add ac_yield_load and ac_yield_tail for delay slot optimization opportunities.

code/duffle/gen/duffle.macs.h

@@ -2,7 +2,7 @@
 #pragma once
 #endif
 // Auto-generated by ps1_meta.lua — DO NOT EDIT
-// Source: C:\projects\Pikuma\ps1\code\duffle\lottes_tape.h
+// Source: C:\projects\pikuma\ps1\code\duffle\lottes_tape.h
 // Component atoms (MipsAtomComp_(ac_*)) -> macro variants (mac_*)
 
 #ifndef WORD_COUNT
@@ -15,6 +15,10 @@
  *  These do NOT yield. They are expanded inline inside Tape Atoms.
  * ---------------------------------------------------------------------------*/
 // The 'Yield' sequence for Tape Atoms (mac_yield).
+//   - mac_yield() is the safe default for atom-endings: 4 words, BD-slot of jr is mandatory nop.
+//   - mac_yield_load() + mac_yield_tail():
+//     - unconditional branch: mac_yield_load fills the branch's BD-slot (replaces a nop);
+//     - mac_yield_tail runs at the branch target (does NOT re-load R_AtomJmp).
 #define mac_yield(...) \
 	load_word(R_AtomJmp, R_TapePtr, 0) \
 ,	add_ui_self(         R_TapePtr, S_(MipsCode)) \
@@ -22,6 +26,18 @@
 ,	nop
 WORD_COUNT(mac_yield, 4)
 
+/* atom_dbg_skip */
+#define mac_yield_load(...) \
+	load_word(R_AtomJmp, R_TapePtr, 0)
+WORD_COUNT(mac_yield_load, 1)
+
+/* atom_dbg_skip */
+#define mac_yield_tail(...) \
+	add_ui_self(R_TapePtr, S_(MipsCode)) \
+,	jump_reg(  R_AtomJmp) \
+,	nop
+WORD_COUNT(mac_yield_tail, 3)
+
 /* atom_dbg_skip */
 /* Words: 3; Loads 3 S2 indices from the face array */
 #define mac_load_tri_indices(...) \

code/duffle/gen/duffle.offsets.h

@@ -1,5 +1,5 @@
 // Auto-generated by ps1_meta.lua (passes/offsets.lua) — DO NOT EDIT
-// Source: C:\projects\Pikuma\ps1\code\duffle\lottes_tape.h
+// Source: C:\projects\pikuma\ps1\code\duffle\lottes_tape.h
 #pragma once
 
 #pragma region lottes_tape