[clang] [CIR][AArch64] Lower NEON vzip intrinsics (PR #193658)

[Andrzej Warzyński via cfe-commits]

================
@@ -2774,8 +2774,31 @@ CIRGenFunction::emitAArch64BuiltinExpr(unsigned builtinID, const CallExpr *expr,
   case NEON::BI__builtin_neon_vtrnq_v:
   case NEON::BI__builtin_neon_vuzp_v:
   case NEON::BI__builtin_neon_vuzpq_v:
+    cgm.errorNYI(expr->getSourceRange(),
+                 std::string("unimplemented AArch64 builtin call: ") +
+                     getContext().BuiltinInfo.getName(builtinID));
+    return mlir::Value{};
   case NEON::BI__builtin_neon_vzip_v:
-  case NEON::BI__builtin_neon_vzipq_v:
+  case NEON::BI__builtin_neon_vzipq_v: {
+    ops[1] = builder.createBitcast(ops[1], ty);
+    ops[2] = builder.createBitcast(ops[2], ty);
+    // Adding a bitcast here as Ops[0] might be a void pointer.
+    mlir::Value baseAddr =
+        builder.createBitcast(ops[0], builder.getPointerTo(ty));
+    mlir::Value sv;
+    for (unsigned vi = 0; vi != 2; ++vi) {
+      llvm::SmallVector<int64_t, 16> indices;
+      for (unsigned i = 0, e = ty.getSize(); i != e; i += 2) {
+        indices.push_back((i + vi * e) >> 1);
+        indices.push_back(((i + vi * e) >> 1) + e);
+      }
+      cir::ConstantOp idx = builder.getConstInt(loc, builder.getSInt32Ty(), vi);
+      mlir::Value addr = builder.createPtrStride(loc, baseAddr, idx);
+      sv = builder.createVecShuffle(loc, ops[1], ops[2], indices);
+      (void)builder.CIRBaseBuilderTy::createStore(loc, sv, addr);
----------------
banach-space wrote:

Thanks for checking - that's what I suspected, but wasn't sure.

https://github.com/llvm/llvm-project/pull/193658