DSP: Improve multiplier code. Add support for unsigned operation - but temporarily disabled, since it somehow makes things sound even worse :P

git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@3139 8ced0084-cf51-0410-be5f-012b33b47a6e

1 files changed, 107 insertions, 90 deletions

diff --git a/Source/Core/DSPCore/Src/DspIntMultiplier.cpp b/Source/Core/DSPCore/Src/DspIntMultiplier.cpp
index 0a2a0e46c2..54d9c69088 100644
--- a/Source/Core/DSPCore/Src/DspIntMultiplier.cpp
+++ b/Source/Core/DSPCore/Src/DspIntMultiplier.cpp
@@ -28,6 +28,76 @@
 

 namespace DSPInterpreter {

 

+// Sets prod as a side effect.

+// Only MULX family instructions have unsigned support.

+s64 dsp_multiply_conditional_unsigned(u16 a, u16 b)

+{

+	s64 prod;

+#if 0  // Makes games sound horrible. TODO: activate and figure out why - it's been verified through DSPSpy :/

+	if (g_dsp.r[DSP_REG_SR] & SR_MUL_UNSIGNED)

+		prod = (u64)a * (u64)b;  // won't overflow 32-bits

+	else

+#endif

+		prod = (s32)(s16)a * (s32)(s16)b;  // won't overflow 32-bits

+

+	// Conditionally multiply by 2.

+	if ((g_dsp.r[DSP_REG_SR] & SR_MUL_MODIFY) == 0)

+		prod <<= 1;

+

+	// Store the product, and return it, in case the caller wants to read it.

+	dsp_set_long_prod(prod);

+	return prod;

+}

+

+// Sets prod as a side effect.

+s64 dsp_multiply(u16 a, u16 b)

+{

+	s64 prod;

+	prod = (s32)(s16)a * (s32)(s16)b;  // won't overflow 32-bits

+

+	// Conditionally multiply by 2.

+	if ((g_dsp.r[DSP_REG_SR] & SR_MUL_MODIFY) == 0)

+		prod <<= 1;

+

+	// Store the product, and return it, in case the caller wants to read it.

+	dsp_set_long_prod(prod);

+	return prod;

+}

+

+s64 dsp_multiply_add(u16 a, u16 b)

+{

+	s64 prod = (s32)(s16)a * (s32)(s16)b;  // won't overflow 32-bits

+

+	// Conditionally multiply by 2.

+	if ((g_dsp.r[DSP_REG_SR] & SR_MUL_MODIFY) == 0)

+		prod <<= 1;

+

+	// Add the original prod value.

+	prod += dsp_get_long_prod();

+

+	// Store the product, and return it, in case the caller wants to read it.

+	dsp_set_long_prod(prod);

+	return prod;

+}

+

+s64 dsp_multiply_sub(u16 a, u16 b)

+{

+	s64 prod = (s32)(s16)a * (s32)(s16)b;  // won't overflow 32-bits

+

+	// Conditionally multiply by 2.

+	if ((g_dsp.r[DSP_REG_SR] & SR_MUL_MODIFY) == 0)

+		prod <<= 1;

+

+	// Subtract from the original prod value.

+	prod = dsp_get_long_prod() - prod;

+

+	// Store the product, and return it, in case the caller wants to read it.

+	dsp_set_long_prod(prod);

+	return prod;

+}

+

+

+

 // CLRP

 // 1000 0100 xxxx xxxx

 // Clears product register $prod.

@@ -79,8 +149,8 @@ void addpaxz(const UDSPInstruction& opc)
 	u8 sreg = (opc.hex >> 9) & 0x1;

 

 	s64 prod = dsp_get_long_prod() & ~0x0ffff;

-	s64 ax_h = dsp_get_long_acx(sreg);

-	s64 acc = (prod + ax_h) & ~0x0ffff;

+	s64 ax = dsp_get_long_acx(sreg);

+	s64 acc = (prod + ax) & ~0xffff;

 

 	dsp_set_long_acc(dreg, acc);

 

@@ -110,13 +180,9 @@ void movpz(const UDSPInstruction& opc)
 // secondary accumulator $axS (treat them both as signed).

 void mulc(const UDSPInstruction& opc)

 {

-	// math new prod

 	u8 sreg = (opc.hex >> 11) & 0x1;

 	u8 treg = (opc.hex >> 12) & 0x1;

-

-	s64 prod = dsp_get_acc_m(sreg) * dsp_get_ax_h(treg) * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

-

+	s64 prod = dsp_multiply(dsp_get_acc_m(sreg), dsp_get_ax_h(treg));

 	Update_SR_Register64(prod);

 }

 

@@ -134,8 +200,7 @@ void mulcmvz(const UDSPInstruction& opc)
 	// update prod

 	u8 sreg  = (opc.hex >> 12) & 0x1;

 	u8 treg  = (opc.hex >> 11) & 0x1;

-	s64 Prod = (s64)dsp_get_acc_m(sreg) * (s64)dsp_get_ax_h(treg) * GetMultiplyModifier();

-	dsp_set_long_prod(Prod);

+	dsp_multiply(dsp_get_acc_m(sreg), (s64)dsp_get_ax_h(treg));

 

 	// update acc

 	u8 rreg = (opc.hex >> 8) & 0x1;

@@ -153,19 +218,18 @@ void mulcmvz(const UDSPInstruction& opc)
 // possible mistake in duddie's doc axT.h rather than axS.h

 void mulcmv(const UDSPInstruction& opc)

 {

-	s64 tempProd = dsp_get_long_prod();

+	s64 old_prod = dsp_get_long_prod();

 

 	// update prod

 	u8 sreg  = (opc.hex >> 12) & 0x1;

 	u8 treg  = (opc.hex >> 11) & 0x1;

-	s64 prod = (s64)dsp_get_acc_m(sreg) * (s64)dsp_get_ax_h(treg) * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

+	dsp_multiply(dsp_get_acc_m(sreg), (s64)dsp_get_ax_h(treg));

 

 	// update acc

 	u8 rreg = (opc.hex >> 8) & 0x1;

-	dsp_set_long_acc(rreg, tempProd);

+	dsp_set_long_acc(rreg, old_prod);

 

-	Update_SR_Register64(tempProd);

+	Update_SR_Register64(old_prod);

 }

 

 // MULCAC $acS.m, $axT.h, $acR

@@ -175,17 +239,16 @@ void mulcmv(const UDSPInstruction& opc)
 // register before multiplication to accumulator $acR.

 void mulcac(const UDSPInstruction& opc)

 {

-	s64 TempProd = dsp_get_long_prod();

+	s64 old_prod = dsp_get_long_prod();

 

 	// update prod

 	u8 sreg  = (opc.hex >> 12) & 0x1;

 	u8 treg  = (opc.hex >> 11) & 0x1;

-	s64 Prod = (s64)dsp_get_acc_m(sreg) * (s64)dsp_get_ax_h(treg) * GetMultiplyModifier();

-	dsp_set_long_prod(Prod);

+	dsp_multiply(dsp_get_acc_m(sreg), (s64)dsp_get_ax_h(treg));

 

 	// update acc

 	u8 rreg = (opc.hex >> 8) & 0x1;

-	s64 acc = TempProd + dsp_get_long_acc(rreg);

+	s64 acc = old_prod + dsp_get_long_acc(rreg);

 	dsp_set_long_acc(rreg, acc);

 

 	Update_SR_Register64(acc);

@@ -199,10 +262,7 @@ void mulcac(const UDSPInstruction& opc)
 void mul(const UDSPInstruction& opc)

 {

 	u8 sreg  = (opc.hex >> 11) & 0x1;

-	s64 prod = (s64)dsp_get_ax_h(sreg) * (s64)dsp_get_ax_l(sreg) * GetMultiplyModifier();

-

-	dsp_set_long_prod(prod);

-

+	s64 prod = dsp_multiply(dsp_get_ax_h(sreg), dsp_get_ax_l(sreg));

 	// FIXME: no update in duddie's docs

 	Update_SR_Register64(prod);

 }

@@ -221,8 +281,7 @@ void mulac(const UDSPInstruction& opc)
 

 	// calculate new prod

 	u8 sreg = (opc.hex >> 11) & 0x1;

-	s64 prod = dsp_get_ax_l(sreg) * dsp_get_ax_h(sreg) * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

+	s64 prod = dsp_multiply(dsp_get_ax_l(sreg), dsp_get_ax_h(sreg));

 

 	// FIXME: no update in duddie's docs

 	Update_SR_Register64(prod);

@@ -241,13 +300,7 @@ void mulmv(const UDSPInstruction& opc)
 	s64 acc = prod;

 	dsp_set_long_acc(rreg, acc);

 

-	s64 val1 = (s16)g_dsp.r[sreg + DSP_REG_AXL0];

-	s64 val2 = (s16)g_dsp.r[sreg + DSP_REG_AXH0];

-

-	prod = val1 * val2 * GetMultiplyModifier();

-

-	dsp_set_long_prod(prod);

-

+	prod = dsp_multiply(dsp_get_ax_l(sreg), dsp_get_ax_h(sreg));

 	Update_SR_Register64(prod);

 }

 

@@ -268,9 +321,7 @@ void mulmvz(const UDSPInstruction& opc)
 	dsp_set_long_acc(rreg, acc);

 

 	// math prod

-	prod = (s64)(s16)g_dsp.r[DSP_REG_AXL0 + sreg] * (s64)(s16)g_dsp.r[DSP_REG_AXH0 + sreg] * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

-

+	prod = dsp_multiply(dsp_get_ax_l(sreg), dsp_get_ax_h(sreg));

 	Update_SR_Register64(prod);

 }

 

@@ -283,12 +334,10 @@ void mulx(const UDSPInstruction& opc)
 	u8 sreg = ((opc.hex >> 12) & 0x1);

 	u8 treg = ((opc.hex >> 11) & 0x1);

 

-	s64 val1 = (sreg == 0) ? dsp_get_ax_l(0) : dsp_get_ax_h(0);

-	s64 val2 = (treg == 0) ? dsp_get_ax_l(1) : dsp_get_ax_h(1);

-

-	s64 prod = val1 * val2 * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

+	u16 val1 = (sreg == 0) ? dsp_get_ax_l(0) : dsp_get_ax_h(0);

+	u16 val2 = (treg == 0) ? dsp_get_ax_l(1) : dsp_get_ax_h(1);

 

+	s64 prod = dsp_multiply_conditional_unsigned(val1, val2);

 	Update_SR_Register64(prod);

 }

 

@@ -308,12 +357,10 @@ void mulxac(const UDSPInstruction& opc)
 	u8 sreg = (opc.hex >> 12) & 0x1;

 	u8 treg = (opc.hex >> 11) & 0x1;

 

-	s64 val1 = (sreg == 0) ? dsp_get_ax_l(0) : dsp_get_ax_h(0);

-	s64 val2 = (treg == 0) ? dsp_get_ax_l(1) : dsp_get_ax_h(1);

-

-	s64 prod = val1 * val2 * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

+	u16 val1 = (sreg == 0) ? dsp_get_ax_l(0) : dsp_get_ax_h(0);

+	u16 val2 = (treg == 0) ? dsp_get_ax_l(1) : dsp_get_ax_h(1);

 

+	s64 prod = dsp_multiply_conditional_unsigned(val1, val2);

 	Update_SR_Register64(prod);

 }

 

@@ -333,12 +380,10 @@ void mulxmv(const UDSPInstruction& opc)
 	u8 sreg = (opc.hex >> 12) & 0x1;

 	u8 treg = (opc.hex >> 11) & 0x1;

 

-	s64 val1 = (sreg == 0) ? dsp_get_ax_l(0) : dsp_get_ax_h(0);

-	s64 val2 = (treg == 0) ? dsp_get_ax_l(1) : dsp_get_ax_h(1);

-

-	s64 prod = val1 * val2 * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

+	s16 val1 = (sreg == 0) ? dsp_get_ax_l(0) : dsp_get_ax_h(0);

+	s16 val2 = (treg == 0) ? dsp_get_ax_l(1) : dsp_get_ax_h(1);

 

+	s64 prod = dsp_multiply_conditional_unsigned(val1, val2);

 	Update_SR_Register64(prod);

 }

 

@@ -360,12 +405,10 @@ void mulxmvz(const UDSPInstruction& opc)
 	u8 sreg = (opc.hex >> 12) & 0x1;

 	u8 treg = (opc.hex >> 11) & 0x1;

 

-	s64 val1 = (sreg == 0) ? dsp_get_ax_l(0) : dsp_get_ax_h(0);

-	s64 val2 = (treg == 0) ? dsp_get_ax_l(1) : dsp_get_ax_h(1);

-

-	prod = val1 * val2 * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

+	u16 val1 = (sreg == 0) ? dsp_get_ax_l(0) : dsp_get_ax_h(0);

+	u16 val2 = (treg == 0) ? dsp_get_ax_l(1) : dsp_get_ax_h(1);

 

+	prod = dsp_multiply_conditional_unsigned(val1, val2);

 	Update_SR_Register64(prod);

 }

 

@@ -379,13 +422,10 @@ void maddx(const UDSPInstruction& opc)
 	u8 sreg = (opc.hex >> 9) & 0x1;

 	u8 treg = (opc.hex >> 8) & 0x1;

 

-	s64 val1 = (sreg == 0) ? dsp_get_ax_l(0) : dsp_get_ax_h(0);

-	s64 val2 = (treg == 0) ? dsp_get_ax_l(1) : dsp_get_ax_h(1);

-

-	s64 prod = dsp_get_long_prod();

-	prod += val1 * val2 * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

+	u16 val1 = (sreg == 0) ? dsp_get_ax_l(0) : dsp_get_ax_h(0);

+	u16 val2 = (treg == 0) ? dsp_get_ax_l(1) : dsp_get_ax_h(1);

 

+	s64 prod = dsp_multiply_add(val1, val2);

 	Update_SR_Register64(prod);

 }

 

@@ -399,13 +439,10 @@ void msubx(const UDSPInstruction& opc)
 	u8 sreg = (opc.hex >> 9) & 0x1;

 	u8 treg = (opc.hex >> 8) & 0x1;

 

-	s64 val1 = (sreg == 0) ? dsp_get_ax_l(0) : dsp_get_ax_h(0);

-	s64 val2 = (treg == 0) ? dsp_get_ax_l(1) : dsp_get_ax_h(1);

-

-	s64 prod = dsp_get_long_prod();

-	prod -= val1 * val2 * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

+	u16 val1 = (sreg == 0) ? dsp_get_ax_l(0) : dsp_get_ax_h(0);

+	u16 val2 = (treg == 0) ? dsp_get_ax_l(1) : dsp_get_ax_h(1);

 

+	s64 prod = dsp_multiply_sub(val1, val2);

 	Update_SR_Register64(prod);

 }

 

@@ -419,13 +456,7 @@ void maddc(const UDSPInstruction& opc)
 	u32 sreg = (opc.hex >> 9) & 0x1;

 	u32 treg = (opc.hex >> 8) & 0x1;

 

-	s64 val1 = dsp_get_acc_m(sreg);

-	s64 val2 = dsp_get_ax_h(treg);

-

-	s64 prod = dsp_get_long_prod();

-	prod += val1 * val2 * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

-

+	s64 prod = dsp_multiply_add(dsp_get_acc_m(sreg), dsp_get_ax_h(treg));

 	Update_SR_Register64(prod);

 }

 

@@ -439,18 +470,10 @@ void msubc(const UDSPInstruction& opc)
 	u32 sreg = (opc.hex >> 9) & 0x1;

 	u32 treg = (opc.hex >> 8) & 0x1;

 

-	s64 val1 = dsp_get_acc_m(sreg);

-	s64 val2 = dsp_get_ax_h(treg);

-

-	s64 prod = dsp_get_long_prod();

-	prod -= val1 * val2 * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

-

+	s64 prod = dsp_multiply_sub(dsp_get_acc_m(sreg), dsp_get_ax_h(treg));

 	Update_SR_Register64(prod);

 }

 

-

-

 // MADD $axS.l, $axS.h

 // 1111 001s xxxx xxxx

 // Multiply low part $axS.l of secondary accumulator $axS by high part

@@ -460,10 +483,7 @@ void madd(const UDSPInstruction& opc)
 {

 	u8 sreg = (opc.hex >> 8) & 0x1;

 

-	s64 prod = dsp_get_long_prod();

-	prod += (s64)dsp_get_ax_l(sreg) * (s64)dsp_get_ax_h(sreg) * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

-

+	s64 prod = dsp_multiply_add(dsp_get_ax_l(sreg), dsp_get_ax_h(sreg));

 	Update_SR_Register64(prod);

 }

 

@@ -476,10 +496,7 @@ void msub(const UDSPInstruction& opc)
 {

 	u8 sreg = (opc.hex >> 8) & 0x1;

 

-	s64 prod = dsp_get_long_prod();

-	prod -= (s64)dsp_get_ax_l(sreg) * (s64)dsp_get_ax_h(sreg) * GetMultiplyModifier();

-	dsp_set_long_prod(prod);

-

+	s64 prod = dsp_multiply_sub(dsp_get_ax_l(sreg), dsp_get_ax_h(sreg));

 	Update_SR_Register64(prod);

 }