So, I've been playing a bit with geometry shaders recently and I've found a very interesting bug, let me show you the code example:

struct Vert2Geom
{
	float4 mPosition : SV_POSITION;
	float2 mTexCoord : TEXCOORD0;
	float3 mNormal : TEXCOORD1;
	float4 mPositionWS : TEXCOORD2;
};

struct Geom2Frag
{
	float4 mPosition : SV_POSITION;
	nointerpolation float4 mAABB : AABB;
	float3 mNormal : TEXCOORD1;
	float2 mTexCoord : TEXCOORD0;
	nointerpolation uint mAxis : AXIS;
	float3 temp : TEXCOORD2;
};

...

[maxvertexcount(3)]
void GS(triangle Vert2Geom input[3], inout TriangleStream<Geom2Frag> output)
{
  ...
}

So, as soon as I have this Geom2Frag structure - there is a crash, to be precise - the only message I get is:

D3D12: Removing Device.

Now, if Geom2Frag last attribute is just type of float2 (hence structure is 4 bytes shorter), there is no crash and everything works as should. I tried to look at limitations for Shader Model 5.1 profiles - and I either overlooked one for geometry shader outputs (which is more than possible - MSDN is confusing in many ways ... but 64 bytes limit seems way too low), or there is something iffy that shader compiler does for me.

Any ideas why this might happen?

If you got a device removed, then it's likely that the GPU or driver crashed on you. You can try to confirm this by running on another GPU if you have one (you can try using enabling your integrated Intel GPU if you're on a desktop), or by selecting the WARP adapter. 

You may also want to try compiling as gs_5_0 instead of gs_5_1 and seeing if the generated DXBC is any different. I've seen a few bugs specific to the *s_5_1 profile, mostly around arrays. Or of course you can always move on to the new fancy new shader compiler, and see if that works better for you.  

EDIT: I also assumed that you've already run with the validation layer enabled, and checked for errors. You can also try enabling the GPU validator, since that can potentially catch some issues that aren't caught by the normal validation layer.

@MJP 

I tried both - gs_5_0 (no success with this one) and also enabling GPU validator. GPU validator told me nothing, the crash was exactly the same.

What is interesting, is that enabling GPU validator introduced a random error of CreateCommittedResource failing - it fails with: 

0x887a0005 - The video card has been physically removed from the system, or a driver upgrade for the video card has occurred. The application should destroy and recreate the device. For help debugging the problem, call GetDeviceRemovedReason.

So I went ahead and called GetDeviceRemovedReason

0x887a0007 - The device failed due to a badly formed command. This is a run-time issue; The application should destroy and recreate the device.

The funny thing is, this error is only experienced on random with SetEnableGPUBasedValidation set to TRUE.

Now I've checked where it was crashing also - and it was during the load phase on CreateCommittedResource, and the parameters seems valid (and same each run, the crash is random though!). Weird...

I haven't tried the new compiler, but it might be worth trying (I did try to pass flags to D3DCompileFromFile to disable optimizations, etc. Without any success or message.

@CortexDragon

I thought the same so I gave it a shot, changed structure to:

struct Geom2Frag
{
	float4 mPosition : SV_POSITION;
	nointerpolation float4 mAABB : AABB;
	float4 mNormal : TEXCOORD1;
	float4 mTexCoord : TEXCOORD0;
	nointerpolation uint4 mAxis : AXIS;
	float4 temp : TEXCOORD2;
};

This way it's 96 bytes, and there shouldn't be any alignment problems. And yes, it still crashes.

I'm trying one additional thing - I noticed that I'm currently not running recent AMD drivers (the update was released about 7 days back, I'm still on older version). Let me quick-try the driver update.

And dang! Magic happened. I updated the driver to version 17.11.2 from the 17.11.1 (I believe driver number is actually another one, that the one that is showed in Radeon application) -> and magically it works!

To be precise BOTH things work - GPU based validation no longer seems to crash, and passing out the structure from geometry shader also doesn't seem to crash anymore.