat hexagons.
Now I'm trying to figure out how to increase performance. I figured drawing the whole grid at once could be faster, but I have no idea how to do that, since each tile can have a different height, and will most likely have a different texture than a neighboring tile.
I'd really appreciate some help on this, because I'd like to get started on the actual game ^.^

Assuming your hex grid is static you can just spin over all your hexagons once, generate their geometry, and append everything to one (or more, if you have more than 2^16 vertices) large VBO that you can draw in one go.
As far as textures go you may be able to use a texture atlas.

I'm currently learning OpenGL as well, I've produced one Android OpenGL application called 'Cloud Stream' where I found similar issues with performance.
As a general answer to performance concerns, there are a few things which can help. The Vertex pipeline for Graphics at a hardware level apparently is more efficient when passed larger amounts of vertices at once. Calling glVertexPointer for each Hex tile is not as efficient as calling it once with the vertices of all tiles.
This makes things harder to code as you essentially draw all your tiles at once, but it does seem to speed things up a bit. In my application, all of the clouds are drawn in the one call.
Other avenues to try would be to save the Vertice positions in a VBO which I found to be quite tricky, at least it was when trying to cater for 2.1 users. These days things might be easier, I'm not sure. With that the idea is to save the Vertice array for your tile into Video Memory and you get back a pointer like you do with your Textures. As you can imagine, not sending your Vertice Array Buffer up each frame speeds things up a little for each tile draw. Even if things aren't static its a good approach as I doubt things are changing for each frame.
Another suggestion I came across online was to use Short instead of Float for your Vertices. And then to change the scale of your finished rendering to get your desired size. This would lower the size of your vertices and speed things up a little... not a huge amount but still worth trying I would say.
One last thing I would like to add, if you end up using any Transparency... be aware that you must paint back to front for it to work which has a performance impact as well. If you draw front to back, the Rendering engine automatically knows not to draw when coordinates are not visible... drawing back to front means everything is drawn. Keep this in mind and always try to draw front to back when possible.
Good luck with your game, I'd love to know how you got on... I'm just starting my game and I'm quite excited to start. If you haven't already come across this... I think it's worth a read. http://www.codeproject.com/KB/graphics/hexagonal_part1.aspx