Support building engine with multiple graphics API

[1vanK]

No description provided.

[weitjong]

I am sorry. I don’t have any spare time for the project now. It is difficult for me to see what went wrong just by looking at those commits, especially when a few commit IDs now become invalid (probably due to force push or something). I also promise myself not to touch those old CMake build scripts anymore.

…

On 8 Apr 2022, at 5:35 AM, 1vanK ***@***.***> wrote:  @weitjong any ideas? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.

[1vanK]

Building:

cmake.exe repo -B build_vs -G "Visual Studio 17" -A x64^
 -D URHO3D_OPENGL=1 -D URHO3D_D3D9=1 -D URHO3D_D3D11=1^
 -D URHO3D_LIB_TYPE=STATIC -D URHO3D_STATIC_RUNTIME=0 -D URHO3D_HASH_DEBUG=0^
 -D URHO3D_ANGELSCRIPT=1 -D URHO3D_FORCE_AS_MAX_PORTABILITY=0 -D URHO3D_LUA=1 -D URHO3D_LUAJIT=1 -D URHO3D_GENERATEBINDINGS=0^
 -D URHO3D_SAMPLES=1 -D URHO3D_TOOLS=0 -D URHO3D_DOCS=0^
 -D URHO3D_NETWORK=1 -D URHO3D_DATABASE_SQLITE=1 -D URHO3D_TRACY_PROFILING=0 -D URHO3D_PHYSICS2D=1 -D URHO3D_URHO2D=1

Running:

18_CharacterDemo_d.exe -opengl
18_CharacterDemo_d.exe -d3d9
18_CharacterDemo_d.exe -d3d11

[eugeneko]

Just curious, what's the performance impact on Sample 20 (Debug and Release)?
Maybe I'll test that myself but I don't know when I will be able to.

[1vanK]

I measured the difference in performance in 99_Benchmark_d.exe and noticed a slight drop in performance. I don't remember the specific numbers, it was something like a percent (10-20 fps with fps > 1500).

[eugeneko]

Sample 20 with instancing off shows 15% performace degradation in both debug and release. Still not that much given the nature of the sample and conditions.

[1vanK]

I will check it. As an optimization we can use function pointers instead of using if (gapi == x) then call func_x();

[eugeneko]

Note that I have disabled instancing to measure draw command queuing itself (which was affected by PR).
With instancing on there are very few draw calls and probably no difference.

[1vanK]

Can you test again (with latest commit, but i'm not sure if it matters), because

Old debug:

New debug:

Old release:

New release (fps better, it's just measurement error):

I just changed in example 20

void HugeObjectCount::SetupViewport()
{
    auto* renderer = GetSubsystem<Renderer>();

    // Set up a viewport to the Renderer subsystem so that the 3D scene can be seen
    SharedPtr<Viewport> viewport(new Viewport(context_, scene_, cameraNode_->GetComponent<Camera>()));
    renderer->SetViewport(0, viewport);
    renderer->SetDynamicInstancing(false); // Added
}

Also I do not pay attention to Avg because it is too unstable, but only on the first column

[eugeneko]

Okay, I just tried it and got totally different numbers, maybe some weird shit in drivers on this PC. I tested OpenGL btw.
Anyway, I don't think this impact matters, as I said. Even 15% in artificial corner case is okay.

[1vanK]

I tested OpenGL btw

The same. API is printed on bottom of Debug Hud

[1vanK]

Mingw x64 release

Old:

New: