X-Git-Url: https://git.gir.st/Chiptunes.git/blobdiff_plain/51f432934fa3a300ca5daada786444c64e037a23..4466dd8b6842dcd98a892ece212799e475c2b1c2:/foo.c

diff --git a/foo.c b/foo.c
index 0d6754d..ac7d6ad 100644
--- a/foo.c
+++ b/foo.c
@@ -2,10 +2,6 @@
 #include "fakeasm.h"
 typedef unsigned char u8;
 
-u8 data[] = {   
-	0x84, 0x9d, 0xb0, 0x69, 0x9d, 0x84, 0x69, 0x58,
-	0x75, 0x8c, 0xb0, 0x69, 0x8c, 0x75, 0x69, 0x58
-};
 u8 zero;	//r16
 u8 acc;		//r17
 u8 i0;		//r18
@@ -16,14 +12,23 @@ u8 n;		//r22
 u8 s;		//r23
 u8 _;		//r24
 		//r25
-u8 t;/*==Ml*/	//r26 (Xlo)
-u8 x;/*==Mh*/	//r27 (Xhi)
+u8 x;/*==Ml*/	//r26 (Xlo)
+u8 t;/*==Mh*/	//r27 (Xhi)
 		//r28
 		//r29
-/*fakestack_l*/	//r30 (Zlo)
-/*fakestack_h*/	//r31 (Zhi)
+void *Z;	//r30 (Zlo)
+/*...*/		//r31 (Zhi)
 #define Mh x //mod3 vars
 #define Ml t // -"-
+
+// .section .data
+u8 data[] = {
+	/*.byte*/ 0x84, 0x9d, 0xb0, 0x69, 0x9d, 0x84, 0x69, 0x58,
+	/*.byte*/ 0x75, 0x8c, 0xb0, 0x69, 0x8c, 0x75, 0x69, 0x58
+};
+
+// .section .text
+
 //http://homepage.divms.uiowa.edu/~jones/bcd/mod.shtml
 void mod3(void) {
 	// mod3(Mh.Ml) -> t
@@ -55,6 +60,27 @@ void mod3(void) {
 	RET
 	#undef tmp
 }
+//.macro definitions for mul-tree:
+#define always(_bit) //nop; for when a test() is not necessary (see tree)
+#define never(_bit)  //nop; for when a test() is not necessary (see tree)
+#define test(_bit,_jmpto) \
+	SBRC	(t, _bit) \
+	RJMP	(_jmpto)
+#define shift16 \
+	LSR	(a2) \
+	ROR	(a1)
+#define shift8 /*top three bits don't need to be corrrect, so save cycles by not carrying*/ \
+	LSR	(a1)
+#define shift0 //nop; last shift is common
+#define add_shift16 \
+	ADD	(a1, i0) \
+	ADC	(a2, i1, carry) \
+	shift16
+#define add_shift8 /*ditto with carrying*/ \
+	ADD	(a1, i0) \
+	shift8
+#define add_shift0 /*last shift is common*/ \
+	ADD	(a1, i0)
 void g(void) {
 	// g(i, t) -> t
 	// tempvars: `x` and `_`
@@ -63,20 +89,136 @@ void g(void) {
 	MOV	(tmp, i2)
 	ANDI	(tmp, 3)
 	TST	(tmp)
-	#undef tmp
-	BREQ	(skip)
+	CPSE	(tmp, zero)
 	SUBI	(t, -8)
-	skip:
-	t = data[t];
-	/*MOV X_hi==x, data_hi
-	  MOV X_lo==t, data_lo
-	  ADD X_lo, t
-	  ADC X_hi, zero
-	  LD  t, X         */
-	t &= 0xfd; //hint
-	t = (((i1&0x1f)<<8|i0)*t)>>8; //TODO
-	t &= 0x1e; //hint
-	RET //TODO: replace CALL/RET with IJMP?
+	#undef tmp
+
+	#define tmp _
+	MOV	(tmp, t) //NOTE: must move value away from `t`, as that is also hi(X)
+	tmp = data[tmp];/*
+	LDI	Xhi, hi8(data)
+	LDI	Xlo, lo8(data)
+	ADD	Xlo, tmp  ;<-- the offset (formerly `t`) into data[]
+	ADC	Xhi, zero
+	LD	tmp, X */
+	MOV	(t, tmp)
+	#undef tmp
+
+	#define a1 x
+	#define a2 _
+	#define a0 t
+	CLR	(a2)
+	CLR	(a1)
+
+	/* decision tree multiplication saves cycles and (hopefully) reduces code size
+	                     _xxx?
+	                 /           \
+	           _xx?0                _xx1?
+	             |                    |
+	           _x?00                _x?01
+	         /       \             /     \
+	    _?000         _?100      _?001    _?101
+	    /   \         /   \        |      /   \
+	 _0000 _1000   _0100 _1100   _1001 _0101 _1101
+	   |     |       |     |       |     |     |
+	  ...   ...     ...   ...     ...   ...   ...
+	   |     |       |     |       |     |     |
+	   B0    58     84    8C      69     75    9D   */
+		test	(0, m____1)
+	m____0:	shift16
+		never	(1)
+	m___00:	shift16
+		test	(2, m__100)
+	m__000:	shift16
+		test	(3, m_1000)
+	m_0000:	shift16
+		always	(4)
+		add_shift16
+		always	(5)
+		add_shift8
+		never	(6)
+		shift8
+		always	(7)
+		add_shift0
+		RJMP	(end_mul) // calc'd 0xb0
+
+	m_1000:	add_shift16
+		always	(4)
+		add_shift16
+		never	(5)
+		shift8
+		always	(6)
+		add_shift8
+		never	(7)
+		shift0
+		RJMP	(end_mul) // calc'd 0x58
+
+	m__100:	add_shift16
+		test	(3, m_1100)
+	m_0100:	shift16
+		RJMP	(upper_8) //'ll calc 0x84
+
+	m_1100:	add_shift16
+	upper_8: /* used twice, so deduplicated */
+		never	(4)
+		shift16
+		never	(5)
+		shift8
+		never	(6)
+		shift8
+		always	(7)
+		add_shift0
+		RJMP	(end_mul) // calc'd 0x8c
+
+	m____1: add_shift16
+		never	(1)
+	m___01:	shift16
+		test	(2, m__101)
+	m__001:	shift16
+		always	(3)
+	m_1001:	add_shift16
+		never	(4)
+		shift16
+		always	(5)
+		add_shift8
+		always	(6)
+		add_shift8
+		never	(7)
+		shift0
+		RJMP	(end_mul) // calc'd 0x69
+
+	m__101:	add_shift16
+		test	(3, m_1101)
+	m_0101:	shift16
+		always	(4)
+		add_shift16
+		always	(5)
+		add_shift8
+		always	(6)
+		add_shift8
+		never	(7)
+		shift0
+		RJMP	(end_mul) // calc'd 0x75
+
+	m_1101:	add_shift16
+		always	(4)
+		add_shift16
+		never	(5)
+		shift8
+		never	(6)
+		shift8
+		always	(7)
+		add_shift0
+		// calc'd 0x9d
+
+	end_mul:
+		LSR (a1) //final shift is a common operation for all
+
+	MOV	(t, a1) //TODO: use a1 in main() directly
+	#undef a0
+	#undef a1
+	#undef a2
+	RET //TODO: replace CALL/RET with IJMP? (requires undoing goto-mul-hack)
 };
 
 int main(void) {
@@ -111,6 +253,7 @@ int main(void) {
 		MOV	(t, n)
 		RCALL	g();
 		SWAP	(t)
+		ANDI	(t, 1)
 		MOV	(acc, t)
 
 		//voice 2:
@@ -211,18 +354,17 @@ int main(void) {
 		RCALL	g();
 		LSR	(t)
 		ANDI	(t, 3)
-		MOV	(x, s)
-		INC	(x)
+		INC	(s)
 		#define tmp _
-                MOV	(tmp, x)
+                MOV	(tmp, s)
 		LSR	(tmp)
-                ADD	(tmp, x)
+                ADD	(tmp, s)
 		ROR	(tmp)
 		LSR	(tmp)
 		LSR	(tmp)
-                ADD	(tmp, x)
+                ADD	(tmp, s)
 		ROR	(tmp)
-                ADD	(tmp, x)
+                ADD	(tmp, s)
 		ROR	(tmp)
 		LSR	(tmp)
 		LSR	(tmp)
@@ -232,8 +374,8 @@ int main(void) {
 
 		putchar(acc<<4); //TODO
 		SUBI	(i0, -1)
-		ADC	(i1, zero, !i0)
-		ADC	(i2, zero, !i0&&!i1)
-		ADC	(i3, zero, !i0&&!i1&&!i2)
+		ADC	(i1, zero, !i0) //XXX: must use "sbci i1,-1" in the assembly version
+		ADC	(i2, zero, !i0&&!i1) //          sbci i2,-1
+		ADC	(i3, zero, !i0&&!i1&&!i2) //     sbci i3,-1
 	}
 }