In addition to moving the code to a separate class, this also
improves flushing behaviour by limiting the amount of memory
being initialized rather than the number of commands executed.
- Implements move semantics for thread objects properly.
This is necessary in order to avoid closing the thread
handle multiple times, and to avoid passing invalid
pointers to the thread function in some cases.
- More closely emulates the behaviour of std::thread.
* [build] do not use shared_library/objects property with MSVC
* [util] use ./com/com_include.h instead of windef.h
It is required for Windows 10 SDK.
* [util] store thread procedure lambda in std::function
* [dxgi] fix annoying MSVC warning
warning C4099: 'IDXGIVkInteropDevice': type name first seen using 'class' now seen using 'struct'
This reverts commit d744a3cd69.
Reportely has an unexpected negative impact on frame times in
The Witcher 3 and potentially other games on some setups.
In my winelib builds, I get unresolved dxvk::D3D11ImmediateContext::MinFlushIntervalUs symbol. I'm not sure why it doesn't inline this constexpr, but it doesn't seem useful to expose that in header anyway.
Wine needs to setup each thread that has an access to Windows APIs. It means that in winelib builds, we can't let standard C++ library create threads and need to use Wine for that instead. I wrote a thin wrapper around Windows thread functions so that the rest of code just has to use new dxvk::thread class instead of std::thread.
std::wstring is problematic, because wchar_t on other platforms might not be what we need. -fshort-wchar can mitigate that partially, but it's more problematic for stdc++ classes.
Also, using UTF-8 for convertion may not be what's intended, esp. if such strings end up being passed back to system API (eg. open file). The patch uses CP_ACP on mingw build (which is how Windows APIs will interpret it) or CP_UNIXCP on Wine (which is Wine extension to convert to whatever glibc and other host libs expect).
It's also needed for the next patch.
In some situations, calling Flush without any commands that
could be flushed would unnecessarily prevent the implicit
flush heuristic from kicking in properly.
Apparently we need those to make sure that the initial layout
transition itself is valid. We don't need to add shader access
though, since we cannot access attachments inside shaders.
Fixes validation errors and some unnecessary cruft. We don't need
visibility operations in the external-to-renderpass dependency
because that part is handled by regular barriers, an execution
dependency is enough (but still required to handle write-after-read.
If a game clears the depth and stencil aspects of a depth-stencil
buffer separately, we must not override the load op and clear value
of the previously set aspect. Fixes a rendering issue in Hitman
Absolution.
Philip Rebohle
Mikhail Paulyshka
Jacek Caban