D3D11: Use appropriate shader for PEEK_Z.

[magumagu]

Trying to use GetDepthMatrixProgram outside of
TCacheEntry::FromRenderTarget is a bad idea, so don't. Instead, use a
shader which only copies the input.

Fixes lens flare depth test in Twilight Princess. See
http://code.google.com/p/dolphin-emu/issues/detail?id=5999 .

[Sonicadvance1]

Maybe there should be a comment as to why it uses the Color program instead of the depth program for PEEK_Z.

[neobrain]

Why is it a bad idea to use GetDepthMatrixProgram outside of TCacheEntry::FromRenderTarget?

As far as I can see it should be possible to use the depth matrix program in other places, too. That said, there's no denying that the current depth matrix program is simply incorrect, in particular due to precision loss by multiplying with 16777215.0f/16777216.0f (also, the d3d code is setting the shader constants such that blue and green components get calculated per pixel even though the render target is R32F). It should be possible to change it to use integers though.

[magumagu]

Well, there's a couple issues here. One, there isn't any reason to use a pixel shader to swizzle the output for a single pixel. Two, (which I think is the cause of the issue) Renderer::AccessEFB uses a DXGI_FORMAT_R32_FLOAT texture, where GetDepthMatrixProgram expects something more like DXGI_FORMAT_R8G8B8A8_UNORM.

I mean, I'm sure the rounding is broken too, but I'm pretty sure that isn't the issue here.

[neobrain]

I was going to object against the fact that GetDepthMatrixProgram supposedly expects an R8G8B8A8_UNORM format, but you made a point when you mentioned there's no need to "use a pixel shader".

[neobrain]

I don't think the usage of GetColorCopyProgram needs to be commented any further.

LGTM. I'll let someone else have the last vote on GetColorCopyProgram commenting.

[neobrain]

Apparently no one else is interested in commenting. Merging.