VKD3D-Proton 3.0.1 brings many Linux gaming enhancements for Direct3D 12 via Vulkan

That is seemingly the final launch earlier than VK_EXT_descriptor_heap lands.
There are some sensible the explanation why that is 3.0.1 as a substitute of three.1,
however numbers do not actually matter that a lot anyway.

A ton of labor on descriptor heaps have been taking place within the background as effectively,
however that’s not included on this launch.

Options

• D3D12 view instancing is now experimentally supported.
  It’s enabled for the one recognized sport that requires it, Crimson Desert (1.04+).
• Implement VK_EXT_present_timing, permitting for clean body pacing for SyncInterval > 1 when supported.
• Assist Impartial Units characteristic.
• Add assist for brand spanking new AGS WMMA ops required for FSR4 Ray Reconstruction and Denoiser, as utilized in Crimson Desert.
• Expose new interfaces for extra up-to-date NVAPI and expose some assist for NVAPI shader intrinsics.
  With up-to-date dxvk-nvapi, the seller extensions for Shader Execution Reordering ought to work for instance.
• Expose new interfaces for AMD AGS and expose some assist for AGS shader intrinsics, as utilized in Crimson Desert.

Fixes

• Permit SV_PrimitiveID to be learn in additional shader phases as anticipated.
• Varied fixes for NVIDIA Reflex to enhance efficiency in some circumstances. [NVIDIA contrib]

  • Moderately than making an attempt to lock NVAPI current ID to our current ID, these can now be decoupled.
    This helps frame-gen particularly, since we will now sign this example correctly.

• Varied fixes to make Turnip go vkd3d-proton check suite.
• Varied fixes in dxil-spirv to make new video games not crash in shader compiler.
• Repair regression in F1 2019/2020 the place previous workaround in previous compiler was not retained in new code.
• Misc different bug fixes.

Efficiency

A lot of the efficiency work for this launch revolved round giving some like to cell chips which rely
on tiling for optimum efficiency:

• Implement deferred clears and discards to doubtlessly enhance efficiency on e.g. Turnip.
• Reap the benefits of render go suspend-resume on cell GPUs, which may doubtlessly enhance efficiency quite a bit.
  NOTE: This depends upon driver and sport to “do the precise factor”. It doesn’t magically make issues go sooner.
• Transfer question pool initialization and resolves round to keep away from breaking render passes as a lot as potential.
• Rework how MSAA resolves work to make it simpler for drivers to optimize.

Different efficiency enhancements:

• Implement a batched system for complicated ExecuteIndirect the place present cut up command record optimization didn’t work.
  Improves GPU sure efficiency in varied video games that spam ExecuteIndirect with state updates like Crimson Desert, Starfield and Halo Infinite.
• Implement SM 6.4 dot2add with VK_VALVE_shader_mixed_float_dot_product if supported.
• Lastly use a correct Vulkan TRANSFER queue for D3D12 COPY queues if supported as a substitute of Vulkan COMPUTE queue.
  RADV doesn’t expose this by default but, however a TRANSFER queue is now used on NVIDIA by default,
  which ought to enhance efficiency when streaming belongings.

Workarounds

• Introduce a mechanism to use workarounds based mostly on shader entry level identify as a substitute of hash.
  This isn’t at all times accessible, however ought to keep away from churn in e.g. Wuthering Waves and Crimson Desert
  that are shifting targets w.r.t. shader hashes.
• Workaround DispatchRays the place sport forgets to set root signature accurately.
• Workaround misc sport bugs in:

  • Shadows of the Tomb Raider
  • Rise of the Tomb Raider
  • Spider Man 2
  • Loss of life Stranding 2
  • Guardians of the Galaxy
  • REANIMAL
  • Crimson Desert
  • … and others I most likely missed to say

• Workaround {hardware} bug on RDNA4 with SV_ShadingRate on affected Mesa variations.

  • Fastened in 26.1+. Workaround is disabled on these variations.

• Take away NULL SMEM PRT HW workaround on AMD in future Mesa 26.2+.

  • Resolves a 1.5 yr previous nightmare in Monster Hunter Wilds.

• Add efficiency workaround for NVIDIA kernel when ready for ID3D12Fences.
• Permit RDNA1 GPUs to faux that they assist barycentrics and VRS. Permits Crimson Desert to run on RDNA1.
• Take away international submission lock workaround on NVIDIA.

Misc

Take away legacy paths for VK_NV_device_generated_commands{,_compute} and compute fallback for NV_dgcc.
No related driver is impacted by this cleanup.