0
votes

When I multiply matrices on CPU

D3DXMATRIX mb = mbb.matWorld * mbb.matView * mbb.matProj;

D3D11_BUFFER_DESC bd;
ZeroMemory(&bd, sizeof(bd));
bd.Usage = D3D11_USAGE_DEFAULT;
bd.ByteWidth = sizeof(D3DXMATRIX);
bd.BindFlags = D3D11_BIND_CONSTANT_BUFFER;
bd.CPUAccessFlags = 0;

D3D11_SUBRESOURCE_DATA initData;
ZeroMemory(&initData, sizeof(initData));
initData.pSysMem = &mb;
//*****************************************

cbuffer ConstantBuffer : register(cb0)
{
    matrix mat;

float4 vs(float4 pos : POSITION) : SV_POSITION
{
    float4 ret = mul(pos, mat);
        return ret;
}

It works great. But when I do it on GPU

struct Buffer
{
    D3DXMATRIX mat[3];
};

Buffer b;
// b[..] = ....

D3D11_BUFFER_DESC bd;
ZeroMemory(&bd, sizeof(bd));
bd.Usage = D3D11_USAGE_DEFAULT;
bd.ByteWidth = sizeof(D3DXMATRIX) * 3;
bd.BindFlags = D3D11_BIND_CONSTANT_BUFFER;
bd.CPUAccessFlags = 0;

D3D11_SUBRESOURCE_DATA initData;
ZeroMemory(&initData, sizeof(initData));
initData.pSysMem = &b;
//*****************************************

cbuffer ConstantBuffer : register(cb0)
{
    matrix matWorld;
    matrix matView;
    matrix matProj;
}

float4 vs(float4 pos : POSITION)
{
    float4 ret = mul(pos, matWorld);
    ret = mul(ret, matView);
    ret = mul(ret, matProj);

    return ret;
}

Nothing is rendered on the screen. What's wrong?

I've been debugging with PIX and it shows that in both versions post-vs vertices are the same.

I noticed that it's caused by projection matrix. For my projection matrix I do:

D3DXMatrixPerspectiveFovLH(&mbb.matProj, D3DX_PI / 2, (float) init_params.width / init_params.height, 0.01f, 100.0f)

FIXED

I have to transpose matrices before passing them to vertex shader.

1
You should be able to use PIX to step through the shader and inspect all of your constant registers.IronMensan

1 Answers

0
votes

Also, you can always do it switch the parameter order in mul();

mul(matrix, vector);

This way the multiplication will be done correctly due to mul() interpreting "matrix" as row major.