112 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -S -passes=slp-vectorizer -pass-remarks-output=%t | FileCheck %s3; RUN: cat %t | FileCheck -check-prefix=REMARK %s4; RUN: opt < %s -S -aa-pipeline=basic-aa -passes=slp-vectorizer -pass-remarks-output=%t | FileCheck %s5; RUN: cat %t | FileCheck -check-prefix=REMARK %s6 7target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"8target triple = "aarch64--linux-gnu"9 10; REMARK-LABEL: Function: gather_multiple_use11; REMARK: Args:12; REMARK-NEXT: - String: 'Vectorized horizontal reduction with cost '13; REMARK-NEXT: - Cost: '-8'14;15; REMARK-NOT: Function: gather_load16 17define internal i32 @gather_multiple_use(i32 %a, i32 %b, i32 %c, i32 %d) {18; CHECK-LABEL: @gather_multiple_use(19; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> poison, i32 [[C:%.*]], i32 020; CHECK-NEXT: [[TMP2:%.*]] = insertelement <4 x i32> [[TMP1]], i32 [[A:%.*]], i32 121; CHECK-NEXT: [[TMP3:%.*]] = insertelement <4 x i32> [[TMP2]], i32 [[B:%.*]], i32 222; CHECK-NEXT: [[TMP4:%.*]] = insertelement <4 x i32> [[TMP3]], i32 [[D:%.*]], i32 323; CHECK-NEXT: [[TMP5:%.*]] = lshr <4 x i32> [[TMP4]], splat (i32 15)24; CHECK-NEXT: [[TMP6:%.*]] = and <4 x i32> [[TMP5]], splat (i32 65537)25; CHECK-NEXT: [[TMP7:%.*]] = mul nuw <4 x i32> [[TMP6]], splat (i32 65535)26; CHECK-NEXT: [[TMP8:%.*]] = add <4 x i32> [[TMP7]], [[TMP4]]27; CHECK-NEXT: [[TMP9:%.*]] = xor <4 x i32> [[TMP8]], [[TMP7]]28; CHECK-NEXT: [[TMP10:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP9]])29; CHECK-NEXT: ret i32 [[TMP10]]30;31 %tmp00 = lshr i32 %a, 1532 %tmp01 = and i32 %tmp00, 6553733 %tmp02 = mul nuw i32 %tmp01, 6553534 %tmp03 = add i32 %tmp02, %a35 %tmp04 = xor i32 %tmp03, %tmp0236 %tmp05 = lshr i32 %c, 1537 %tmp06 = and i32 %tmp05, 6553738 %tmp07 = mul nuw i32 %tmp06, 6553539 %tmp08 = add i32 %tmp07, %c40 %tmp09 = xor i32 %tmp08, %tmp0741 %tmp10 = lshr i32 %b, 1542 %tmp11 = and i32 %tmp10, 6553743 %tmp12 = mul nuw i32 %tmp11, 6553544 %tmp13 = add i32 %tmp12, %b45 %tmp14 = xor i32 %tmp13, %tmp1246 %tmp15 = lshr i32 %d, 1547 %tmp16 = and i32 %tmp15, 6553748 %tmp17 = mul nuw i32 %tmp16, 6553549 %tmp18 = add i32 %tmp17, %d50 %tmp19 = xor i32 %tmp18, %tmp1751 %tmp20 = add i32 %tmp09, %tmp0452 %tmp21 = add i32 %tmp20, %tmp1453 %tmp22 = add i32 %tmp21, %tmp1954 ret i32 %tmp2255}56 57@data = global [6 x [258 x i8]] zeroinitializer, align 158define void @gather_load(ptr noalias %ptr) {59; CHECK-LABEL: @gather_load(60; CHECK-NEXT: [[ARRAYIDX182:%.*]] = getelementptr inbounds i16, ptr [[PTR:%.*]], i64 161; CHECK-NEXT: [[ARRAYIDX183:%.*]] = getelementptr inbounds i16, ptr [[PTR]], i64 262; CHECK-NEXT: [[ARRAYIDX184:%.*]] = getelementptr inbounds i16, ptr [[PTR]], i64 363; CHECK-NEXT: [[ARRAYIDX185:%.*]] = getelementptr inbounds i16, ptr [[PTR]], i64 464; CHECK-NEXT: [[ARRAYIDX149:%.*]] = getelementptr inbounds [6 x [258 x i8]], ptr @data, i64 0, i64 1, i64 065; CHECK-NEXT: [[L0:%.*]] = load i8, ptr [[ARRAYIDX149]], align 166; CHECK-NEXT: [[CONV150:%.*]] = zext i8 [[L0]] to i1667; CHECK-NEXT: [[ADD152:%.*]] = add i16 10, [[CONV150]]68; CHECK-NEXT: [[ARRAYIDX155:%.*]] = getelementptr inbounds [6 x [258 x i8]], ptr @data, i64 0, i64 2, i64 169; CHECK-NEXT: [[L1:%.*]] = load i8, ptr [[ARRAYIDX155]], align 170; CHECK-NEXT: [[CONV156:%.*]] = zext i8 [[L1]] to i1671; CHECK-NEXT: [[ADD158:%.*]] = add i16 20, [[CONV156]]72; CHECK-NEXT: [[ARRAYIDX161:%.*]] = getelementptr inbounds [6 x [258 x i8]], ptr @data, i64 0, i64 3, i64 273; CHECK-NEXT: [[L2:%.*]] = load i8, ptr [[ARRAYIDX161]], align 174; CHECK-NEXT: [[CONV162:%.*]] = zext i8 [[L2]] to i1675; CHECK-NEXT: [[ADD164:%.*]] = add i16 30, [[CONV162]]76; CHECK-NEXT: [[ARRAYIDX167:%.*]] = getelementptr inbounds [6 x [258 x i8]], ptr @data, i64 0, i64 4, i64 377; CHECK-NEXT: [[L3:%.*]] = load i8, ptr [[ARRAYIDX167]], align 178; CHECK-NEXT: [[CONV168:%.*]] = zext i8 [[L3]] to i1679; CHECK-NEXT: [[ADD170:%.*]] = add i16 40, [[CONV168]]80; CHECK-NEXT: store i16 [[ADD152]], ptr [[ARRAYIDX182]], align 281; CHECK-NEXT: store i16 [[ADD158]], ptr [[ARRAYIDX183]], align 282; CHECK-NEXT: store i16 [[ADD164]], ptr [[ARRAYIDX184]], align 283; CHECK-NEXT: store i16 [[ADD170]], ptr [[ARRAYIDX185]], align 284; CHECK-NEXT: ret void85;86 %arrayidx182 = getelementptr inbounds i16, ptr %ptr, i64 187 %arrayidx183 = getelementptr inbounds i16, ptr %ptr, i64 288 %arrayidx184 = getelementptr inbounds i16, ptr %ptr, i64 389 %arrayidx185 = getelementptr inbounds i16, ptr %ptr, i64 490 %arrayidx149 = getelementptr inbounds [6 x [258 x i8]], ptr @data, i64 0, i64 1, i64 091 %l0 = load i8, ptr %arrayidx149, align 192 %conv150 = zext i8 %l0 to i1693 %add152 = add i16 10, %conv15094 %arrayidx155 = getelementptr inbounds [6 x [258 x i8]], ptr @data, i64 0, i64 2, i64 195 %l1 = load i8, ptr %arrayidx155, align 196 %conv156 = zext i8 %l1 to i1697 %add158 = add i16 20, %conv15698 %arrayidx161 = getelementptr inbounds [6 x [258 x i8]], ptr @data, i64 0, i64 3, i64 299 %l2 = load i8, ptr %arrayidx161, align 1100 %conv162 = zext i8 %l2 to i16101 %add164 = add i16 30, %conv162102 %arrayidx167 = getelementptr inbounds [6 x [258 x i8]], ptr @data, i64 0, i64 4, i64 3103 %l3 = load i8, ptr %arrayidx167, align 1104 %conv168 = zext i8 %l3 to i16105 %add170 = add i16 40, %conv168106 store i16 %add152, ptr %arrayidx182, align 2107 store i16 %add158, ptr %arrayidx183, align 2108 store i16 %add164, ptr %arrayidx184, align 2109 store i16 %add170, ptr %arrayidx185, align 2110 ret void111}112