A64: Implement TRN1

src/frontend/A64/decoder/a64.inc

@@ -515,7 +515,7 @@ INST(USRA_1,                 "USRA",                                      "01111
 
 // Data Processing - FP and SIMD - SIMD Permute
 //INST(UZP1,                   "UZP1",                                      "0Q001110zz0mmmmm000110nnnnnddddd")
-//INST(TRN1,                   "TRN1",                                      "0Q001110zz0mmmmm001010nnnnnddddd")
+INST(TRN1,                   "TRN1",                                      "0Q001110zz0mmmmm001010nnnnnddddd")
 INST(ZIP1,                   "ZIP1",                                      "0Q001110zz0mmmmm001110nnnnnddddd")
 //INST(UZP2,                   "UZP2",                                      "0Q001110zz0mmmmm010110nnnnnddddd")
 //INST(TRN2,                   "TRN2",                                      "0Q001110zz0mmmmm011010nnnnnddddd")

src/frontend/A64/translate/impl/simd_permute.cpp

@@ -8,6 +8,42 @@
 
 namespace Dynarmic::A64 {
 
+bool TranslatorVisitor::TRN1(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) {
+    if (!Q && size == 0b11) {
+        return ReservedValue();
+    }
+
+    const size_t datasize = Q ? 128 : 64;
+    const size_t esize = 8 << size.ZeroExtend();
+
+    const IR::U128 m = V(datasize, Vm);
+    const IR::U128 n = V(datasize, Vn);
+
+    const IR::U128 result = [&] {
+        switch (esize) {
+        case 8:
+        case 16:
+        case 32: {
+            // Create a mask of elements we care about (e.g. for 8-bit 0x00FF00FF00FF00FF)
+            const u64 mask_element = Common::Ones<u64>(esize);
+            const u64 mask_value = Common::Replicate<u64>(mask_element, esize * 2);
+
+            const IR::U128 mask = ir.VectorBroadcast(64, I(64, mask_value));
+            const IR::U128 anded_m = ir.VectorAnd(m, mask);
+            const IR::U128 anded_n = ir.VectorAnd(n, mask);
+            return ir.VectorOr(ir.VectorLogicalShiftLeft(esize * 2, anded_m, esize), anded_n);
+        }
+        case 64: {
+        default:
+            return ir.VectorSetElement(esize, n, 1, ir.VectorGetElement(esize, m, 0));
+        }
+        }
+    }();
+
+    V(datasize, Vd, result);
+    return true;
+}
+
 bool TranslatorVisitor::ZIP1(bool Q, Imm<2> size, Vec Vm, Vec Vn, Vec Vd) {
     if (size == 0b11 && !Q) {
         return ReservedValue();