* This avoids having to go throug the pool checking when we know
concretely what type we're looking up, we just need to cast to the
wrapped type and fetch it.
* Where available or allowed, we replay the captured user markers.
* When we're doing partial replay and we only push some of the markers
then stop, we keep track of how many were pushed so we can pop them
again and keep things balanced.
* Also for renderdoc internal book-keeping around the frame replay
itself we explicitly annotate the renderdoc internals
* On vulkan since there are no queue-level markers, we fire single-shot
command buffers that do nothing but push/pop/insert a marker.
* Instead of hardcoding which extension functions we fetch on replay,
we do the same as on capture and fetch all functions for the
extensions that are actually enabled.
* Unverified as yet, but the former is relatively straightforward and
not serialised, and the latter is quite similar to the NV dedicated
allocation extension.
* Drop support for now deprecated KHX sharing extensions:
- VK_KHX_external_memory
- VK_KHX_external_memory_capabilities
- VK_KHX_external_memory_win32
- VK_KHX_external_memory_fd
- VK_KHX_external_semaphore
- VK_KHX_external_semaphore_capabilities
- VK_KHX_external_semaphore_win32
- VK_KHX_external_semaphore_fd
- VK_KHX_win32_keyed_mutex
* And add support for the KHR promoted versions (which are almost the
same, with some minor changes).
After selecting an application to launch on Android, inspect it to see if
it contains the RenderDoc layer and required permissions. If it does not,
display a warning similar to desktop. When clicked, if only the layer was
missing, offer to patch the APK, uninstall, and reinstall, with the
warning that it doesn't work for all applications (or at all for GLES).
Also provides pointers to how to package the layer yourself.
The process works by using the host temp directory to pull the APK and
modify it. If the steps fail for any reason, the log is populated and
patching is halted.
This allows the library to be used as a Vulkan layer and shared
object for GLES debugging. It was already being renamed in the
distribution scripts. We've been renaming by hand for local
Vulkan development.
For commands that write to stdout multiple times before writing
stderr, the second ReadFile would hang indefinitely. This was
seen using "adb pull" with a sufficiently large APK (~20MB).
This code was inadvertently changing the working environment
of applications that called it. Hasn't been a problem until
we started calling scripts which need PATH to remain good.
* Every time clang-format applied to this file, it would add a new \
on the end of the first line of the comment. Splitting it into two
single-line C comments works around this behaviour.
* When statically linking libstdc++ by default the symbols are all
default visibility. This causes a problem if you statically link an
old libstdc++ then inject librenderdoc.so, and the application starts
using the old libstdc++ instead of the system's.
* Really we only want to use the static one for calls from
librenderdoc.so. In theory -Wl,--exclude-libs,libstdc++ should do this
for us, but it starts to cause bizarre crashes.
* Instead, we have to resort to a manual linker script which whitelists
which symbols should be visible. Normally we can do this just for our
own symbols with -fvisibility=hidden and __attribute__((visibility))
but that doesn't work for linked symbols apparently.
baldurk