Digger1955/dxvk-gplasync-lowlatency DXVK-GPLALL-2.6.6-1

DXVK-GPLAsync-LowLatency 2.6.6-1 (DXVK-GPLALL 2.6.6-1)

on GitHub

DXVK-GPLAsync-LowLatency 2.6.6-1 (DXVK-GPLALL 2.6.6-1)

Based on DXVK 2.6.2 and commits from DXVK 2.7, DXVK 2.7.1 and later, DXVK GPLAsync 2.6.2 and DXVK GPLAsync 2.7-patch, DXVK Low Latency 2.7.1-commit 9659672.

It consists of:

• DXVK-GPLALL-GCC-WinMacLinux-SSE2-O3-LTO 2.6.6-1
• DXVK-GPLALL-GCC-WinMacLinux-SSE4.2-O3-LTO-GENERIC 2.6.6-1
• DXVK-GPLALL-GCC-WinMacLinux-SSE4.2-O3-LTO-INTEL 2.6.6-1
• DXVK-GPLALL-MSVC-Windows-SSE2-O1_Oi_Ob3-LTCG 2.6.6-1
• DXVK-GPLALL-MSVC-Windows-AVX2-O1_Oi_Ob3-LTCG-AMD64 2.6.6-1
• DXVK-GPLALL-MSVC-Windows-SSE4.2-O1_Oi_Ob3-LTCG-INTEL64 2.6.6-1
• DXVK Native-GPLALL-GCC-SSE2-O3-LTO 2.6.6-1
• DXVK Native-GPLALL-GCC-SSE4.2-O3-LTO-GENERIC 2.6.6-1
• DXVK Native-GPLALL-GCC-SSE4.2-O3-LTO-INTEL 2.6.6-1
• Source Code
• dxvk.conf

Release Notes:

IMPORTANT: DXVK-GPLALL 2.6.x version will be maintained for GPUs/drivers that do not have VK_KHR_maintenance5 and/or maxPushConstantsSize=256. Commits from DXVK, DXVK GPLAsync, DXVK Low Latency will be ported to DXVK-GPLALL 2.6.x, if possible. Also, DXVK-GPLALL 2.6.x will be recompiled from time to time, when new compiler versions with significant improvements are out.

Fourth maintenance release of DXVK-GPLALL 2.6.x branch.

IMPORTANT: Please refer to DXVK-GPLALL Wiki for information about changes to the recommended configuration of Low Latency Frame Pacing Options.

• Significant changes:
  • DXVK: [dxvk] Throttle resource eviction
    • Improves performance under extreme memory pressure.
  • DXVK Low Latency: Use calibrated device timestamps for each submit
    • Significant precision and smoothness improvements for GPU Drivers that support VK_EXT_calibrated_timestamps. Note: VK_EXT_calibrated_timestamps was used instead of VK_KHR_calibrated_timestamps due to larger coverage of devices and drivers.
    • IMPORTANT: timestampValidBits check can't be implemented, due to DXVK 2.6.x codebase. In order to check, whether GPU has timestampValidBits = 64, use Vulkan GPUInfo website, section "Queue Families". Example - NVIDIA GeForce GT730M.
    • IMPORTANT: On devices that do not have timestampValidBits = 64 (Adreno/Mail/PowerVR etc. mobile GPUs, Intel iGPUs, Intel ARC A GPU Series, possibly more) DXVK-GPLALL will crash applications randomly, if dxvk.framePace is set to dxvk.framePace = "low-latency" (default value) or dxvk.framePace = "low-latency-vrr-x". In order to use DXVK-GPLALL 2.6.6-1 on such devices set dxvk.framePace = "max-frame-latency" or dxvk.framePace = "min-latency".
  • DXVK Low Latency: Minor optimizations
    • CPU-bound performance improvements.
  • DXVK Low Latency: Threaded sleep
    • Significant smoothness improvements, minor performance and latency improvement.
• DXVK-GPLALL changes:
  • GCC: Added -ffunction-sections, -fdata-sections, -Wl,--gc-sections flags.
    • Has the same effect as /Gw, /GF, /Gy, /OPT:REF,ICF for MSVC - reduces DLLs size and potentially improves performance.
  • GCC-SSE4.2-O3-LTO-INTEL: Added -mbmi, -mbmi2, -mmovbe flags.
    • Allows compiler usage of BMI, BMI2 and MOVBE instructions, which improve CPU-bound performance. These instructions are non-AVX part of x86-64-v3 set and do not change DXVK-GPLALL requirements specified in Builds Reference Guide.
      • Note: MSVC-AVX2-O1_Oi_Ob3-LTCG-AMD64 build uses all of these instructions as a part of /arch:AVX2. MSVC-SSE4.2-O1_Oi_Ob3-LTCG-INTEL64 build does not use these instructions, due to /arch:SSE4.2, and there is no way to specify usage of separate instruction in MSVC compiler.
  • GCC-SSE4.2-O3-LTO-GENERIC: Added -mbmi, -mmovbe flags.
    • Allows compiler usage of BMI and MOVBE instructions, which improve CPU-bound performance. These instructions are non-AVX part of x86-64-v3 set and do not change DXVK-GPLALL requirements specified in Builds Reference Guide.
      • Note: -mbmi2 flag is not used, in order to avoid performance regression on AMD CPUs up to and including Zen2, due to usage of BMI2 PDEP/PEXT instructions.
  • MSVC: Changed optimization level from /O2 to /O1 with /Oi.
    • CPU-bound performance improvements. For DXVK-type workloads (DirectX-to-Vulkan translation layer) on out-of-order CPUs (all x86-64 CPUs) fitting a code segment into the instruction cache is preffered to fastest code. Older CPUs with slower and smaller cache will see the most significant performance improvements.
      • Note: For GCC, change from -O3 to -Os leads to compilation failure, probably due to alignment issues.
  • GCC: Added -Wno-deprecated flag to silence not very useful compilation warnings about -mtune=x86-64 usage.
  • MSVC: Set /Qpar-report:1 and /Qvec-report:1 flags to decrease compilation logs size.
  • MSVC and MSVC-AVX2-O1_Oi_Ob3-LTCG-AMD64: Additions of default MSVC flags, in order to have an explicit way to change MSVC defaults - no functional changes.
  • Clang: Added -fwhole-program-vtables compiler flag.
    • Requires FullLTO, enables single-implementation devirtualization and virtual constant propagation, improves CPU-bound performance.
  • GCC/MSVC/Clang: Added explanatory comments to compiler optimization flags.

Next maintenance releases will have less changes, due to more significant incompatibilities between DXVK 2.6.x and next DXVK versions.

List of ported and adapted upstream DXVK commits:

584041f - [util] Add a frame cap for The Evil Within 2

c774d93 - [include] Fix Windows typedefs in native headers

5c3437c - [d3d11] Fix surface map flags check

a26a504 - [d3d11] Add D3D11Device::RegisterDeviceRemovedEvent() stub

b776f63 - [dxvk] Fix VVL error about vkGetPhysicalDeviceMultisamplePropertiesEXT

c10c75e - [d3d8] Readjust ZBIAS factor

aa7f779 - Support for MW:R HMW-Mod

08925ba - Add comment for AMD AGS crash in HMW-Mod

77d57f2 - [util] Add infinite warfare / iw7-mod support

8710c80 - [dxgi,dxvk] Add support for P010/P016 textures

a5d2ded - [d3d9] Allow cursor clears on hidden HW cursors

9d5dd98 - [d3d9] Add a tiny amount of padding to texture mapping data - [d3d9] Add a tiny amount of padding to texture mapping data

252bdd8 - [util] Remove BlazBlue textureMemory workaround - [d3d9] Add a tiny amount of padding to texture mapping data

c27ef1c - [util] Spoof GPU for Sims 3

fe277ae - Correct a few native API data types

eab5145 - [d3d11] Construct D3D11DXGISurface with only ID3D11Resource. - [d3d11] Implement subresource surface support.

0bf876e - [d3d11] Implement subresource surface support. - [d3d11] Implement subresource surface support.

4bbe487 - [util] Fix crash on resolution change in Pirates

c8d5d8e - [dxgi] Do not update target frame rate when PRESENT_TEST is set

aadc4b4 - meson: Fix glfw dependency request to use correct pkgconfig name

9f97d4c - native: Add define for DECLARE_INTERFACE_IID

31c5433 - [dxvk] Throttle resource eviction

d222d2c - [util] Disable direct image mapping for Total War Pharaoh Dynasties

b02cac8 - [meta] FIFA regex adjustment

4357d63 - [util] Enable zeroMappedMemory for Dawn of War DE

fae3c43 - [meta] Expand the W40K DoW DE regex

6f3a758 - [util] Enable zeroMappedMemory for Nioh 2

1ebe643 - [CI] Run steamrt container with --user 1001

1676dca - [dxvk] Fix build failure with Clang 22

List of ported and adapted DXVK Low Latency commits:

9b15fdc - [dxvk] Switch to implicit VRR frame pacing - Switch to implicit VRR pacing

• Note: No functional changes compared to 691cab5

075d652 - [dxvk] Drop the V-Sync requirement for VRR pacing - Switch to implicit VRR pacing

• Note: No functional changes compared to b7e76e6

1a928cd - [dxvk] Disable V-Sync buffer tracking

cf650d7 - [dxvk] Add VK_KHR_calibrated_timestamps - Use calibrated device timestamps for each submit

• Note: VK_EXT_calibrated_timestamps was used instead of VK_KHR_calibrated_timestamps due to larger coverage of devices and drivers.

ae4db00 - [dxvk] Improve GPU runtime estimation - Use calibrated device timestamps for each submit

177919e - [util] Add ringbuffer allocator - Use calibrated device timestamps for each submit

b10ff87 - [dxvk] Add query pools to the frame pacer - Use calibrated device timestamps for each submit

28a142e - [dxvk] Write device timestamp at the end of each submit - Use calibrated device timestamps for each submit

76592cf - [dxvk] Use device timestamps for frame pacing - Use calibrated device timestamps for each submit

96f0acf - [dxvk] Check VK_KHR_calibrated_timestamps support for frame pacing - Use calibrated device timestamps for each submit

• Note: VK_EXT_calibrated_timestamps was used instead of VK_KHR_calibrated_timestamps due to larger coverage of devices and drivers.
• IMPORTANT: timestampValidBits check can't be implemented, due to DXVK 2.6.x codebase. In order to check, whether GPU has timestampValidBits = 64, use Vulkan GPUInfo website, section "Queue Families". Example - NVIDIA GeForce GT730M
• IMPORTANT: On devices that do not have timestampValidBits = 64 (Adreno/Mail/PowerVR etc. mobile GPUs, Intel iGPUs, Intel ARC A GPU Series, possibly more) DXVK-GPLALL will crash applications randomly, if dxvk.framePace is set to dxvk.framePace = "low-latency" (default value) or dxvk.framePace = "low-latency-vrr-x". In order to use DXVK-GPLALL 2.6.6-1 on such devices set dxvk.framePace = "max-frame-latency" or dxvk.framePace = "min-latency".

55558cb - [dxvk] Fix GPU runtime in GpuProgress

330afbd - [dxvk] Improve robustness to stutters

40e778a - [dxvk] Enable calibrated timestamps only for low-latency modes - Minor optimizations

• IMPORTANT: This commit makes possible for DXVK-GPLALL to work on devices that do not have timestampValidBits = 64 (Adreno/Mail/PowerVR etc. mobile GPUs, Intel iGPUs, Intel ARC A GPU Series, possibly more) in dxvk.framePace = "max-frame-latency" or dxvk.framePace = "min-latency" modes.

159ff57 - [dxvk] Fix ringbuffer allocator bug - Minor optimizations

eeb9e83 - [dxvk] Reduce memory footprint of latency markers - Minor optimizations

e554b1b - [dxvk] Optimize DxvkSharedAllocationCache::freeAllocation - Minor optimizations

c42092a - [dxvk] Optimize DxvkCsChunkPool - Minor optimizations

b81475c - [dxvk] Optimize sync::Fence - Minor optimizations

96daf3c - [util] Add sync::AtomicSignal based on WaitOnAddress / futex - Threaded sleep

5311d42 - [dxvk] Add ThreadedSleep - Threaded sleep

6ccd7b4 - [dxvk] Add frame duration measurement for app-thread - Jitter display

9b85b1e - [dxvk] Add jitter tracking - Jitter display

07d5b3e - [hud] Change initialization location of latency hud items - Jitter display

9659672 - [hud] Add jitter hud item - Jitter display