wip

code/duffle/gen/macs.h

@@ -175,6 +175,60 @@ WORD_COUNT(mac_gte_sqr_v3, 8)
 ,	shift_aright_var(r_dz, r_dz, r_shift)
 WORD_COUNT(mac_gte_gpf_scale, 13)
 
+#define mac_apply_matrix_lv(r_mtx, r_vec, r_out, r_t0, r_t1, r_t2) \
+	load_word(r_t0, r_mtx,  0) \
+,	nop \
+,	gte_mv_to_ctrl_r(r_t0, gte_cr_RT11_Code) \
+,	load_word(r_t0, r_mtx,  4) \
+,	nop \
+,	gte_mv_to_ctrl_r(r_t0, gte_cr_RT12_Code) \
+,	load_word(r_t0, r_mtx,  8) \
+,	nop \
+,	gte_mv_to_ctrl_r(r_t0, gte_cr_RT13_Code) \
+,	load_word(r_t0, r_mtx, 12) \
+,	nop \
+,	gte_mv_to_ctrl_r(r_t0, gte_cr_RT21_Code) \
+,	load_half_u(r_t0, r_mtx, 16) \
+,	nop \
+,	gte_mv_to_ctrl_r(r_t0, gte_cr_RT22_Code) \
+,	nop2	/* Load PACKED pos into V0 (libgte SVECTOR layout).
+	 * r_vec points to atom-0-staged packed data ((pos.y << 16) | pos.x at +0, pos.z at +4).
+	 * LWC2 base register MUST be the pointer r_vec, NOT the loaded value r_t0. */ \
+,	load_word(r_t0, r_vec, 0) \
+,	nop \
+,	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.
+	 * 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. */ \
+,	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) \
+,	nop \
+,	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, 31)
+
+#define mac_trans_matrix(r_mtx, r_off, r_t0, r_t1) \
+	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_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_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_t0, r_mtx, O_(MT3_S2S4,t[2]))
+WORD_COUNT(mac_trans_matrix, 12)
+
 #define mac_gcmd_push(cmd, reg_transfer, reg_base, port) \
 	load_upper_i(reg_transfer, u4_hi(cmd)) \
 ,	or_i_self(   reg_transfer, u4_lo(cmd))	/* load_upper_i(reg_transfer, cmd >> 16),	// or_i_self(   reg_transfer, cmd & 0xFFFF), */ \

code/duffle/gte.atom.c

@@ -83,6 +83,90 @@ FI_ Slice_MipsCode ac_gte_gpf_scale(AtomBuilder_R ab, U4 r_sx, U4 r_sy, U4 r_sz,
 	shift_aright_var(r_dz, r_dz, r_shift),
 })
 
+/* ─── APPLY MATRIX LV (libgte ApplyMatrixLV port) ───
+ * Atom component — auto-generates mac_apply_matrix_lv Mac composer macro.
+ * Uses GTE RTPS (cv=1, sf=1, v=0) with lwc2-loaded V0/VZ0 inputs.
+ * Per PSX-SPX `geometrytransformationenginegte.md` lines 416-418:
+ *   IR1 = MAC1 = (TRX*1000h + RT11*VX0 + RT12*VY0 + RT13*VZ0) SAR (sf*12)
+ *   IR2 = MAC2 = (TRY*1000h + RT21*VX0 + RT22*VY0 + RT23*VZ0) SAR (sf*12)
+ *   IR3 = MAC3 = (TRZ*1000h + RT31*VX0 + RT32*VY0 + RT33*VZ0) SAR (sf*12)
+ * RTPS uses the FULL row of the rotation matrix (not just diagonal like MVMVA with mx=0).
+ * libgte's `gte_ApplyMatrix` calls `gte_rtv0()` = RTPS cv=1 v=0 mx=0.
+ * Per `gte.h` line 405 the body sets cv=3 (BK, zero-initialized) so no TR contribution.
+ *
+ * Operands:
+ *   r_mtx  : MT3_S2S4* (matrix pointer)
+ *   r_vec  : U4 (pointer to PACKED V0 data — (pos.y << 16) | pos.x at +0, pos.z at +4)
+ *   r_out  : V3_S4* (output pointer; MAC1/2/3 stored here)
+ *   r_t0/1/2 : 3 GPR codes for matrix load + intermediate state
+ * Words: ~26.  Clobbers: r_t0, r_t1, r_t2 (C2 $0..$4, VXY0/VZ0, MAC1/2/3, SXY0/1/2). */
+FI_ Slice_MipsCode ac_apply_matrix_lv(AtomBuilder_R ab
+	,	U4 r_mtx, U4 r_vec, U4 r_out
+	,	U4 r_t0,  U4 r_t1,  U4 r_t2
+) MipsAtomComp_Proc_(ac_apply_matrix_lv, ab, {
+	/* Load MATRIX rows into GTE RT11..RT33 (libgte convention: ctc2 to C2 $0..$4 in order).
+	 * load_half_u zero-extends the last word so RT33 = m[2][2] and TRX = 0. */
+	load_word(r_t0, r_mtx,  0), nop,
+	gte_mv_to_ctrl_r(r_t0, gte_cr_RT11_Code),
+	load_word(r_t0, r_mtx,  4), nop,
+	gte_mv_to_ctrl_r(r_t0, gte_cr_RT12_Code),
+	load_word(r_t0, r_mtx,  8), nop,
+	gte_mv_to_ctrl_r(r_t0, gte_cr_RT13_Code),
+	load_word(r_t0, r_mtx, 12), nop,
+	gte_mv_to_ctrl_r(r_t0, gte_cr_RT21_Code),
+	load_half_u(r_t0, r_mtx, 16), nop,
+	gte_mv_to_ctrl_r(r_t0, gte_cr_RT22_Code),
+	nop2,
+
+	/* Load PACKED pos into V0 (libgte SVECTOR layout).
+	 * r_vec points to atom-0-staged packed data ((pos.y << 16) | pos.x at +0, pos.z at +4).
+	 * LWC2 base register MUST be the pointer r_vec, NOT the loaded value r_t0. */
+	load_word(r_t0, r_vec, 0), nop,
+	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.
+	 * 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. */
+	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),
+	nop,
+	store_word(r_t0, r_out, 0),
+	store_word(r_t1, r_out, 4),
+	store_word(r_t2, r_out, 8),
+})
+
+/* ─── 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.
+ * Words: 12.  Clobbers: r_t0, r_t1. */
+FI_ Slice_MipsCode ac_trans_matrix(AtomBuilder_R ab
+	,	U4 r_mtx, U4 r_off
+	,	U4 r_t0,  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_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_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_t0, r_mtx, O_(MT3_S2S4,t[2])),
+})
+
 #pragma endregion MACs (Mips Atom Components)
 
 #pragma region Atom Procs

code/duffle/gte.h

@@ -391,6 +391,19 @@ enum { _C2_TX_SUBS_ = 0
 	* The wedge alias is the 3D complement interpretation of the same 3 scalars (MAC1..MAC3). */
 #define gte_cmdw_mvmva  (gte_cmd_base | enc_gte_cmd(gte_cmd_mvmva))
 
+/* MVMVA with sf=1 (12-bit shift), cv=3 (no translation), v=3 (IR): for ApplyMatrixLV.
+ * Reads input from IR1/2/3 (loaded via mtc2 rt, C2_IRx). MAC1/2/3 = (RT row · IR) >> 12.
+ * The 12-bit shift produces values like R*pos >> 12, matching the libgte C-side ApplyMatrixLV output.
+ * Note: PCSX-Redux's MVMVA interpretation differs from the spec on some matrix layouts.
+ */
+#define gte_cmdw_mvmva_ir    (gte_cmd_base | enc_gte_sf(1) | enc_gte_cv(3) | enc_gte_v(3) | enc_gte_cmd(gte_cmd_mvmva))
+#define gte_cmdw_mvmva_no_tr gte_cmdw_mvmva_ir
+
+/* RTPS with sf=1 (12-bit shift, no translation): matches the output of libgte's
+ * ApplyMatrixLV when the GTE pipeline expects R*pos >> 12. The shift produces
+ * values like (-270, 710, 1713) which match the C11 reference path. */
+#define gte_cmdw_rtps_sf1   (gte_cmd_base | enc_gte_sf(1) | enc_gte_cv(3) | enc_gte_cmd(gte_cmd_rtps))
+
 /* SQR / GPF cosmetic-bits compat helpers.
  * Each command's `_compat` macro ORs in the `fake_cmd` field value libgte happens to emit.
  * The hardware ignores these bits (per PSX-SPX line 48). */

code/hello_camera/hello_camera.c

@@ -369,8 +369,7 @@ void update(PrimitiveArena* pa, U4* ordering_buf)
 
 		pos = smem.cam.pos; mul_v3s4(& pos, v3s4(-1,-1,-1)); // RGA(Lengyel): -eye in world coordinates (spatial bulk only; implicit weight is dropped).
 
-		mul_m3s2_v3s4(& smem.cam.look_at, & pos, & off);
+		mul_m3s2_v3s4(& smem.cam.look_at, & pos, & off); trans_m3s2(   & smem.cam.look_at,        & off);
-		trans_m3s2(   & smem.cam.look_at,        & off);
 	}
 
 	// Draw cube