I'm having an issue rendering my sprites in DX11. Atm for debugging convenience I've not transformed the vertices into screen space when outputting from the vertex shader. So at the moment I use a vertex shader & a pixel shader which just samples the diffuse texture... the basics.

Using a triangle strip, the sprite seems to come out flipped vertically(EDIT). So it does not appear unless I set my rasterizerstate.FrontCounterClockwise to 1.

Here's setting up my view port:
[source lang="cpp"]
//Setup the viewport for rendering.
m_oViewPort.Width = (float)m_poWindow->GetClientWidth(); //1024 in this case
m_oViewPort.Height = (float)m_poWindow->GetClientHeight(); //768 in this case
m_oViewPort.MinDepth = 0.0f;
m_oViewPort.MaxDepth = 1.0f;
m_oViewPort.TopLeftX = 0.0f;
m_oViewPort.TopLeftY = 0.0f;
[/source]

and here is some PIX debug information:


Following this for the triangle strip winding direction:


I'm confused as to why the sprite renders flipped vertically(EDIT). Can anyone shed some light on my problem?

EDIT: Thought I might as well throw this in...

[source lang="cpp"]
//Shader Model 4.0

Texture2D DiffuseTexture : register( t0 );

SamplerState DiffuseTextureSampler : register( s0 );

cbuffer cbViewProj : register( b0 )
{
float3x3 ViewProjection;
}

cbuffer cbWorld : register( b1 )
{
float3x3 World;
};

struct VS_INPUT
{
float2 Pos : POSITION;
float2 Tex : TEXCOORD0;
};

struct PS_INPUT
{
float4 Pos : SV_POSITION;
float2 Tex : TEXCOORD0;
};

PS_INPUT VS( VS_INPUT input )
{
PS_INPUT output;

output.Pos = float4(input.Pos.x, input.Pos.y, 0.0f, 1.0f);
output.Tex = input.Tex;

return output;
}

float4 PS( PS_INPUT input) : SV_Target
{
return DiffuseTexture.Sample(DiffuseTextureSampler, input.Tex );
}[/source]

You have supplied your vertices in counter-clockwise order: (-32, 32), (-32, ,-32), (32, 32) so your triangles are effectively facing backwards.

Either reorder the vertices (-32, 32), (32, 32), (-32, -32) or the indices (0, 2, 1).

I must be having a mental blank...

TopLeft XY = 0,0
BottomRight XY = 1024, 768

So sprites...
BottomLeft = -32, 32 (0)
TopLeft = -32, -32 (1)
BottomRight = 32, 32 (2)
TopRight = 32, -32 (3)

Which according to the above image is how the vertices should be ordered? This is exactly how I set up my sprites in dx9 and it rendered perfect...

Lastly, when I did manually reorder the vertices during testing like you said, the sprite rendered but was still flipped Vertically(EDIT) (could tell by the texture)

Keep in mind that the viewport-coordinates are in pixels with the topleft pixel being (0, 0) and the bottomright (1024, 768) (in this case), whereas the vertex-positions are screen space coordinates that range from top (-1.f, 1.f) to bottom (1.f, -1.f).

Thx for pointing that out. I realise now not projecting my vertices into screen space was the issue (since the projection unflipped them).

Re-wrote my shader:

[source lang="cpp"]//Shader Model 4.0

Texture2D DiffuseTexture : register( t0 );

SamplerState DiffuseTextureSampler : register( s0 );

cbuffer cbViewProj : register( b0 )
{
float3x3 ViewProjection;
}

cbuffer cbWorldColour : register( b1 )
{
float3x3 World;
float4 vMeshColor;
};

struct VS_INPUT
{
float2 Pos : POSITION;
float2 Tex : TEXCOORD0;
};

struct PS_INPUT
{
float4 Pos : SV_POSITION;
float2 Tex : TEXCOORD0;
};

PS_INPUT VS( VS_INPUT input )
{
PS_INPUT output;
float3 oVec = float3(input.Pos.x, input.Pos.y, 1.0f);
oVec = mul(oVec, ViewProjection);

output.Pos = float4(oVec.x, oVec.y, 0.0f, 1.0f);
output.Tex = input.Tex;

return output;
}

float4 PS( PS_INPUT input) : SV_Target
{
return DiffuseTexture.Sample( DiffuseTextureSampler, input.Tex );
}[/source]
Though now I have another issue /sigh. My verts don't seem to be projecting correctly.

Given that this is all I've changed, that the PIX debug info in my first post is still valid (E.G the vertex values are entering in that order) why would I get this output:


In memory, the ViewProjection buffer looks like this:


So going by my first post...
Prim1Vert0 = -32, 32
Prim1Vert1 = -32, -32

Output for Prim1Vert0 = -0.063, 0.083
Output for Prim1Vert1 = -0.063, 0.083

Considering they have two different values for the y component the output should also by different. I manually wrote out the matrix math and it appears something is not working as expected with the "oVec = mul(oVec, ViewProjection);" line. Though considering how the ViewProjection buffer looks, I'm not sure why it isn't working.

EDIT: As I'm noob and idk how to test the values otherwise... I instead outputted the individual values from ViewProjection to make sure they were mapped correctly.
Using the above buffer as a reference:
ViewProjection._m00 = 0
ViewProjection._m01 = 4
ViewProjection._m02 = 8

I need to rearrange my matrix?

Vertices 0, 1 and 4, 5 have identical SV_POSITION values, so you're rendering a line.

I acknowledged that in my last post... I'm unsure why they have identical positions, especially when different vertex values are being processed.

EDIT: Furthermore... does that matrix layout (mentioned in my last post) seem correct? I thought that was the problem. It (from my perspective) appears correct in memory but accessing the individual member variables reveals the values are in the wrong position? Or perhaps I'm misinterpreting

EDIT: More testing.... if the values on the left represent how my 3x3 matrix looks in the buffer, the red values on the right represent how they are accessed in the hlsl shader.




The 3 floats at the bottom are just padding to make sure the buffer size reaches a multiple of 16... but I'm a little confused why it is being accessed this way. It is as if it is a row major Matrix4x4?

I've done a bit of research and am now working under the assumption the float3x3 becomes fragmented when being stored in the registers (as it is packed in groups of 4 floats).

Now it seems even if my application is completely 2D I should still use a 4x4 matrix due to this annoying behavior. Is this correct, or is there some workaround I am unaware of?

True, the matrix is stored in memory as a 3 * 4 component float. When you Map() the resource, you'll have to adjust your sysmem copy accordingly.

For a 2D game there is no real need to upload the entire matrix though. For a non-rotating camera all you need is the inverse of its translation and the '2 / width' and ' 2 / height' (the screen space 'stretch') components of its orthographic projection, which can both be combined in a single float4. As long as your game doesn't depend on the depth buffer, you'll only have to make sure you output w = 1 from the vertex shader for correct clipping.

float4 camvp = { camera.x, camera.y, projection[0][0], projection[1][1] };

output.Pos = float4((oVec.x - camvp.x) * camvp.z, (oVec.y - camvp.y) * camvp.w, 0.0, 1.0);

Ty for your reply. As you said I ended up manually padding my struct on the cpu. Though I'm going to try and use my camera to zoom and rotate ;) I should have done more research on the hlsl padding though... I had come from using the dx9 effects framework which I guess must have accounted for the padding.