Why would you even need to read them back? In order to get them into the cbuffer, you must have had them somewhere CPU-side to begin with. Cache and use that copy instead.

Tell us more about the process.

So when you say sprite model extractor do you mean like a 2d sprite, rendering to the screen, and then you somehow want to generate a model out of that? (Vertices... Indices...)?

And you said you're reading the "Constant Buffer", on the cpu side to get the variables. There I have to fully agree with mhagain, if you've written to them, you must have the data you mapped into them on the cpu side somewhere, maybe try to save them. And if I'm not mistaken, it sounds like (Maybe I'm wrong) you're trying to change some variables inside the constant buffer on the gpu side, see the problem with that is that it's impossible (Please correct me if I'm wrong), and that's the reason why they were named 'constant' buffers.

One way that you could get variables from the gpu is somehow writing them to the output, could be a render target, then on the cpu side extract the variables from the render target and process them there. You could even use a structured buffer, send the structure to the Gpu, and then you may edit (write to) this buffer, and you can write to any index of it. (It's usually an array).

PS. GPU to CPU operations aren't usually the fastest.

But give us some more info, then we could get closer to the solution.

EDIT:

"In addition, is there any API to debug the shader? Like the windows debug API?"

There is, Intel GPA is very good at this (well, it works for me), another option may be to enable the debug flag when creating the shader, then later on in either visual graphics debugger or pix, find the shader object, click on it, then somewhere you'll be able to open it and even sometimes debug it (This is possible with the visual graphics debugger, not sure about pix though).

Hope this helps somehow.

-MIGI0027

Please be aware that my knowledge on this is very limited, though it seems to work for me. So if you see anything wrong or complete nonsense, please tell me so I don't send the wrong message to fs1.

If you already know the layout of the buffer, like:

struct CBUFFER
{
    D3DXMATRIX view, proj, world;
};

Maybe you could create a staging buffer with the same size as the origional one. Then with the staging, do something like this to copy the data:

DeviceContext->CopyResource(staging, buffer);

And remember that this staging buffer has different parameters than the original one, it should include this (the creation flags):

bufferDesc.Usage = D3D11_USAGE_STAGING;
bufferDesc.BindFlags = 0;
bufferDesc.CPUAccessFlags = D3D11_CPU_ACCESS_READ;

So when you've copied the data from the original buffer to the staging, you may begin reading from it:

D3D11_MAPPED_SUBRESOURCE ms;
engine.devcon->Map((ID3D11Buffer*)staging, NULL, D3D11_MAP_READ, NULL, &ms);

CBUFFER cbData;
memcpy(&cbData, ms.pData, sizeof(cbData)); // I think the syntax looks somewhat like this

engine.devcon->Unmap((ID3D11Buffer*)staging, NULL);  

Now you have an instance of the data which you should be able to read from ( the cbData ).

Now this may not work if you don't know the layout/structure of the buffer, in that case I'll have to think a bit more.

Hope this is useful!

-MIGI0027