jqueuniet/dm2-tools
1
0
Fork 0
Code Issues 2 Pull requests Projects Releases Packages Wiki Activity Actions

Validate the lighting-stripe palette model and decode the cls4 selector #259

Merged
jqueuniet merged 0 commits from refs/pull/259/head into main 2026-07-17 11:32:24 +02:00
Conversation 0 Commits - Files changed -
jqueuniet commented 2026-07-17 11:16:56 +02:00 (Migrated from codeberg.org)
Copy link

Closes #214.

Validation (primary goal)

The lighting-stripe hypothesis is confirmed on both axes; the live ADDR_PALSEL trace fallback is unnecessary.

  • Re-running the stripe-aware matcher (from #213) on the session-1 captures already resolved CRAM line 1 in all five (four exact bank-stripe hits, one modulo a two-slot flame overlay).
  • A fresh annotated capture session then confirmed the indexing half: bank = map_graphics_style (three banks observed; the starting cave gives a certain map-0/style-2 → bank-2 correlation) and stripe = light level (stripe 1↔4 series in a single room, line 1 byte-identical across its two bright states). Details in docs/formats/palette-lut.md § "VALIDATED".
  • Side findings: the torch glow is a separate CRAM line-2 palette; the slots-9/10 flame flicker overlay recurs byte-identically; banks 1 and 5 are byte-identical in the archive.

Secondary target: cls4 selector + jump-table rule

  • The FTL BSS jump-table entries are function-entry CODE offsets, and jsr disp(a5) maps as entry = (disp − 0x22)/8 — the 32-byte above-A5 params block precedes the runtime table. Live-proven with two breakpoints; the same rule holds on Sega CD and Amiga with matching entry numbers (no merged-table anomaly). Earlier (disp−2)/8 labels were systematically +4 off; the GDAT resolver is 0x2abc0, not 0x1e78e.
  • The cls4 LUT selector splits into an id-taking boolean gate (entry 496, live-sampled on a keyhole) and the pointer-taking selector cls4 = table[record->byte4 >> 4] (entry 492), plus the simplified-sensor Text path cls4 = (word2 >> 3) & 0xFF. The supposed ABI mismatch was the numbering artifact.

Tooling

  • palette_oracle_match now labels candidates with their cls tuples (banks read directly as cls2), with a regression test pinning the six bank ENT1 indices.
  • tooling/palette-oracle/run_capture.sh is a real capture launcher: interactive mode plus a --fifo mode that bridges the BlastEm debugger through a PTY via socat (its REPL hangs on a non-TTY stdin).

Remaining work is tracked in the decoder-wiring follow-up issue.

Closes #214. ## Validation (primary goal) The lighting-stripe hypothesis is **confirmed** on both axes; the live `ADDR_PALSEL` trace fallback is unnecessary. - Re-running the stripe-aware matcher (from #213) on the session-1 captures already resolved CRAM line 1 in all five (four exact bank-stripe hits, one modulo a two-slot flame overlay). - A fresh annotated capture session then confirmed the indexing half: bank = `map_graphics_style` (three banks observed; the starting cave gives a certain map-0/style-2 → bank-2 correlation) and stripe = light level (stripe 1↔4 series in a single room, line 1 byte-identical across its two bright states). Details in `docs/formats/palette-lut.md` § "VALIDATED". - Side findings: the torch glow is a separate CRAM line-2 palette; the slots-9/10 flame flicker overlay recurs byte-identically; banks 1 and 5 are byte-identical in the archive. ## Secondary target: cls4 selector + jump-table rule - The FTL BSS jump-table entries are function-entry CODE offsets, and `jsr disp(a5)` maps as `entry = (disp − 0x22)/8` — the 32-byte above-A5 params block precedes the runtime table. Live-proven with two breakpoints; the same rule holds on Sega CD and Amiga with matching entry numbers (no merged-table anomaly). Earlier `(disp−2)/8` labels were systematically +4 off; the GDAT resolver is `0x2abc0`, not `0x1e78e`. - The cls4 LUT selector splits into an id-taking boolean gate (entry 496, live-sampled on a keyhole) and the pointer-taking selector `cls4 = table[record->byte4 >> 4]` (entry 492), plus the simplified-sensor Text path `cls4 = (word2 >> 3) & 0xFF`. The supposed ABI mismatch was the numbering artifact. ## Tooling - `palette_oracle_match` now labels candidates with their cls tuples (banks read directly as `cls2`), with a regression test pinning the six bank ENT1 indices. - `tooling/palette-oracle/run_capture.sh` is a real capture launcher: interactive mode plus a `--fifo` mode that bridges the BlastEm debugger through a PTY via socat (its REPL hangs on a non-TTY stdin). Remaining work is tracked in the decoder-wiring follow-up issue.
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
jqueuniet/dm2-tools!259
No description provided.