temporary profile bypass

[256shadesofgrey]

In v1 it is possible to set framegen to 1x (i.e. no frame gen) to do a quick a/b test of what the game would feel like with and without framegen. In the v2 dev17 build that I just installed that is no longer possible. The minimum number that can be set for the multiplier is 2x. Now the only way to disable a profile is to delete the game .exe entry from the active applications which requires typing in the game .exe name again when I want to reenable it. And also it doesn't work without a game restart.

Hence my request is to either return the ability to go down to 1x for the multiplier, or add a simple enable/disable toggle somewhere so that we can do quick a/b testing without having to remember the game .exe name or any other profile configurations.

[llitz]

@PancakeTAS - unsure if you are interested, but I have this working in my branch, with a few other changes that ended up being required as I was testing with wayland and xwayland apps (vulkan and via zink). Windows resizing also works.
The only edge case I have is when going from a lower multiplier to a higher one (up to 4, I also locked that down) with certain apps - in specific glxgears (zink + xwayland) requires a window resize to trigger the application to use a new swapchain. I have likely done something wrong, but it seems stable and I haven't seem vram leaking

In the end I needed 6 changes in there (from the 8 patches), one was fixing what I assume is a typo and the last is just exposing config values, but it works with the defaults I set in previous patches

https://github.com/llitz/lsfg-vk/tree/patches

[PancakeTAS]

The reason I haven't added this yet, is because most games just freeze when there's a VK_ERROR_OUT_OF_DATE_KHR.. for some reason.

Been meaning to look into it, but seeing how your changes are using precisely the same logic as in lsfg-vk 1.x.y (which has the freezing issue), I don't think I'll use em. Also there's going to be a virtual swapchain, so there'd have to be quite a lot of refactoring.. but I try to write the virtual swapchain in a way where there's no need to recreate the swapchain anyways.

[llitz]

I had that issue before, but the deferral code got rid of the freeze.

There are two mitigations for the freeze:

1 - deferral: this ensures the old swapchain stays in usewhichbacoids the freeze
2 - reservation: if the memory is pre-allocated, we just change our side of the backend and it doesn't require a full swapchain change, allowing to go from 1x to 4x. It does cost memory upfront though, hence why I kept the default at 1

I tried using hint instead before I did the deferral, but it didn't work well in one of the tests. The code still isn't perfect, there's some misdetection on when a swapchain stops being used, but I don't have any leaks or crashes or, freezes (that drove me insane trying to test with guild wars 2).

The one detail I haven't looked into yet, but it also affects the code without the patch, is why gw2 doesn't start in Wayland mode with lsfg-vk, but that seems to be completely unrelated to anything here

P.s.: Forgot to say I went into this route because I also had crashes on the 1.x codebase

[AMA147000]

I stumbled upon that issue today... So until it gets fixed, here is a github gist that solves the issue but requires restarting the game for it to work as it adds .disabled at the end of the vulkan layer filename. Also it only works globally on all profiles (if you want you can enable it run a game then disable it and run the other, but yeah...)

I know I can just build @llitz's version, but I created the script before finding that issue, so if anyone wants that solution here you go I guess