[d3d9] Fix issues with presentation

[K0bin]

• Do GDI blit for SWAPEFFECT_COPY_VSYNC too (which is missing from the mingw headers on my system)
• Dont swap buffers when using SWAPEFFECT_COPY

- Blit the backbuffer contents into the frontbuffer, so GetFrontBufferData returns the expected data
- Replace noExplicitFrontBuffer with that frontbuffer blit (this needs testing)

• Dont create an additional front buffer for SWAPEFFECT_COPY and windowed SWAPEFFECT_DISCARD with 1 backbuffer. The latter behaves identical to SWAPEFFECT_COPY in my testing, it even supports partial presentation.
• In those cases, GetFrontBufferData will just return the contents of the backbuffer. That's technically wrong ofc but I don't think adding in another blit in windowed mode is worth it. It's only wrong for SWAPEFFECT_COPY and DISCARD in windowed mode. Fullscreen SWAPEFFECT_COPY is probably very rare because that kind of defeats the purpose as partial presentation doesn't work in fullscreen. GetFrontBufferData was broken in windowed mode anyway because Windows takes a screenshot of the entire screen when you call it and we just used to return the contents of the front buffer.

Fixes #2732
Fixes #2240
Fixes #2568
Fixes #3238
Fixes #1476
Fixes #1481
Fixes #2456
Fixes #2542
Fixes #3005
Fixes #3208
Fixes #3238
Fixes #3206
Fixes #3141
Fixes #3250
Fixes #1554
Fixes #2756
Fixes #2915

Closes #2749

[pchome]

imho using const swapEffectCopy everywhere would be cleaner a bit

const bool swapEffectCopy = m_presentParams.SwapEffect == D3DSWAPEFFECT_COPY || m_presentParams.SwapEffect == D3DSWAPEFFECT_COPY_VSYNC;

[K0bin]

I wrote a test application to figure out how this is supposed to behave:

WINDOWED:
FLIP + 5 buffers => has additional front buffer
FLIP + 1 buffers => has additional front buffer
DISCARD + 5 buffers => has additional front buffer, contents of the backbuffers stay valid
DISCARD + 1 buffers => does not have additional front buffer, contents of the backbuffers stay valid
FLIPEX + 5 buffers => has additional front buffer
FLIPEX + 1 buffers => has additional front buffer
COPY + 1 buffers => does not have additional front buffer

DISCARD always preserves the contents of the back buffer.

Partial Present works for COPY and DISCARD + 1 buffer.
GetFrontBufferData takes a screenshot of the entire monitor in windowed mode.

FULLSCREEN
FLIP + 5 buffers => has additional front buffer
FLIP + 1 buffers => has additional front buffer
DISCARD + 5 buffers => has additional front buffer, contents of the backbuffers stay valid
DISCARD + 1 buffers => has additional front buffer, contents of the backbuffers stay valid
FLIPEX + 5 buffers => has additional front buffer
FLIPEX + 1 buffers => has additional front buffer
COPY + 1 buffers => does not have additional front buffer

DISCARD always preserves the contents of the back buffer.

Partial Present does not work in fullscreen mode.
GetFrontBufferData works as expected.

[w-flo]

Replace noExplicitFrontBuffer with that frontbuffer blit (this needs testing)

FWIW, I tested Zusi 3 with your latest patches applied on top of master, and it seems to work without issues. 🐸 Zusi 3 needed the noExplicitFrontBuffer config because it uses DISCARD swapeffect & 1 backbuffer. Thanks for working to get rid of that noExplicitFrontBuffer hack!

Edit: Still works on latest patches without the frontbuffer blit (b5f03a8)

[arodland]

Tested FlexRadio SmartSDR, which is a WPF app that has absolutely horrible performance with regular wine and broken (mostly black) display with dxvk 2.1 and master.

This build makes it run just like it does on wined3d — visually correct, but near-unusably slow. I guess it was too much to hope for a magic fix :)

[Blisto91]

Yes this implementation is expected to be slow

[Blisto91]

First round of testing based on most of the maybe Partial Present tracked issues

#2732 Fixed
#2240 Fixed
#2568 (Part of issue) Fixed
#2278 Not fixed
#1476 Fixed
#2334 Untested (couldn't get it running)
#1481 Fixed
#2456 Fixed
#2542 Seems fixed or at least improved. dotnet install and (x)Wayland can cause issues
#2860 Untested
#3005 Fixed based on apitrace
#3116 Not fixed
#3208 Fixed

[Blisto91]

Testing of front buffer issues and related
#3238 Fixed
#3206 Fixed based on apitrace
#3141 Fixed
#3250 Fixed. Thanks @CME42
#1554 Fixed based on apitrace

Honorable mentions
#2756 Fixed
#2915 Fixed

Edit: added a extra honorable mention

[misyltoad]

@K0bin Do you want to add all of these issues to your commit with a Closes: so we can have a super statisfying clean sweep? :-)

[misyltoad]

Minor nits.

[w-flo]

The very first bug report about the front buffer stuff might have been #1368, which this PR fixes properly instead of special-casing the workaround, so if you want to add it to the super long Closes: line 🥇 …