1 ; RUN: opt -S -passes=loop-vectorize -force-vector-width=4 -force-vector-interleave=1 < %s | FileCheck %s
3 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
6 ; Check that the need for overflow check prevents vectorizing a loop with tiny
7 ; trip count (which implies opt for size).
8 ; CHECK-LABEL: @func_34
9 ; CHECK-NOT: vector.scevcheck
10 ; CHECK-NOT: vector.body:
12 define void @func_34() {
17 %storemerge2 = phi i32 [ 0, %bb1 ], [ %_tmp2300, %bb67 ]
18 %sext = shl i32 %storemerge2, 16
19 %_tmp2299 = ashr exact i32 %sext, 16
20 %_tmp2300 = add nsw i32 %_tmp2299, 1
21 %_tmp2310 = trunc i32 %_tmp2300 to i16
22 %_tmp2312 = icmp slt i16 %_tmp2310, 3
23 br i1 %_tmp2312, label %bb67, label %bb68
29 ; Check that a loop under opt-for-size is vectorized, w/o checking for
31 ; NOTE: Some assertions have been autogenerated by utils/update_test_checks.py
32 define void @scev4stride1(ptr noalias nocapture %a, ptr noalias nocapture readonly %b, i32 %k) #0 {
33 ; CHECK-LABEL: @scev4stride1(
34 ; CHECK-NEXT: for.body.preheader:
35 ; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
37 ; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[K:%.*]], i64 0
38 ; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
39 ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
41 ; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
42 ; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
43 ; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 0
44 ; CHECK-NEXT: [[TMP4:%.*]] = mul nsw <4 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
45 ; CHECK-NEXT: [[TMP5:%.*]] = extractelement <4 x i32> [[TMP4]], i32 0
46 ; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i32 [[TMP5]]
47 ; CHECK-NEXT: [[TMP7:%.*]] = extractelement <4 x i32> [[TMP4]], i32 1
48 ; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[B]], i32 [[TMP7]]
49 ; CHECK-NEXT: [[TMP9:%.*]] = extractelement <4 x i32> [[TMP4]], i32 2
50 ; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[B]], i32 [[TMP9]]
51 ; CHECK-NEXT: [[TMP11:%.*]] = extractelement <4 x i32> [[TMP4]], i32 3
52 ; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i32, ptr [[B]], i32 [[TMP11]]
53 ; CHECK-NEXT: [[TMP13:%.*]] = load i32, ptr [[TMP6]], align 4
54 ; CHECK-NEXT: [[TMP14:%.*]] = load i32, ptr [[TMP8]], align 4
55 ; CHECK-NEXT: [[TMP15:%.*]] = load i32, ptr [[TMP10]], align 4
56 ; CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr [[TMP12]], align 4
57 ; CHECK-NEXT: [[TMP17:%.*]] = insertelement <4 x i32> poison, i32 [[TMP13]], i32 0
58 ; CHECK-NEXT: [[TMP18:%.*]] = insertelement <4 x i32> [[TMP17]], i32 [[TMP14]], i32 1
59 ; CHECK-NEXT: [[TMP19:%.*]] = insertelement <4 x i32> [[TMP18]], i32 [[TMP15]], i32 2
60 ; CHECK-NEXT: [[TMP20:%.*]] = insertelement <4 x i32> [[TMP19]], i32 [[TMP16]], i32 3
61 ; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 [[TMP0]]
62 ; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds i32, ptr [[TMP21]], i32 0
63 ; CHECK-NEXT: store <4 x i32> [[TMP20]], ptr [[TMP22]], align 4
64 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
65 ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
66 ; CHECK-NEXT: [[TMP24:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1024
67 ; CHECK-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !0
68 ; CHECK: middle.block:
69 ; CHECK-NEXT: br i1 true, label [[FOR_END_LOOPEXIT:%.*]], label [[SCALAR_PH]]
72 ; CHECK: for.end.loopexit:
73 ; CHECK-NEXT: ret void
79 %i.07 = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]
80 %mul = mul nsw i32 %i.07, %k
81 %arrayidx = getelementptr inbounds i32, ptr %b, i32 %mul
82 %0 = load i32, ptr %arrayidx, align 4
83 %arrayidx1 = getelementptr inbounds i32, ptr %a, i32 %i.07
84 store i32 %0, ptr %arrayidx1, align 4
85 %inc = add nuw nsw i32 %i.07, 1
86 %exitcond = icmp eq i32 %inc, 1024
87 br i1 %exitcond, label %for.end.loopexit, label %for.body
93 attributes #0 = { optsize }