Improve iplBoard and iplChannelSelect matching #223

Merged
coah80 merged 2 commits from agent/w0928/board into main 2026-09-28 05:09:53 +00:00
coah80 commented 2026-09-28 05:09:44 +00:00 (Migrated from github.com)

exact-function improvements for iplBoard and iplChannelSelect, both still NonMatching. partly ported from old luna salvage branches, without their hacks

unit before after
iplBoard 88/94 fns, code 15824/20276 93/94, code 19144/20276, data still 896/896
iplChannelSelect 97/102 fns, code 22616/25668 101/102, code 24736/25668
  • iplBoard: early return rewrites of calc / edge_arrow, split condition in stt_wait_fadeaway, explicit next-pointer loop in stt_normal, a small static inline VEC3 add for calc_board_object. dropped the salvage's string blob pointer math and volatile locals. appendRecord is left at 97.68 (the original keeps 3 interrupt states on the stack, only volatile gets there so i left it)
  • iplChannelSelect: ctor initializer list, drag delta as a VEC2 with a zero speed branch (moveDrag), HermiteIntp<VEC3>/<f32>::get specializations in the cpp (fixes setChanZoomOrtho and calcDiskLayout), named ortho locals in setChannelScissor. didnt take the salvage's novtable define, lbl_ data, volatile float loads or register keywords

gates (fresh orchestrator build on top of #222)

  • no function or unit measure regresses anywhere (full report diff against main)
  • no new asm / lbl_ / volatile / register in the diff
  • clean ninja build/43U/ok passes, main.dol sha1 26116613f624061ba99c8d1a299aaa6efa85670d
  • full ninja passes
exact-function improvements for iplBoard and iplChannelSelect, both still NonMatching. partly ported from old luna salvage branches, without their hacks | unit | before | after | |---|---|---| | iplBoard | 88/94 fns, code 15824/20276 | 93/94, code 19144/20276, data still 896/896 | | iplChannelSelect | 97/102 fns, code 22616/25668 | 101/102, code 24736/25668 | - iplBoard: early return rewrites of `calc` / `edge_arrow`, split condition in `stt_wait_fadeaway`, explicit next-pointer loop in `stt_normal`, a small static inline VEC3 add for `calc_board_object`. dropped the salvage's string blob pointer math and volatile locals. `appendRecord` is left at 97.68 (the original keeps 3 interrupt states on the stack, only volatile gets there so i left it) - iplChannelSelect: ctor initializer list, drag delta as a VEC2 with a zero speed branch (`moveDrag`), `HermiteIntp<VEC3>/<f32>::get` specializations in the cpp (fixes `setChanZoomOrtho` and `calcDiskLayout`), named ortho locals in `setChannelScissor`. didnt take the salvage's novtable define, `lbl_` data, volatile float loads or `register` keywords gates (fresh orchestrator build on top of #222) - no function or unit measure regresses anywhere (full report diff against main) - no new asm / `lbl_` / volatile / register in the diff - clean `ninja build/43U/ok` passes, main.dol sha1 `26116613f624061ba99c8d1a299aaa6efa85670d` - full ninja passes
Sign in to join this conversation.