- Avoids a silly situation where a texture is discarded and an identical copy created immediately afterward.
Unfortunately allocating memory blocks is really slow so avoid it as much as possible.
- Each desc manages its own lifetime now instead of relying on global timestamp check
- Fixes situation where same object remains active without update for long
- Some texture accesses require image compositing steps to assemble the requested image from existing subresources.
Handle all the common routines in a unified manner to avoid having one broken path (e.g mipmap gather not supporting bitcast operations)
- Currently this conversion is being done on the CPU to reuse as much code as possible.
The expectation is that this almost never happens, so there is not point in increasing maintenance burden by adding compute paths
- Tracks which kind of raster was done (Z-ordered vs linear) throughout the application.
- This allows to identify if data is in the expected format or not.
- Do not write to the texture until previous draw call is completed using it.
- This is usually not much of a problem until blending operations come into play.
SRC coordinates
- When extracting a 1x1 texture from another texture of a different
format, width conversion can result in a dimension of 0 if the
extracted texel is not a full texel in SRC
- Adds the same optimization/simplification steps to complex image
transfer routines. Whenever possible, multi-step transfers are collapsed
into a single operation.
- Queueing commands on the offloader is a good idea but unfortunately
page faults can still happen causing a cyclic dependency and eventual
deadlock. Characterized by a vk::wait_for_event timed out error
accompanied by severe hitching.
- Drain the fault-able commands before pushing a submit operation to the
queue. If a fault is in progress, bypass the queue system and submit
raw. Technically this is incorrect but there isn't much that can be
done about it right now.
- A few nagging issues remain, specifically that partial command stream
largely caused by poor synchronization structures for partial CS flush
and also the fact that occlusion map entries wait on a command buffer
and not an EID!
