1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -S -loop-vectorize -mtriple=powerpc64le-unknown-unknown \
3; RUN:   -force-target-max-vector-interleave=1 -mcpu=pwr9 < %s | FileCheck %s
4define dso_local void @test(i32* %Arr, i32 signext %Len) {
5; CHECK-LABEL: @test(
6; CHECK-NEXT:  entry:
7; CHECK-NEXT:    [[CMP1:%.*]] = icmp slt i32 0, [[LEN:%.*]]
8; CHECK-NEXT:    br i1 [[CMP1]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_COND_CLEANUP:%.*]]
9; CHECK:       for.body.lr.ph:
10; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[LEN]], 4
11; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
12; CHECK:       vector.ph:
13; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i32 [[LEN]], 4
14; CHECK-NEXT:    [[N_VEC:%.*]] = sub i32 [[LEN]], [[N_MOD_VF]]
15; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
16; CHECK:       vector.body:
17; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
18; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> undef, i32 [[INDEX]], i32 0
19; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> undef, <4 x i32> zeroinitializer
20; CHECK-NEXT:    [[INDUCTION:%.*]] = add <4 x i32> [[BROADCAST_SPLAT]], <i32 0, i32 1, i32 2, i32 3>
21; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
22; CHECK-NEXT:    [[TMP1:%.*]] = sext i32 [[TMP0]] to i64
23; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[ARR:%.*]], i64 [[TMP1]]
24; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[TMP2]], i32 0
25; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i32* [[TMP3]] to <4 x i32>*
26; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i32>, <4 x i32>* [[TMP4]], align 4
27; CHECK-NEXT:    [[TMP5:%.*]] = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> [[WIDE_LOAD]])
28; CHECK-NEXT:    [[TMP6:%.*]] = sext i32 [[TMP0]] to i64
29; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, i32* [[ARR]], i64 [[TMP6]]
30; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, i32* [[TMP7]], i32 0
31; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i32* [[TMP8]] to <4 x i32>*
32; CHECK-NEXT:    store <4 x i32> [[TMP5]], <4 x i32>* [[TMP9]], align 4
33; CHECK-NEXT:    [[INDEX_NEXT]] = add i32 [[INDEX]], 4
34; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
35; CHECK-NEXT:    br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !0
36; CHECK:       middle.block:
37; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 [[LEN]], [[N_VEC]]
38; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_COND_FOR_COND_CLEANUP_CRIT_EDGE:%.*]], label [[SCALAR_PH]]
39; CHECK:       scalar.ph:
40; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_LR_PH]] ]
41; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
42; CHECK:       for.cond.for.cond.cleanup_crit_edge:
43; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]
44; CHECK:       for.cond.cleanup:
45; CHECK-NEXT:    br label [[FOR_END:%.*]]
46; CHECK:       for.body:
47; CHECK-NEXT:    [[I_02:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_INC:%.*]] ]
48; CHECK-NEXT:    [[IDXPROM:%.*]] = sext i32 [[I_02]] to i64
49; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[ARR]], i64 [[IDXPROM]]
50; CHECK-NEXT:    [[TMP11:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
51; CHECK-NEXT:    [[TMP12:%.*]] = call i32 @llvm.bswap.i32(i32 [[TMP11]])
52; CHECK-NEXT:    [[IDXPROM1:%.*]] = sext i32 [[I_02]] to i64
53; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[ARR]], i64 [[IDXPROM1]]
54; CHECK-NEXT:    store i32 [[TMP12]], i32* [[ARRAYIDX2]], align 4
55; CHECK-NEXT:    br label [[FOR_INC]]
56; CHECK:       for.inc:
57; CHECK-NEXT:    [[INC]] = add nsw i32 [[I_02]], 1
58; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[INC]], [[LEN]]
59; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND_FOR_COND_CLEANUP_CRIT_EDGE]], !llvm.loop !2
60; CHECK:       for.end:
61; CHECK-NEXT:    ret void
62;
63entry:
64  %cmp1 = icmp slt i32 0, %Len
65  br i1 %cmp1, label %for.body.lr.ph, label %for.cond.cleanup
66
67for.body.lr.ph:                                   ; preds = %entry
68  br label %for.body
69
70for.cond.for.cond.cleanup_crit_edge:              ; preds = %for.inc
71  br label %for.cond.cleanup
72
73for.cond.cleanup:                                 ; preds = %for.cond.for.cond.cleanup_crit_edge, %entry
74  br label %for.end
75
76for.body:                                         ; preds = %for.body.lr.ph, %for.inc
77  %i.02 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.inc ]
78  %idxprom = sext i32 %i.02 to i64
79  %arrayidx = getelementptr inbounds i32, i32* %Arr, i64 %idxprom
80  %0 = load i32, i32* %arrayidx, align 4
81  %1 = call i32 @llvm.bswap.i32(i32 %0)
82  %idxprom1 = sext i32 %i.02 to i64
83  %arrayidx2 = getelementptr inbounds i32, i32* %Arr, i64 %idxprom1
84  store i32 %1, i32* %arrayidx2, align 4
85  br label %for.inc
86
87for.inc:                                          ; preds = %for.body
88  %inc = add nsw i32 %i.02, 1
89  %cmp = icmp slt i32 %inc, %Len
90  br i1 %cmp, label %for.body, label %for.cond.for.cond.cleanup_crit_edge
91
92for.end:                                          ; preds = %for.cond.cleanup
93  ret void
94}
95
96; Function Attrs: nounwind readnone speculatable willreturn
97declare i32 @llvm.bswap.i32(i32)
98