* Debugging shaders with UAV access could(likely will) have side-effects on
those UAV contents, so when ContinueDebug needs to fetch UAV contents we
should replay back to before the draw to fetch them.
* This replay only happens once per ContinueDebug, and UAV contents are cached
globally so we only do it once per UAV that's accessed, so the impact is low.
* In heavy D3D11 workloads the refcounting overhead especially during fast
binding changes was significant. Refactoring the refcounting to work on a
different model and deferring destruction of objects removes most of the
overhead.
* This helps catches cases where a discarded image is accidentally used and in
many cases may still have valid data. Particularly on Vulkan this is relevant
for DONT_CARE renderpass load and store ops.
The rules for merging semantics into an array were not strict enough.
If either the type, interpolation mode, or size is different, or if it
is not using a register's x component, then we can't combine it. Also,
the rules for marking a semantic as array length 1 were too strict,
resulting in some semantics packing into other registers incorrectly.
* This doesn't test every edge case, but it is a simple smoke test that ensures
any format can at least get consistent accurate results out.
* Further work is needed to test the different failures, overdraws, primitive
IDs, and all the other specific checks in the history.
SV_PrimitiveID is only valid as a pixel shader input if there's either
no geometry shader, or the geometry shader emits SV_PrimitiveID. With
D3D11, the debug layer will complain but rendering will proceed. With
D3D12, the PSO fails to compile. For both APIs, pixel shader debugging
now checks whether this is true, and disables its usage if it isn't.
Added tests for various valid scenarios with SV_PrimitiveID.
* It's not particularly scalable and can be brittle to driver changes, and we
can use targeted specific pixel tests to check what we really want - to see if
the output has rendered correctly.
* Overlay tests still check files directly - this is a future refactor to
remove.
D3D11 may default to a different adapter than D3D12. For the sharing
test, both adapters must be D3D12 compatible. Tell D3D11 to use the
adapter that D3D12 is using to ensure compatibility.
Also fixed a memory leak in the test.
On laptops with integrated+discrete adapters, adapter 0 may not
support D3D12, so a default create device call will fail. Adapters
other than the 0th one need to be specified explicitly.
* On some GPUs the stencil output comes in all channels, not just green, and
it's unclear if this is legal or not. To be safe just mask out the green we
expect.
* On all APIs structs are aligned up to 16-byte alignment, and on D3D as with
arrays then elements coming after the struct can be packed into the padding
after the last struct element.
* These tests ensure that texture rendering works correctly for all different
types of texture types, and for all formats, across different APIs, including
across a remote-proxy connection.
baldurk