XMMatrixTranspose 访问违规读取位置 0xFFFFFFFF

Question

I've got a DirectX 11 Project and I'm using the Beginning DirectX 11 Game Programming book going through all the chapters. I've got my Camera in place and it has been working with no crashes however recently when debugging my Obj Loader I've found that it keeps crashing on the XMMatrixTranspose line, I've swapped my view matrix and projection matrix round to see if it was specific to my view matrix which its not.

Here is the block of code that is causing the error

vMatrix = camera.GetViewMatrix();
vMatrix = XMMatrixTranspose(vMatrix);

pMatrix = XMMatrixPerspectiveFovLH(XM_PIDIV4, (float)(screenWidth / screenHeight), 0.01f, 100.0f);
pMatrix = XMMatrixTranspose(pMatrix);

This is found in my DX11::Render function. This is called in my main.cpp game loop as seen below.

MSG msg = { 0 };

while (msg.message != WM_QUIT)
{
    if (PeekMessage(&msg, 0, 0, 0, PM_REMOVE))
    {
        TranslateMessage(&msg);
        DispatchMessage(&msg);
    }
    else
    {
        directX->Update(0.0f);
        directX->Render();
    }
}

directX->Shutdown();

All of my variables have memory and this code has been working before, when it started to do this it would do it very intermittently, but now it seems to be doing it all the time.

Any help and guidance would be amazing! I've been researching for hours and can't find anything that works and I've had this issue for about 7 hours! Its gotta go!!

Thanks in advance.

UPDATE It has now gone back to intermittently crashing

UPDATE 2 I'm also using xnamath not DXMath

Answer 1

You should first read MSDN Programmer's Guide, specifically the type usage guidelines .

DirectXMath and XNAMath are basically the same library. XNAMath is just the old version of it which was last updated in 2011. You should move to using DirectXMath. See this blog post .

The most likely problem is that you are trying to use XMMATRIX and/or XMMVECTOR types declared on the heap (ie in a class you use new on) which will not be properly aligned for x86 (32-bit) and ARM builds. It happens to line up for x64 native code, but that's mostly a happy accident.

In other words the following code will result in an AV in x86 (32-bit) and ARM depending on exactly how memory is laid out, but will work for x64 native:

struct A
{
    XMMATRIX m;
};

A a = new A;
a->m = XMMatrixIdentity();

To make it work robustly on all architectures, you need to use:

struct A
{
    XMFLOAT4X4 m;
};

A a = new A;
XMStoreFloat4x4(&A->m, XMMatrixIdentity());

You'd use XMFLOAT4X4 or similar types which do not require alignment in your classes, and then use explicit Load/Save functions. Note that you can also make use of more compact forms like XMFLOAT4X3 depending on your matrix content.

Alternatively, you can make use of the SimpleMath wrapper in the DirectX Tool Kit which does exactly that only "automagically" through the power of C++ conversion operators and constructors.

struct A
{
    Matrix m;
};

A a = new A;
A->m = XMMatrixIdentity();
// Note this is redundant
// as the Matrix default constructor sets it to identity already.

This functionality is not part of the base library because it has some performance implications. By making it part of the SimpleMath wrapper instead, it allows users to opt in to an easier model of use while preventing accidental use of slower conversion paths for everyone else.