Without mods, Nier Replicant runs faster when going above 60 FPS.
The game had an official patch that implemented a framerate limiter to prevent this. This limiter is terrible, because it's not a stable 60 FPS, but a weird 57-58 FPS. There's no way to disable this in-game, it has to be done by editing a config file or through a mod.
So, why remove the default frame limiter from DXVK?
- In the default case (a user plays the game as it is), it does nothing, since 57-58 is lower than 60.
- If a user is going out of their way to edit the config file, why would they assume that DXVK already provides a frame limiter? They are going to follow [a guide](https://www.pcgamingwiki.com/wiki/NieR_Replicant#Framerate_limited_to_57.7E58_FPS) that says to set up a frame limiter, it doesn't say "set up a frame limiter, unless you are using DXVK, in that case it's already there".
- They are using [Special K in order to use a mod to play at high refresh rates at normal speed](https://wiki.special-k.info/SpecialK/Custom/Replicant). In this case, DXVK's default limiter is harmful, since it is not documented that it's there by default.
Since this default limiter is useless in the first two cases and harmful in the third, I think it should be removed.
The alternative would be to document this (e.g. in PCGamingWiki), but the instructions wouldn't look pretty... "After following the instructions to use Special K in order to play at higher framerates at normal speed, if are using DXVK/Proton, also do these things to disable its default 60 FPS cap: [...]"
Especially because that the game isn't broken in the default case, I don't think DXVK should tamper with these things in a way that requires documentation to revert.
Tested Special K's mod to play at higher refresh rates on Linux.
Drastically limits the amount of descriptor memory we allocate in situations
where an application renders without presenting anything to a swap chain.
The new limit is a bit tight for some real-world use cases (e.g. Ashes of the Singularity),
but at worst we will start calling vkAllocateDescriptorSets once per set and draw.
This is not a legal optimization inside non-uniform control flow due
to Vulkan's extremely permissive convergence rules, and apparently
breaks on Nvidia as a result.
Mesa drivers already do the same thing internally anyway.
When targeting ARM64EC, both __x86_64__ and _M_X86_64 are defined but
not all x86 intrinsics are present, treat EC as regular ARM64 so the
native intrinsics are used instead.
Dirt 5 does not crash without working ags anymore and Ethan Carter Redux also starts fine without a spoof.
This allows the built in AMD ags in Proton 9 to be used for these games.
The Hitman 3 config is redundant as it doesn't allow RT to be enabled without Nvapi anyway.
The uses deferred contexts for rendering if driver command lists are enabled,
but when AMDAGS is loaded, it will also unconditionally use MultiDrawIndirect
functions. Since the AGS version in use does not support deferred contexts,
this breaks rendering, so we will have to force it into the immediate context
path.
Testing also shows slightly higher performance (~3-5%) with this path in
CPU-bound scenarios.
Non-explicit conversion operators in general can participate in very
surprising conversion chains. Explicit bool operator is a good place to
start with, because even with explicit they do get automatic contextual
conversion in a lot of places, e.g., if conditions.
When assigning to a BOOL (which is an uint in disguise) and using explicit
bool conversion operators (introduced in a latter commit) an explicit cast
is required.
* The Settlers submits (possibly incorrectly) an SRV to ClearUnorderedAccessViewUint. The static_cast in the function does not translate correctly and crashes.
Native D3D11 behavior is to ignore the bad parameter entirely. It does not clear the SRV nor does it fault or even error with the DEBUG validator.
When a fence has been missed, we can avoid locking *most* of the time
via the double-checked locking pattern. We still lock before a second
check in case the scheduler caused us to miss the fence. If the
scheduler did cause us to miss the fence, we can drop the lock prior to
executing the callback function, as a second micro-optimization.
Signed-off-by: Richard Yao <richard.yao@alumni.stonybrook.edu>
Game engine physics/speed for Sonic CD is tied to frame rate so limit max frame rate to 60 fps. Otherwise, the game runs too quickly for high refresh rate monitors.
HDR in Control (a patch released by a developer post-launch, not
actually in the game sadly) tries to set a video mode with
DXGI_FORMAT_R16G16B16A16_FLOAT.
This seemingly works on Windows, and based on FindClosestMode etc
documentaton, this seems required to work for any format that scanout
it supported for.
It's really not like the bpp is meaningful on Windows with the
distinction of 8bit and 10bit not working in GDI modes at all.
Nor does it end up actually setting anything on Linux/Deck where
modesets are emulated.
So, treat DXGI_FORMAT_R16G16B16A16_FLOAT as 32bpp so the
FindClosestMatchingMode and EnterFullscreenMode calls succeed.
