intiial review on: resolve_look_at__input_and_sub_proc

code/duffle/gen/macs.h

@@ -60,8 +60,8 @@ WORD_COUNT(mac_yield_tail, 3)
 
 /* atom_dbg_skip */
 #define mac_load_v2s2(rs_x, rs_y, r_base, offset) \
-	load_half( rs_x, r_base, O_(V3_S2,x)) \
+	load_half( rs_x, r_base, offset + O_(V3_S2,x)) \
-,	load_half( rs_y, r_base, O_(V3_S2,y))
+,	load_half( rs_y, r_base, offset + O_(V3_S2,y))
 WORD_COUNT(mac_load_v2s2, 2)
 
 /* atom_dbg_skip */
@@ -72,9 +72,9 @@ WORD_COUNT(mac_store_v2s2, 2)
 
 /* atom_dbg_skip */
 #define mac_load_v3s4(rs_x, rs_y, rs_z, r_base, offset) \
-	load_word( rs_x, r_base, O_(V3_S4,x)) \
+	load_word( rs_x, r_base, offset + O_(V3_S4,x)) \
-,	load_word( rs_y, r_base, O_(V3_S4,y)) \
+,	load_word( rs_y, r_base, offset + O_(V3_S4,y)) \
-,	load_word( rs_z, r_base, O_(V3_S4,z))
+,	load_word( rs_z, r_base, offset + O_(V3_S4,z))
 WORD_COUNT(mac_load_v3s4, 3)
 
 /* atom_dbg_skip */

code/duffle/lottes_tape.h

@@ -100,6 +100,8 @@ enum {
 	// S 0-7
 };
 
+typedef U2 Reg; // Register parameter used with atom or atom component procedures
+
 typedef U4 const MipsCode; // Underlying type to mips asm words.
 typedef Slice_(MipsCode);
 
@@ -150,7 +152,7 @@ typedef Slice_(MipsAtom);
 typedef Slice_MipsAtom Tape;
 
 /* The 'Exit' Atom */
-atom_dbg_skip MipsAtom_(tape_exit) { jump_reg(rret_addr), nop };
+atom_dbg_skip MipsAtom_(tape_exit) { jump_reg(R_RA), nop };
 
 // TODO(Ed): When we have a substantial workload/throughput, profile each of these to see impact at ABI boundaries.
 

code/duffle/math.atom.c

@@ -10,8 +10,8 @@ ATOM_FILE_DEBUGGER_LINE_MARKER(math_atom_c);
 #pragma region MACs (Mips Atom Component)
 
 FI_ Slice_MipsCode ac_load_v2s2(AtomBuilder_R ab, U4 rs_x, U4 rs_y, U4 r_base, U4 offset) atom_dbg_skip MipsAtomComp_Proc_(ac_load_v2s2, ab, {
-	load_half( rs_x, r_base, O_(V3_S2,x)),
+	load_half( rs_x, r_base, offset + O_(V3_S2,x)),
-	load_half( rs_y, r_base, O_(V3_S2,y)),
+	load_half( rs_y, r_base, offset + O_(V3_S2,y)),
 })
 
 FI_ Slice_MipsCode ac_store_v2s2(AtomBuilder_R ab, U4 rt_x, U4 rt_y, U4 base, U4 offset) atom_dbg_skip MipsAtomComp_Proc_(ac_store_v2s2, ab, {
@@ -20,16 +20,22 @@ FI_ Slice_MipsCode ac_store_v2s2(AtomBuilder_R ab, U4 rt_x, U4 rt_y, U4 base, U4
 })
 
 FI_ Slice_MipsCode ac_load_v3s4(AtomBuilder_R ab, U4 rs_x, U4 rs_y, U4 rs_z, U4 r_base, U4 offset) atom_dbg_skip MipsAtomComp_Proc_(ac_load_v3s4, ab, {
-	load_word( rs_x, r_base, O_(V3_S4,x)),
+	load_word( rs_x, r_base, offset + O_(V3_S4,x)),
-	load_word( rs_y, r_base, O_(V3_S4,y)),
+	load_word( rs_y, r_base, offset + O_(V3_S4,y)),
-	load_word( rs_z, r_base, O_(V3_S4,z)),
+	load_word( rs_z, r_base, offset + O_(V3_S4,z)),
 })
+// TODO(Ed): we could generate these mappings properly..
+#define ac_load_p3s4 ac_load_v3s4
+#define mac_load_p3s4 mac_load_v3s4
 
 FI_ Slice_MipsCode ac_store_v3s4(AtomBuilder_R ab, U4 rt_x, U4 rt_y, U4 rt_z, U4 base, U4 offset) atom_dbg_skip MipsAtomComp_Proc_(ac_store_v3s4, ab, {
 	store_word(rt_x, base, offset + O_(V3_S4,x)),
 	store_word(rt_y, base, offset + O_(V3_S4,y)),
 	store_word(rt_z, base, offset + O_(V3_S4,z)),
 })
+// TODO(Ed): we could generate these mappings properly..
+#define ac_store_p3s4 ac_store_v3s4
+#define mac_store_p3s4 mac_store_v3s4
 
 FI_ Slice_MipsCode ac_sub_v3s4(AtomBuilder_R ab, U4 rds_x, U4 rds_y, U4 rds_z, U4 rt_x,  U4 rt_y,  U4 rt_z) atom_dbg_skip MipsAtomComp_Proc_(ac_sub_v3s4, ab, {
 	sub_s(rds_x, rds_x, rt_x),

code/duffle/mips.atom.c

@@ -20,14 +20,14 @@ ATOM_FILE_DEBUGGER_LINE_MARKER(mips_atom_c);
  *   6. sp += 8
  */
 internal MipsAtom_(mips_flush_icache) {
-	add_ui(rstack_ptr, rstack_ptr, -MipsStackAlignment), // sp -= 8
+	add_ui(R_SP, R_SP, -MipsStackAlignment), // sp -= 8
-	store_word(rret_addr, rstack_ptr, S_(U4)),           // sw  $ra,   4($sp) 
+	store_word(R_RA, R_SP, S_(U4)),           // sw  $ra,   4($sp) 
-	add_ui(rret_0, rdiscard, bios_flushcache),           // addiu $a0, $0, 0x44 
+	add_ui(R_V0, R_0, bios_flushcache),           // addiu $a0, $0, 0x44 
-	add_ui(rtmp_0, rdiscard, bios_table_addr),           // addiu $t0, $0, 0xA0 
+	add_ui(R_T0, R_0, bios_table_addr),           // addiu $t0, $0, 0xA0 
-	jump_link(rtmp_0, rret_addr), nop,                   // jalr  $t0, $ra, BD slot
+	jump_link(R_T0, R_RA), nop,                   // jalr  $t0, $ra, BD slot
-	load_word(rret_addr, rstack_ptr, S_(U4)),            // lw   $ra, 4($sp) 
+	load_word(R_RA, R_SP, S_(U4)),            // lw   $ra, 4($sp) 
-	jump_reg(rret_addr),                                 // jr   $ra 
+	jump_reg(R_RA),                                 // jr   $ra 
-	add_ui(rstack_ptr, rstack_ptr, MipsStackAlignment),  // sp += 8 (BD) 
+	add_ui(R_SP, R_SP, MipsStackAlignment),  // sp += 8 (BD) 
 	mac_yield(),
 };
 

code/duffle/mips.h

@@ -136,31 +136,31 @@ enum {
 
 /* Semantic Aliases for MIPS Registers (O32 ABI) */
 
-	, rdiscard      = R_0  /* Hardwired to 0 */
+	// , rdiscard      = R_0  /* Hardwired to 0 */
-	, rasm_tmp      = R_AT /* Assembler temporary (destroyed by some assembler pseudoinstructions!) */
+	// , rasm_tmp      = R_AT /* Assembler temporary (destroyed by some assembler pseudoinstructions!) */
-	, rret_0        = R_V0 /* Function return value */
+	// , rret_0        = R_V0 /* Function return value */
-	, rret_1        = R_V1 /* Second return value (e.g., 64-bit) */
+	// , rret_1        = R_V1 /* Second return value (e.g., 64-bit) */
-	, rarg_0        = R_A0 /* First function argument */
+	// , rarg_0        = R_A0 /* First function argument */
-	, rarg_1        = R_A1 /* Second function argument */
+	// , rarg_1        = R_A1 /* Second function argument */
-	, rarg_2        = R_A2 /* Third function argument */
+	// , rarg_2        = R_A2 /* Third function argument */
-	, rarg_3        = R_A3 /* Fourth function argument */
+	// , rarg_3        = R_A3 /* Fourth function argument */
-	, rtmp_0        = R_T0 /* Temporary (Caller saved) */
+	// , rtmp_0        = R_T0 /* Temporary (Caller saved) */
-	, rtmp_1        = R_T1 /* Temporary (Caller saved) */
+	// , rtmp_1        = R_T1 /* Temporary (Caller saved) */
-	, rtmp_2        = R_T2 /* Temporary (Caller saved) */
+	// , rtmp_2        = R_T2 /* Temporary (Caller saved) */
-	, rtmp_3        = R_T3 /* Temporary (Caller saved) */
+	// , rtmp_3        = R_T3 /* Temporary (Caller saved) */
-	, rtmp_4        = R_T4 /* Temporary (Caller saved) — common GTE base pointer */
+	// , rtmp_4        = R_T4 /* Temporary (Caller saved) — common GTE base pointer */
-	, rtmp_9        = R_T9 /* Temporary (Caller saved) — common GTE base pointer */
+	// , rtmp_9        = R_T9 /* Temporary (Caller saved) — common GTE base pointer */
-	, rstatic_0     = R_S0 /* Static (Callee saved, preserved across calls) */
+	// , rstatic_0     = R_S0 /* Static (Callee saved, preserved across calls) */
-	, rstatic_1     = R_S1
+	// , rstatic_1     = R_S1
-	, rstatic_2     = R_S2
+	// , rstatic_2     = R_S2
-	, rstatic_3     = R_S3
+	// , rstatic_3     = R_S3
-	, rstatic_4     = R_S4
+	// , rstatic_4     = R_S4
-	, rstatic_5     = R_S5
+	// , rstatic_5     = R_S5
-	, rstatic_6     = R_S6
+	// , rstatic_6     = R_S6
-	, rstatic_7     = R_S7
+	// , rstatic_7     = R_S7
-	, rsaved_0      = R_S0 /* Alias for rstatic_0 (alternate vocabulary) */
+	// , rsaved_0      = R_S0 /* Alias for rstatic_0 (alternate vocabulary) */
-	, rstack_ptr    = R_SP /* Stack Pointer */
+	// , rstack_ptr    = R_SP /* Stack Pointer */
-	, rret_addr     = R_RA /* Return Address (populated by JAL) */
+	// , rret_addr     = R_RA /* Return Address (populated by JAL) */
 
 /* --- MIPS CPU Opcodes (Bits 31-26) --- */
 
@@ -453,7 +453,7 @@ enum { _BitOffsets = 0
 #define shift_amount(rd, rt, n)     shift_lleft(rd, rt, n)
 
 /* nop — sll $0, $0, 0 */
-#define nop  shift_lleft(rdiscard, rdiscard, 0)
+#define nop  shift_lleft(R_0, R_0, 0)
 #define nop2 nop, nop
 
 // li_s — load signed 16-bit immediate into GPR (addiu rt, $0, imm — sign-extends).

code/duffle/pad.c

@@ -36,13 +36,13 @@ NI_ void pad_bios_init_start(PadBiosRaw* raw0, PadBiosRaw* raw1)
 	 *   $t2 = 0xB0 (BIOS B-table address) */
 	asm volatile(
 		asm_words(
-			or_u(    rarg_2, rarg_1, rdiscard),               /* $a2 = $a1 = raw1 */
+			or_u(    R_A2, R_A0, R_0),                 /* $a2 = $a1 = raw1 */
-			add_ui(  rarg_1, rdiscard, bios_pad_buffer_size), /* $a1 = 0x22 */
+			add_ui(  R_A1, R_0, bios_pad_buffer_size), /* $a1 = 0x22 */
-			add_ui(  rarg_3, rdiscard, bios_pad_buffer_size), /* $a3 = 0x22 */
+			add_ui(  R_A3, R_0, bios_pad_buffer_size), /* $a3 = 0x22 */
-			add_ui(  rtmp_1, rdiscard, bios_init_pad_2),      /* $t1 = 0x12 */
+			add_ui(  R_T1, R_0, bios_init_pad_2),      /* $t1 = 0x12 */
-			add_ui(  rtmp_2, rdiscard, bios_btable_addr),     /* $t2 = 0xB0 */
+			add_ui(  R_T2, R_0, bios_btable_addr),     /* $t2 = 0xB0 */
-			call_reg(rtmp_2),                                 /* jalr $t2, $ra */
+			call_reg(R_T2),                            /* jalr $t2, $ra */
-			nop                                               /* BD slot */
+			nop                                        /* BD slot */
 		)
 		asm_rpins, r_use(p0), r_use(p1)
 		asm_clobber: 
@@ -62,9 +62,9 @@ NI_ void pad_bios_init_start(PadBiosRaw* raw0, PadBiosRaw* raw1)
 	/* B(13h) StartPAD2() — no args. The BIOS preserves $sp. */
 	asm volatile(
 		asm_words(
-			add_ui(  rtmp_1, rdiscard, bios_start_pad_2), /* $t1 = 0x13 */
+			add_ui(  R_T1, R_0, bios_start_pad_2), /* $t1 = 0x13 */
-			add_ui(  rtmp_2, rdiscard, bios_btable_addr), /* $t2 = 0xB0 (re-load) */
+			add_ui(  R_T2, R_0, bios_btable_addr), /* $t2 = 0xB0 (re-load) */
-			call_reg(rtmp_2),                             /* jalr $t2, $ra */
+			call_reg(R_T2),                             /* jalr $t2, $ra */
 			nop                                           /* BD slot */
 		)
 		asm_clobber:

code/hello_camera/hello_camera.atom.c

@@ -188,26 +188,17 @@ typedef Struct_(ResolveLookAtScratch) {
  */
 
 typedef Struct_(Binds_ResolveLookAtSub) {
-	U4 target;  /* U4 (C-side P3_S4* — read by atom 0 directly; NOT a scratchpad address) */
+	P3_S4* target;  /* U4 (C-side P3_S4* — read by atom 0 directly; NOT a scratchpad address) */
-	U4 eye;     /* U4 (C-side P3_S4* — read by atom 0 directly; staged into scratchpad by atom 0) */
+	P3_S4* eye;     /* U4 (C-side P3_S4* — read by atom 0 directly; staged into scratchpad by atom 0) */
-	U4 up_in;   /* U4 (C-side V3_S4* — read by atom 0 directly; staged into scratchpad by atom 0) */
+	V3_S4* up_in;   /* U4 (C-side V3_S4* — read by atom 0 directly; staged into scratchpad by atom 0) */
-	U4 scratchpad;
+	ResolveLookAtScratch* scratchpad;
 };
 
 /* Atom 0 in the bundle: input_and_sub. Stages C-side inputs into the scratchpad and computes fwd = target - eye.
- * Inputs (C-side pointers popped from the tape):
- *   r_target_ptr : P3_S4* (C-side struct; atom 0 reads target.x/y/z directly)
- *   r_eye_ptr    : P3_S4* (C-side struct; staged into scratchpad at +96/+100/+104)
- *   r_up_in_ptr  : V3_S4* (C-side struct; staged into scratchpad at +128/+132/+136)
- *   r_scratch    : R_ResolveScratch (R_T4) — scratch base, read by atoms 1-6
- * 
- * Bind-pop layout:
- *   Binds_ResolveLookAtSub
  * Staging work:
  *   * Stage eye.x/y/z   → scratch (for atom 6's translation column)
  *   * Stage up_in.x/y/z → scratch (for atom 2's outer-product operand)
  *   * Compute fwd = target - eye, store fwd.x/y/z → scratch+0/+4/+8 (for atom 1)
- *
  * GPR codes (assigned by resolve_look_at_init):
  *   r_target_ptr : R_T0
  *   r_eye_ptr    : R_T1
@@ -219,56 +210,35 @@ typedef Struct_(Binds_ResolveLookAtSub) {
  *   r_tmp3       : R_T7 (stage eye/up_in + load target.y)
  *   R_AT         : hardcoded (load eye.y / eye.z / target.z)
  *   R_V0         : hardcoded (load eye.z / target.z)
- *
  * Pool cost: 8 GPRs + R_T4 (carrier) + R_AT + R_V0 (hardcoded) = 11 GPRs.
  */
-internal MipsAtom* resolve_look_at__input_and_sub_proc(AtomArena_R aa,	U4 r_scratch
+internal MipsAtom* resolve_look_at__input_and_sub_proc(AtomArena_R aa,	
+			// TODO(Ed): We can resolve scratch at anytime its fixed to a specific address.
+			U4 r_scratch
 	,	U4 r_target_ptr,U4 r_eye_ptr, U4 r_up_in_ptr
 	,	U4 r_tmp0,      U4 r_tmp1,    U4 r_tmp2, U4 r_tmp3
 ) MipsAtom_Proc_(resolve_look_at__input_and_sub, aa, {
-	/* Pop the 3 C-side pointers + scratch_base from the tape. */
 	load_word(r_target_ptr, R_TapePtr, O_(Binds_ResolveLookAtSub,target)),
 	load_word(r_eye_ptr,    R_TapePtr, O_(Binds_ResolveLookAtSub,eye)),
 	load_word(r_up_in_ptr,  R_TapePtr, O_(Binds_ResolveLookAtSub,up_in)),
 	load_word(r_scratch,    R_TapePtr, O_(Binds_ResolveLookAtSub,scratchpad)),
 	add_ui_self(            R_TapePtr, S_(Binds_ResolveLookAtSub)),
 
-	/* Stage eye.x/y/z into the scratchpad (atom 6 reads these for the translation
+	// Stage eye.x/y/z into the scratchpad (atom 6 reads these for the translation column).
-	 * column). Reuse r_tmp0/r_tmp1/r_tmp2. Offsets via O_(ResolveLookAtScratch,*). */
+	mac_load_p3s4( r_tmp0, r_tmp1, r_tmp2, r_eye_ptr, 0),
-	load_word(r_tmp0, r_eye_ptr, O_(P3_S4,x)),
+	mac_store_p3s4(r_tmp0, r_tmp1, r_tmp2, r_scratch, O_(ResolveLookAtScratch,eye)),
-	load_word(r_tmp1, r_eye_ptr, O_(P3_S4,y)),
-	load_word(r_tmp2, r_eye_ptr, O_(P3_S4,z)),
-	nop, /* load-delay */
-	store_word(r_tmp0, r_scratch, O_(ResolveLookAtScratch,eye.x)),
-	store_word(r_tmp1, r_scratch, O_(ResolveLookAtScratch,eye.y)),
-	store_word(r_tmp2, r_scratch, O_(ResolveLookAtScratch,eye.z)),
 
-	/* Stage up_in.x/y/z into the scratchpad (atom 2 reads these for the outer
+	/* Stage up_in.x/y/z into the scratchpad. */
-	 * product with uz). Reuse r_tmp0/r_tmp1/r_tmp2. */
+	mac_load_p3s4( r_tmp0, r_tmp1, r_tmp2, r_up_in_ptr, 0),
-	load_word(r_tmp0, r_up_in_ptr, O_(V3_S4,x)),
+	mac_store_p3s4(r_tmp0, r_tmp1, r_tmp2, r_scratch, O_(ResolveLookAtScratch,up_in)),
-	load_word(r_tmp1, r_up_in_ptr, O_(V3_S4,y)),
-	load_word(r_tmp2, r_up_in_ptr, O_(V3_S4,z)),
-	nop, /* load-delay */
-	store_word(r_tmp0, r_scratch, O_(ResolveLookAtScratch,up_in.x)),
-	store_word(r_tmp1, r_scratch, O_(ResolveLookAtScratch,up_in.y)),
-	store_word(r_tmp2, r_scratch, O_(ResolveLookAtScratch,up_in.z)),
 
 	/* Compute fwd = target - eye. */
-	load_word(r_tmp0, r_target_ptr, O_(P3_S4,x)),
+	mac_load_p3s4(r_tmp0, r_tmp1, r_tmp2, r_target_ptr, 0),
-	load_word(r_tmp1, r_target_ptr, O_(P3_S4,y)),
+	mac_load_p3s4(r_tmp3, R_AT, R_V0, r_eye_ptr, 0),
-	load_word(r_tmp2, r_target_ptr, O_(P3_S4,z)),
+	mac_sub_v3s4(
-	load_word(r_tmp3, r_eye_ptr,    O_(P3_S4,x)),
+		r_tmp0, r_tmp1, r_tmp2,
-	load_word(R_AT,   r_eye_ptr,    O_(P3_S4,y)),
+		r_tmp3, R_AT,   R_V0),
-	load_word(R_V0,   r_eye_ptr,    O_(P3_S4,z)),
+	mac_store_v3s4(r_tmp0, r_tmp1, r_tmp2, r_scratch, O_(ResolveLookAtScratch,fwd)),
-	nop, /* load-delay */
-	sub_u(r_tmp0, r_tmp0, r_tmp3),
-	sub_u(r_tmp1, r_tmp1, R_AT),
-	sub_u(r_tmp2, r_tmp2, R_V0),
-
-	/* Store fwd.x/y/z (atom 1 reads these as the normalize src). */
-	store_word(r_tmp0, r_scratch, O_(ResolveLookAtScratch,fwd.x)),
-	store_word(r_tmp1, r_scratch, O_(ResolveLookAtScratch,fwd.y)),
-	store_word(r_tmp2, r_scratch, O_(ResolveLookAtScratch,fwd.z)),
 
 	mac_yield()
 })