[Bug 55107] New: DirectX programs cannot run on dedicated GPU in windowed mode for hybrid-graphics laptop

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=55107

Bug ID: 55107
           Summary: DirectX programs cannot run on dedicated GPU in
                    windowed mode for hybrid-graphics laptop
           Product: Wine
           Version: 8.10
          Hardware: x86-64
                OS: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: -unknown
          Assignee: [e-mail]
          Reporter: [e-mail]
      Distribution: ---

Years ago, I created a Direct3D program with statically-linked C library. This
program has minimal library dependency and has been tested working on all
versions of Windows with DirectX 9 supported. You can download it at:
https://github.com/xuancong84/solid-rubix-cube/raw/master/Release/svcube.exe

I typically run this programs in 4 modes of Wine:
1. full-screen mode without dedicated GPU, `wine svcube.exe`
2. full-screen mode with dedicated GPU, `__NV_PRIME_RENDER_OFFLOAD=1
__GLX_VENDOR_LIBRARY_NAME=nvidia wine svcube.exe`
3. windowed mode without dedicated GPU, cmdline same as 1 but with 'emulate a
virtual desktop' checked
4. windowed mode with dedicated GPU, cmdline same as 2 but with 'emulate a
virtual desktop' checked

From Wine 5.0 onwards (not tested on older versions), not a single version of
Wine (including both wine and winehq, including -stable -staging and -devel)
can pass all the 4 modes. In the latest winehq-stable (version
8.0.1~bullseye-1, on MX-Linux21/Debian11), the program can run successfully in
1,2, and 3, but crashes on 4 with the following error:
```
002c:fixme:winediag:LdrInitializeThunk wine-staging 7.22 is a testing version
containing experimental patches.
002c:fixme:winediag:LdrInitializeThunk Please mention your exact version when
filing bug reports on winehq.org.
0084:fixme:hid:handle_IRP_MN_QUERY_ID Unhandled type 00000005
0084:fixme:hid:handle_IRP_MN_QUERY_ID Unhandled type 00000005
0084:fixme:hid:handle_IRP_MN_QUERY_ID Unhandled type 00000005
0084:fixme:hid:handle_IRP_MN_QUERY_ID Unhandled type 00000005
MESA-INTEL: warning: Performance support disabled, consider sysctl
dev.i915.perf_stream_paranoid=0
MESA-INTEL: warning: Performance support disabled, consider sysctl
dev.i915.perf_stream_paranoid=0
MESA-INTEL: warning: Performance support disabled, consider sysctl
dev.i915.perf_stream_paranoid=0
0110:fixme:ver:GetCurrentPackageId (006CFAF0 00000000): stub
006c:fixme:imm:ImeSetActiveContext (0000000000010026, 0): stub
006c:fixme:imm:ImmReleaseContext (0000000000010020, 0000000000010026): stub
0110:fixme:imm:ImeSetActiveContext (00010072, 1): stub
0110:fixme:imm:ImmReleaseContext (0002006A, 00010072): stub
X Error of failed request:  BadMatch (invalid parameter attributes)
  Major opcode of failed request:  156 (NV-GLX)
  Minor opcode of failed request:  43 ()
  Serial number of failed request:  409
  Current serial number in output stream:  410
```

I have NVIDIA CUDA installed (Driver Version: 530.30.02, CUDA Version: 12.1).
And PyTorch/Tensorflow can run with/without GPU decoding. I am pretty sure my
Nvidia GLX is working properly because Mode 2 can run successfully with the
process appearing in `nvidia-smi` and taking GPU usage.