Color bleed with Reshade when fullscreen

[ovitus]

On the latest nightly build of FBNeo (v1.0.0.03), this happens when I go fullscreen using Reshade with CRT-Guest-Advanced. I've seen it with other emulators also, as described here. It doesn't happen with the version of FBNeo for Fightcade (v0.2.96.44-55). I was trying the latest version since I don't play online often and wanted to overclock to fix the slowdown inherent in some Neo Geo games.

[barbudreadmon]

Which blitter is this ? Does it happen with all blitters ?
I sort of remember reading that fc uses a different blitter by default.

[ovitus]

Reshade only works with DirectX 9 and up (along with OpenGL and Vulkan). It happens with both Alt and Experimental

[dinkc64]

My pc can't even run reshade, bummer

[barbudreadmon]

FBNeo/src/intf/video/win32/vid_directx9.cpp

Lines 1266 to 1279 in 753ed64

	// Clear the swapchain's buffers
	if (nVidFullscreen) {
	for (int i = 0; i < 3; i++) {
	pD3DDevice->Clear(0, NULL, D3DCLEAR_TARGET, D3DCOLOR_XRGB(0, 0, 0), 1.0f, 0);
	pD3DDevice->Present(NULL, NULL, NULL, NULL);
	}
	} else {
	RECT rect;
	GetClientScreenRect(hVidWnd, &rect);
	rect.top += nMenuHeight; rect.bottom += nMenuHeight;
	pD3DDevice->Clear(0, NULL, D3DCLEAR_TARGET, D3DCOLOR_XRGB(0, 0, 0), 1.0f, 0);
	pD3DDevice->Present(&rect, &rect, NULL, NULL);
	}

That code seems to clear the framebuffer, but i think it's only called at init and never between frames ? That might be the problem ?

[barbudreadmon]

I can confirm fc is making that Clear call in their 2 dx9 rendering functions :

https://github.com/fightcadeorg/fightcade-fbneo/blob/b986cb7f2a4910aeee9d3da00ed69474d3e1e3c4/src/intf/video/win32/vid_directx9.cpp#L1247

https://github.com/fightcadeorg/fightcade-fbneo/blob/b986cb7f2a4910aeee9d3da00ed69474d3e1e3c4/src/intf/video/win32/vid_directx9.cpp#L2063

So this missing call is extremely likely to be the culprit, i'll let the real experts implement a proper fix though since i'm not a windows user and can't test those fixes.

[barbudreadmon]

I managed to test ee91d5a under linux+wine and didn't notice any regression, so i'm gonna assume it's safe enough to be committed as a test fix.

[ovitus]

I tested that with Windows 11 and it's still occurring

[barbudreadmon]

Hmmm, maybe the full code block from #2088 (comment) is required instead of the one-liners i added in ee91d5a ?

[barbudreadmon]

I reverted my previous commit, someone with a proper testing environment for this will have to look into it.

[ovitus]

Thanks, let me know if I can help at all with testing

[dinkc64]

if someone wants to shoot me a lowprofile video card for my sff (small form factor) pc, I'd do it