Repository navigation
Fix glowing foliage stems and branches (register marker collisions) - #759
Merged
NotRexed merged 1 commit intoNov 9, 2026
Merged
Conversation
Survey the instruction's original operand spans before allocating renaming markers. Incremental native_high tracking allowed an early destination marker to alias a later source, corrupting mesh lighting calculations. Add a game-free FFMA regression that checks effective registers with and without the uniform bank, and run it in source CI. Assisted-by: OpenAI Codex Co-Authored-By: OpenAI Codex <noreply@openai.com>
pelicanux
force-pushed
the
fix/native-register-marker-collisions
branch
from
November 8, 2026 20:31
4d68af0 to
add0937
Compare
pelicanux
changed the title
This was referenced Nov 9, 2026
Open
Closed
Closed
Closed
rakefire-impala
pushed a commit
to rakefire-impala/Hookshot-w-Mesh-Shader-Fallback
that referenced
this pull request
Nov 9, 2026
…nd (bright red/green foliage stems) (cherry picked from commit deaf4ff)
rakefire-impala
added a commit
to rakefire-impala/Hookshot-w-Mesh-Shader-Fallback
that referenced
this pull request
Nov 9, 2026
…e new switch lines Item 11 now has the cause. The "stem" culprits were depth passes; the stems' colour pass computed its light-probe lighting wrong because of a translator bug on every GPU (upstream NotRexed#757, NotRexed#874, NotRexed#986), which upstream PR NotRexed#759 fixed. Positions and triangles read back in one frame were bit-identical. Also notes the hearts glitch as upstream NotRexed#993 and documents "read" and "info". Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
rakefire-impala
added a commit
to rakefire-impala/Hookshot-w-Mesh-Shader-Fallback
that referenced
this pull request
Nov 9, 2026
Black hearts, the minimap glitch and the slow unpause on 4 GB cards: the cause and the fix, needing more testing. Known limits now say NotRexed#759 is cherry-picked onto a branch 170 commits behind upstream main (with upstream's open AMD stripes report), and that stutters are not diagnosed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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.
Fix bright red/green stems in nearby foliage by keeping register-renaming markers separate from every original operand.
Marker allocation checks
native_highwhile it is still being populated. A marker chosen for an early field can therefore alias a later source. Ordinary GPRs belowreserved_basecan collide as well. The fix surveys the original operand spans first, then excludes them from marker allocation, preserving the existing pair/quad handling.The regression generates synthetic FFMA instructions and checks the effective destination and all three sources after REGMAP decoding. It covers both reserved-bank configurations, later sources and repeated destination/addend operands. The existing CI runner discovers it automatically; no workflow changes are needed.
Validation:
71e59e3, with the same earlier fixes on both sides: Kokiri stems retain natural colors and leaves remain visible. Tested on RX 9070, RADV/Mesa 26.2.6, with frame generation disabled; a further run without overlay or diagnostic instrumentation confirmed the result.Addresses the glowing-branch symptom in #667. Only the lowering fix, its synthetic regression and the test-runner entry are included.