* It seems like on nv windows we need to explicitly rebind the main context to
the main thread and give each worker thread its own window, to prevent the
worker thread from being unable to bind its context sometimes.
* The block index may change between capture and replay, so serialising the
index alone is unstable similar to locations. Program initial states already
serialise by name, but if a capture contains a block binding change mid-frame
this could serialise wrongly.
* 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.
* We preserve each API's interpretation of bit order for packed formats like
RGBA4 or R5G6B5 when displaying the raw data in the UI, but when we need to
proxy it or save to disk, we always transform to D3D's order as standard.
* This allows us to proxy them reliably because we always have a standard bit
order and APIs that need a different order transform when fetching data to the
standard format, or setting proxy data from the standard format.
* This is a very big blunt hammer for fixing the problem of multithreaded
submission from GL. Every GL call checks to see if the context changed (which
would only happen from a thread switch to a different context) and if detected
it inserts a manual MakeCurrent call equivalent.
* It's slow to capture (when this happens - checking is not particularly slow)
and slow to replay, but it's functional which is an improvement.
* 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.
* This is needed particularly for integer textures which are incomplete if
sampled (even with a non-sampling instruction) with the default LINEAR
filtering.
baldurk