* When displaying mip 0 of a texture at less than 100% zoom we linear sampling,
but we don't want to linear sample across slices. Adding a half pixel offset
in z ensures we sample precisely on the slice itself.
* We force FetchUAV/FetchSRV to always return something, and it can be empty if
the resource is unbound meaning we get proper OOB behaviour for it. The D3D12
implementation still prints an error message for registers that don't
correspond to the root signature at all, but for D3D11 we silently allow it.
* 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.
baldurk