Skip to content

Correct GPU timing readback and transmitted-shadow reconstruction - #156

Draft
proggeramlug wants to merge 2 commits into
codex/windows-engine-planfrom
codex/windows-profiler-integrity
Draft

Correct GPU timing readback and transmitted-shadow reconstruction#156
proggeramlug wants to merge 2 commits into
codex/windows-engine-planfrom
codex/windows-profiler-integrity

Conversation

@proggeramlug

@proggeramlug proggeramlug commented Sep 10, 2026

Copy link
Copy Markdown
Contributor

GPU profiling on Radeon/Vulkan returned the preceding frame's timestamps, and transmitted shadows reconstructed receiver positions with an incorrectly laid-out inverse camera matrix. The profiler now copies resolved queries in a following submission and rejects incomplete measurements; the shadow pass now uploads the same transposed inverse used by SSR/PT, placing the colored shadow beneath its caster.

The shadow regression isolates direct lighting from asynchronous GI and automatically retains failed captures. Its original camera, light, material, frame counts, and assertions are unchanged. The isolated old implementation fails on Radeon; the correction produces 3,217 affected pixels, and a swapped red/green attenuation control is rejected.

Validation: the profiler implementation passes all 487 shared unit tests, 90 golden tests and remaining enabled local runtime suites, contracts, lint, and the quality lane (25 quality-runner tests). Its current-frame regression rejects all 12 old Vulkan samples and passes on Radeon Vulkan and DX12. Hosted Metal reports the test as ok, but a timestamp-capability early return is possible, so Metal GPU timestamp validation remains unproven. The shadow follow-up at fa93690 passes the complete local shared suite and all 23 hosted checks, including the colored-shadow regression on Metal. The two scheduled physical-hardware jobs remain skipped. Existing ignores and external-scene omissions are explicit in the reports.

Both SSGI shader variants were remeasured with the corrected profiler in 20 isolated runs, each with 120 complete GPU frames. The stationary fix adds a median 0.008598 ms at 256 square and 0.043532 ms at 1280x720 for the five probe passes. Earlier timing observations are superseded; image baselines and budgets are unchanged.

Stacked on draft #155. Related: #128, #135, #149, #153. The two Windows portable-image discrepancies, HD temporal stability, representative-scene and other hardware acceptance remain open.

Profiler report | Shadow correction | Raw evidence

@coderabbitai

coderabbitai Bot commented Sep 10, 2026

Copy link
Copy Markdown

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

  • Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

@proggeramlug proggeramlug changed the title Repair current-frame GPU profiling and reject incomplete timing evidence Correct GPU timing readback and transmitted-shadow reconstruction Sep 10, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant