What is the purpose of the array aspect of IASetVertexBuffers?

Instancing needs an extra vertex buffer for the per-instance data.

It's also the case that sometimes non-interleaved input is the cleanest solution. Specific example: I'm doing some keyframe interpolation in part of my code. Vertex buffer slot 0 gets positions and normals for previous keyframe, vertex buffer slot 1 gets positions and normals for current keyframe, vertex buffer slot 2 gets everything else. Much more performant than interpolating on the CPU and building dynamic data for each model (not to mention needing about 10% lines of code by comparison).

Third example: stencil shadows. When drawing stencil shadows you just need positions, so you split them off into a buffer bound to slot 0 and put everything else into a buffer bound to slot 1. Instant bandwidth saving.

The only example I can think of is in non-interleaved data scenarios, but that doesn't really seem like something that should be encouraged through the API. :\

Why not? Interleaved isn't always the best way to lay out your vertex data.

[quote name='Sent1nel' timestamp='1331145203' post='4920145']
The only example I can think of is in non-interleaved data scenarios, but that doesn't really seem like something that should be encouraged through the API. :\

Why not? Interleaved isn't always the best way to lay out your vertex data.
[/quote]

There is nothing wrong with non-interleaved data per se, it just don't have the same performance characteristics as interleaved data (see Game Programming Gems 1 or here). I just find it odd that Direct3D seems to expose the use of vertex buffer arrays whereas, as far as I can tell, OpenGL does not.

The array signature of that function is simply to allow you to optimise your D3D layer by making less function calls, if you want to.

e.g. instead of making multiple calls like:

[font=courier new,courier,monospace]SetVertexBuffer( 0, buffer0 );[/font]
[font=courier new,courier,monospace]SetVertexBuffer( 1, buffer1 );[/font]
They're giving you the option to combine those calls together like:
[font=courier new,courier,monospace]SetVertexBuffers( 0, 2, {buffer0, buffer1} );[/font]

Though you're still free to make one call per binding too if you want to. Many parts of D3D11 are written this way, allowing you to make less D3D function calls in many parts of the API if you choose to.

Worth noting as well that the Apple link provided has a nice section beginning "an exception to this rule is..."

The API for setting multiple vertex buffers in one call is not even remotely unusual for D3D11 - it's completely consistent with comparable APIs for setting constant buffers and shader resources, for example. The fact that OpenGL doesn't expose a similar API signifies nothing - they're different APIs and do things in subtly different ways (OpenGL for example still doesn't decouple vertex layout from data/buffer specification, whereas D3D has been doing that for ages, and is much cleaner as a result).

At the risk of sounding like an idiot...

What exactly is being talked about here? There is no ID3D11DeviceContext::IASetVertexBuffers, only ID3D11DeviceContext::IASetVertexBuffer, and I see nowhere where it has the ability to bind an array of buffers.



EDIT: D'oh. I was thinking of IASetIndexBuffer for some reason. Probably because that's where OP's link pointed to.

http://msdn.microsof...6(v=vs.85).aspx

void IASetVertexBuffers(
[in] UINT StartSlot,
[in] UINT NumBuffers,
[in] ID3D11Buffer *const *ppVertexBuffers,
[in] const UINT *pStrides,
[in] const UINT *pOffsets
);

There is nothing wrong with non-interleaved data per se, it just don't have the same performance characteristics as interleaved data (see Game Programming Gems 1 or here).

I would be careful about drawing broad performance conclusions from extremely old source material, or a few performance guidelines written for some very specific hardware. You also have to consider a broader picture here...even if interleaved data is better for one particular draw call it might not be the best choice overall if it lets you save even more performance for a different draw call.