[Solved] Simplest Vertex/Fragment Shaders using QGLWidget Derivation

Hornsj2

In relation to a little project I've been working on I'm trying to get a test harness for shaders running.

Can anyone spot the problem with the following code? All I get is what appears to be an OpenGL window painted with the clear color. I'm using the core profile with OpenGL version 3.0 to test the following code.

@#include "gltarget.h"
#include <qdebug.h>

GlTarget::GlTarget(QWidget* parent ) : QGLWidget(parent)
,m_shaderProgram(new QGLShaderProgram(this))
{
makeCurrent();
if( glewInit() != GLEW_OK )
qDebug() << "Can't initialize glew";
if ( ! GLEW_VERSION_3_0 )
{
qDebug() << "OpenGL version 3 is not supported";
exit( 1 );
}
}

void GlTarget::initializeGL()
{

glClearColor(0.0, 0.0, 0.0, 0.0);
glEnable(GL_DEPTH_TEST);

m_shaderProgram.addShaderFromSourceCode(QGLShader::Vertex,
"#version 130\n"
"in highp vec4 vertex;\n"
"in mediump mat4 matrix;\n"
"void main(void)\n"
"{\n"
" gl_Position = matrix * vertex;\n"
"}");
m_shaderProgram.addShaderFromSourceCode(QGLShader::Fragment,
"#version 130\n"
"uniform mediump vec4 color;\n"
"out mediump vec4 out_color;\n"
"void main(void)\n"
"{\n"
" out_color = color;\n"
"}");
m_shaderProgram.link();
m_shaderProgram.bind();

}

void GlTarget::resizeGL(int width, int height)
{
glViewport(0, 0, (GLint)width, (GLint)height);
}

void GlTarget::paintGL()
{
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);

int vertexLocation = m_shaderProgram.attributeLocation("vertex");
int matrixLocation = m_shaderProgram.attributeLocation("matrix");
int colorLocation = m_shaderProgram.uniformLocation("color");

QVector3D triangleVertices[] = {
QVector3D(60.0f, 10.0f, 0.0f),
QVector3D(110.0f, 110.0f, 0.0f),
QVector3D(10.0f, 110.0f, 0.0f)
};

QMatrix4x4 pmvMatrix;
pmvMatrix.ortho(rect());

m_shaderProgram.enableAttributeArray(vertexLocation);
m_shaderProgram.setAttributeArray(vertexLocation, triangleVertices);
m_shaderProgram.setUniformValue(matrixLocation, pmvMatrix);
m_shaderProgram.setUniformValue(colorLocation, QColor(0, 255, 0, 255));

glDrawArrays(GL_TRIANGLES, 0, 3);

m_shaderProgram.disableAttributeArray(vertexLocation);
}@

ZapB

No time to look in detail today but I may get a chance over the weekend.

Hornsj2

Thanks because I am completely stumped.

I'm using what are essentially "pass-through" shaders copied from the qt example in the 4.7 QGLShaderProgram.

The projection is orthographic and the viewport should contain the geometry but nothing shows up!

The shaders compile and link, the attributes are passed in as they should be. but no triangle!

ZapB

It's not something daft like the camera is co-planar withthe triangle is it? I notice that you triangles vertices are all in the plane z=0. Try offsetting them a little perhaps?

Hornsj2

Not a bad guess but I think since the view volume goes by default from -1 to 1 in qt's orthographic matrix I should be able to see something at 0.

ZapB

OK I can shed a little more light on this now I think. It has taken quite a bit of digging today hence the delay.

Short story is, don't use the CoreProfile, use CompatibilityProfile instead for now.

Here is the full story...for info I am using OpenGL 3.3 since my card does not support OpenGL 4.x but it is essentially the same in this case...

I constructed my own little test harness which I managed to get working with an OpenGL 2 context. When switching to OpenGL 3.3 context and running with the compatibility profile all is still fine. However, running with a QGLFormat I noticed the following output when a call to QGLShaderProgram::addShaderFromSourceFile() is made:

@QGLShaderProgram: shader programs are not supported@

That seems wrong to me so I decided to delve into Qt itself. I have built a copy of Qt from the master branch of git (the public repo) in debug mode so that I could step into the above function. Doing so and see where the rabbit hole goes I wound up at this snippter of code:

@
QGLExtensions::Extensions QGLExtensions::currentContextExtensions()
{
QGLExtensionMatcher extensions(reinterpret_cast<const char *>(glGetString(GL_EXTENSIONS)));
Extensions glExtensions;

if (extensions.match("GL_ARB_texture_rectangle"))
    glExtensions |= TextureRectangle;
if (extensions.match("GL_ARB_multisample"))
    glExtensions |= SampleBuffers;
if (extensions.match("GL_SGIS_generate_mipmap"))
    glExtensions |= GenerateMipmap;
if (extensions.match("GL_ARB_texture_compression"))
    glExtensions |= TextureCompression;
if (extensions.match("GL_EXT_texture_compression_s3tc"))
    glExtensions |= DDSTextureCompression;
if (extensions.match("GL_OES_compressed_ETC1_RGB8_texture"))
    glExtensions |= ETC1TextureCompression;
if (extensions.match("GL_IMG_texture_compression_pvrtc"))
    glExtensions |= PVRTCTextureCompression;
if (extensions.match("GL_ARB_fragment_program"))
    glExtensions |= FragmentProgram;
if (extensions.match("GL_ARB_fragment_shader"))
    glExtensions |= FragmentShader;
....

@

The problem lies in the call to

@glGetString(GL_EXTENSIONS)@

since this is now removed from OpenGL >= 3.1. Instead the following functions should be used:

@
glGetInteger(GL_NUM_EXTENSIONS)
glGetStringi(GL_EXTENSIONS, i)
@

where i is the index of the extension to query and 0 <= i < glGetInteger(GL_NUM_EXTENSIONS).

When the above extension matching code fails (due to using the CoreProfile) this in turn makes Qt's OpenGL support think that no extensions are available which is of course broken.

So the crux of the problem seems to be the way that Qt is querying for OpenGL extensions when using OpenGL 3.0 or higher (although OpenGL 3.0 still supports the old method too).

Therefore to fix this properly a patch to Qt itself is required. I'll have a quick bash at doing this now but I am short on time today (as always). Hopefully it will be a fairly simple fix although I think some refactoring of the QGLExtensionMatcher class will be needed.

I have filed a "bug":http://bugreports.qt.nokia.com/browse/QTBUG-18529 for this. Please vote on it to increase the odds of it getting fixed in time for Qt 4.8.0.

I'll post back here and on the bug if I get anywhere with a fix.

Hornsj2

Hey, great find. You sure do have a lot more knowledge and willpower in finding these bugs. I don't think I would have thought to dig into the framework. I'm sure everyone from Nokia on down appreciates it.

Thanks.

ZapB

I'm still digging into Qt's OpenGL support to see where the function entry point addresses are resolved but I am out of time for today - taking the wife out since it is Mother's Day here ;-)

I'll see if I can get any further with a patch tomorrow evening.

If you use the CompatibilityProfile does your simple pair of shaders and test harness work?

Hornsj2

!http://joshuahorns.com/Untitled.png(My ScreenShot)!

That's what I get... just a "clear color" screen. You were able to get shader code running? Could you post that? I might just give up and use pure opengl calls directly. I'm probably doing something very dumb with QGLShaderProgram but I can't for the life of me figure it out.

ZapB

OK here's what I have for my simple test harness at present. Toggling which profile I use causes it to work or give a blank clear colour like you have. BTW what graphics card and driver are you using?

main.cpp:

@
#include "shaderwidget.h"

#include <QApplication>
#include <QGLFormat>

#include <GL/glut.h>

int main( int argc, char* argv[] )
{
QApplication a( argc, argv );
glutInit( &argc, argv );

QGLFormat glFormat;
glFormat.setVersion( 3, 3 );
glFormat.setProfile&#40; QGLFormat::CompatibilityProfile &#41;;
//glFormat.setProfile&#40; QGLFormat::CoreProfile &#41;;
glFormat.setSampleBuffers( true );

ShaderWidget w( glFormat );
w.show();

return a.exec&#40;&#41;;

}
@

shaderwidget.h:
@
#ifndef SHADERWIDGET_H
#define SHADERWIDGET_H

#include <QGLWidget>

#include <QGLShaderProgram>

class ShaderWidget : public QGLWidget
{
Q_OBJECT
public:
ShaderWidget( const QGLFormat& format, QWidget* parent = 0 );
~ShaderWidget();

protected:
virtual void initializeGL();
virtual void resizeGL( int w, int h );
virtual void paintGL();

virtual void keyPressEvent( QKeyEvent* e );

private:
QGLShaderProgram m_shader;

float m_theta;
float m_phi;

};

#endif // SHADERWIDGET_H
@

shaderwidget.cpp:

@
#include <GL/glew.h>

#include "shaderwidget.h"

#include <QCoreApplication>
#include <QKeyEvent>

#include <GL/glut.h>

ShaderWidget::ShaderWidget( const QGLFormat& format, QWidget* parent )
: QGLWidget( format, parent ),
m_shader(),
m_theta( 0.0f ),
m_phi( 0.0f )
{
QGLFormat glFormat = QGLWidget::format();
qDebug() << QString( "OpenGL Version = %1, %2" )
.arg( glFormat.majorVersion() )
.arg( glFormat.minorVersion() );

if ( !glFormat.sampleBuffers() )
    qWarning() << "Could not enable sample buffers";

qDebug() << "OpenGL context valid =" << context()->isValid();

}

ShaderWidget::~ShaderWidget()
{
}

void ShaderWidget::initializeGL()
{
// Set the clear color to black
glClearColor( 0.0f, 0.0f, 0.0f, 1.0f );

// Set the drawing color to green
glColor3f( 0.9f, 0.9f, 0.0f );

bool result;
result = m_shader.addShaderFromSourceFile&#40; QGLShader::Vertex, "simple.vert" &#41;;
if ( !result )
    qDebug() << m_shader.log();
result = m_shader.addShaderFromSourceFile&#40; QGLShader::Fragment, "simple.frag" &#41;;
if ( !result )
    qDebug() << m_shader.log();
result = m_shader.link();
if ( !result )
    qDebug() << m_shader.log();

}

void ShaderWidget::resizeGL( int w, int h )
{
// Prevent a divde by zero
if ( h == 0 )
h = 1;

// Set the viewport to window dimensions
glViewport( 0, 0, w, h );

// Reset the coordinate system
glMatrixMode( GL_PROJECTION );
glLoadIdentity();

// Establish the clipping volume by setting up an perspective projection
double aspectRatio = double( w ) / double( h );
double verticalViewingAngle = 45.0;
gluPerspective( verticalViewingAngle, aspectRatio, 1.0, 425.0 );

glMatrixMode( GL_MODELVIEW );
glLoadIdentity();

}

void ShaderWidget::paintGL()
{
// Clear the buffer with the current clearing color
glClear( GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT );

glPushMatrix();
glTranslatef( 0.0f, 0.0f, -200.0 );
glRotatef( m_theta, 1.0f, 0.0f, 0.0f );
glRotatef( m_phi,   0.0f, 1.0f, 0.0f );

m_shader.bind();

glutSolidTeapot( 40 );

glPopMatrix();

}

void ShaderWidget::keyPressEvent( QKeyEvent* e )
{
switch ( e->key() )
{
case Qt::Key_Escape:
QCoreApplication::instance()->quit();
break;

    case Qt::Key_Left:
        m_phi += 1.0f;
        updateGL();
        break;

    case Qt::Key_Right:
        m_phi -= 1.0f;
        updateGL();
        break;

    case Qt::Key_Up:
        m_theta += 1.0f;
        updateGL();
        break;

    case Qt::Key_Down:
        m_theta -= 1.0f;
        updateGL();
        break;

    default:
        QGLWidget::keyPressEvent( e );
}

}
@

The vertex shader, simple.vert:

@
varying vec4 secondaryColor;

void main( void )
{
// Normal MVP transform
vec4 clipCoord = gl_ModelViewProjectionMatrix * gl_Vertex;
gl_Position = clipCoord;

// Copy primary color
gl_FrontColor = gl_Color;

// Calculate NDC
vec3 ndc = clipCoord.xyz / clipCoord.w;

// Map from [-1,1] to [0,1] before outputting
secondaryColor = vec4( ( ndc * 0.5 ) + 0.5, 1.0 );

}
@

and finally the fragment shader:

@
varying vec4 secondaryColor;

void main(void)
{
// Mix primary and secondary colors 50:50
gl_FragColor = mix( gl_Color, vec4( vec3( secondaryColor ), 1.0 ), 0.5 );
}
@

I know that this still uses the older-style shader code but I wanted to start from something that I know works and then slowly move it towards a pure OpenGL >= 3.1 Core Profile appproach and see what breaks on the way. That is how I found this bug in extension wrangling for OpenGL 3.1 or newer.

Does the above work for you?

Hornsj2

I suspect it will work. I'll try it after I get home from work tonight.

The "problem" is that, as you know, the core profile removes deprecated functionality. In the case of the code above the following will have been removed ( my goal was to keep my code "pure" and not have any deprecated functions and simply turning on compatibility seems to not work with QGLShaderProgram):

The matrix stack, and by extension all matrix related manipulation such as the following:
@ Reset the coordinate system
glMatrixMode( GL_PROJECTION );
glLoadIdentity();

// Establish the clipping volume by setting up an perspective projection
double aspectRatio = double( w ) / double( h );
double verticalViewingAngle = 45.0;
gluPerspective( verticalViewingAngle, aspectRatio, 1.0, 425.0 );

glMatrixMode( GL_MODELVIEW );
glLoadIdentity()
@

and this
@
glPushMatrix();
glTranslatef( 0.0f, 0.0f, -200.0 );
glRotatef( m_theta, 1.0f, 0.0f, 0.0f );
glRotatef( m_phi, 0.0f, 1.0f, 0.0f );
@
You already touched on it but I'll toss it out there again. Some built in shader variables such as gl_FragColor are deprecated / removed in the core profile.

The fact that the above works and my code doesn't, even though I turn on the compatibility profile, leads me to believe QGLShaderProgram can't be used in my project. For completeness I'll test my tessellation shader additions with your harness and make a merge request.

I've never used Gitorious but I signed up and it wants a public key. If I understand the directions I can generate a public key with Putty. Is that right? Also, once that's done where do I store the key? Is it to be placed in a certificate store? Also, are there rules as far as comments or will Nokia clean that up?

ZapB

[quote author="Hornsj2" date="1301913471"]I suspect it will work. I'll try it after I get home from work tonight.

The "problem" is that, as you know, the core profile removes deprecated functionality. In the case of the code above the following will have been removed ( my goal was to keep my code "pure" and not have any deprecated functions and simply turning on compatibility seems to not work with QGLShaderProgram):
[/quote]

Yes I know, but as I said I wanted to start off from a starting point that I know works with Qt and QGLShaderProgram. I was then going to remove the old fixed functionality calls and modify the shaders to use the Core Profile when I found this bug in Qt's extension mangling support so I did not go any further yet.

[quote author="Hornsj2" date="1301913471"]
The fact that the above works and my code doesn't, even though I turn on the compatibility profile, leads me to believe QGLShaderProgram can't be used in my project.
[/quote]

Not yet, because of the problem reported above, namely that QGLShaderProgram believes that it cannot use shaders when you are using the Core Profile due to this bug.

[quote author="Hornsj2" date="1301913471"]
For completeness I'll test my tessellation shader additions with your harness and make a merge request.
[/quote]

It should work if you use the compatibility mode - except if there is a bug in your shaders of course ;-)

[quote author="Hornsj2" date="1301913471"]
I've never used Gitorious but I signed up and it wants a public key. If I understand the directions I can generate a public key with Putty. Is that right? Also, once that's done where do I store the key? Is it to be placed in a certificate store?
[/quote]

Sounds about right for generating the key. It's been ages since I used PuTTY to generate a key. Have a google for it, should be easy to find instructions.

[quote author="Hornsj2" date="1301913471"]
Also, are there rules as far as comments or will Nokia clean that up?
[/quote]

You need to follow Qt's "coding style guidelines":https://qt.gitorious.org/qt/pages/QtCodingStyle. There is more info "here ":https://qt.gitorious.org/qt/pages/Home.

Good luck.

Hornsj2

Thanks for the reply.

I might not have thought it through, but I think it might go deeper than changing the logic guarding the use of shaders. Again I realize I might be telling you something you already have thought of.

It would seem that, in addition to changing the method for checking extensions, maybe the function pointer declarations in Qt's OpenGL support need to be looked at. I do know there are some preprocessor guards that rely on that check in Qt. These guards encapsulate OpenGL function pointer definitions and defines. I ran into that with my tessellation work.

ZapB

Yes they do. As a starter for 10 I am looking at where to resolve the function pointer for glGetStringi() so that I can at least query the extensions ;-)

Also, the logic in QGLShaderProgram needs to be amended since shaders are part of the specification and not provided by extensions for newer OpenGL versions.

There is quite some work to do to properly support OpenGL 3.1 or newer.

Hornsj2

Great. It sounds like I should integrate my changes into your changes if you will be reworking that part of the code. I had to add tessellation support, in part, there.

ZapB

I have spoken to sroedal (Qt OpenGL dude) on irc this afternoon and he is committing a patch that is very similar to what I wrote yesterday afternoon. It allows for proper parsing of OpenGL extensions when using the Core profile. It should hopefully hit Qt master branch on gitorious within a day or so.

I'll try applying it locally tonight and see if QGLShader then starts working or if further changes are needed to make it realise that shaders are supported.

I also asked about having Qt resolve the Core profile entry points and the response was thet using GLEW is the current advice - which is what we had been doing anyway. I still think it would be nice that if we ask Qt for a OpenGL 3.2 Core profile context that it then resolves the function pointers for the Core profile entry points for us without having to use GLEW. But that is another problem I guess.

I'll keep you updated.

Hornsj2

Your posted example with fixed function works fine. I'll try using the entire shader pipeline.

It's normally slow going for me during the week. Weekends are my most productive time for these projects.

By "entire shader pipeline" I mean vert,geom, tess control, tess eval, frag.

ZapB

Cool. I'll look forward to seeing your example. I have not used the tesselation shaders yet so I would be keen to see a nice little example. I probably won't get back to this until the weekend either but I'll keep an eye out for that patch hitting Qt master.

Hornsj2

Promises promises... yeah I got caught today moving stuff. I WILL have the shaders tested by tomorrow night.

Hornsj2

By the way my tessellation tests will be posted in the proper thread.