[CIR][CIRGen][Builtin][Neon] Lower neon_vpadd_v and neon_vpaddq_v into llvm intrinsic (llvm#960)

This PR refactored Neon Built in code in clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp a bit to make it cleaner.

Also changed RUNOption of test file clang/test/CIR/CodeGen/AArch64/neon-arith.c to make test more concise, and easy to compare against OG (to compare, just remove -fclangir from llvm gen part of RUN, and the test should still pass)

Author: ghehg

clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp

@@ -3192,7 +3192,7 @@ static mlir::Value buildArmLdrexNon128Intrinsic(unsigned int builtinID,
   }
 }
 
-mlir::Value buildNeonCall(unsigned int builtinID, CIRGenFunction &cgf,
+mlir::Value buildNeonCall(CIRGenBuilderTy &builder,
                           llvm::SmallVector<mlir::Type> argTypes,
                           llvm::SmallVectorImpl<mlir::Value> &args,
                           llvm::StringRef intrinsicName, mlir::Type funcResTy,
@@ -3209,7 +3209,6 @@ mlir::Value buildNeonCall(unsigned int builtinID, CIRGenFunction &cgf,
   if (shift > 0)
     llvm_unreachable("Argument shift NYI");
 
-  CIRGenBuilderTy &builder = cgf.getBuilder();
   for (unsigned j = 0; j < argTypes.size(); ++j) {
     if (isConstrainedFPIntrinsic) {
       assert(!MissingFeatures::buildConstrainedFPCall());
@@ -3239,6 +3238,24 @@ static int64_t getIntValueFromConstOp(mlir::Value val) {
       .getSExtValue();
 }
 
+/// This function `buildCommonNeonCallPattern0` implements a common way
+//  to generate neon intrinsic call that has following pattern:
+//  1. There is a need to cast result of the intrinsic call back to
+//     expression type.
+//  2. Function arg types are given, not deduced from actual arg types.
+static mlir::Value
+buildCommonNeonCallPattern0(CIRGenFunction &cgf, std::string &intrincsName,
+                            llvm::SmallVector<mlir::Type> argTypes,
+                            llvm::SmallVectorImpl<mlir::Value> &ops,
+                            mlir::Type funcResTy, const clang::CallExpr *e) {
+  CIRGenBuilderTy &builder = cgf.getBuilder();
+  mlir::Value res =
+      buildNeonCall(builder, std::move(argTypes), ops, intrincsName, funcResTy,
+                    cgf.getLoc(e->getExprLoc()));
+  mlir::Type resultType = cgf.ConvertType(e->getType());
+  return builder.createBitcast(res, resultType);
+}
+
 mlir::Value CIRGenFunction::buildCommonNeonBuiltinExpr(
     unsigned builtinID, unsigned llvmIntrinsic, unsigned altLLVMIntrinsic,
     const char *nameHint, unsigned modifier, const CallExpr *e,
@@ -3303,18 +3320,25 @@ mlir::Value CIRGenFunction::buildCommonNeonBuiltinExpr(
   default:
     llvm::errs() << getAArch64SIMDIntrinsicString(builtinID) << " ";
     llvm_unreachable("NYI");
-  case NEON::BI__builtin_neon_vqadd_v:
-    mlir::Value res = buildNeonCall(builtinID, *this, {vTy, vTy}, ops,
-                                    (intrinicId != altLLVMIntrinsic)
-                                        ? "llvm.aarch64.neon.uqadd"
-                                        : "llvm.aarch64.neon.sqadd",
-                                    vTy, getLoc(e->getExprLoc()));
-    mlir::Type resultType = ConvertType(e->getType());
-    // AArch64 intrinsic one-element vector type cast to
-    // scalar type expected by the builtin
-    return builder.createBitcast(res, resultType);
+
+  case NEON::BI__builtin_neon_vpadd_v:
+  case NEON::BI__builtin_neon_vpaddq_v: {
+    std::string intrincsName = mlir::isa<mlir::FloatType>(vTy.getEltType())
+                                   ? "llvm.aarch64.neon.faddp"
+                                   : "llvm.aarch64.neon.addp";
+    return buildCommonNeonCallPattern0(*this, intrincsName, {vTy, vTy}, ops,
+                                       vTy, e);
     break;
   }
+  case NEON::BI__builtin_neon_vqadd_v: {
+    std::string intrincsName = (intrinicId != altLLVMIntrinsic)
+                                   ? "llvm.aarch64.neon.uqadd"
+                                   : "llvm.aarch64.neon.sqadd";
+    return buildCommonNeonCallPattern0(*this, intrincsName, {vTy, vTy}, ops,
+                                       vTy, e);
+    break;
+  }
+  }
   return nullptr;
 }