I wrote a simple program using OpenGL 4.3 which displays a triangle, quadrilateral, and pentagon. I defined my vertices in the following array:
vec2 vertices[NumPoints] = {
vec2(-1, -0.75), vec2(-0.75, -0.25), vec2(-0.5, -0.75), //Triangle
vec2(0, -0.25), vec2(0.5, -0.25), vec2(0.5, -0.75), vec2(0, -0.75), //Quad
vec2(0.25, 0.25), vec2(0.5, 0.5), vec2(0.75, 0.25), vec2(0.65, 0), vec2(0.35, 0) // pentagon
};
For the sake of brevity I'll omit most of the boilerplate code. In my display function I have the following code:
glDrawArrays(GL_TRIANGLES, 0, 3); // draw the points
glDrawArrays(GL_TRIANGLE_FAN, 3, 4); //quad
glDrawArrays(GL_TRIANGLE_FAN, 7, 5); //polygon
Everything works fine and there isn't any problems. However, it seems rather tedious and almost impossible to create complex scenes if you need know exactly how many vertices you need upfront. Am I missing something here? Also, if needed to create a circle, how would I do that using just GL_TRIANGLES?
In a real world application you will have scene management, with multiple objects in the scene and multiple sub-objects for each object. Objects are responsible for generating their vertex data and corresponding drawing calls and schedule them appropriately.
For example, you can have a cube object that has a single property - edge length. From that single property you generate the full set of vertices required to render a cube.
You can also chose to convert the cube primitive to another compatible object, for example a box primitive where you have 3 properties - height, width and depth, or even an arbitrary polygonal mesh that is made of faces, which are made of edges which are made of vertices.
It is a good idea to sort the different scene objects in such an order to allow minimizing the number of draw calls, which is the typical bottleneck 3D graphics struggles with. Combined with instancing and adaptive LOD you can get significant performance improvements.
For the circle - in primitive mode the most efficient way to draw it is using a triangle fan. But if you convert the circle primitive to a polygonal mesh, you could render regular triangles. The number of vertices, needed to draw the circle will grow. With triangle fan, you need 3 vertices for the first triangle and then only 1 additional vertex for every additional segment, with regular triangles you will need the full 3 vertices for every segment of the circle.
Am I missing something here?
Yes. You can allocate memory dynamically and read data from files. That's how any real world program deals with this kind of things. You'll have some scene management structure which allows to load a scene and objects from files. The file itself will contain some metadata, as number of faces, vertices, etc. which can be used to prepare the data structures at runtime.
Related
I have a rectangle and a circle. Of the circle i have all points' coordinates because i calculate them to draw it using math rules. the rectangle in drawed using two triangles so 4 vertices. Now these are free to translate and route in the plan and i want to determinate when one of them touch the other one. so I thought that this happens when one of the coordinates of one of them is the same that one of the others of the other object. The problem is that i haven't an array of all coordinates of the rectangle. Is there a method that return all coordinates that a drawed triangles and not only the vertices' ones in OpenGL?
There is method to record coords and commands supplied to OpenGL, using stencil buffer, but that's a rather inefficient way, because you would need to decompile commands inside buffer.
If you didn't had an array of coordinates , you already used the most inefficient way to supply geometry to OpenGL:
glBegin(...);
glVertex3f(...);
glVertex3f(...);
...
glVertex3f(...);
glEnd();
The more efficient way to do that is to use vertex buffer, which automatically requires to have array of coordinates. With large amount of vertices, VBO methods is times faster than vertex by vertex copying.
OpenGL doesn't store the coordinates you've supplied to it any longer than it is required, i.e. until rasterization. Whole goal of OpenGL is to create image on screen, not to solve some abstract tasks.
I want to render multiple 3D cubes from one vbo. Each cube has a uniform color.
At this time, I create a vbo where each vertex has a color information.
Is it posible to upload only one color for a one shape (list of verticies)?
I'm also want to mix GL_TRIANGLES and GL_LINES in the glDrawElements-method of the same shader. Is it posible?
//Edit : I only have OpenGL 2.1. Later I want to build this project on Android.
//Edit 2:
I want to render a large count of cubes (up to 150.000). One cube has 24 verticies of geometry and color and 34 indices. Now my idea is to create some vbo's (maybe 50) and share out the cubes to the vbo's. I hope that this minimizes the overhead.
Drawing lots of cubes
Yes, if you want to draw a bunch of cubes, you can specify the color for each cube once.
Create a VBO containing the vertexes for one cube.
// cube = 36 vertexes with glDrawArrays(GL_TRIANGLES)
vbo1 = [v1] [v2] [v3] ... [v36]
Create another VBO with the view matrix and color for each cube, and use an attribute divisor of 1. (You can use the same vbo, but I would use a separate one.)
vbo2 = [cube 1 mat, color] [cube 2 mat, color] ... [cube N mat, color]
Call glDrawElementsInstanced() or glDrawArraysInstanced(). This will draw the cube over and over again.
Alternatively, you can use glUniform() for each cube, but this will limit the number of cubes you can draw. The above method will let you draw thousands, easily.
Mixing GL_TRIANGLES and GL_LINES
You will have to call glDraw????() once for each type of primitive. You can use the same shader for both times, if you like.
Regarding your questions :
Is it possible to upload only one color for one shape ?
Yes , you can use a uniform instead of a vertex attribute(ofc this means changes in more places). However, you will need to set the uniform for each shape, and have a different drawcall for each differently colored shape .
Is it possible to mix GL_TRIANGLES and GL_LINES in the glDrawElements ?
Yes and no. Yes , but you will need a new drawcall (which is obvious). You cannot do on the same drawcall some shapes with GL_TRIANGLES and some shapes with GL_LINES.
In pseudocode this will look like this :
draw shapes 1,2,10 from the vbo using color red and GL_TRIANGLES
draw shapes 3,4,6 from the vbo using color blue and GL_LINES
draw shapes 7,8,9 from the vb using color blue and GL_TRIANGLES
With OpenGL 2.1, I don't think there's a reasonable way of specifying the color only once per cube, and still draw everything in a single draw call.
The most direct approach is that, instead of having the color attribute in a VBO, you specify it directly before the draw call. Assuming that you're using generic vertex attributes, where you would currently have:
glEnableVertexAttribArray(colorLoc);
glVertexAttripPointer(colorLoc, ...);
you do this:
glDisableVertexAttribArray(colorLoc);
glVertexAttrib3f(colorLoc, r, g, b);
where glDisableVertexAttribArray() is only needed if the array was previously enabled for the location.
The big disadvantage is that you can only draw cubes with the same color in one draw call. In the extreme case, that's one draw call per cube. Of course if you have multiple cubes with the same color, you could still batch those into a single draw call.
You wonder whether this is more efficient than having a color for each vertex in the VBO? Impossible to say in general. You'll always get the same answer in cases like this: Try both, and benchmark. I'm skeptical that you will find it beneficial. In my experience, it's fairly rare for fetching vertex data to be a major performance bottleneck. So cutting out one attribute will likely no give you much of a gain. On the other hand, making many small draw calls absolutely can (and often will) hurt performance.
There is one option you can use that is sort of a hybrid. I'm not necessarily recommending it, but just in the interest of brainstorming. If you use a fairly limited number of colors, you can use a single scalar attribute in the VBO that encodes a "color index". Then in the vertex shader, you can use a texture lookup to translate the "color index" to the actual color.
The really good options are beyond OpenGL 2.1. #DietrichEpp nicely explained instanced rendering, which is an elegant solution for cases like this.
And no, you can not have lines and triangles in the same draw call. Even the most flexible draw calls in OpenGL 4.x, like glDrawElementsIndirect(), still take only one primitive type.
I'm migrating our graphics ending from using the old fixed pipeline functions to making use of the programmable pipeline. Our simplest model is just a collection of points in space where each point can be represented by different shapes. One of these being a cube.
I'm basing my code off the cube example from the OpenGL superbible.
In this example the cubes are placed at somewhat random places whereas I will have a fixed lit of points in space. I'm wondering if there is a way to pass that list to my shader so that a cube is drawn at each point vs looping through the list and calling glDrawElements each time. Is that even worth the trouble (performance wise)?
PS we are limited to OpenGL 3.3 functionality.
Is that even worth the trouble (performance wise)?
Probably yes, but try to profile nonetheless.
What you are looking for is instanced rendering, take a look at glDrawElementsInstanced and glVertexAttribDivisor.
What you want to do is store the 8 vertices of a generic cube (centered on the origin) in one buffer, and also store the coordinates of the center of each cube in another vertex attribute buffer.
Then you can use glDrawElementsInstanced to draw N cubes taking the vertices from the first buffer, and translating them in the shader using the specific position stored in the second buffer.
Something like this:
glVertexAttribPointer( vertexPositionIndex, /** Blah .. */ );
glVertexAttribPointer( cubePositionIndex, /** Blah .. */ );
glVertexAttribDivisor( cubePositionIndex, 1 ); // Advance one vertex attribute per instance
glDrawElementsInstanced( GL_TRIANGLES, 36, GL_UNSIGNED_BYTE, indices, NumberOfCubes );
In your vertex shader you need two attributes:
vec3 vertexPosition; // The coordinates of a vertex of the generic cube
vec3 cubePosition; // The coordinates of the center the specific cube being rendered
// ....
vec3 vertex = vertexPosition + cubePosition;
Obviously you can have also a buffer to store the size of each cube, or another one for the orientation, the idea remains the same.
In your example every cube uses its own model matrix per frame.
If you want to keep that you need multiple drawElements calls.
If some cubes don't move (don't need a per frame model matrix) you should combine these cubes into one VBO.
I have to support some legacy code which draws point clouds using the following code:
glEnableClientState(GL_VERTEX_ARRAY);
glVertexPointer(3, GL_FLOAT, 0, (float*)cloudGlobal.data());
glEnableClientState(GL_NORMAL_ARRAY);
glNormalPointer(GL_FLOAT, 0, (float*)normals.data());
glDrawArrays(GL_POINTS, 0, (int)cloudGlobal.size());
glFinish();
This code renders all vertices regardless of the angle between normal and the "line of sight". What I need is draw only vertices whose normals are directed towards us.
For faces this would be called "culling", but I don't know how to enable this option for mere vertices. Please suggest.
You could try to use the lighting system (unless you already need it for shading). Set ambient color alpha to zero, and then simply use alpha test to discard the points with zero alpha. You will probably need to set quite high alpha in diffuse color in order to avoid half-transparent points, in case alpha blending is required to antialiass the points (to render discs instead of squares).
This assumes that the vertices have normals (but since you are talking about "facing away", I assume they do).
EDIT:
As correctly pointed out by #derhass, this will not work.
If you have cube-map textures, perhaps you can copy normal to texcoord and perform lookup of alpha from a cube-map (also in combination with the texture matrix to take camera and point cloud transformations into account).
Actually in case your normals are normalized, you can scale them using the texture matrix to [-0.49, +0.49] and then use a simple 1D (or 2D) bar texture (half white, half black - incl. alpha). Note that counterintuitively, this requires texture wrap mode to be left as default GL_REPEAT (not clamp).
If your point clouds have shape of some closed objects, you can still get similar behavior even without cube-map textures by drawing a dummy mesh with glColorMask(0, 0, 0, 0) (will only write depth) that will "cover" the points that are facing away. You can generate this mesh also as a group of quads that are placed behind the points in the opposite direction of their normal, and are only visible from the other side than the points are supposed to be visible, thus covering them.
Note that this will only lead to visual improvement (it will look like the points are culled), not performance improvement.
Just out of curiosity - what's your application and why do you need to avoid shaders?
Since GL_QUADS has been removed from OpenGL 3.1 and above, what is the fastest way to draw lots of quads without using it? I've tried several different methods (below) and have ranked them on speed on my machine, but I was wondering if there is some better way, since the fastest way still seems wasteful and inelegant. I should mention that in each of these methods I'm using VBOs with interleaved vertex and texture coordinates, since I believe that to be best practice (though I may be wrong). Also, I should say that I can't reuse any vertices between separate quads because they will have different texture coordinates.
glDrawElements with GL_TRIANGLE_STRIP using a primitive restart index, so that the index array looks like {0, 1, 2, 3, PRI, 4, 5, 6, 7, PRI, ...}. This takes in the first 4 vertices in my VBO, treats them as a triangle strip to make a rectangle, and then treats the next 4 vertices as a separate strip. The problem here is just that the index array seems like a waste of space. The nice thing about GL_QUADS in earlier versions of OpenGL is that it automatically restarts primitives every 4 vertices. Still, this is the fastest method I can find.
Geometry shader. I pass in 1 vertex for each rectangle and then construct the appropriate triangle strip of 4 vertices in the shader. This seems like it would be the fastest and most elegant, but I've read, and now seen, that geometry shaders are not that efficient compared to passing in redundant data.
glDrawArrays with GL_TRIANGLES. I just draw every triangle independently, reusing no vertices.
glMultiDrawArrays with GL_TRIANGLE_STRIP, an array of all multiples of 4 for the "first" array, and an array of a bunch of 4's for the "count" array. This tells the video card to draw the first 4 starting at 0, then the first 4 starting at 4, and so on. The reason this is so slow, I think, is that you can't put these index arrays in a VBO.
You've covered all the typical good ways, but I'd like to suggest a few less typical ones that I suspect may have higher performance. Based on the wording of the question, I shall assume that you're trying to draw an m*n array of tiles, and they all need different texture coordinates.
A geometry shader is not the right tool to add and remove vertices. It's capable of doing that, but it's really intended for cases when you actually change the number of primitives you're rendering dynamically (e.g. shadow volume generation). If you just want to draw a whole bunch of adjacent different primitives with different texture coordinates, I suspect the absolute fastest way would be to use tessellation shaders. Just pass in a single quad and have the tessellator compute texture coordinates procedurally.
A similar and more portable method would be to look up each quad's texture coordinate. This is trivial: say you're drawing 50x20 quads, you would have a 50x20 texture that stores all your texture coordinates. Tap this texture in your vertex program (or perhaps more efficiently in your geometry program) and send the result in a varying to the fragment program for actual rendering.
Note that in both of the above cases, you can reuse vertices. In the first method, the intermediate vertices are generated on the fly. In the second, the vertices' texture coordinates are replaced in the shader with cached values from the texture.