Fix PerfQuery inconsistencies across backends.

[AdmiralCurtiss]

This was discovered by someone on Twitter:
https://twitter.com/sup39x1207/status/1632813805213454337
I was promised a PR on Discord a few days ago but it never showed up, so here I am with it instead so it doesn't get forgotten. @TellowKrinkle

We're still not accurate to hardware but let's at least fix this obvious error...

dolphin/Source/Core/VideoBackends/OGL/OGLPerfQuery.cpp

Line 83 in a6b1b2f

return result;

dolphin/Source/Core/VideoBackends/Vulkan/VKPerfQuery.cpp

Line 125 in a6b1b2f

return result / 4;

dolphin/Source/Core/VideoBackends/D3D/D3DPerfQuery.cpp

Line 99 in a6b1b2f

return result;

dolphin/Source/Core/VideoBackends/D3D12/D3D12PerfQuery.cpp

Line 132 in a6b1b2f

return result / 4;

[Dentomologist]

Once these changes are added GetQueryResult will have the same implementation in all the real backends, and it doesn't seem to me that there's likely to be a reason for them to diverge in the future.

Given that, is there a reason we shouldn't just use that implementation in PerfQueryBase to prevent future problems of this type, and let the couple specialized backends that need a different implementation keep their overrides?

[TellowKrinkle]

One more mismatch

dolphin/Source/Core/VideoBackends/OGL/OGLPerfQuery.cpp

Lines 171 to 172 in 019bde6

	if (g_ActiveConfig.iMultisamples > 1)
	result /= g_ActiveConfig.iMultisamples;

Is only in the OGL and Metal renderers
I would think this is the correct thing to do (as in, perf query results are per sample and not per shader invocation), but it would be good to test MSAA and verify

Edit: Test done, OpenGL is correct, we need to divide by the number of samples: https://twitter.com/sup39x1207/status/1635109919606337536

[AdmiralCurtiss]

Like that?

[TellowKrinkle]

Like that?

Yep