使用QGLShaderProgram将自定义类型（结构）从Qt传递到GLSL

Question

I defined a struct for light parameters which contains two vectors. The struct is defined in both C++ and GLSL in an analogous way (note: QVector3D encapsulates 3 float s, not double s):

C++ host program:

struct LightParameters {
    QVector3D pos;
    QVector3D intensity;
};

Fragment Shader:

struct LightParameters {
    vec3 pos;
    vec3 intensity;
};

In the fragment shader, I also define the following uniforms. The numbers of lights is limited to 8, so the uniform array has a constant size (but only numLights are actually used):

const int maxLights = 8;
uniform int numLights;
uniform LightParameters lights[maxLights];

In the host program, I define the lights and pass them to the shader ( QGLShaderProgram ):

static const int maxLights = 8;
LightParameters lights[maxLights];

int numLights = 1;
lights[0].pos = {0, 1, 0};
lights[0].intensity = {1, 1, 1};

shaderProgram->bind();
shaderProgram->setUniformValue("numLights", numLights);
shaderProgram->setUniformValueArray("lights", reinterpret_cast<GLfloat*>(lights), maxLights, 6);
shaderProgram->release();

The 6 in the last parameter should indicate that each array element uses 6 GLfloat s from the raw float array, aka " tuple size ".

According to the Qt documentation, it should not be possible to use a tuple size of 6 (only 1, 2, 3, 4 are allowed). On the other side, QGLShaderProgram allows to pass in a matrix with up to 4x4 entries. GLSL also allows structs for uniforms which have more than 4 floats in them ( according to this example ).

This leads to the question:

How can I pass structs of floats / vectors as a uniform array using Qt? If it's not possible using Qt, I will call the GL function to do this, but I'm interested if Qt can do this for me conveniently.

Of course I tried the code above anyway. It results in all vectors in the struct being set to 0.

When I drop the second parameter in the struct and use a tuple size of 3, the code still does not work! However, when just using vec3 in the shader (still the struct with one QVector3D in the host program, tuple size = 3), it works.

So the problem seems to be that custom types are handled differently in the OpenGL host program API, but Qt doesn't seem to support it. I currently use two uniform arrays (one for the pos, one for the intensity) as a workaround.

Answer 1

Qt has nothing to do with this.

You cannot pass arrays of structs as though it were an array of values . If you have a uniform array whose type is not a basic type , then every array index and every member of that struct has a separate uniform location with a separate uniform name . Therefore, you must pass each member separately .

You must do this:

shaderProgram->setUniformValue("lights[0].pos", lights[0].pos);
shaderProgram->setUniformValue("lights[0].intensity", lights[0].intensity);
shaderProgram->setUniformValue("lights[1].pos", lights[1].pos);
shaderProgram->setUniformValue("lights[1].intensity", lights[1].intensity);
shaderProgram->setUniformValue("lights[2].pos", lights[2].pos);
shaderProgram->setUniformValue("lights[2].intensity", lights[2].intensity);

And so forth.

Or you could just use a proper uniform block and dispense with all of that.

Answer 2

I use a template function:

template<class T> void setUniformArrayValue(QOpenGLShaderProgram *program,
                                       const QString& arrayName,
                                       const QString& varName,
                                       int index,
                                       const T& value)
{
    const char* name = QString("%1[%2].%3")
            .arg(arrayName)
            .arg(index)
            .arg(varName)
            .toStdString().c_str();
    program->setUniformValue(name, value);
}

and in the program :

for (int i = 0; i < m_lights.size(); i++) {
    setUniformArrayValue<QColor>(program, "lights", "ambient", i, m_lights[i].ambient());
    ...
}