nvngx_dlssd 310.9.1 (SL 2.14.1): WRITE_AFTER_WRITE sync hazard on NGX-internal images during first Vulkan evaluate; 310.7.0 is clean

Open
#126 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty
5/5
Estimated time
Over a week
Newbie friendliness
25/100
Issue type
Bug
Clarity
Mostly clear
Activity status
Active
Tech stack
c

Research direction

Reproduce the first Vulkan evaluate using NGX_VULKAN_CREATE_DLSSD_EXT1 and NGX_VULKAN_EVALUATE_DLSSD_EXT with synchronization validation enabled. Compare nvngx_dlssd.dll 310.9.1 against 310.7.0 and inspect the reported nv.ngx.dlssd.resource hazards. Done means the newer DLL produces no WRITE_AFTER_WRITE validation messages.

Written by the indexing model from the issue text.

Description

Environment

  • RTX 5090, driver 610.88, Windows 11, Vulkan 1.4.341
  • Vulkan SDK 1.4.357 validation layer with synchronization validation enabled
  • nvngx_dlssd.dll 310.9.1 from the Streamline 2.14.1 release zip, NGX used directly (NGX_VULKAN_CREATE_DLSSD_EXT1 / NGX_VULKAN_EVALUATE_DLSSD_EXT, not through the interposer)
  • Ray Reconstruction with DLUnified denoise mode, packed roughness, linear depth, IsHDR | MVLowRes, render preset F (the same happens with preset left at default)

What happens
On the first evaluate of a run, the synchronization validation layer reports two hazards, both between commands recorded by NGX on images created and named by NGX:

vkCmdClearColorImage(): WRITE_AFTER_WRITE hazard detected. vkCmdClearColorImage writes to VkImage 0x...[nv.ngx.dlssd.resource], which was previously written during an image layout transition initiated by vkCmdPipelineBarrier[nv.ngx.dlssd.Evaluate].

(two instances, on two different nv.ngx.dlssd.resource images). Later evaluates in the same run report nothing.

Expected
No hazards: the barrier that transitions these internal images presumably needs the transfer-write access/stage of the following clear in its destination scope (or the clear should come before the transition).

Regression
The same application, scene and settings with nvngx_dlssd.dll 310.7.0 (Streamline 2.12.0) report zero validation messages.

Impact
Applications that treat any validation message as a failure (e.g. CI or headless test runs) have to filter these by object name until a fixed DLL ships.

Dominant language
C
Stars
1.4k
Forks
174
PR merge metrics
No merged PRs in 30d

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from NVIDIA-RTX/Streamline

All issues in NVIDIA-RTX/Streamline

Similar issues

More C issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.