The "files" system expects other files to be #included, which won't make
sense for GLSL sources, which are expected to be concatenated. Use the
concatenated string as a single "main.glsl" file.
* In some cases there was an unpredictable crash when starting a replay
loop, doing a full frame replay immediately after a previous replay.
There were no debug messages, and the debug layer fixed it - as did
enabling single flush validate. It seems to me some timing/overloading
issue, and doing this sync should be relatively harmless so we use
this to avoid the crash.
* Technically the resolve doesn't allow format conversion at all, so
instead of resolving directly to the swapchain image we need a middle-
man image of the same format as we use internally for the MSAA target
(RGBA8_SRGB) to resolve to, then blit from that to the actual
backbuffer.
* ARB_texture_view only works on textures allocated with
ARB_texture_storage so we have to check for both extensions and then
use texture storage to allocate space for the destination array
texture.
* For now, resource replacement implicitly means shader edits, but other
replacements could also affect the results of PostVS (e.g. buffer or
state changes). Trying to capture all this state in the cache key is
not very useful.
* The PostVS cache will be re-filled immediately after, as any
replacement does a force-refresh of the current event.
* AMD reports image required sizes with some variation even for the same
creation parameters. This can mean the same image is reported with
less required size during capture compared to replay, meaning memory
binds won't work.
* To get around this, we pad up the requirements to alignment * 4 and
fudge to try and encompass any possible variation.
* This can happen if the resource was not included in the capture, but
there was a descriptor passed (i.e. not NULL desc which is handled
separately) which specifies some parameters, but leaves the format
as UNKNOWN.
* Not doing this could taint the state and even cause invalid read-write
conflicts to change bindings vs. what should happen, if a RT is left
bound when it shouldn't, and then an SRV is bound as NULL because it
would conflict.
* Intel's drivers in particular seem buggy when trying to reflect
shaders, so there's no point in reflecting shaders that aren't from
the capture as all we do is add the potential for crashes.
* VULKAN_LAYER_FOLDER points outside the prefix to /etc in absolute, so
usually if you customise one you'll want to customise the other.
* This is especially true if you're pointing the install prefix at some
non-root location, and you want make install to succeed instead of
failing to write to /etc.
baldurk