[llvm] [AArch64] Add @llvm.experimental.vector.match (PR #101974)

[Paul Walker via llvm-commits]

================
@@ -6364,6 +6364,91 @@ SDValue AArch64TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
         DAG.getNode(AArch64ISD::CTTZ_ELTS, dl, MVT::i64, CttzOp);
     return DAG.getZExtOrTrunc(NewCttzElts, dl, Op.getValueType());
   }
+  case Intrinsic::experimental_vector_match: {
+    SDValue ID =
+        DAG.getTargetConstant(Intrinsic::aarch64_sve_match, dl, MVT::i64);
+
+    auto Op1 = Op.getOperand(1);
+    auto Op2 = Op.getOperand(2);
+    auto Mask = Op.getOperand(3);
+
+    EVT Op1VT = Op1.getValueType();
+    EVT Op2VT = Op2.getValueType();
+    EVT ResVT = Op.getValueType();
+
+    assert((Op1VT.getVectorElementType() == MVT::i8 ||
+            Op1VT.getVectorElementType() == MVT::i16) &&
+           "Expected 8-bit or 16-bit characters.");
+    assert(Op1VT.getVectorElementType() == Op2VT.getVectorElementType() &&
+           "Operand type mismatch.");
+    assert(!Op2VT.isScalableVector() && "Search vector cannot be scalable.");
+
+    // Note: Currently Op1 needs to be v16i8, v8i16, or the scalable versions.
+    // In the future we could support other types (e.g. v8i8).
+    assert(Op1VT.getSizeInBits().getKnownMinValue() == 128 &&
+           "Unsupported first operand type.");
----------------
paulwalker-arm wrote:

By this point all types must be legal, so this assert shouldn't be needed?

Out of interest, do you know what is special about `v8i8`? From a high level the general flow of insert into an SVE reg -> match -> extract the result seems like it would just work.

https://github.com/llvm/llvm-project/pull/101974