Add None WSI

[expipiplus1]

This, of course, is unable to create windows.

TODO:

• VK_EXT_headless surface support? (I think this could be a separate PR if desired)
• Testing
• Does this violate assumptions elsewhere? (Nothing that doesn't require createSurface succeeding first)

[expipiplus1]

I couldn't find any tests or builds for the non-windows wsi. Do these exist somewhere I've not checked?

[pchome]

idk,

#define DXVK_WSI_NONE 1
#include <d3d11.h>
#include <cstdio>

int main(int argc, char** argv) {
  ID3D11Device* device;
  ID3D11DeviceContext* ctx;
  D3D_FEATURE_LEVEL level = D3D_FEATURE_LEVEL_11_1;
  D3D_FEATURE_LEVEL level2 = D3D_FEATURE_LEVEL_11_1;
  HRESULT hr = D3D11CreateDevice(nullptr, D3D_DRIVER_TYPE_HARDWARE, nullptr, 0, &level, 1, D3D11_SDK_VERSION, &device, &level2, &ctx);

  if (FAILED(hr)) {
    printf("ko\n");
    return 1;
  }

  // TODO: do something
  // ...
  ctx->Flush();

  printf("done\n");
  return 0;
}

Run
DXVK_LOG_LEVEL=debug DXVK_LOG_PATH=none ./native_test_wsi_none

Output

info:  Game: native_test_wsi_none
info:  DXVK: v2.1
info:  Vulkan: Found vkGetInstanceProcAddr in libvulkan.so @ 0x7f009c8bb920
info:  Built-in extension providers:
info:    No WSI
...

[expipiplus1]

@pchome sure, writing such a test isn't hard, but there doesn't seem to be any infrastructure from which to call the test.

[misyltoad]

I did have some tests at one point -- but currently unmerged. I think they are in the sample branch.

[expipiplus1]

in that case, if there aren't tests for the less-trivial glfw and sdl2 paths in main, perhaps it's ok to defer tests here?

[misyltoad]

Yeah sure.

Should probably hook up headless surface though.

[expipiplus1]

ok, I've put a call to vkCreateHeadlessSurfaceEXT in there.

[expipiplus1]

@Joshua-Ashton is this good to merge?

[expipiplus1]

FWIW in terms of "testing", I've been using this since april without issue.

[flibitijibibo]

@expipiplus1, do you have time to review the conversation in #3451 to see how this MR would best fit into an "official" WSI system for dxvk? Once we have a working system it should be easy to include a None implementation as part of that work.

[expipiplus1]

@flibitijibibo It looks as though it would fit in with the API @Joshua-Ashton proposed there. As long as nothing requires getting a window or monitor then it's all fine. CreateSurface here just ignores the window parameter.

I would be keen to get this merged, and can commit to porting it to the IDXVKWsi when it lands.

[flibitijibibo]

That makes sense to me - the absolute earliest I'll be able to look at DxvkWsi is December (if that; SDL_ActionSet work starts January so it's a tiny window), so this should probably go in first so that the implementer knows for sure that all backends still work correctly.

[flibitijibibo]

The dynamic WSI patch series can now be found at #3738. Definitely give this a try whenever you have time; it's mostly the same at the backend level but with a C++ interface to allow for multiple simultaneous implementations in the official binary.

[flibitijibibo]

This got stepped on by #3738, but the good news is that this should now be possible to support in the same binary as SDL/GLFW.