36 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=instcombine -S | FileCheck %s3 4target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"5target triple = "armv8-arm-none-eabi"6 7; Turning a table lookup intrinsic into a shuffle vector instruction8; can be beneficial. If the mask used for the lookup is the constant9; vector {7,6,5,4,3,2,1,0}, then the back-end generates rev6410; instructions instead.11 12define <8 x i8> @tbl1_8x8(<8 x i8> %vec) {13; CHECK-LABEL: @tbl1_8x8(14; CHECK-NEXT: entry:15; CHECK-NEXT: [[TMP0:%.*]] = shufflevector <8 x i8> [[VEC:%.*]], <8 x i8> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>16; CHECK-NEXT: ret <8 x i8> [[TMP0]]17;18entry:19 %vtbl1 = call <8 x i8> @llvm.arm.neon.vtbl1(<8 x i8> %vec, <8 x i8> <i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)20 ret <8 x i8> %vtbl121}22 23; Bail the optimization if a mask index is out of range.24define <8 x i8> @tbl1_8x8_out_of_range(<8 x i8> %vec) {25; CHECK-LABEL: @tbl1_8x8_out_of_range(26; CHECK-NEXT: entry:27; CHECK-NEXT: [[VTBL1:%.*]] = call <8 x i8> @llvm.arm.neon.vtbl1(<8 x i8> [[VEC:%.*]], <8 x i8> <i8 8, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)28; CHECK-NEXT: ret <8 x i8> [[VTBL1]]29;30entry:31 %vtbl1 = call <8 x i8> @llvm.arm.neon.vtbl1(<8 x i8> %vec, <8 x i8> <i8 8, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)32 ret <8 x i8> %vtbl133}34 35declare <8 x i8> @llvm.arm.neon.vtbl1(<8 x i8>, <8 x i8>)36