1*9880d681SAndroid Build Coastguard Worker; RUN: llc < %s -march=arm64 -aarch64-neon-syntax=apple -mcpu=cyclone | FileCheck %s 2*9880d681SAndroid Build Coastguard Worker; RUN: llc < %s -march=arm64 -aarch64-neon-syntax=apple -O0 -mcpu=cyclone | FileCheck %s --check-prefix=CHECK-FAST 3*9880d681SAndroid Build Coastguard Worker 4*9880d681SAndroid Build Coastguard Workerdefine <16 x i8> @foo(<16 x i8> %a) nounwind optsize readnone ssp { 5*9880d681SAndroid Build Coastguard Worker; CHECK: uaddlv.16b h0, v0 6*9880d681SAndroid Build Coastguard Worker; CHECK: rshrn.8b v0, v0, #4 7*9880d681SAndroid Build Coastguard Worker; CHECK: dup.16b v0, v0[0] 8*9880d681SAndroid Build Coastguard Worker; CHECK: ret 9*9880d681SAndroid Build Coastguard Worker 10*9880d681SAndroid Build Coastguard Worker; CHECK-FAST: uaddlv.16b 11*9880d681SAndroid Build Coastguard Worker; CHECK-FAST: rshrn.8b 12*9880d681SAndroid Build Coastguard Worker; CHECK-FAST: dup.16b 13*9880d681SAndroid Build Coastguard Worker %tmp = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v16i8(<16 x i8> %a) nounwind 14*9880d681SAndroid Build Coastguard Worker %tmp1 = trunc i32 %tmp to i16 15*9880d681SAndroid Build Coastguard Worker %tmp2 = insertelement <8 x i16> undef, i16 %tmp1, i32 0 16*9880d681SAndroid Build Coastguard Worker %tmp3 = tail call <8 x i8> @llvm.aarch64.neon.rshrn.v8i8(<8 x i16> %tmp2, i32 4) 17*9880d681SAndroid Build Coastguard Worker %tmp4 = shufflevector <8 x i8> %tmp3, <8 x i8> undef, <16 x i32> zeroinitializer 18*9880d681SAndroid Build Coastguard Worker ret <16 x i8> %tmp4 19*9880d681SAndroid Build Coastguard Worker} 20*9880d681SAndroid Build Coastguard Worker 21*9880d681SAndroid Build Coastguard Workerdeclare <8 x i8> @llvm.aarch64.neon.rshrn.v8i8(<8 x i16>, i32) nounwind readnone 22*9880d681SAndroid Build Coastguard Workerdeclare i32 @llvm.aarch64.neon.uaddlv.i32.v16i8(<16 x i8>) nounwind readnone 23