Repository navigation
Preserve shader precompile backoff across starts - #800
Closed
monsterlag wants to merge 4 commits into
Closed
monsterlag wants to merge 4 commits into
monsterlag wants to merge 4 commits into
Conversation
Keep exhausted startup protection instead of deleting notes and retrying every cached compute program on later starts. Scope the marker to existing program and driver identity, and migrate legacy exhausted counts through one controlled attempt. Assisted-by: OpenAI Codex
Keep the released prediction, compute variant preparation and handoff code. The fixture extracts only through the actual preparation entry call so it remains independent of the released notes-reader shape. Assisted-by: OpenAI Codex
Give native subprocesses the same verified temporary base as Python. Keep the C++ ownership guard and every policy assertion intact when Python cached paths differ from the CRT environment lookup. Assisted-by: OpenAI Codex
Owner
|
Closing to keep the queue clear — please rebase on current main and reopen; thanks! |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The refreshed branch is published at eda9dd9. This PR remains closed; the review and checks below refer to that current-main branch. Hosted CI for the refreshed head awaits reopening.
Two interrupted startup precompiles currently drop the notes and delete their failure marker. Later starts try cached compute programs again even with no notes, so the same startup failure can repeat. Keep a versioned interruption count scoped to the shader-program stamp, GPU, driver, pipeline-cache UUID and the presence of
RX_VK_NOOPT. After two interruptions, preserve the notes and use ordinary lazy builds until that identity changes. A completed attempt clears its own marker; legacy markers get one scoped retry.The refreshed branch is based on main
b2765ba6through a normal merge. Upstream background preparation, parallel program-cache reading and guarded compute creation remain intact. The fixture supplies the current cache-reader dependencies; all eleven original behavior checks are unchanged.Validation: