ir: Add opcodes for left signed saturated shifts

src/backend/x64/emit_x64_vector.cpp

@@ -3572,6 +3572,70 @@ void EmitX64::EmitVectorSignedSaturatedNeg64(EmitContext& ctx, IR::Inst* inst) {
     });
 }
 
+// MSVC requires the capture within the saturate lambda, but it's
+// determined to be unnecessary via clang and GCC.
+#ifdef __clang__
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wunused-lambda-capture"
+#endif
+template <typename T, typename U = std::make_unsigned_t<T>>
+static bool VectorSignedSaturatedShiftLeft(VectorArray<T>& dst, const VectorArray<T>& data, const VectorArray<T>& shift_values) {
+    static_assert(std::is_signed_v<T>, "T must be signed.");
+
+    bool qc_flag = false;
+
+    constexpr size_t bit_size_minus_one = Common::BitSize<T>() - 1;
+
+    const auto saturate = [bit_size_minus_one](T value) {
+        return static_cast<T>((static_cast<U>(value) >> bit_size_minus_one) + (U{1} << bit_size_minus_one) - 1);
+    };
+
+    for (size_t i = 0; i < dst.size(); i++) {
+        const T element = data[i];
+        const T shift = std::clamp<T>(static_cast<T>(Common::SignExtend<8>(shift_values[i] & 0xFF)),
+                                      -static_cast<T>(bit_size_minus_one), std::numeric_limits<T>::max());
+
+        if (element == 0) {
+            dst[i] = 0;
+        } else if (shift < 0) {
+            dst[i] = static_cast<T>(element >> -shift);
+        } else if (static_cast<U>(shift) > bit_size_minus_one) {
+            dst[i] = saturate(element);
+            qc_flag = true;
+        } else {
+            const T shifted = element << shift;
+
+            if ((shifted >> shift) != element) {
+                dst[i] = saturate(element);
+                qc_flag = true;
+            } else {
+                dst[i] = shifted;
+            }
+        }
+    }
+
+    return qc_flag;
+}
+#ifdef __clang__
+#pragma clang diagnostic pop
+#endif
+
+void EmitX64::EmitVectorSignedSaturatedShiftLeft8(EmitContext& ctx, IR::Inst* inst) {
+    EmitTwoArgumentFallbackWithSaturation(code, ctx, inst, VectorSignedSaturatedShiftLeft<s8>);
+}
+
+void EmitX64::EmitVectorSignedSaturatedShiftLeft16(EmitContext& ctx, IR::Inst* inst) {
+    EmitTwoArgumentFallbackWithSaturation(code, ctx, inst, VectorSignedSaturatedShiftLeft<s16>);
+}
+
+void EmitX64::EmitVectorSignedSaturatedShiftLeft32(EmitContext& ctx, IR::Inst* inst) {
+    EmitTwoArgumentFallbackWithSaturation(code, ctx, inst, VectorSignedSaturatedShiftLeft<s32>);
+}
+
+void EmitX64::EmitVectorSignedSaturatedShiftLeft64(EmitContext& ctx, IR::Inst* inst) {
+    EmitTwoArgumentFallbackWithSaturation(code, ctx, inst, VectorSignedSaturatedShiftLeft<s64>);
+}
+
 void EmitX64::EmitVectorSub8(EmitContext& ctx, IR::Inst* inst) {
     EmitVectorOperation(code, ctx, inst, &Xbyak::CodeGenerator::psubb);
 }

src/frontend/ir/ir_emitter.cpp

@@ -1646,6 +1646,21 @@ U128 IREmitter::VectorSignedSaturatedNeg(size_t esize, const U128& a) {
     return {};
 }
 
+U128 IREmitter::VectorSignedSaturatedShiftLeft(size_t esize, const U128& a, const U128& b) {
+    switch (esize) {
+    case 8:
+        return Inst<U128>(Opcode::VectorSignedSaturatedShiftLeft8, a, b);
+    case 16:
+        return Inst<U128>(Opcode::VectorSignedSaturatedShiftLeft16, a, b);
+    case 32:
+        return Inst<U128>(Opcode::VectorSignedSaturatedShiftLeft32, a, b);
+    case 64:
+        return Inst<U128>(Opcode::VectorSignedSaturatedShiftLeft64, a, b);
+    }
+    UNREACHABLE();
+    return {};
+}
+
 U128 IREmitter::VectorSub(size_t esize, const U128& a, const U128& b) {
     switch (esize) {
     case 8:

src/frontend/ir/ir_emitter.h

@@ -278,6 +278,7 @@ public:
     U128 VectorSignedSaturatedNarrowToSigned(size_t original_esize, const U128& a);
     U128 VectorSignedSaturatedNarrowToUnsigned(size_t original_esize, const U128& a);
     U128 VectorSignedSaturatedNeg(size_t esize, const U128& a);
+    U128 VectorSignedSaturatedShiftLeft(size_t esize, const U128& a, const U128& b);
     U128 VectorSub(size_t esize, const U128& a, const U128& b);
     Table VectorTable(std::vector<U128> values);
     U128 VectorTableLookup(const U128& defaults, const Table& table, const U128& indices);

src/frontend/ir/microinstruction.cpp

@@ -369,6 +369,10 @@ bool Inst::WritesToFPSRCumulativeSaturationBit() const {
     case Opcode::VectorSignedSaturatedNeg16:
     case Opcode::VectorSignedSaturatedNeg32:
     case Opcode::VectorSignedSaturatedNeg64:
+    case Opcode::VectorSignedSaturatedShiftLeft8:
+    case Opcode::VectorSignedSaturatedShiftLeft16:
+    case Opcode::VectorSignedSaturatedShiftLeft32:
+    case Opcode::VectorSignedSaturatedShiftLeft64:
     case Opcode::VectorUnsignedSaturatedAccumulateSigned8:
     case Opcode::VectorUnsignedSaturatedAccumulateSigned16:
     case Opcode::VectorUnsignedSaturatedAccumulateSigned32:

src/frontend/ir/opcodes.inc

@@ -422,6 +422,10 @@ OPCODE(VectorSignedSaturatedNeg8,                          U128,           U128
 OPCODE(VectorSignedSaturatedNeg16,                         U128,           U128                                                            )
 OPCODE(VectorSignedSaturatedNeg32,                         U128,           U128                                                            )
 OPCODE(VectorSignedSaturatedNeg64,                         U128,           U128                                                            )
+OPCODE(VectorSignedSaturatedShiftLeft8,                    U128,           U128,           U128                                            )
+OPCODE(VectorSignedSaturatedShiftLeft16,                   U128,           U128,           U128                                            )
+OPCODE(VectorSignedSaturatedShiftLeft32,                   U128,           U128,           U128                                            )
+OPCODE(VectorSignedSaturatedShiftLeft64,                   U128,           U128,           U128                                            )
 OPCODE(VectorSub8,                                         U128,           U128,           U128                                            )
 OPCODE(VectorSub16,                                        U128,           U128,           U128                                            )
 OPCODE(VectorSub32,                                        U128,           U128,           U128                                            )