Implemented USAD8 and USADA8

src/backend_x64/emit_x64.cpp

@@ -1672,6 +1672,10 @@ void EmitX64::EmitPackedSaturatedSubS16(IR::Block&, IR::Inst* inst) {
     EmitPackedOperation(code, reg_alloc, inst, &Xbyak::CodeGenerator::psubsw);
 }
 
+void EmitX64::EmitPackedAbsDiffSumS8(IR::Block&, IR::Inst* inst) {
+    EmitPackedOperation(code, reg_alloc, inst, &Xbyak::CodeGenerator::psadbw);
+}
+
 static void DenormalsAreZero32(BlockOfCode* code, Xbyak::Xmm xmm_value, Xbyak::Reg32 gpr_scratch) {
     using namespace Xbyak::util;
     Xbyak::Label end;

src/frontend/ir/ir_emitter.cpp

@@ -408,6 +408,10 @@ Value IREmitter::PackedSaturatedSubS16(const Value& a, const Value& b) {
     return Inst(Opcode::PackedSaturatedSubS16, {a, b});
 }
 
+Value IREmitter::PackedAbsDiffSumS8(const Value& a, const Value& b) {
+    return Inst(Opcode::PackedAbsDiffSumS8, {a, b});
+}
+
 Value IREmitter::TransferToFP32(const Value& a) {
     return Inst(Opcode::TransferToFP32, {a});
 }
@@ -654,4 +658,3 @@ Value IREmitter::Inst(Opcode op, std::initializer_list<Value> args) {
 
 } // namespace IR
 } // namespace Dynarmic
-

src/frontend/ir/ir_emitter.h

@@ -153,6 +153,7 @@ public:
     Value PackedSaturatedAddS16(const Value& a, const Value& b);
     Value PackedSaturatedSubU16(const Value& a, const Value& b);
     Value PackedSaturatedSubS16(const Value& a, const Value& b);
+    Value PackedAbsDiffSumS8(const Value& a, const Value& b);
 
     Value TransferToFP32(const Value& a);
     Value TransferToFP64(const Value& a);

src/frontend/ir/opcodes.inc

@@ -95,6 +95,7 @@ OPCODE(PackedSaturatedAddU16,   T::U32,         T::U32,         T::U32
 OPCODE(PackedSaturatedAddS16,   T::U32,         T::U32,         T::U32                          )
 OPCODE(PackedSaturatedSubU16,   T::U32,         T::U32,         T::U32                          )
 OPCODE(PackedSaturatedSubS16,   T::U32,         T::U32,         T::U32                          )
+OPCODE(PackedAbsDiffSumS8,      T::U32,         T::U32,         T::U32                          )
 
 // Floating-point operations
 OPCODE(TransferToFP32,          T::F32,         T::U32                                          )

src/frontend/translate/translate_arm/parallel.cpp

@@ -66,6 +66,27 @@ bool ArmTranslatorVisitor::arm_USAX(Cond cond, Reg n, Reg d, Reg m) {
     return InterpretThisInstruction();
 }
 
+bool ArmTranslatorVisitor::arm_USAD8(Cond cond, Reg d, Reg m, Reg n) {
+    if (d == Reg::PC || n == Reg::PC || m == Reg::PC)
+        return UnpredictableInstruction();
+    if (ConditionPassed(cond)) {
+        auto result = ir.PackedAbsDiffSumS8(ir.GetRegister(n), ir.GetRegister(m));
+        ir.SetRegister(d, result);
+    }
+    return true;
+}
+
+bool ArmTranslatorVisitor::arm_USADA8(Cond cond, Reg d, Reg a, Reg m, Reg n){
+    if (d == Reg::PC || n == Reg::PC || m == Reg::PC)
+        return UnpredictableInstruction();
+    if (ConditionPassed(cond)) {
+        auto tmp = ir.PackedAbsDiffSumS8(ir.GetRegister(n), ir.GetRegister(m));
+        auto result = ir.AddWithCarry(ir.GetRegister(a), tmp, ir.Imm1(0));
+        ir.SetRegister(d, result.result);
+    }
+    return true;
+}
+
 bool ArmTranslatorVisitor::arm_USUB8(Cond cond, Reg n, Reg d, Reg m) {
     if (d == Reg::PC || n == Reg::PC || m == Reg::PC)
         return UnpredictableInstruction();

src/frontend/translate/translate_arm/translate_arm.h

@@ -214,14 +214,8 @@ struct ArmTranslatorVisitor final {
     bool arm_SEL(Cond cond, Reg n, Reg d, Reg m);
 
     // Unsigned sum of absolute difference functions
-    bool arm_USAD8(Cond cond, Reg d, Reg m, Reg n) {
-        UNUSED(cond, d, m, n);
-        return InterpretThisInstruction();
-    }
-    bool arm_USADA8(Cond cond, Reg d, Reg a, Reg m, Reg n) {
-        UNUSED(cond, d, a, m, n);
-        return InterpretThisInstruction();
-    }
+    bool arm_USAD8(Cond cond, Reg d, Reg m, Reg n);
+    bool arm_USADA8(Cond cond, Reg d, Reg a, Reg m, Reg n);
 
     // Packing instructions
     bool arm_PKHBT(Cond cond, Reg n, Reg d, Imm5 imm5, Reg m);

tests/arm/fuzz_arm.cpp

@@ -961,6 +961,29 @@ TEST_CASE("Fuzz ARM parallel instructions", "[JitX64]") {
     }
 }
 
+TEST_CASE("Fuzz ARM sum of absolute differences", "[JitX64]") {
+    auto validate_d_m_n = [](u32 inst) -> bool {
+        return Bits<16, 19>(inst) != 15 &&
+               Bits<8, 11>(inst) != 15 &&
+               Bits<0, 3>(inst) != 15;
+    };
+    auto validate_d_a_m_n = [&](u32 inst) -> bool {
+        return validate_d_m_n(inst) &&
+               Bits<12, 15>(inst) != 15;
+    };
+
+    const std::array<InstructionGenerator, 2> differences_instructions = {{
+        InstructionGenerator("cccc01111000dddd1111mmmm0001nnnn", validate_d_m_n), // USAD8
+        InstructionGenerator("cccc01111000ddddaaaammmm0001nnnn", validate_d_a_m_n), // USADA8
+    }};
+
+    SECTION("Sum of Absolute Differences (Differences)") {
+        FuzzJitArm(1, 1, 10000, [&differences_instructions]() -> u32 {
+            return differences_instructions[RandInt<size_t>(0, differences_instructions.size() - 1)].Generate();
+        });
+    }
+}
+
 TEST_CASE( "SMUAD", "[JitX64]" ) {
     Dynarmic::Jit jit{GetUserCallbacks()};
     code_mem.fill({});