[Dd7to9] SCP: Containment Breach Multiplayer - FPS dives off a cliff

[LokeYourC3PH]

Hello dear dxwrapper devs,

I am having a tiny issue with dxwrapper: It kills the FPS in SCP: Containment Breach Multiplayer. Why? I am not sure, judge for yourselves. We wanted to see if we could get the game to run on DirectX9, get some major improvements for older systems and Laptops that have iGPUs (since DirectX7 handles this badly), and generally be able to perhaps get more out of the game by using dxwrapper. The problem is that, as already mentioned, the exact opposite is happening.

No dxwrapper at all, native DirectX7 (FPS locked to 120, but can achieve about 300+ when unlocked):

Using dxwrapper:
With more config options aside from just "Dd7to9 = 1":

Without any other options, only "Dd7to9 = 1":

As you can see, dxwrapper makes the framerate dive straight off a cliff, basically quartering it in many cases. Which options I enable, apparently don't matter, since all the ones I thought could make a difference, make none at all, and in some cases even cause Memory Access Violations (crashing the game).

Log file:
dxwrapper-game.log

Is there any possible solution to this? dgVooDoo causes this to happen as well, but it does increase FPS to some extent on systems with iGPUs or really weak systems. On mine with an RTX 4070 Super, both seem to just nuke my framerate.

PS:
Just for giggles, I also tried chaining it with DXVK, which causes the FPS to drop down to 20, almost becoming a slideshow.

[elishacloud]

This particular game performs Blts using a color key directly to the backbuffer in video memory. In dxwrapper (and probably other wrappers), that requires copying the backbuffer to system memory, performing the Blt, and then copying it back to video memory.

dxwrapper does cache the surface in system memory and can perform multiple Blts before copying the data back, which helps somewhat and is likely why it is faster than DXVK in certain cases.

The real solution would be to Blt the data onto an alpha texture first, and then draw that texture to the backbuffer. This approach would eliminate the need to copy the backbuffer to system memory and back, which is the main source of the slowdown.

[LokeYourC3PH]

This particular game performs Blts using a color key directly to the backbuffer in video memory. In dxwrapper (and probably other wrappers), that requires copying the backbuffer to system memory, performing the Blt, and then copying it back to video memory.

dxwrapper does cache the surface in system memory and can perform multiple Blts before copying the data back, which helps somewhat and is likely why it is faster than DXVK in certain cases.

The real solution would be to Blt the data onto an alpha texture first, and then draw that texture to the backbuffer. This approach would eliminate the need to copy the backbuffer to system memory and back, which is the main source of the slowdown.

I assume that's not something I can do anything with or about, so... What now?

[elishacloud]

I assume that's not something I can do anything with or about, so... What now?

Correct. I need to build this feature. Once I build it then there will be nothing for you to do.