Could be an ANV bug, but on Intel the game doesn't render geometry properly at all and ultimately crashes with:
/usr/i686-w64-mingw32/include/c++/13.1.0/array:202: constexpr std::array<_Tp, _Nm>::value_type& std::array<_Tp, _Nm>::operator[](size_type) [with _Tp = dxvk::DxsoSemantic; unsigned int _Nm = 32; reference = dxvk::DxsoSemantic&; size_type = unsigned int]: Assertion '__n < this->size()' failed.
Trace captured with dxvk on Intel: RT3_Intel.trace.tar.xz
We just fixed the Intel crash at least.
Apparently it crashes on ANV inside vkCreateGraphicsPipeline now.
There are a bunch of validation errors that need to get investigated.
RT3 validation errors.txt
This should work with dxbc-spirv.
Nope, still doing the terrain shuffle (see the large patch at the bottom).
apparently I misunderstood the issue because the trace is borked and all the discussion was around an ANV crash.
Anyway, debugged this now and the game is requiring fixed-function vertex shaders to produce identical math to their own custom programmable shaders, which is completely insane.
Nothing extracted yet.
This is going to be a rather hard one to describe, so I've captured a video. The TLDR of it is that when scrolling through some of the closest zoom levels on the game map, the terrain textures sort of shuffle around in a crazy dance.
I've captured a trace, but since the game does "wonderful" things with buffers, apitrace has mostly caught garbage. The issue can be seen in the trace as well though, among the madness.
Software information
Railroad Tycoon 3
System information
Apitrace file(s)
RT3.trace.tar.xz
Log files
RT3.log