Depends on how your functions are written, I guess. They're not doing anything magical; you could easily replicate the code yourself. However, to write math code that is both fast and correct is a colossal pain and takes some care. I've corrected core math routines in a variety of code bases before because they had subtle or not so subtle mistakes in functions that were not properly tested.

D3DX is guaranteed to be correct, and reasonably fast. That counts for a lot.

P.S. A lot of the D3DX functions are coded in the header, if you look in the right places.

If you're not experienced at writing a math library, then don't expect to write one that's going to be better than D3DX. It's more likely you'll end up with something slower and buggier.

I'll also point out that there's a newer replacement for D3DX math, called DirectXMath.

Probably worth noting that the D3DX library has been deprecated; fine if you want to continue using the older SDK, but it'd be better to switch over earlier if you ever plan on updating. The newer SDK still has the most of the same functionality, just a few bits and pieces here and there that work slightly differently e.g. XMMatrixMultiply instead of D3DXMatrixMultiply.

Probably worth noting that the D3DX library has been deprecated; fine if you want to continue using the older SDK, but it'd be better to switch over earlier if you ever plan on updating. The newer SDK still has the most of the same functionality, just a few bits and pieces here and there that work slightly differently e.g. XMMatrixMultiply instead of D3DXMatrixMultiply.

Also the newer one is using intrinsics in the back end to make use of SSE2 which is in 99% of the CPU's nowadays. Btw if you turn SSE2 code generation on in the compiler settings your own matrix code will also be converted to make use of SSE2, however the XM versions will most likely be faster.

I once tried to make my own small matrix math library, because I needed just few basic operations and didn't want to include the whole D3DX for it. When I implemented matrix multiplication I was quite surprised that it indeed was slower (I don't remember how much, but it was quite noticable).

And multiplication is a very simple algorithm, there really isn't much room for mistakes.

The reason is that functions like D3DXMatrixMultiply are written directly in asm and are using SSE.

http://www.gamedev.net/topic/430741-d3dxmatrixmultiply-implimentation/

The D3DX ones actually aren't using SSE, that's why XNAmath and the later rename to DirectxMath library is created as that is using SSE2 intrinsics to do most of it's operations. As the D3DX and XNAmath definitions of the primitive types are substantially different when you actually look at the headers that implement it. The XNA one is doing all kinds of stuff to figure out whether or not SSE is support or not and if it is it switches to __m128 members in most structures.

DirectxMath and XNAmath are the same library and should be included with the June SDK, could be you have to include the xnamath.h as the header for it as the rename happened more or less arround the time that DirectX SDK was changed to be a component of the Windows SDK.

If you just transpose the matrices you wrote yourself before you feed them to D3D they should just work as that is how you switch from column major to row major which D3D9 likes.

You could also look into the DirectX Tool Kit, which provides a lot of functionality similar to XNA, but for C++. This includes their "SimpleMath" library, which is a thin wrapper over DXMath/XMath that provides a simpler interface that might be more to your liking. It also provides useful things like sprite fonts, sprite batching, a shader effects framework with common effects, and more.

You could also look into the DirectX Tool Kit, which provides a lot of functionality similar to XNA, but for C++. This includes their "SimpleMath" library, which is a thin wrapper over DXMath/XMath that provides a simpler interface that might be more to your liking. It also provides useful things like sprite fonts, sprite batching, a shader effects framework with common effects, and more.

The effect system is already in the SDK you just have to compile it yourself, this was done so you could tweak it to your own likings instead of being stuck with a massive lib that you had to link in. You can find this in the Samples directory under C++ and then Effects11.