2
votes

I'm writing a program which displays a MS3D model using DirectX, and unfortunately, the result shows nothing on the screen. When I use the Graphics Debugger from Visual Studio 13, I notice that the pixel shader is missing from the pipeline, as it is shown in the below picture

image

This is my pixel shader source code:

 cbuffer SkinningTransforms
 {
    matrix WorldMatrix; 
    matrix ViewProjMatrix;
 };
//--------------------------------------------------------------------------------
// Inter-stage structures
//--------------------------------------------------------------------------------
struct VS_INPUT
{
    float3  position        : POSITION;
    int4    bone            : BONEID;
    float4  weights         : BONEWEIGHT;
    float3  normal          : NORMAL;
    float3  tangent         : TANGENT;
    float2  tex             : TEXCOORD;
};
//--------------------------------------------------------------------------------
struct VS_OUTPUT
{
    float4 position         : SV_Position;
    float3 normal           : NORMAL;
    float3 light            : LIGHT;
    float2 tex              : TEXCOORDS;
};

Texture2D       ColorTexture : register( t0 );           
SamplerState    LinearSampler : register( s0 );

//--------------------------------------------------------------------------------
VS_OUTPUT VSMAIN( in VS_INPUT input )
{
    VS_OUTPUT output;
    //Transform vertex and pass them to the pixel shader
    return output;
}

//--------------------------------------------------------------------------------
float4 PSMAIN( in VS_OUTPUT input ) : SV_Target
{
    // Calculate the lighting
    float3 n = normalize( input.normal );
    float3 l = normalize( input.light );


    float4 texColor = ColorTexture.Sample( LinearSampler, input.tex );

    float4 color = texColor * (max(dot(n,l),0) + 0.05f );
    return( color );
}

As I was known from Graphics Debugger, all of the graphics event are right. I listed in below important events, which might be relating to Pixel Shader:

106:(obj:4) ID3D11Device::CreateDepthStencilView(obj:24,NULL,obj:25)*
108:(obj:5) ID3D11DeviceContext::OMSetRenderTargets(8,{obj:1,NULL,NULL,NULL,NULL,NULL,NULL,NULL},obj:25)*
109:(obj:5) ID3D11DeviceContext::ClearRenderTargetView(obj:1,addr:21)*
111:(obj:5) ID3D11DeviceContext::ClearDepthStencilView(obj:25,1,1.000f,0)*
119:(obj:4) ID3D11Device::CreateSamplerState(addr:24,obj:27)*
134:(obj:4) ID3D11Device::CreatePixelShader(addr:27,21056,NULL,obj:30)*
135:CreateObject(D3D11 Pixel Shader,obj:30)
136:(obj:5) ID3D11DeviceContext::PSSetShader(obj:30,NULL,0)*
137:(obj:5) ID3D11DeviceContext::PSSetSamplers(0,1,{obj:27})*
139:(obj:4) ID3D11Device::CreateTexture2D(addr:28,addr:5,obj:31)*
140:CreateObject(D3D11 Texture2D,obj:31)
142:(obj:4) ID3D11Device::CreateShaderResourceView(obj:31,NULL,obj:32)*
143:CreateObject(D3D11 Shader Resource View,obj:32)
144:(obj:5) ID3D11DeviceContext::PSSetShaderResources(0,1,{obj:32})*
146:(obj:4) ID3D11Device::CreateRasterizerState(addr:29,obj:33)*
147:CreateObject(D3D11 Rasterizer State,obj:33)
152:(obj:5) ID3D11DeviceContext::RSSetState(obj:33)*
154:(obj:5) ID3D11DeviceContext::RSSetViewports(1,addr:30)*
156:(obj:4) ID3D11Device::CreateBlendState(addr:11,obj:34)*
157:CreateObject(D3D11 Blend State,obj:34)
159:(obj:5) ID3D11DeviceContext::OMSetBlendState(obj:34,addr:31,-1)*
162:(obj:4) ID3D11Device::CreateDepthStencilState(addr:32,obj:35)*
163:CreateObject(D3D11 Depth-Stencil State,obj:35)
165:(obj:5) ID3D11DeviceContext::OMSetDepthStencilState(obj:35,0)*

I debugged all of the function in the above list, and all of them return OK. Nothing wrong. My question is what is the reason the pixex shader is missing from pipleline, which in turn may result in the empty screen.

2
Did you eventually figure out the problem? I have the same issue (PS is missing in pipeline stage, but is active in the object view of the immediate context, so it is not the problem described in msdn.microsoft.com/en-us/library/jj191650.aspx). Of course, the image is black...Christoph
Same exact issue, verified RS state & output from VS. I did notice that event though RSSetViewports is called, the viewport on the draw call is Not Set. I'm still investigating this.Tamir Daniely
Ok, for me it happened because I optimistically gave the viewport a max depth of 100.0f when it should be 1.0f at most. Even though I disabled depth clipping in the rasterizer this prevents RSSetViewports from setting the the viewport at all and no error is thrown. Check in the graphics debugger whether the viewport on the Device Context is actually set.Tamir Daniely

2 Answers

3
votes

Adding to the other answers, constant buffer organization can be the cause of this problem. In my case, the pixel shader was missing from the pipeline but also the vertex shader wasn't transforming the vertices correctly. Upon inspection it was revealed that the world matrix had incorrect values because the boolean value at the top of the constant buffer was causing data misalignment. HLSL packs data into 16 byte boundries which are so called vectors that have 4 components. A boolean is 4 bytes which is the same with a float.

cbuffer cbPerObject : register( b1 )
{
    bool gUseTexture ;

    row_major float4x4 gWorld ;
    row_major float4x4 gWorldInvTranspose ;
    row_major float4x4 gWorldViewProj ;
    row_major float4x4 gTexTransform ;
    Material gMaterial ;
} ;

So in the constant buffer above, the boolean + the first 3 components of the first row of the world matrix gets mapped to the first vector and this causes everything to get shifted by 3 components, misaligning the world matrix (and the other matrices following and possibly other data).

Two possible solutions :

  1. Move the boolean to the end of the structure. I did this and it worked.
  2. Add a 3-component sized padding variable between the world matrix and the boolean. I tried this by adding an XMFLOAT3 in the c++ structure and a float3 in HLSL. This worked too.

Long story short, pay attention to HLSL packing.

EDIT : At the time I thought these methods worked, as all variables except the boolean had correct values. I didn't use the bool at the time so I assumed that was fine too. Turns out it's not.

HLSL bools and c++ bools have different sizes. HLSL bools are 4 bytes, whereas c++ bools are implementation defined (1 byte on my machine for example). Anyways, they will most likely be different and it causes problems.

Either use Windows BOOL type or another appropriately-sized value like an int or a uint.

Take a look at https://gamedev.stackexchange.com/a/22605.

Also the second to last post here explains the situation clearly (this link is referenced in the answer in the gamedev link above also).

Beware though because packing is still an issue. Even if you use a BOOL or a uint or whatever, if you place it in the beginning in the above structure as before, you will get incorrect values in your constant buffer. So take both of these issues (data alignment and the boolean problem) into consideration when working with constant buffers.

1
votes

As I wrote in my comment, I had a similar problem.

In my case, the pixel shader was correctly bound (see http://msdn.microsoft.com/en-us/library/jj191650.aspx). Furthermore, I ensured by debugging the vertex shader that the result of the Transformation should be visible and hence should generate visible fragments.

In this case (which seems to be the same you describe), make sure that your rasteriser state is correct. You might want to check that it is actually set (using the graphics object view of the immediate context) and that it lets your geometry through. For debugging purposes, I found it helpful disabling backface culling. I use

D3D11_RASTERIZER_DESC rasterDesc;
ZeroMemory(&rasterDesc, sizeof(rasterDesc));
rasterDesc.CullMode = D3D11_CULL_MODE::D3D11_CULL_NONE;
rasterDesc.FillMode = D3D11_FILL_MODE::D3D11_FILL_SOLID;