ARM: implement multiplication carry flag algorithm

src/nba/src/arm/handlers/arithmetic.inl

@@ -3,6 +3,18 @@
  *
  * Licensed under GPLv3 or any later version.
  * Refer to the included LICENSE file.
+ *
+ * Multiplication carry flag algorithm has been altered from its original form according to its GPL-compatible license, as follows:
+ *
+ * Copyright (C) 2024 zaydlang, calc84maniac
+ *
+ * This software is provided 'as-is', without any express or implied warranty. In no event will the authors be held liable for any damages arising from the use of this software.
+ *
+ * Permission is granted to anyone to use this software for any purpose, including commercial applications, and to alter it and redistribute it freely, subject to the following restrictions:
+ *
+ *   1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required.
+ *   2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software.
+ *   3. This notice may not be removed or altered from any source distribution.
  */
 
 void SetZeroAndSignFlag(u32 value) {
@@ -11,7 +23,7 @@ void SetZeroAndSignFlag(u32 value) {
 }
 
 template<bool is_signed = true>
-void TickMultiply(u32 multiplier) {
+bool TickMultiply(u32 multiplier) {
   u32 mask = 0xFFFFFF00;
 
   bus.Idle();
@@ -28,6 +40,98 @@ void TickMultiply(u32 multiplier) {
     mask <<= 8;
     bus.Idle();
   }
+
+  // Return true if full ticks used.
+  return mask == 0;
+}
+
+bool MultiplyCarrySimple(u32 multiplier) {
+  // Carry comes directly from final injected booth carry bit.
+  // Final booth addend is negative only if upper 2 bits are 10.
+  return (multiplier >> 30) == 2;
+}
+
+bool MultiplyCarryLo(u32 multiplicand, u32 multiplier, u32 accum = 0) {
+  // Set low bit of multiplicand to cause negation to invert the upper bits.
+  // This bit cannot propagate to the resulting carry bit.
+  multiplicand |= 1;
+
+  // Optimized first iteration.
+  u32 booth = (s32)(multiplier << 31) >> 31;
+  u32 carry = multiplicand * booth;
+  u32 sum = carry + accum;
+
+  int shift = 29;
+  do {
+    // Process 8 multiplier bits using 4 booth iterations.
+    for (int i = 0; i < 4; i++, shift -= 2) {
+      // Get next booth factor (-2 to 2, shifted left by 30-shift).
+      u32 next_booth = (s32)(multiplier << shift) >> shift;
+      u32 factor = next_booth - booth;
+      booth = next_booth;
+      // Get scaled value of booth addend.
+      u32 addend = multiplicand * factor;
+      // Accumulate addend with carry-save add.
+      accum ^= carry ^ addend;
+      sum += addend;
+      carry = sum - accum;
+    }
+  } while (booth != multiplier);
+
+  // Carry flag comes from bit 31 of carry-save adder's final carry.
+  return carry >> 31;
+}
+
+template<bool sign_extend>
+bool MultiplyCarryHi(u32 multiplicand, u32 multiplier, u32 accum_hi = 0) {
+  // Only last 3 booth iterations are relevant to output carry.
+  // Reduce scale of both inputs to get upper bits of 64-bit booth addends
+  // in upper bits of 32-bit values, while handling sign extension.
+  if (sign_extend) {
+    multiplicand = (s32)multiplicand >> 6;
+    multiplier = (s32)multiplier >> 26;
+  } else {
+    multiplicand >>= 6;
+    multiplier >>= 26;
+  }
+  // Set low bit of multiplicand to cause negation to invert the upper bits.
+  // This bit cannot propagate to the resulting carry bit.
+  multiplicand |= 1;
+
+  // Pre-populate magic bit 61 for carry.
+  u32 carry = ~accum_hi & 0x20000000;
+  // Pre-populate magic bits 63-60 for accum (with carry magic pre-added).
+  u32 accum = accum_hi - 0x08000000;
+
+  // Get factors for last 3 booth iterations.
+  u32 booth0 = (s32)(multiplier << 27) >> 27;
+  u32 booth1 = (s32)(multiplier << 29) >> 29;
+  u32 booth2 = (s32)(multiplier << 31) >> 31;
+  u32 factor0 = multiplier - booth0;
+  u32 factor1 = booth0 - booth1;
+  u32 factor2 = booth1 - booth2;
+
+  // Get scaled value of 3rd-last booth addend.
+  u32 addend = multiplicand * factor2;
+  // Finalize bits 61-60 of accum magic using its sign.
+  accum -= addend & 0x10000000;
+  // Get scaled value of 2nd-last booth addend.
+  addend = multiplicand * factor1;
+  // Finalize bits 63-62 of accum magic using its sign.
+  accum -= addend & 0x40000000;
+
+  // Get carry from carry-save add in bit 61 and propagate it to bit 62.
+  u32 sum = accum + (addend & 0x20000000);
+  // Subtract out carry magic to get actual accum magic.
+  accum -= carry;
+
+  // Get scaled value of last booth addend.
+  addend = multiplicand * factor0;
+  // Add to bit 62 and propagate carry.
+  sum += addend & 0x40000000;
+
+  // Cancel out accum magic bit 63 to get carry bit 63.
+  return (sum ^ accum) >> 31;
 }
 
 u32 ADD(u32 op1, u32 op2, bool set_flags) {

src/nba/src/arm/handlers/handler16.inl

@@ -183,12 +183,18 @@ void Thumb_ALU(u16 instruction) {
       break;
     }
     case ThumbDataOp::MUL: {
-      TickMultiply(state.reg[dst]);
+      u32 lhs = state.reg[src];
+      u32 rhs = state.reg[dst];
+      bool full = TickMultiply(rhs);
       pipe.access = Access::Code | Access::Nonsequential;
 
-      state.reg[dst] *= state.reg[src];
+      state.reg[dst] = lhs * rhs;
       SetZeroAndSignFlag(state.reg[dst]);
-      state.cpsr.f.c = 0;
+      if (full) {
+        state.cpsr.f.c = MultiplyCarrySimple(rhs);
+      } else {
+        state.cpsr.f.c = MultiplyCarryLo(lhs, rhs);
+      }
       break;
     }
     case ThumbDataOp::BIC: {

src/nba/src/arm/handlers/handler32.inl

@@ -247,15 +247,22 @@ void ARM_Multiply(u32 instruction) {
   auto rhs = GetReg(op2);
   auto result = lhs * rhs;
 
-  TickMultiply(rhs);
+  bool full = TickMultiply(rhs);
 
+  u32 accum = 0;
   if (accumulate) {
-    result += GetReg(op3);
+    accum = GetReg(op3);
+    result += accum;
     bus.Idle();
   }
 
   if (set_flags) {
     SetZeroAndSignFlag(result);
+    if (full) {
+      state.cpsr.f.c = MultiplyCarrySimple(rhs);
+    } else {
+      state.cpsr.f.c = MultiplyCarryLo(lhs, rhs, accum);
+    }
   }
 
   SetReg(dst, result);
@@ -273,7 +280,7 @@ void ARM_MultiplyLong(u32 instruction) {
   int dst_lo = (instruction >> 12) & 0xF;
   int dst_hi = (instruction >> 16) & 0xF;
 
-  s64 result;
+  u64 result;
 
   pipe.access = Access::Code | Access::Nonsequential;
   state.r15 += 4;
@@ -284,18 +291,21 @@ void ARM_MultiplyLong(u32 instruction) {
   if (sign_extend) {
     result = s64(s32(lhs)) * s64(s32(rhs));
   } else {
-    result = s64(u64(lhs) * u64(rhs));
+    result = u64(lhs) * u64(rhs);
   }
 
-  TickMultiply<sign_extend>(rhs);
+  bool full = TickMultiply<sign_extend>(rhs);
   bus.Idle();
 
+  u32 accum_lo = 0;
+  u32 accum_hi = 0;
   if (accumulate) {
-    s64 value = GetReg(dst_hi);
+    accum_lo = GetReg(dst_lo);
+    accum_hi = GetReg(dst_hi);
 
-    value <<= 16;
-    value <<= 16;
-    value  |= GetReg(dst_lo);
+    u64 value = accum_hi;
+    value <<= 32;
+    value |= accum_lo;
 
     result += value;
     bus.Idle();
@@ -306,6 +316,11 @@ void ARM_MultiplyLong(u32 instruction) {
   if (set_flags) {
     state.cpsr.f.n = result_hi >> 31;
     state.cpsr.f.z = result == 0;
+    if (full) {
+      state.cpsr.f.c = MultiplyCarryHi<sign_extend>(lhs, rhs, accum_hi);
+    } else {
+      state.cpsr.f.c = MultiplyCarryLo(lhs, rhs, accum_lo);
+    }
   }
 
   SetReg(dst_lo, result & 0xFFFFFFFF);