CODEWITHSUNDEEP
javaandroidopengl-es
alexrnov
alexrnov

Reputation: 2534

Several separate vertex buffer objects

For rendering a 3D object, four separate vertex buffers are created: for vertices, indices, texture coordinates and normals:

private final int[] VBO = new int[4]; // array for vertex buffer objects

private void createVertexBuffers() {
  VBO[0] = 0; VBO[1] = 0; VBO[2] = 0; VBO[3] = 0;

  GLES20.glGenBuffers(4, VBO, 0);
  bufferVertices.position(0);
  GLES20.glBindBuffer(GLES20.GL_ARRAY_BUFFER, VBO[0]);
  GLES20.glBufferData(GLES20.GL_ARRAY_BUFFER, VERTEX_STRIDE * NUMBER_VERTICES,
        bufferVertices, GLES20.GL_STATIC_DRAW); // VBO for vertex

  bufferTextureCoordinates.position(0);
  GLES20.glBindBuffer(GLES20.GL_ARRAY_BUFFER, VBO[1]);
  GLES20.glBufferData(GLES20.GL_ARRAY_BUFFER, TEXTURE_STRIDE * NUMBERS_TEXTURES,
        bufferTextureCoordinates, GLES20.GL_STATIC_DRAW); // VBO for texture coordinates

  bufferNormals.position(0);
  GLES20.glBindBuffer(GLES20.GL_ARRAY_BUFFER, VBO[2]);
  GLES20.glBufferData(GLES20.GL_ARRAY_BUFFER, VERTEX_STRIDE * NUMBER_NORMALS,
        bufferNormals, GLES20.GL_STATIC_DRAW); // VBO for normals

  bufferIndices.position(0);
  GLES20.glBindBuffer(GLES20.GL_ELEMENT_ARRAY_BUFFER, VBO[3]);
  GLES20.glBufferData(GLES20.GL_ELEMENT_ARRAY_BUFFER, INT_SIZE * NUMBER_INDICES,
        bufferIndices, GLES20.GL_STATIC_DRAW); // VBO for indices
}

The 3D-objects themselves are also many, respectively, the number of buffers additionally is increasing. Question: Is it a normal approach to use many separate buffers? In mobile apps? I would be grateful for the answers.

Note: looked at similar questions, but I still have uncertainty.

Upvotes: 1

Views: 444

Answers (1)

solidpixel
solidpixel

Reputation: 12069

Interleaved attributes (array of structs) are generally more efficient than completely de-interleaved ones (struct of arrays). The reason for this is that you're less likely to load a whole cache line and then only use one value from it.

However recent mobile implementations still like some level of deinterleaving. For tile-based GPUs it's common to process position computation first, and then only process the rest of the vertex shader if the vertex contributes to a visible triangle. For this you want two packed buffer regions one for all attributes that contribute to position computation, and one for everything else.

As always this comes with caveats. If packing as an array of structs forces a lot of padding elements to correct alignment, that rapidly eats into the benefits.

Upvotes: 1

Related Questions