DirectX to OpenGL matrix issue

Started by
8 comments, last by RandyDeb 9 years, 6 months ago

Recently i started porting my engine to linux (Testing on Ubuntu 14.04)

So next to my DX11 and DX9 render systems, i started to implement an OpenGL render system.

I'm new to OpenGL, so it's a big learning phase.

Currenly, the OpenGL support simple rendering using shaders, vertex buffers and index buffers (only shaders, no FFP)

When rendering the same code (same object, same matrices) on DX9 or DX11, the result is the same (they are using the same shader as well)

But when rendering with the OpenGL render system on Windows, the position of the cube is different.

Even when running the same code with the OpenGL render system on Windows and Linux, the result is different.

See the attached screenshots

DX11 rendering (or DX9): The cube is in the middle of the screen

[attachment=24020:DX11.png]

OpenGL rendering on Windows: The position of the cube is different (higher and to the right)

[attachment=24021:GL_Windows.png]

OpenGL rendering on Linux: Camera closer to the cube???

[attachment=24022:GL_Linux.PNG]

I have no clue what is causing this, specially cause the result between windows and linux using OpenGL is also different.

Probably some matrix issue, but i searched the net and it seems when using shaders, the same matrices for DX and OpenGL can be used.

Currently i'm stuck on this for about 3 days now, reading about matrices, the difference between DirectX/OpenGL, looking at different SDK's, ...

So, i hope someone can see or know what i have been doing wrong ;)

Below all code related to the matrices and shader code:

This is the base matrix code for the cube rendering

For OpenGL i commented the matModelViewProj.Transpose() line.

But on linux, when commenting the line of not, the result stays the same.


	NpMatrix matWorld = NpMatrixIdentity();
	NpMatrix matProj = NpMatrixPerspectiveFovLH(NP_PIDIV4, (NpFloat32)m_displayWidth / (NpFloat32)m_displayHeight, 0.1f, 100.0f);
	NpMatrix matView = NpMatrixLookAtLH(NpVector3(3.0f, 5.0f, -6.0f), NpVector3(0.0f, 0.0f, 0.0f), NpVector3(0.0f, 1.0f, 0.0f));
	NpMatrix matModelViewProj = matWorld * matView * matProj;
	matModelViewProj.Transpose();

D3D11 Shader constant code:


	MatrixData bufferData;
	bufferData.WVP = matModelViewProj;
	m_d3dImmediateContext->UpdateSubresource(cBuffer, 0, NULL, &bufferData, 0, 0);

D3D9 Shader constant code:


	float paramData[16];
	memcpy(&paramData[0], (NpFloat32*)matModelViewProj, sizeof(float) * 16);
	m_d3dDevice->SetVertexShaderConstantF(0, paramData, 4);

GL Shader constant code:


	GLfloat paramData[16];
	memcpy(&paramData[0], (NpFloat32*)matModelViewProj, sizeof(GLfloat) * 16);
	glUniformMatrix4fv(paramModelViewProjection, 1, GL_FALSE, paramData);

DX Vertex Shader:


cbuffer matrixBuffer : register(b0)
{
    float4x4 matModelViewProjection;
};

struct VSIn
{
	float3 Pos : POSITION;
	float4 Col : COLOR;
};

struct VSOut
{
	float4 Pos : SV_POSITION;
	float4 Col : COLOR0;
};

VSOut main(VSIn input)
{
	VSOut output;
	output.Pos = mul(float4(input.Pos, 1.0), matModelViewProjection);
	output.Col = input.Col;
	return output;
}

DX Pixel Shader:


struct VSOut
{
	float4 Pos : SV_POSITION;
	float4 Col : COLOR0;
};

float4 main(VSOut input) : SV_TARGET
{
	return input.Col;
}

OpenGL Vertex Shader:


uniform mat4 matModelViewProjection;

attribute vec3 vert_POSITION0;
attribute vec4 vert_COLOR0;

varying vec4 frag_SV_POSITION;
varying vec4 frag_COLOR0;

void main()
{
	frag_SV_POSITION = matModelViewProjection * vec4(vert_POSITION0, 1.0);
	frag_COLOR0 = vert_COLOR0;
	gl_Position = frag_SV_POSITION;
}

OpenGL Fragment Shader:


varying vec4 frag_SV_POSITION;
varying vec4 frag_COLOR0;

void main()
{
	gl_FragColor = frag_COLOR0.rgba;
}

The OpenGL shaders have "#version 120" added to the top by the render system when loading the shaders.

And just to make sure, my matrix functions



NP_INLINE NpMatrix NPAPI NpMatrixPerspectiveLH(const NpFloat32 width, const NpFloat32 height, const NpFloat32 nearPlane, const NpFloat32 farPlane)
{
	NpFloat32 dualNearPlane = nearPlane + nearPlane;
    NpFloat32 range = farPlane / (farPlane - nearPlane);

	NpMatrix matrix;
	matrix.m[0][0] = dualNearPlane / width;
	matrix.m[0][1] = 0.0f;
	matrix.m[0][2] = 0.0f;
	matrix.m[0][3] = 0.0f;
	matrix.m[1][0] = 0.0f;
	matrix.m[1][1] = dualNearPlane / height;
	matrix.m[1][2] = 0.0f;
	matrix.m[1][3] = 0.0f;
	matrix.m[2][0] = 0.0f;
	matrix.m[2][1] = 0.0f;
	matrix.m[2][2] = range;
	matrix.m[2][3] = 1.0f;
	matrix.m[3][0] = 0.0f;
	matrix.m[3][1] = 0.0f;
	matrix.m[3][2] = -range * nearPlane;
	matrix.m[3][3] = 0.0f;
	return matrix;
}



NP_INLINE NpMatrix NPAPI NpMatrixLookAtLH(const NpVector3 position, const NpVector3 target, const NpVector3 upDirection)
{
	NpVector3 direction = target - position;
	return NpMatrixLookToLH(position, direction, upDirection);
}



NP_INLINE NpMatrix NPAPI NpMatrixLookToLH(const NpVector3 position, const NpVector3 direction, const NpVector3 upDirection)
{
	NpVector3 zAxis = direction.GetNormalized();
	NpVector3 xAxis = upDirection.Cross(zAxis).GetNormalized();
	NpVector3 yAxis = zAxis.Cross(xAxis);

	NpMatrix matrix;
	matrix.m[0][0] = xAxis.x;
	matrix.m[0][1] = xAxis.y;
	matrix.m[0][2] = xAxis.z;
	matrix.m[0][3] = -xAxis.Dot(position);
	matrix.m[1][0] = yAxis.x;
	matrix.m[1][1] = yAxis.y;
	matrix.m[1][2] = yAxis.z;
	matrix.m[1][3] = -yAxis.Dot(position);
	matrix.m[2][0] = zAxis.x;
	matrix.m[2][1] = zAxis.y;
	matrix.m[2][2] = zAxis.z;
	matrix.m[2][3] = -zAxis.Dot(position);
	matrix.m[3][0] = 0.0f;
	matrix.m[3][1] = 0.0f;
	matrix.m[3][2] = 0.0f;
	matrix.m[3][3] = 1.0f;
	matrix.Transpose();
	return matrix;
}



	NP_INLINE void NPAPI Transpose()
	{
		NpMatrix temp;
		temp.m[0][0] = m[0][0];
		temp.m[0][1] = m[2][0];
		temp.m[0][2] = m[0][1];
		temp.m[0][3] = m[2][1];
		temp.m[1][0] = m[1][0];
		temp.m[1][1] = m[3][0];
		temp.m[1][2] = m[1][1];
		temp.m[1][3] = m[3][1];
		temp.m[2][0] = m[0][2];
		temp.m[2][1] = m[2][2];
		temp.m[2][2] = m[0][3];
		temp.m[2][3] = m[2][3];
		temp.m[3][0] = m[1][2];
		temp.m[3][1] = m[3][2];
		temp.m[3][2] = m[1][3];
		temp.m[3][3] = m[3][3];

		m[0][0] = temp.m[0][0];
		m[0][1] = temp.m[1][0];
		m[0][2] = temp.m[0][1];
		m[0][3] = temp.m[1][1];
		m[1][0] = temp.m[0][2];
		m[1][1] = temp.m[1][2];
		m[1][2] = temp.m[0][3];
		m[1][3] = temp.m[1][3];
		m[2][0] = temp.m[2][0];
		m[2][1] = temp.m[3][0];
		m[2][2] = temp.m[2][1];
		m[2][3] = temp.m[3][1];
		m[3][0] = temp.m[2][2];
		m[3][1] = temp.m[3][2];
		m[3][2] = temp.m[2][3];
		m[3][3] = temp.m[3][3];
	}

Advertisement

Seems to me like you're not setting the viewport: https://www.khronos.org/opengles/sdk/docs/man/xhtml/glViewport.xml

In your Ubuntu screenshot, this is obvious, because the default viewport is (0, 0, 1, 1), so only the center of your cube shows through it.

In your Windows example, it's not clear what's happening - OpenGL probably takes the viewport size from the window's client area size, when you , but then you resize the window, and the old viewport no longer matches the new window size?

Check the examples in this thread for how to implement a resize callback and use it to set the viewport: http://www.gamedev.net/topic/661407-why-triangle-is-not-drawing/

Your implementation of Transpose() is absolutely unreasonable.
Unreasonably slow, and unreasonably complex and hard-to-follow. What did simplicity and efficiency do to make you hate them so?


	/**
	 * Swap two numbers.
	 *
	 * \param _fLeft The left operand.
	 * \param _fRight The right operand.
	 */
	template <typename _tType>
	NP_INLINE void NPAPI Swap( _tType &_fLeft, _tType &_fRight )
	{
		_tType fTemp = _fLeft;
		_fLeft = _fRight;
		_fRight = fTemp;
	}


	NP_INLINE void NPAPI Transpose()
	{
		Swap( m[0][1], m[1][0] );
		Swap( m[0][2], m[2][0] );
		Swap( m[0][3], m[3][0] );
		Swap( m[1][2], m[2][1] );
		Swap( m[1][3], m[3][1] );
		Swap( m[2][3], m[3][2] );
	}

The problem lies within m_displayWidth and m_displayHeight. It’s clear from the images that you are setting these values to the window (or…something?) width/height, not the client width/height.

You need to set these values to the actual client RECT of the window.

Show the code for creating the window and setting these values, and also any viewport code.

L. Spiro

I restore Nintendo 64 video-game OST’s into HD! https://www.youtube.com/channel/UCCtX_wedtZ5BoyQBXEhnVZw/playlists?view=1&sort=lad&flow=grid

The matrix transpose is based on the DirectXMath code. Thanks for your code :)

For the DX9/11 and OpenGL on windows, during the initialization of the render system, i adjust the window size



	// Setup the window size
	RECT windowRect;
	SetRect(&windowRect, 0, 0, m_displayWidth, m_displayHeight);
	if (m_windowedMode) {
		AdjustWindowRect(&windowRect, WS_OVERLAPPEDWINDOW, FALSE);
	}
	else {
		AdjustWindowRect(&windowRect, WS_POPUPWINDOW, FALSE);
	}
	
	// Resize the window
	if (m_windowedMode) {
		SetWindowLong(m_engine->GetWindow()->GetHWND(), GWL_STYLE, WS_OVERLAPPEDWINDOW);
		SetWindowPos(m_engine->GetWindow()->GetHWND(), HWND_TOP, 0, 0, windowRect.right - windowRect.left, windowRect.bottom - windowRect.top, SWP_FRAMECHANGED | SWP_NOMOVE | SWP_NOSIZE | SWP_SHOWWINDOW);
	}
	else {
		SetWindowLong(m_engine->GetWindow()->GetHWND(), GWL_STYLE, WS_POPUPWINDOW);	
		SetWindowPos(m_engine->GetWindow()->GetHWND(), HWND_TOP, 0, 0, windowRect.right - windowRect.left, windowRect.bottom - windowRect.top, SWP_FRAMECHANGED | SWP_NOMOVE | SWP_NOSIZE | SWP_SHOWWINDOW);
	}

As for linux, i don't adjust



	// Create the window
	m_window = XCreateWindow(m_display, RootWindow(m_display, m_visual->screen), 0, 0, 640, 480, 0, m_visual->depth, InputOutput, m_visual->visual, CWBorderPixel | CWColormap | CWEventMask | CWOverrideRedirect, &attributes);
	

(the display width/height settings are also set to 640x480 for this test)

D3D11 viewport:


	// Setup the viewport
	D3D11_VIEWPORT viewport;
	viewport.Width = (FLOAT)m_displayWidth;
	viewport.Height = (FLOAT)m_displayHeight;
	viewport.MinDepth = 0.0f;
	viewport.MaxDepth = 1.0f;
	viewport.TopLeftX = 0.0f;
	viewport.TopLeftY = 0.0f;

GL Viewport


glViewport(0, 0, m_displayWidth, m_displayHeight);

Ah, fixed the OpenGL windows issue.

There was an issue with the windows resizing (used SWP_NOSIZE)

After double checking everything, i see i changed my config to 800x600 while the window was created using 640x480 and the window was not resized correctly.

I will do some searching on the X11 window to see how i should resize :)

Thanks for the answers

There is clearly some kind of discrepancy here.

The reason Direct3D 11 is working is because the viewport and back back buffer are set to the same size, that size is not the same size as the window, and at some point it is being scaled to fit the screen. I would normally accuse IDirect3DDevice9::Present() or IDXGISwapChain::Present() but they do not do smooth scaling. Notice the anti-aliasing. Unless you have multi-sampling enabled, which you did not mention (but would internally be a different-sized buffer being scaled down).

You need to be looking for things related to this, not matrices. The problem is not related to the matrices or shaders.

L. Spiro

[EDIT]

Late post, after issue was solved.

[/EDIT]

I restore Nintendo 64 video-game OST’s into HD! https://www.youtube.com/channel/UCCtX_wedtZ5BoyQBXEhnVZw/playlists?view=1&sort=lad&flow=grid

There is no multi sampling or anything enabled, im using just very basic render settings.

The windows part is working now for DirectX and OpenGL, but as for linux, the issue remains the same :(


#elif defined(NP_PLATFORM_LINUX)
	// Resize the window
	XMoveResizeWindow(m_engine->GetWindow()->GetXDisplay(), m_engine->GetWindow()->GetXWindow(), 0, 0, m_displayWidth, m_displayHeight);

	// Create the context
	m_context = glXCreateContext(m_engine->GetWindow()->GetXDisplay(), m_engine->GetWindow()->GetXVisualInfo(), NULL, True);
	if (!m_context) {
		printf("Failed to create the GLX context\n");
	}

	// Make the current context
	if (!glXMakeCurrent(m_engine->GetWindow()->GetXDisplay(), m_engine->GetWindow()->GetXWindow(), m_context)) {
		printf("Failed to set the GLX context\n");
	}

	// Get the drawable
	m_drawable = glXGetCurrentDrawable();
#endif

	// Initialize the render API
	glewExperimental = GL_TRUE;
	glewInit();

	GLfloat viewportData[4];
	glGetFloatv(GL_VIEWPORT, viewportData);
	printf("viewport: %f; %f; %f; %f\n", viewportData[0], viewportData[1], viewportData[2], viewportData[3]);

	glViewport(0, 0, m_displayWidth, m_displayHeight);

Console output: viewport: 0.000000; 0.000000; 640.000000; 480.000000

m_displayWidth = 640

m_displayHeight = 480



const NpUInt32 NPAPI CLinuxWindow::InitializeClass()
{
	// Set the error handler
#if defined(NP_DEBUG)
	XSetErrorHandler(CLinuxWindow_XErrorHandler);
#endif

	// Get the default display and screen
	m_display = XOpenDisplay(0);
	m_screen = DefaultScreen(m_display);

	// Get the visual
	int visualNumber = 0;
	XVisualInfo visualTemplate;
	visualTemplate.screen = m_screen;
	visualTemplate.depth = 24;
	while (m_visual == NULL && (visualTemplate.depth >= 16)) {
		m_visual = XGetVisualInfo(m_display, VisualScreenMask | VisualDepthMask, &visualTemplate, &visualNumber);
		visualTemplate.depth -= 8;
	}
	printf("Choosen visual: %ld\n", m_visual->visualid);
 
	// Create the color map
	Colormap colormap = XCreateColormap(m_display, RootWindow(m_display, m_visual->screen), m_visual->visual, AllocNone);

	XSetWindowAttributes attributes;
	attributes.colormap = colormap;
	attributes.border_pixel = 0;
	attributes.event_mask = StructureNotifyMask | FocusChangeMask | ExposureMask;
	attributes.event_mask |= PointerMotionMask | ButtonPressMask | KeyPressMask | ButtonReleaseMask | KeyReleaseMask;
	attributes.override_redirect = false; //= m_fullscreen;

	// Create the window
	m_window = XCreateWindow(m_display, RootWindow(m_display, m_visual->screen), 0, 0, 640, 480, 0, m_visual->depth, InputOutput, m_visual->visual, CWBorderPixel | CWColormap | CWEventMask | CWOverrideRedirect, &attributes);
	
	// Show the window
	XMapRaised(m_display, m_window);

	// Add the destroy message
	Atom wmDelete;
	wmDelete = XInternAtom(m_display, CLinuxWindow_WM_DELETE_WINDOW, True);
	XSetWMProtocols(m_display, m_window, &wmDelete, 1);

	// Operation successful
	return NP_OK;
}

ps: The result for rendering 2D without matrices was the same for windows and linux


I will do some searching on the X11 window to see how i should resize smile.png

Err..

Don't use X11 directly. It's going to be replaced with other systems later on, where X11 will be running in emulation mode for a while until removed.

Use some library that abstracts all this away and does all the window handling for you on ALL operating systems.

For example SDL2.

Off topic from your actual problem, but there's one slight difference in D3D/GL matrices - D3D's final NDC z coords range from 0 to 1, and GL's from -1 to 1.
So without any changes, your game will end up wasting 50% of your depth buffer precision in the GL version. To fix this, you just need to modify the projection matrix to scale in z by 2 and offset by -1 so you're projecting into the full z range.

If you're using a GL-oriented math library, it will construct it's projection matrices like this by default, so you'd have to make the opposite scale/bias z adjustments to get D3D to work (the error would be a misbehaving near-clip plane, appearing too far out).

I made the SDL2 implementation for my render window on linux, but still having the same behavior, while the windows opengl result is now the same as DX

So i'm not sure whats missing (except for the Window/OpenGL context, all code is the same for windows/linux)

=== EDIT ===

Never mind ... got it working ;)

Seems the shaders in my working directory on Linux were not up to date. Guess my next step is working out a better FileManager before continue implementing OpenGL

Thanks for all the help for solving the issue

This topic is closed to new replies.

Advertisement