pain

code/duffle/gen/macs.h

@@ -199,11 +199,14 @@ WORD_COUNT(mac_gte_gpf_scale, 13)
 ,	gte_lw(C2_VXY0, r_vec, 0) \
 ,	load_word(r_t0, r_vec, 4) \
 ,	nop \
-,	gte_lw(C2_VZ0, r_vec, 4)	/* RTPS cv=1 (no translation accumulation; BK is zero-initialized), sf=0, v=0, mx=0.
+,	gte_lw(C2_VZ0, r_vec, 4)	/* RTPS cv=3 (no translation contribution: TRX/TRY/TRZ are zeroed, BK is zero-initialized),
+	 * sf=1 (integer, no shift = full 32-bit R*pos product), v=0 (uses V0 input), mx=0 (rotation matrix).
 	 * MAC1 = RT11*V0.x + RT12*V0.y + RT13*V0.z + 0
 	 * MAC2 = RT21*V0.x + RT22*V0.y + RT23*V0.z + 0
 	 * MAC3 = RT31*V0.x + RT32*V0.y + RT33*V0.z + 0
-	 * Side effect: RTPS also writes SXY0/1/2 and SZ0..SZ3 (perspective projection). Ignored. */	/* gte_cmdw_rtps_no_tr,	/* Read MAC1/2/3 → out. */ */ \
+	 * Side effect: RTPS also writes SXY0/1/2 and SZ0..SZ3 (perspective projection). Ignored.
+	 * Note: gte_cmdw_rtps_no_tr (the cv=3 alias) was removed from gte.h. Use gte_cmdw_rtps_sf1. */ \
+,	gte_cmdw_rtps_sf1	/* Read MAC1/2/3 → out. */ \
 ,	gte_mv_from_data_r(r_t0, C2_MAC1) \
 ,	gte_mv_from_data_r(r_t1, C2_MAC2) \
 ,	gte_mv_from_data_r(r_t2, C2_MAC3) \
@@ -211,22 +214,16 @@ WORD_COUNT(mac_gte_gpf_scale, 13)
 ,	store_word(r_t0, r_out, 0) \
 ,	store_word(r_t1, r_out, 4) \
 ,	store_word(r_t2, r_out, 8)
-WORD_COUNT(mac_apply_matrix_lv, 30)
+WORD_COUNT(mac_apply_matrix_lv, 31)
 
-#define mac_trans_matrix(r_mtx, r_off, r_t0, r_t1) \
+#define mac_trans_matrix(r_mtx, r_off, r_t1) \
-	load_word(r_t0, r_mtx,  O_(MT3_S2S4,t[0])) \
+	load_word(r_t1, r_off, O_(V3_S4,x)) \
-,	load_word(r_t1, r_off, O_(V3_S4,x)) \
+,	store_word(r_t1, r_mtx, O_(MT3_S2S4,t[0])) \
-,	add_u(r_t0, r_t0, r_t1) \
-,	store_word(r_t0, r_mtx, O_(MT3_S2S4,t[0])) \
-,	load_word(r_t0, r_mtx,  O_(MT3_S2S4,t[1])) \
 ,	load_word(r_t1, r_off, O_(V3_S4,y)) \
-,	add_u(r_t0, r_t0, r_t1) \
+,	store_word(r_t1, r_mtx, O_(MT3_S2S4,t[1])) \
-,	store_word(r_t0, r_mtx, O_(MT3_S2S4,t[1])) \
-,	load_word(r_t0, r_mtx,  O_(MT3_S2S4,t[2])) \
 ,	load_word(r_t1, r_off, O_(V3_S4,z)) \
-,	add_u(r_t0, r_t0, r_t1) \
+,	store_word(r_t1, r_mtx, O_(MT3_S2S4,t[2]))
-,	store_word(r_t0, r_mtx, O_(MT3_S2S4,t[2]))
+WORD_COUNT(mac_trans_matrix, 6)
-WORD_COUNT(mac_trans_matrix, 12)
 
 #define mac_gcmd_push(cmd, reg_transfer, reg_base, port) \
 	load_upper_i(reg_transfer, u4_hi(cmd)) \

code/duffle/gte.atom.c

@@ -126,12 +126,14 @@ FI_ Slice_MipsCode ac_apply_matrix_lv(AtomBuilder_R ab
 	load_word(r_t0, r_vec, 4), nop,
 	gte_lw(C2_VZ0, r_vec, 4),
 
-	/* RTPS cv=1 (no translation accumulation; BK is zero-initialized), sf=0, v=0, mx=0.
+	/* RTPS cv=3 (no translation contribution: TRX/TRY/TRZ are zeroed, BK is zero-initialized),
+	 * sf=1 (integer, no shift = full 32-bit R*pos product), v=0 (uses V0 input), mx=0 (rotation matrix).
 	 * MAC1 = RT11*V0.x + RT12*V0.y + RT13*V0.z + 0
 	 * MAC2 = RT21*V0.x + RT22*V0.y + RT23*V0.z + 0
 	 * MAC3 = RT31*V0.x + RT32*V0.y + RT33*V0.z + 0
-	 * Side effect: RTPS also writes SXY0/1/2 and SZ0..SZ3 (perspective projection). Ignored. */
+	 * Side effect: RTPS also writes SXY0/1/2 and SZ0..SZ3 (perspective projection). Ignored.
-	// gte_cmdw_rtps_no_tr,
+	 * Note: gte_cmdw_rtps_no_tr (the cv=3 alias) was removed from gte.h. Use gte_cmdw_rtps_sf1. */
+	gte_cmdw_rtps_sf1,
 
 	/* Read MAC1/2/3 → out. */
 	gte_mv_from_data_r(r_t0, C2_MAC1),
@@ -145,26 +147,21 @@ FI_ Slice_MipsCode ac_apply_matrix_lv(AtomBuilder_R ab
 
 /* ─── TRANS MATRIX (libgte TransMatrix port) ───
  * Atom component — auto-generates mac_trans_matrix Mac composer macro.
- * m->t += off.  Uses 2 GPRs via per-axis load-delay-slot pattern.
+ * m->t = v (struct copy; libgte's TransMatrix at 0x8001a540 is just 3 store_words, no GTE, no add).
- * Words: 12.  Clobbers: r_t0, r_t1. */
+ * Uses 1 GPR (r_t1 = off value) per axis; per-axis load-delay-slot pattern.
+ * Words: 9.  Clobbers: r_t1. */
 FI_ Slice_MipsCode ac_trans_matrix(AtomBuilder_R ab
 	,	U4 r_mtx, U4 r_off
-	,	U4 r_t0,  U4 r_t1
+	,	U4 r_t1
 ) MipsAtomComp_Proc_(ac_trans_matrix, ab, {
-	load_word(r_t0, r_mtx,  O_(MT3_S2S4,t[0])),
 	load_word(r_t1, r_off, O_(V3_S4,x)),
-	add_u(r_t0, r_t0, r_t1),
+	store_word(r_t1, r_mtx, O_(MT3_S2S4,t[0])),
-	store_word(r_t0, r_mtx, O_(MT3_S2S4,t[0])),
 
-	load_word(r_t0, r_mtx,  O_(MT3_S2S4,t[1])),
 	load_word(r_t1, r_off, O_(V3_S4,y)),
-	add_u(r_t0, r_t0, r_t1),
+	store_word(r_t1, r_mtx, O_(MT3_S2S4,t[1])),
-	store_word(r_t0, r_mtx, O_(MT3_S2S4,t[1])),
 
-	load_word(r_t0, r_mtx,  O_(MT3_S2S4,t[2])),
 	load_word(r_t1, r_off, O_(V3_S4,z)),
-	add_u(r_t0, r_t0, r_t1),
+	store_word(r_t1, r_mtx, O_(MT3_S2S4,t[2])),
-	store_word(r_t0, r_mtx, O_(MT3_S2S4,t[2])),
 })
 
 #pragma endregion MACs (Mips Atom Components)

code/hello_camera/hello_camera.c

@@ -360,9 +360,12 @@ void update(PrimitiveArena* pa, U4* ordering_buf)
 		uz      = scratch->uz;
 		right   = scratch->right;
 		ux      = scratch->ux;
+		up      = scratch->up;     // ← atom 4's output (replaces C-side cross_v3s4)
+		uy      = scratch->uy;     // ← atom 5's output (replaces C-side normalize_v3s4)
 
 		// cross_v3s4(& uz, & ux, & up); normalize_v3s4(& up, & uy);
 		
+		// Atom 6 not yet enabled: populate look_at.m[][] from ux/uy/uz here.
 		smem.cam.look_at.m[0][0] = ux.x; smem.cam.look_at.m[0][1] = ux.y; smem.cam.look_at.m[0][2] = ux.z;
 		smem.cam.look_at.m[1][0] = uy.x; smem.cam.look_at.m[1][1] = uy.y; smem.cam.look_at.m[1][2] = uy.z;
 		smem.cam.look_at.m[2][0] = uz.x; smem.cam.look_at.m[2][1] = uz.y; smem.cam.look_at.m[2][2] = uz.z;