Difference between casted vec and ivec

OpenGL
1

I was trying to apply skinning to a mesh and I used a code sample from the GLTF file format overview (cannot insert link) as help to compute the skinning matrix.
There is something that I dont understand : in the following code, why is the float-to-int cast necessary ?
Here are two versions, with and without cast.
When using the first version, with some float-to-int casts, it works. But with the second code version, using the ivec type for the “joint” attribute, the behaviour is different.

#version 440 core

layout(location = 0) in vec3 pos;
layout(location = 1) in vec3 normal;
layout(location = 2) in vec4 joint;
layout(location = 3) in vec4 jointWeight;

(...)

uniform mat4 jointMatrices[64];

void main()
{
    mat4 combinedJointMatrix =
        jointWeight[0] * jointMatrices[int(joint[0])] +
        jointWeight[1] * jointMatrices[int(joint[1])] +
        jointWeight[2] * jointMatrices[int(joint[2])] +
        jointWeight[3] * jointMatrices[int(joint[3])];

Why can’t we use ivec instead, like this ?

#version 440 core

layout(location = 0) in vec3 pos;
layout(location = 1) in vec3 normal;
layout(location = 2) in ivec4 joint;
layout(location = 3) in vec4 jointWeight;

(...)

uniform mat4 jointMatrices[64];

void main()
{
    mat4 combinedJointMatrix =
        jointWeight[0] * jointMatrices[joint[0]] +
        jointWeight[1] * jointMatrices[joint[1]] +
        jointWeight[2] * jointMatrices[joint[2]] +
        jointWeight[3] * jointMatrices[joint[3]];
2

Who says that it is necessary? You can index arrays with floats; the system will convert them to integers.

Well, yeah. Your VS attribute is no longer a float, it’s an integer. Any apparent differences is probably due to how you set up your vertex attributes in OpenGL. Which you didn’t show.

3

Sorry for not showing all the code, I thought it might not be relevant.
I set up the attributes using the information from a GLTF file (and a library called tinygltf).

static void setVertexAttribPointerFromGltf(int attributeIndex, const tinygltf::Accessor& accessor, const tinygltf::BufferView& bufferView)
{
    glEnableVertexAttribArray(attributeIndex);
    glVertexAttribPointer(attributeIndex,
        getAttributeNumComponents(accessor.type), accessor.componentType,
        accessor.normalized, bufferView.byteStride, (void*)bufferView.byteOffset);
}

“accessor.componentType” says it is unsigned short.

4

For attribIndex == 2?

What are:

  1. getAttributeNumComponents(accessor.type),
  2. accessor.componentType, and
  3. accessor.normalized

for attribIndex = 2, before and after your shader change above?

5

You need to use glVertexAttribIPointer (note the extra “I”) for integer attributes. If you use glVertexAttribPointer, the GPU will be instructed to perform integer-to-float conversion on the data, but the shader will be interpreting the values as integers.

6

Oh I see. I thought wheither glVertexAttribPointer performed conversion to float or not depended on the type parameter. I will try using glVertexAttribIPointer and I will report on the result.

7

Well, it won’t perform conversion if the type parameter is GL_FLOAT. The result is always float (more precisely, a vec4), the type parameter tells it what it’s converting from.

8

I have tried setting the attribute with glVertexAttribIPointer and using it as ivec in the shader and it works as intended. Thanks for the help.

9

This topic was automatically closed 183 days after the last reply. New replies are no longer allowed.