XEffects - Reloaded - New Release (V 1.4)

wing64 · Post by **wing64** » Thu Feb 05, 2009 2:53 am

1. If i have 2 or 3 light my game will pay with FPS slow down (1 light in my big scene FPS = 1 on geForce6100

).
BlindSide Can u suggest me the way of mod XEffect support far away objects ?
2. I founded article in gamedev.net Soft-Edged Shadows by Anirudh.S Shastry (http://www.gamedev.net/reference/articl ... le2193.asp)
but i try port to irrlicht in Step 1: Rendering the shadow map my result is a white canvas texture ??? but in example is a blue color

vertex and pixel shader i copy and paste to irrlicht

vertex shader

Code: Select all

struct VSOUTPUT_SHADOW
{
   float4 vPosition    : POSITION;
   float  fDepth       : TEXCOORD0;
};

VSOUTPUT_SHADOW VS_Shadow( float4 inPosition : POSITION )
{
   // Output struct
   VSOUTPUT_SHADOW OUT = (VSOUTPUT_SHADOW)0;
   // Output the transformed position
   OUT.vPosition = mul( inPosition, g_matLightViewProj );
   // Output the scene depth
   OUT.fDepth = OUT.vPosition.z;
   return OUT;
}

pixel shader

Code: Select all

float4  PS_Shadow( VSOUTPUT_SHADOW IN ) : COLOR0
{
   // Output the scene depth
   return float4( IN.fDepth, IN.fDepth, IN.fDepth, 1.0f );
}

What's happend with my code ??? or irrlicht not support shader from that article ?
best regards

FuzzYspo0N · Post by **FuzzYspo0N** » Thu Feb 05, 2009 7:55 am

this is HLSL , meaning directX shader code. if you are running in openGL it wont work, plus check the console for compile errors AND,

g_matLightViewProj <- this needs to be set via OnSetShaderConstants?

are you setting the shader up code side as well?

wing64 · Post by **wing64** » Thu Feb 05, 2009 10:29 am

I have been set to Direct3D9 and code In OnSetConstants function are

Code: Select all

		// View matrix
		vector3df vLightPos = vector3df(  10.0f, 20.0f, -40.0f );
		//vector3df vLightAim = vector3df(   -100.0f,  5.0f,   40.0f );
		vector3df vLightAim = vector3df(   0.0f,  0.0f,   0.0f );
		vector3df g_vUp		= vector3df(   0.0f,  1.0f,   0.0f );

		viewMat.buildCameraLookAtMatrixLH(vLightPos, vLightAim, g_vUp );
		projMat.buildProjectionMatrixPerspectiveFovLH(90*DEGTORAD, 1, 1.0f, 1024.0f );
		worldViewProj = projMat;
		worldViewProj *= viewMat;
		worldViewProj *= driver->getTransform(video::ETS_WORLD);
		services->setVertexShaderConstant("g_matLightViewProj", worldViewProj.pointer(), 16);

code vertex shader

Code: Select all

//--------------------------------------------
// Global variables
//--------------------------------------------
matrix	g_matLightViewProj	: LightViewProjection;

//--------------------------------------------
// Vertex shaders
//--------------------------------------------
struct VSOUTPUT_SHADOW
{
	float4 vPosition	: POSITION;
	float  fDepth		: TEXCOORD0;
};

// Shadow generation vertex shader
VSOUTPUT_SHADOW vertexMain( float4 inPosition : POSITION )
{
	// Output struct
	VSOUTPUT_SHADOW OUT = (VSOUTPUT_SHADOW)0;

	// Output the transformed position	
	OUT.vPosition = mul( inPosition, g_matLightViewProj );

	// Output the scene depth
	OUT.fDepth = OUT.vPosition.z;

	return OUT;
}

code pixel shader

Code: Select all

//--------------------------------------------
// Vertex shaders
//--------------------------------------------
struct VSOUTPUT_SHADOW
{
	float4 vPosition	: POSITION;
	float  fDepth		: TEXCOORD0;
};

//--------------------------------------------
// Pixel shaders
//--------------------------------------------
float4  pixelMain( VSOUTPUT_SHADOW IN ) : COLOR0
{
	// Output the scene depth
	return float4( IN.fDepth, IN.fDepth, IN.fDepth, 1.0 );
}

but result in same old above post (white canvas texture)

porcus · Post by **porcus** » Sat Feb 07, 2009 11:03 pm

Hi BlindSide,

Sorry for the delay but I had no time in the last days.
The problem is that the shadows are not working with
a directional light. You told me that I have to divide the far and nearvalue
by 1000 but if I do this I get only a black screen (which looks evident with so small values).

BlindSide wrote:Directional light is kinda tricky. You may have to divide your near and far values by 1000 when using a directional light to see anything.

With the usual near and farvalues everything (except shadows)
seem to be correct. Why are no shadows visible with directional light ???

wing64 · Post by **wing64** » Fri Feb 13, 2009 8:10 am

I founded another a shadow mapping technique:
Parallel-Split Shadow Maps
http://appsrv.cse.cuhk.edu.hk/~fzhang/pssm_project/
and
http://http.developer.nvidia.com/GPUGem ... _ch10.html
Could i port that technique to XEffect ? Please suggest me.
best regards,

Dorth · Post by **Dorth** » Fri Feb 13, 2009 2:11 pm

And this is why you use a string instead of a char array in anything but the most time or size critical zone (and even then, a fixed buffer can still be bigger than a string if it isn't completely used.)

BlindSide · Post by **BlindSide** » Sat Feb 14, 2009 12:27 am

wing64 wrote:I founded another a shadow mapping technique:
Parallel-Split Shadow Maps
http://appsrv.cse.cuhk.edu.hk/~fzhang/pssm_project/
and
http://http.developer.nvidia.com/GPUGem ... _ch10.html
Could i port that technique to XEffect ? Please suggest me.
best regards,

Yes that should be possible without modifying XEffects, just create one light for each split, and you are free to set the light's view and projection matrices. (If they are not public just make them public, but there should be get/set functions for the matrices).

Dorth wrote:And this is why you use a string instead of a char array in anything but the most time or size critical zone (and even then, a fixed buffer can still be bigger than a string if it isn't completely used.)

I use core::stringc everywhere in this, I'm guessing this post was out of place? But to keep the healthy advice going I'll add that when using strings you should pass them by reference or const reference as much as possible (Be careful of temporary variables!).

porcus wrote:Hi BlindSide,

Sorry for the delay but I had no time in the last days.
The problem is that the shadows are not working with
a directional light. You told me that I have to divide the far and nearvalue
by 1000 but if I do this I get only a black screen (which looks evident with so small values).

Sorry for the late reply, I suggest you look at this tutorial for more information on orthogonal lights (It's for the older XEffects but the same principles can be applied). I hope that helps.

wing64 · Post by **wing64** » Sat Feb 14, 2009 5:27 am

BlindSide wrote:Yes that should be possible without modifying XEffects, just create one light for each split, and you are free to set the light's view and projection matrices. (If they are not public just make them public, but there should be get/set functions for the matrices).

OK but i confuse something for split shadow
1. First i calculate camera plane (with camera scenenode position and target ) this step return zFar plane
2. calculate split area with zFar from step 1
3. loop num of split
3.1 calculate frustum corners from position,target (from camera scenenode) near, far (from step2) this step return 8 point of corrners
3.2 calculate light frustum with 8 point (from step 3.1), light position, target, fov (from SShadowLight class) this step return a new view and projection matrix of light
3.3 render node with view projection matrix (from step 3.2) to shadowdepthRTT
3.4 calculate crop matrix with with view projection matrix (from step 3.2)
3.5 recalculate view and projection with position,target (from camera scenenode) and render it to RT[0] (i set shadowdepthRTT to material texture layer 0)
ok now the screen is

I don't know why shadow not cast on floor or myself ? If somebody know this problem please suggest me

(i wrote this code over 1 week)

code & binary here
http://www.upload-thai.com/download.php ... 88f7c326ab
mirror http://us.share.geocities.com/zaros64/shadow200.zip

ps. geocities link u need manual copy url and paste to address bar or download manager program

best regards,

Dorth · Post by **Dorth** » Sat Feb 14, 2009 5:54 am

The f... I did not post that post in this thread at all...
What's going on with those fora???

BlindSide · Post by **BlindSide** » Sat Feb 14, 2009 10:10 am

wing64 take it one step at a time, first focus on getting the crop matrix working, and then concentrate on splits only after you get it for one light. In the meantime I'll have a look through your code and see if I can spot anything wrong when I have some time.

Dorth, perhaps your reply was meant for this thread?

Dorth · Post by **Dorth** » Sat Feb 14, 2009 4:15 pm

No, it was meant for the fast planet renderer's one, where he had buffer overflow.

wing64 · Post by **wing64** » Sun Feb 15, 2009 4:54 am

BlindSide wrote:wing64 take it one step at a time, first focus on getting the crop matrix working, and then concentrate on splits only after you get it for one light. In the meantime I'll have a look through your code and see if I can spot anything wrong when I have some time.

I hope u have some time for check code. (Help me please....)

Halifax · Post by **Halifax** » Mon Feb 16, 2009 1:27 am

I have a question. After a decode the depth:

Code: Select all

vec4 eDepth = texture2D( DepthMapSampler, gl_TexCoord[0].st );
float dDepth = eDepth.x + ( eDepth.y / 256.0 );

What range is it in?

I've noticed the code:

Code: Select all

void main() 
{
	float depth = gl_TexCoord[0].z / gl_TexCoord[0].x;
	float mulDepth = depth * 256.0 + 2.0;
	float flooredDepth = floor(mulDepth);
		
    gl_FragColor = vec4(flooredDepth / 256.0, mulDepth - flooredDepth, 0.0, 0.0);
}

Where gl_TexCoord[0].z is the clip-space z position, right? So that should be in [0,1]. And then the gl_Texcoord[0].x value is the distance from the near to far clip plane. I'm not quite understanding how it is stored encoded, or how it is stored when it is decoded. Can anyone help me out there?

By the way, all I'm doing right now is trying to render the depth map in a post-processing shader to get an output similar to this gray-like look: http://scien.stanford.edu/class/psych22 ... sDepth.jpg

BlindSide · Post by **BlindSide** » Mon Feb 16, 2009 2:11 am

You don't have to decode it fully. You can just output "eDepth.x" to get that gray look. Yeah it's in the range [0,1].

Halifax · Post by **Halifax** » Mon Feb 16, 2009 2:53 am

BlindSide wrote:You don't have to decode it fully. You can just output "eDepth.x" to get that gray look. Yeah it's in the range [0,1].

Ah okay, thanks BlindSide.

EDIT:

Okay, I tried that, but it appears as though I am receiving an error. First case code:

Code: Select all

uniform sampler2D ColorMapSampler;
uniform sampler2D ScreenMapSampler;
uniform sampler2D DepthMapSampler;

void main()
{
	vec4 color = texture2D( ColorMapSampler, gl_TexCoord[0].st );
	vec4 eDepth = texture2D( DepthMapSampler, gl_TexCoord[0].st );

	//gl_FragColor = vec4( eDepth.x, eDepth.x, eDepth.x, 1.0 );
	gl_FragColor = color;
}

Output:

Second case code:

Code: Select all

uniform sampler2D ColorMapSampler;
uniform sampler2D ScreenMapSampler;
uniform sampler2D DepthMapSampler;

void main()
{
	vec4 color = texture2D( ColorMapSampler, gl_TexCoord[0].st );
	vec4 eDepth = texture2D( DepthMapSampler, gl_TexCoord[0].st );

	gl_FragColor = vec4( eDepth.x, eDepth.x, eDepth.x, 1.0 );
	//gl_FragColor = color;
}

Output:

Can you see anything blantantly wrong with my code? I am calling enableDepthPass with true, and addNodeToDepthPass for all the nodes.