Direct3D 12 to Vulkan project VKD3D-Proton v2.14 out now with various performance improvements

Options

• Implement DXGI body statistics (uncovered by DXVK DXGI).
• Implement a worldwide body price limiter (see VKD3D_FRAME_RATE or DXVK_FRAME_RATE).
  
  Additionally improves conduct of presentation with swap interval > 1 since we use body limiter as an alternative
  
  of duplicated presents now. Additionally permits assist for full-screen body price targets in DXGI which usually would suggest a mode change.
• Implement assist for planar video codecs akin to NV12.
• Implement D24 depth bias accurately now on AMD when VK_EXT_depth_bias_control is supported.
• Expose a brand new command interop interface that enables e.g. dxvk-nvapi to implement DLSS3 body technology.
• Use VK_KHR_compute_shader_derivatives when accessible.
• Use VK_EXT_device_generated_commands when accessible. Expose execute oblique tier 1.1.
• Implement GPU add heap from newest AgilitySDKs. Permits express management over ReBAR as an alternative of heuristic based mostly hacks in video games that use the brand new API.
• Implement ID3DDestructionNotifier. Fixes some explicit video games that count on this to be supported.

Efficiency

• Cut back some VRAM bloat on RDNA2 and three GPUs when VK_MESA_image_alignment_control is uncovered.
• Enhance CPU overhead for video games that question swapchain format assist time and again.
• Take away previous heuristic that most popular 2 frames of latency relying on BufferCount used.
  
  The default on DXGI is 3, and utilizing 2 induced some efficiency points in numerous video games with GPU hunger,
  
  particularly on Deck. VKD3D_SWAPCHAIN_LATENCY_FRAMES remains to be accessible as an override to drive a tighter default.
• Rewrite queue submission logic to deal higher with troublesome submission patterns akin to FSR3 3.1 Body Technology.
  
  On implementations with just one graphics queue, vkd3d-proton will now try and do primary software program scheduling of GPU work.
  
  This may increasingly regress GPU efficiency in another circumstances and VKD3D_CONFIG=no_staggered_submit is a solution to disable this code path.
  
  One notably large enchancment is FF XVI on RADV with FSR 3 frame-gen, with nearly doubled efficiency in some circumstances.
  
  We’re nonetheless awaiting a correct kernel-level repair for this drawback to be totally resolved.
• Rewrite queue submission logic to make use of fewer “dummy” wait/sign submissions.
  
  Works round pathological CPU overhead in amdgpu taking 20ms+ to submit work in some circumstances.
• Rewrite queue submission logic for sparse updates to be extra environment friendly.

Fixes and workarounds

• Rework numerous multi-sampling queries to be extra spec appropriate.
• Workaround bugged MSAA conduct in World of Warcraft.
• Workaround buggy/questionable use of ID3D12PipelineLibrary in FF XVI.
• All the time use native 16-bit integers for min16int. Fixes some real-world bugs the place shaders count on min16int is at all times applied as 16-bit.
• Workaround sport bug resulting in GPU cling in Dragon Age: Veilguard on RADV.
• All the time emit correct floating-point atmosphere modes in DXBC shaders. Fixes glitched eyes in Dragon Age: Veilguard on NV.
• Repair potential use-after-free bug for some sparse useful resource replace circumstances.
• Appropriately validate when software makes an attempt to allocate a too giant descriptor heap.
  
  Fixes Stalker 2 getting into into undefined conduct.
• A variety of misc fixes in dxil-spirv as ordinary.
• Workaround damaged amdgpu zerovram conduct on 6.10+ kernels. Fixes random excessive glitchiness in Helldivers 2 on AMD.
• Workaround NV problem which result in GPU cling when loading a save file in Star Wars: Outlaws.
• Repair copying between BC <-> RGBA photographs in some circumstances.
• Add workaround for a sport bug in The First Descendant which result in damaged cubemap reflections in some circumstances.
• Workaround Cranium & Bones crashing on startup on NV GPUs by disabling Reflex assist.
• Workaround Hunt: Showdown lacking exact qualifiers on vertex shaders, resulting in glitched rendering.
• Workaround poor CPU efficiency in Crimson Lifeless Redemption.

Misc / Debug

• Add assist for instruction_qa_checks. For deep debug, permits us to be notified when NaNs and Infs are generated in shaders.
  
  For inside QA use.
• Add fine-grained management of QA conduct on a per-shader foundation. For narrowing down points.
• Take away a bunch of previous and out of date workarounds for NV drivers. New cutoff is 535 sequence.
• Bump uncovered SDKVersion to 614 to match newest steady AgilitySDK.
• Add an non-obligatory code path to assist DXBC through the official dxilconv library.
  
  This code is just not enabled in launch builds,
  
  and is presently solely meant as a path to make the most of QA instrumentation for DXBC shaders.