[video_core] Restrict macro JIT zero-register skips

Only apply the zero-register ALU skip when the operation is safe to elide without changing carry/result semantics.

This avoids invalid-instruction floods seen with Macro JIT enabled while keeping the optimization for operations where a zero source is harmless.

src/video_core/macro.cpp

@@ -782,13 +782,17 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
     const bool valid_operation = !is_a_zero && !is_b_zero;
     [[maybe_unused]] const bool is_move_operation = !is_a_zero && is_b_zero;
     const bool has_zero_register = is_a_zero || is_b_zero;
-    const bool no_zero_reg_skip = opcode.alu_operation == Macro::ALUOperation::AddWithCarry ||
-                                  opcode.alu_operation == Macro::ALUOperation::SubtractWithBorrow;
+    const bool zero_reg_skip =
+        optimizer.zero_reg_skip && !is_a_zero && is_b_zero &&
+        (opcode.alu_operation == Macro::ALUOperation::Xor ||
+         opcode.alu_operation == Macro::ALUOperation::Or ||
+         (optimizer.can_skip_carry && (opcode.alu_operation == Macro::ALUOperation::Add ||
+                                       opcode.alu_operation == Macro::ALUOperation::Subtract)));
 
     Xbyak::Reg32 src_a;
     Xbyak::Reg32 src_b;
 
-    if (!optimizer.zero_reg_skip || no_zero_reg_skip) {
+    if (!zero_reg_skip) {
         src_a = Compile_GetRegister(opcode.src_a, RESULT);
         src_b = Compile_GetRegister(opcode.src_b, eax);
     } else {
@@ -804,7 +808,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
 
     switch (opcode.alu_operation) {
     case Macro::ALUOperation::Add:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (valid_operation) {
                 add(src_a, src_b);
             }
@@ -822,7 +826,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
         setc(byte[STATE + offsetof(JITState, carry_flag)]);
         break;
     case Macro::ALUOperation::Subtract:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (valid_operation) {
                 sub(src_a, src_b);
                 has_emitted = true;
@@ -841,7 +845,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
         setc(byte[STATE + offsetof(JITState, carry_flag)]);
         break;
     case Macro::ALUOperation::Xor:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (valid_operation) {
                 xor_(src_a, src_b);
             }
@@ -850,7 +854,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
         }
         break;
     case Macro::ALUOperation::Or:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (valid_operation) {
                 or_(src_a, src_b);
             }
@@ -859,7 +863,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
         }
         break;
     case Macro::ALUOperation::And:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (!has_zero_register) {
                 and_(src_a, src_b);
             }
@@ -868,7 +872,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
         }
         break;
     case Macro::ALUOperation::AndNot:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (!is_a_zero) {
                 not_(src_b);
                 and_(src_a, src_b);
@@ -879,7 +883,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
         }
         break;
     case Macro::ALUOperation::Nand:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (!is_a_zero) {
                 and_(src_a, src_b);
                 not_(src_a);