nvapi-d3d: Just error out of GetCurrentSLIState

[K0bin]

Fallout New Vegas seems to think it's running on an SLI system unless we just error out here.

I also tried setting maxNumAFRGroups and numAFRGroups both to 0 but that didn't fix it either.
Additionally I tried just setting numAFRGroups to 0 and keeping maxNumAFRGroups at 1 but that didn't work either.

The Nvidia sample suggests what we have right now is a valid implementation:
bSLIEnabled = (gMaxNumAFRGroups > 1); but it appears that the check in Fallout isn't particularly robust.

So I suggest we just return an error when the function gets called. Alternatively we could return null pointers for the SLI functions in QueryInterface. I don't have a test setup for Nvapi on Windows right now, so I can't check what Nvidia does on Windows these days without a lot of effort.

Fixes: doitsujin/dxvk#4696
ValveSoftware/Proton#356 (comment)

[adamdmoss]

This feels wrong, at least for the purposes of fixing one game... NV docs suggest that an error from the nvapi call should even be interpreted by a game as 'prompt the player to upgrade their driver' which is counterproductive.

https://docs.nvidia.com/gameworks/content/technologies/desktop/sli_using_nvapi.htm

The fix would be to figure out why this game is happy with the result from real nvapi and unhappy with the result from dxvk-nvapi. dxvk-nvapi's implementation looks sensible at first glance but I wouldn't be surprised if a game expects numAFRGroups to be 0 instead of 1.

[K0bin]

but I wouldn't be surprised if a game expects numAFRGroups to be 0 instead of 1.

I tested that. It did not fix it. I guess I need to see what Windows returns.

[K0bin]

Okay I did thorough testing on Windows.

The docs don't list NVAPI_NO_ACTIVE_SLI_TOPOLOGY as a possible return value but that's exactly what I got.
The struct fields do not get initialized except numVRSLIGpus of version 2 of the struct. That one gets set to 0, the others don't get changed. The other errors (incompatible struct, invalid argument) do take precedence, so that part stays as it was.

NvAPI_D3D_ImplicitSLIControl does not return NVAPI_NO_ACTIVE_SLI_TOPOLOGY, regardless of the value passed for implicitSLIControl.

[K0bin]

error: XDG_RUNTIME_DIR is invalid or not set in the environment.

The failed CI test isn't caused by the PR.

[Saancreed]

It is, you just looked at the wrong message:

D3D methods succeed
  GetCurrentSLIState succeeds
  GetCurrentSLIState with unknown struct version returns incompatible-struct-
  version
-------------------------------------------------------------------------------
../../../tests/nvapi_d3d.cpp:52
...............................................................................

../../../tests/nvapi_d3d.cpp:55: FAILED:
  REQUIRE( NvAPI_D3D_GetCurrentSLIState(&unknown, &state) == NVAPI_INCOMPATIBLE_STRUCT_VERSION )
with expansion:
  -113 == -9

===============================================================================
test cases:    20 |    [19](https://github.com/jp7677/dxvk-nvapi/actions/runs/13467194283/job/37635278839?pr=279#step:8:20) passed | 1 failed
assertions: 27591 | 27590 passed | 1 failed

[K0bin]

Should be fixed now.

[adamdmoss]

Nice (and surprising) discoveries.

[jp7677]

Thanks a lot for the research and fix, very cool!

Minor nit, otherwise LGTM! If you are at it, you could tweak the commit message slightly since it feels a bit more graceful now than just "error out". But also ok if you want to keep it like it is.