[CIR][CIRGen][Builtin][Neon] Lower neon_vaddvq_u32 and neon_vaddvq_u64 (#1225)

Author: ghehg

clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp

@@ -2654,6 +2654,7 @@ static mlir::Value emitCommonNeonSISDBuiltinExpr(
   unsigned builtinID = info.BuiltinID;
   clang::CIRGen::CIRGenBuilderTy &builder = cgf.getBuilder();
   mlir::Type resultTy = cgf.convertType(expr->getType());
+  mlir::Type argTy = cgf.convertType(expr->getArg(0)->getType());
   mlir::Location loc = cgf.getLoc(expr->getExprLoc());
 
   switch (builtinID) {
@@ -2672,11 +2673,9 @@ static mlir::Value emitCommonNeonSISDBuiltinExpr(
     llvm_unreachable(" neon_vaddlv_u32 NYI ");
   case NEON::BI__builtin_neon_vaddlvq_s32:
     llvm_unreachable(" neon_vaddlvq_s32 NYI ");
-  case NEON::BI__builtin_neon_vaddlvq_u32: {
-    mlir::Type argTy = cgf.convertType(expr->getArg(0)->getType());
+  case NEON::BI__builtin_neon_vaddlvq_u32:
     return emitNeonCall(builder, {argTy}, ops, "aarch64.neon.uaddlv", resultTy,
                         loc);
-  }
   case NEON::BI__builtin_neon_vaddv_f32:
     llvm_unreachable(" neon_vaddv_f32 NYI ");
   case NEON::BI__builtin_neon_vaddv_s32:
@@ -2692,9 +2691,9 @@ static mlir::Value emitCommonNeonSISDBuiltinExpr(
   case NEON::BI__builtin_neon_vaddvq_s64:
     llvm_unreachable(" neon_vaddvq_s64 NYI ");
   case NEON::BI__builtin_neon_vaddvq_u32:
-    llvm_unreachable(" neon_vaddvq_u32 NYI ");
   case NEON::BI__builtin_neon_vaddvq_u64:
-    llvm_unreachable(" neon_vaddvq_u64 NYI ");
+    return emitNeonCall(builder, {argTy}, ops, "aarch64.neon.uaddv", resultTy,
+                        loc);
   case NEON::BI__builtin_neon_vcaged_f64:
     llvm_unreachable(" neon_vcaged_f64 NYI ");
   case NEON::BI__builtin_neon_vcages_f32:

clang/test/CIR/CodeGen/AArch64/neon-arith.c

@@ -919,3 +919,25 @@ int16_t test_vaddv_s16(int16x4_t a) {
   // LLVM-NEXT:    [[TMP0:%.*]] = trunc i32 [[VADDV_I]] to i16
   // LLVM-NEXT:    ret i16 [[TMP0]]
 }
+
+uint32_t test_vaddvq_u32(uint32x4_t a) {
+  return vaddvq_u32(a);
+
+  // CIR-LABEL: vaddvq_u32
+  // CIR: cir.llvm.intrinsic "aarch64.neon.uaddv" {{%.*}} : (!cir.vector<!u32i x 4>) -> !u32i
+
+  // LLVM-LABEL: test_vaddvq_u32
+  // LLVM:   [[VADDVQ_U32_I:%.*]] = call i32 @llvm.aarch64.neon.uaddv.i32.v4i32(<4 x i32> {{%.*}})
+  // LLVM:   ret i32 [[VADDVQ_U32_I]]
+}
+
+uint64_t test_vaddvq_u64(uint64x2_t a) {
+  return vaddvq_u64(a);
+
+  // CIR-LABEL: vaddvq_u64
+  // CIR: cir.llvm.intrinsic "aarch64.neon.uaddv" {{%.*}} : (!cir.vector<!u64i x 2>) -> !u64i
+
+  // LLVM-LABEL: test_vaddvq_u64
+  // LLVM:   [[VADDVQ_U64_I:%.*]] = call i64 @llvm.aarch64.neon.uaddv.i64.v2i64(<2 x i64> {{%.*}})
+  // LLVM:   ret i64 [[VADDVQ_U64_I]]
+}