1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -loop-vectorize -force-vector-width=4 -mtriple=thumbv8.1m.main-none-none-eabi -mattr=+mve.fp -enable-arm-maskedgatscat -tail-predication=force-enabled -S %s -o - | FileCheck %s
3
4define void @test_stride1_4i32(i32* readonly %data, i32* noalias nocapture %dst, i32 %n) {
5; CHECK-LABEL: @test_stride1_4i32(
6; CHECK:       vector.body:
7; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY:%.*]] ]
8; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i32> undef, i32 [[INDEX]], i32 0
9; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT1:%.*]], <4 x i32> undef, <4 x i32> zeroinitializer
10; CHECK-NEXT:    [[INDUCTION:%.*]] = add <4 x i32> [[BROADCAST_SPLAT2]], <i32 0, i32 1, i32 2, i32 3>
11; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
12; CHECK-NEXT:    [[TMP1:%.*]] = mul nuw nsw i32 [[TMP0]], 1
13; CHECK-NEXT:    [[TMP2:%.*]] = add nuw nsw i32 [[TMP1]], 2
14; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[DATA:%.*]], i32 [[TMP2]]
15; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <4 x i32> [[BROADCAST_SPLAT:%.*]], i32 0
16; CHECK-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 [[TMP0]], i32 [[TMP4]])
17; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[TMP3]], i32 0
18; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32* [[TMP5]] to <4 x i32>*
19; CHECK-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>* [[TMP6]], i32 4, <4 x i1> [[ACTIVE_LANE_MASK]], <4 x i32> undef)
20; CHECK-NEXT:    [[TMP7:%.*]] = add nsw <4 x i32> <i32 5, i32 5, i32 5, i32 5>, [[WIDE_MASKED_LOAD]]
21; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, i32* [[DST:%.*]], i32 [[TMP0]]
22; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, i32* [[TMP8]], i32 0
23; CHECK-NEXT:    [[TMP10:%.*]] = bitcast i32* [[TMP9]] to <4 x i32>*
24; CHECK-NEXT:    call void @llvm.masked.store.v4i32.p0v4i32(<4 x i32> [[TMP7]], <4 x i32>* [[TMP10]], i32 4, <4 x i1> [[ACTIVE_LANE_MASK]])
25; CHECK-NEXT:    [[INDEX_NEXT]] = add i32 [[INDEX]], 4
26; CHECK-NEXT:    [[TMP11:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC:%.*]]
27; CHECK-NEXT:    br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !0
28;
29entry:
30  br label %for.body
31for.body:                                         ; preds = %for.body.preheader, %for.body
32  %i.023 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
33  %mul = mul nuw nsw i32 %i.023, 1
34  %add5 = add nuw nsw i32 %mul, 2
35  %arrayidx6 = getelementptr inbounds i32, i32* %data, i32 %add5
36  %0 = load i32, i32* %arrayidx6, align 4
37  %add7 = add nsw i32 5, %0
38  %arrayidx9 = getelementptr inbounds i32, i32* %dst, i32 %i.023
39  store i32 %add7, i32* %arrayidx9, align 4
40  %inc = add nuw nsw i32 %i.023, 1
41  %exitcond.not = icmp eq i32 %inc, %n
42  br i1 %exitcond.not, label %end, label %for.body
43end:                                 ; preds = %end, %entry
44  ret void
45}
46
47define void @test_stride-1_4i32(i32* readonly %data, i32* noalias nocapture %dst, i32 %n) {
48; CHECK-LABEL: @test_stride-1_4i32(
49; CHECK:       vector.body:
50; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY:%.*]] ]
51; CHECK-NEXT:    [[TMP8:%.*]] = add i32 [[INDEX]], 0
52; CHECK-NEXT:    [[TMP9:%.*]] = mul nuw nsw i32 [[TMP8]], -1
53; CHECK-NEXT:    [[TMP10:%.*]] = add nuw nsw i32 [[TMP9]], 2
54; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i32, i32* [[DATA:%.*]], i32 [[TMP10]]
55; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[TMP11]], i32 0
56; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i32, i32* [[TMP12]], i32 -3
57; CHECK-NEXT:    [[TMP14:%.*]] = bitcast i32* [[TMP13]] to <4 x i32>*
58; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i32>, <4 x i32>* [[TMP14]], align 4
59; CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i32> [[WIDE_LOAD]], <4 x i32> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
60; CHECK-NEXT:    [[TMP15:%.*]] = add nsw <4 x i32> <i32 5, i32 5, i32 5, i32 5>, [[REVERSE]]
61; CHECK-NEXT:    [[TMP16:%.*]] = getelementptr inbounds i32, i32* [[DST:%.*]], i32 [[TMP8]]
62; CHECK-NEXT:    [[TMP17:%.*]] = getelementptr inbounds i32, i32* [[TMP16]], i32 0
63; CHECK-NEXT:    [[TMP18:%.*]] = bitcast i32* [[TMP17]] to <4 x i32>*
64; CHECK-NEXT:    store <4 x i32> [[TMP15]], <4 x i32>* [[TMP18]], align 4
65; CHECK-NEXT:    [[INDEX_NEXT]] = add i32 [[INDEX]], 4
66; CHECK-NEXT:    [[TMP19:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC:%.*]]
67; CHECK-NEXT:    br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !4
68;
69entry:
70  br label %for.body
71for.body:                                         ; preds = %for.body.preheader, %for.body
72  %i.023 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
73  %mul = mul nuw nsw i32 %i.023, -1
74  %add5 = add nuw nsw i32 %mul, 2
75  %arrayidx6 = getelementptr inbounds i32, i32* %data, i32 %add5
76  %0 = load i32, i32* %arrayidx6, align 4
77  %add7 = add nsw i32 5, %0
78  %arrayidx9 = getelementptr inbounds i32, i32* %dst, i32 %i.023
79  store i32 %add7, i32* %arrayidx9, align 4
80  %inc = add nuw nsw i32 %i.023, 1
81  %exitcond.not = icmp eq i32 %inc, %n
82  br i1 %exitcond.not, label %end, label %for.body
83end:                                 ; preds = %end, %entry
84  ret void
85}
86
87define void @test_stride2_4i32(i32* readonly %data, i32* noalias nocapture %dst, i32 %n) {
88;
89; CHECK-LABEL: @test_stride2_4i32(
90; CHECK:       vector.body:
91; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY:%.*]] ]
92; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[INDEX]], 0
93; CHECK-NEXT:    [[TMP3:%.*]] = mul nuw nsw i32 [[TMP2]], 2
94; CHECK-NEXT:    [[TMP4:%.*]] = add nuw nsw i32 [[TMP3]], 2
95; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, i32* [[DATA:%.*]], i32 [[TMP4]]
96; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[TMP5]], i32 0
97; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32* [[TMP6]] to <8 x i32>*
98; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i32>, <8 x i32>* [[TMP7]], align 4
99; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
100; CHECK-NEXT:    [[TMP8:%.*]] = add nsw <4 x i32> <i32 5, i32 5, i32 5, i32 5>, [[STRIDED_VEC]]
101; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, i32* [[DST:%.*]], i32 [[TMP2]]
102; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[TMP9]], i32 0
103; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32* [[TMP10]] to <4 x i32>*
104; CHECK-NEXT:    store <4 x i32> [[TMP8]], <4 x i32>* [[TMP11]], align 4
105; CHECK-NEXT:    [[INDEX_NEXT]] = add i32 [[INDEX]], 4
106; CHECK-NEXT:    [[TMP12:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC:%.*]]
107; CHECK-NEXT:    br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !6
108;
109entry:
110  br label %for.body
111for.body:                                         ; preds = %for.body.preheader, %for.body
112  %i.023 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
113  %mul = mul nuw nsw i32 %i.023, 2
114  %add5 = add nuw nsw i32 %mul, 2
115  %arrayidx6 = getelementptr inbounds i32, i32* %data, i32 %add5
116  %0 = load i32, i32* %arrayidx6, align 4
117  %add7 = add nsw i32 5, %0
118  %arrayidx9 = getelementptr inbounds i32, i32* %dst, i32 %i.023
119  store i32 %add7, i32* %arrayidx9, align 4
120  %inc = add nuw nsw i32 %i.023, 1
121  %exitcond.not = icmp eq i32 %inc, %n
122  br i1 %exitcond.not, label %end, label %for.body
123end:                                 ; preds = %end, %entry
124  ret void
125}
126
127define void @test_stride3_4i32(i32* readonly %data, i32* noalias nocapture %dst, i32 %n) {
128; CHECK-LABEL: @test_stride3_4i32(
129; CHECK:       vector.body:
130; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY:%.*]] ]
131; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
132; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
133; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[INDEX]], 1
134; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[INDEX]], 2
135; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[INDEX]], 3
136; CHECK-NEXT:    [[TMP4:%.*]] = mul nuw nsw <4 x i32> [[VEC_IND]], <i32 3, i32 3, i32 3, i32 3>
137; CHECK-NEXT:    [[TMP5:%.*]] = add nuw nsw <4 x i32> [[TMP4]], <i32 2, i32 2, i32 2, i32 2>
138; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[DATA:%.*]], <4 x i32> [[TMP5]]
139; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x i32> [[BROADCAST_SPLAT:%.*]], i32 0
140; CHECK-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 [[TMP0]], i32 [[TMP7]])
141; CHECK-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0i32(<4 x i32*> [[TMP6]], i32 4, <4 x i1> [[ACTIVE_LANE_MASK]], <4 x i32> undef)
142; CHECK-NEXT:    [[TMP8:%.*]] = add nsw <4 x i32> <i32 5, i32 5, i32 5, i32 5>, [[WIDE_MASKED_GATHER]]
143; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, i32* [[DST:%.*]], i32 [[TMP0]]
144; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[TMP9]], i32 0
145; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32* [[TMP10]] to <4 x i32>*
146; CHECK-NEXT:    call void @llvm.masked.store.v4i32.p0v4i32(<4 x i32> [[TMP8]], <4 x i32>* [[TMP11]], i32 4, <4 x i1> [[ACTIVE_LANE_MASK]])
147; CHECK-NEXT:    [[INDEX_NEXT]] = add i32 [[INDEX]], 4
148; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
149; CHECK-NEXT:    [[TMP12:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC:%.*]]
150; CHECK-NEXT:    br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !8
151;
152entry:
153  br label %for.body
154for.body:                                         ; preds = %for.body.preheader, %for.body
155  %i.023 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
156  %mul = mul nuw nsw i32 %i.023, 3
157  %add5 = add nuw nsw i32 %mul, 2
158  %arrayidx6 = getelementptr inbounds i32, i32* %data, i32 %add5
159  %0 = load i32, i32* %arrayidx6, align 4
160  %add7 = add nsw i32 5, %0
161  %arrayidx9 = getelementptr inbounds i32, i32* %dst, i32 %i.023
162  store i32 %add7, i32* %arrayidx9, align 4
163  %inc = add nuw nsw i32 %i.023, 1
164  %exitcond.not = icmp eq i32 %inc, %n
165  br i1 %exitcond.not, label %end, label %for.body
166end:                                 ; preds = %end, %entry
167  ret void
168}
169
170define void @test_stride4_4i32(i32* readonly %data, i32* noalias nocapture %dst, i32 %n) {
171; CHECK-LABEL: @test_stride4_4i32(
172; CHECK:       vector.body:
173; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY:%.*]] ]
174; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
175; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
176; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[INDEX]], 1
177; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[INDEX]], 2
178; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[INDEX]], 3
179; CHECK-NEXT:    [[TMP4:%.*]] = mul nuw nsw <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
180; CHECK-NEXT:    [[TMP5:%.*]] = add nuw nsw <4 x i32> [[TMP4]], <i32 2, i32 2, i32 2, i32 2>
181; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[DATA:%.*]], <4 x i32> [[TMP5]]
182; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x i32> [[BROADCAST_SPLAT:%.*]], i32 0
183; CHECK-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 [[TMP0]], i32 [[TMP7]])
184; CHECK-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0i32(<4 x i32*> [[TMP6]], i32 4, <4 x i1> [[ACTIVE_LANE_MASK]], <4 x i32> undef)
185; CHECK-NEXT:    [[TMP8:%.*]] = add nsw <4 x i32> <i32 5, i32 5, i32 5, i32 5>, [[WIDE_MASKED_GATHER]]
186; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, i32* [[DST:%.*]], i32 [[TMP0]]
187; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[TMP9]], i32 0
188; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32* [[TMP10]] to <4 x i32>*
189; CHECK-NEXT:    call void @llvm.masked.store.v4i32.p0v4i32(<4 x i32> [[TMP8]], <4 x i32>* [[TMP11]], i32 4, <4 x i1> [[ACTIVE_LANE_MASK]])
190; CHECK-NEXT:    [[INDEX_NEXT]] = add i32 [[INDEX]], 4
191; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
192; CHECK-NEXT:    [[TMP12:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC:%.*]]
193; CHECK-NEXT:    br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP10:!llvm.loop !.*]]
194;
195entry:
196  br label %for.body
197for.body:                                         ; preds = %for.body.preheader, %for.body
198  %i.023 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
199  %mul = mul nuw nsw i32 %i.023, 4
200  %add5 = add nuw nsw i32 %mul, 2
201  %arrayidx6 = getelementptr inbounds i32, i32* %data, i32 %add5
202  %0 = load i32, i32* %arrayidx6, align 4
203  %add7 = add nsw i32 5, %0
204  %arrayidx9 = getelementptr inbounds i32, i32* %dst, i32 %i.023
205  store i32 %add7, i32* %arrayidx9, align 4
206  %inc = add nuw nsw i32 %i.023, 1
207  %exitcond.not = icmp eq i32 %inc, %n
208  br i1 %exitcond.not, label %end, label %for.body
209end:                                 ; preds = %end, %entry
210  ret void
211}
212
213define void @test_stride_loopinvar_4i32(i32* readonly %data, i32* noalias nocapture %dst, i32 %n, i32 %stride) {
214; CHECK-LABEL: @test_stride_loopinvar_4i32(
215; CHECK:       vector.body:
216; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY:%.*]] ]
217; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i32> undef, i32 [[INDEX]], i32 0
218; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT1]], <4 x i32> undef, <4 x i32> zeroinitializer
219; CHECK-NEXT:    [[INDUCTION:%.*]] = add <4 x i32> [[BROADCAST_SPLAT2]], <i32 0, i32 1, i32 2, i32 3>
220; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[INDEX]], 0
221; CHECK-NEXT:    [[TMP2:%.*]] = mul nuw nsw i32 [[TMP1]], [[STRIDE:%.*]]
222; CHECK-NEXT:    [[TMP3:%.*]] = add nuw nsw i32 [[TMP2]], 2
223; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, i32* [[DATA:%.*]], i32 [[TMP3]]
224; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x i32> [[BROADCAST_SPLAT]], i32 0
225; CHECK-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 [[TMP1]], i32 [[TMP5]])
226; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, i32* [[TMP4]], i32 0
227; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i32* [[TMP6]] to <4 x i32>*
228; CHECK-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>* [[TMP7]], i32 4, <4 x i1> [[ACTIVE_LANE_MASK]], <4 x i32> undef)
229; CHECK-NEXT:    [[TMP8:%.*]] = add nsw <4 x i32> <i32 5, i32 5, i32 5, i32 5>, [[WIDE_MASKED_LOAD]]
230; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, i32* [[DST:%.*]], i32 [[TMP1]]
231; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[TMP9]], i32 0
232; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32* [[TMP10]] to <4 x i32>*
233; CHECK-NEXT:    call void @llvm.masked.store.v4i32.p0v4i32(<4 x i32> [[TMP8]], <4 x i32>* [[TMP11]], i32 4, <4 x i1> [[ACTIVE_LANE_MASK]])
234; CHECK-NEXT:    [[INDEX_NEXT]] = add i32 [[INDEX]], 4
235; CHECK-NEXT:    [[TMP12:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC:%.*]]
236; CHECK-NEXT:    br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP12:!llvm.loop !.*]]
237;
238entry:
239  br label %for.body
240for.body:                                         ; preds = %for.body.preheader, %for.body
241  %i.023 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
242  %mul = mul nuw nsw i32 %i.023, %stride
243  %add5 = add nuw nsw i32 %mul, 2
244  %arrayidx6 = getelementptr inbounds i32, i32* %data, i32 %add5
245  %0 = load i32, i32* %arrayidx6, align 4
246  %add7 = add nsw i32 5, %0
247  %arrayidx9 = getelementptr inbounds i32, i32* %dst, i32 %i.023
248  store i32 %add7, i32* %arrayidx9, align 4
249  %inc = add nuw nsw i32 %i.023, 1
250  %exitcond.not = icmp eq i32 %inc, %n
251  br i1 %exitcond.not, label %end, label %for.body
252end:                                 ; preds = %end, %entry
253  ret void
254}
255
256define void @test_stride_noninvar_4i32(i32* readonly %data, i32* noalias nocapture %dst, i32 %n) {
257; CHECK-LABEL: @test_stride_noninvar_4i32(
258; CHECK:       vector.body:
259; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY:%.*]] ]
260; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
261; CHECK-NEXT:    [[VEC_IND2:%.*]] = phi <4 x i32> [ <i32 3, i32 11, i32 19, i32 27>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT3:%.*]], [[VECTOR_BODY]] ]
262; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[INDEX]], 0
263; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[INDEX]], 1
264; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[INDEX]], 2
265; CHECK-NEXT:    [[TMP4:%.*]] = add i32 [[INDEX]], 3
266; CHECK-NEXT:    [[TMP5:%.*]] = mul nuw nsw <4 x i32> [[VEC_IND]], [[VEC_IND2]]
267; CHECK-NEXT:    [[TMP6:%.*]] = add nuw nsw <4 x i32> [[TMP5]], <i32 2, i32 2, i32 2, i32 2>
268; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, i32* [[DATA:%.*]], <4 x i32> [[TMP6]]
269; CHECK-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0i32(<4 x i32*> [[TMP7]], i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)
270; CHECK-NEXT:    [[TMP8:%.*]] = add nsw <4 x i32> <i32 5, i32 5, i32 5, i32 5>, [[WIDE_MASKED_GATHER]]
271; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, i32* [[DST:%.*]], i32 [[TMP1]]
272; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, i32* [[TMP9]], i32 0
273; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i32* [[TMP10]] to <4 x i32>*
274; CHECK-NEXT:    store <4 x i32> [[TMP8]], <4 x i32>* [[TMP11]], align 4
275; CHECK-NEXT:    [[INDEX_NEXT]] = add i32 [[INDEX]], 4
276; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
277; CHECK-NEXT:    [[VEC_IND_NEXT3]] = add <4 x i32> [[VEC_IND2]], <i32 32, i32 32, i32 32, i32 32>
278; CHECK-NEXT:    [[TMP12:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC:%.*]]
279; CHECK-NEXT:    br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP14:!llvm.loop !.*]]
280;
281entry:
282  br label %for.body
283for.body:                                         ; preds = %for.body.preheader, %for.body
284  %i.023 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
285  %stride = phi i32 [ %next.stride, %for.body ], [ 3, %entry ]
286  %mul = mul nuw nsw i32 %i.023, %stride
287  %add5 = add nuw nsw i32 %mul, 2
288  %arrayidx6 = getelementptr inbounds i32, i32* %data, i32 %add5
289  %0 = load i32, i32* %arrayidx6, align 4
290  %add7 = add nsw i32 5, %0
291  %arrayidx9 = getelementptr inbounds i32, i32* %dst, i32 %i.023
292  store i32 %add7, i32* %arrayidx9, align 4
293  %inc = add nuw nsw i32 %i.023, 1
294  %next.stride = add nuw nsw i32 %stride, 8
295  %exitcond.not = icmp eq i32 %inc, %n
296  br i1 %exitcond.not, label %end, label %for.body
297end:                                 ; preds = %end, %entry
298  ret void
299}
300
301define void @test_stride_noninvar2_4i32(i32* readonly %data, i32* noalias nocapture %dst, i32 %n) {
302; CHECK-LABEL: @test_stride_noninvar2_4i32(
303; CHECK-NEXT:  entry:
304; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
305; CHECK:       for.body:
306; CHECK-NEXT:    [[I_023:%.*]] = phi i32 [ [[INC:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]
307; CHECK-NEXT:    [[STRIDE:%.*]] = phi i32 [ [[NEXT_STRIDE:%.*]], [[FOR_BODY]] ], [ 3, [[ENTRY]] ]
308; CHECK-NEXT:    [[MUL:%.*]] = mul nuw nsw i32 [[I_023]], [[STRIDE]]
309; CHECK-NEXT:    [[ADD5:%.*]] = add nuw nsw i32 [[MUL]], 2
310; CHECK-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, i32* [[DATA:%.*]], i32 [[ADD5]]
311; CHECK-NEXT:    [[TMP0:%.*]] = load i32, i32* [[ARRAYIDX6]], align 4
312; CHECK-NEXT:    [[ADD7:%.*]] = add nsw i32 5, [[TMP0]]
313; CHECK-NEXT:    [[ARRAYIDX9:%.*]] = getelementptr inbounds i32, i32* [[DST:%.*]], i32 [[I_023]]
314; CHECK-NEXT:    store i32 [[ADD7]], i32* [[ARRAYIDX9]], align 4
315; CHECK-NEXT:    [[INC]] = add nuw nsw i32 [[I_023]], 1
316; CHECK-NEXT:    [[NEXT_STRIDE]] = mul nuw nsw i32 [[STRIDE]], 8
317; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N:%.*]]
318; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label [[END:%.*]], label [[FOR_BODY]]
319; CHECK:       end:
320; CHECK-NEXT:    ret void
321;
322entry:
323  br label %for.body
324for.body:                                         ; preds = %for.body.preheader, %for.body
325  %i.023 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
326  %stride = phi i32 [ %next.stride, %for.body ], [ 3, %entry ]
327  %mul = mul nuw nsw i32 %i.023, %stride
328  %add5 = add nuw nsw i32 %mul, 2
329  %arrayidx6 = getelementptr inbounds i32, i32* %data, i32 %add5
330  %0 = load i32, i32* %arrayidx6, align 4
331  %add7 = add nsw i32 5, %0
332  %arrayidx9 = getelementptr inbounds i32, i32* %dst, i32 %i.023
333  store i32 %add7, i32* %arrayidx9, align 4
334  %inc = add nuw nsw i32 %i.023, 1
335  %next.stride = mul nuw nsw i32 %stride, 8
336  %exitcond.not = icmp eq i32 %inc, %n
337  br i1 %exitcond.not, label %end, label %for.body
338end:                                 ; preds = %end, %entry
339  ret void
340}
341
342define void @test_stride_noninvar3_4i32(i32* readonly %data, i32* noalias nocapture %dst, i32 %n, i32 %x) {
343; CHECK-LABEL: @test_stride_noninvar3_4i32(
344; CHECK:       vector.body:
345; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY:%.*]] ]
346; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
347; CHECK-NEXT:    [[VEC_IND4:%.*]] = phi <4 x i32> [ [[INDUCTION:%.*]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT5:%.*]], [[VECTOR_BODY]] ]
348; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[INDEX]], 0
349; CHECK-NEXT:    [[TMP4:%.*]] = add i32 [[INDEX]], 1
350; CHECK-NEXT:    [[TMP5:%.*]] = add i32 [[INDEX]], 2
351; CHECK-NEXT:    [[TMP6:%.*]] = add i32 [[INDEX]], 3
352; CHECK-NEXT:    [[TMP7:%.*]] = mul nuw nsw <4 x i32> [[VEC_IND]], [[VEC_IND4]]
353; CHECK-NEXT:    [[TMP8:%.*]] = add nuw nsw <4 x i32> [[TMP7]], <i32 2, i32 2, i32 2, i32 2>
354; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, i32* [[DATA:%.*]], <4 x i32> [[TMP8]]
355; CHECK-NEXT:    [[WIDE_MASKED_GATHER:%.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0i32(<4 x i32*> [[TMP9]], i32 4, <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i32> undef)
356; CHECK-NEXT:    [[TMP10:%.*]] = add nsw <4 x i32> <i32 5, i32 5, i32 5, i32 5>, [[WIDE_MASKED_GATHER]]
357; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i32, i32* [[DST:%.*]], i32 [[TMP3]]
358; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i32, i32* [[TMP11]], i32 0
359; CHECK-NEXT:    [[TMP13:%.*]] = bitcast i32* [[TMP12]] to <4 x i32>*
360; CHECK-NEXT:    store <4 x i32> [[TMP10]], <4 x i32>* [[TMP13]], align 4
361; CHECK-NEXT:    [[INDEX_NEXT]] = add i32 [[INDEX]], 4
362; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], <i32 4, i32 4, i32 4, i32 4>
363; CHECK-NEXT:    [[VEC_IND_NEXT5]] = add <4 x i32> [[VEC_IND4]], [[DOTSPLAT3:%.*]]
364; CHECK-NEXT:    [[TMP14:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC:%.*]]
365; CHECK-NEXT:    br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP16:!llvm.loop !.*]]
366;
367entry:
368  br label %for.body
369for.body:                                         ; preds = %for.body.preheader, %for.body
370  %i.023 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
371  %stride = phi i32 [ %next.stride, %for.body ], [ 3, %entry ]
372  %mul = mul nuw nsw i32 %i.023, %stride
373  %add5 = add nuw nsw i32 %mul, 2
374  %arrayidx6 = getelementptr inbounds i32, i32* %data, i32 %add5
375  %0 = load i32, i32* %arrayidx6, align 4
376  %add7 = add nsw i32 5, %0
377  %arrayidx9 = getelementptr inbounds i32, i32* %dst, i32 %i.023
378  store i32 %add7, i32* %arrayidx9, align 4
379  %inc = add nuw nsw i32 %i.023, 1
380  %next.stride = add nuw nsw i32 %stride, %x
381  %exitcond.not = icmp eq i32 %inc, %n
382  br i1 %exitcond.not, label %end, label %for.body
383end:                                 ; preds = %end, %entry
384  ret void
385}
386
387declare i32 @llvm.experimental.vector.reduce.add.v4i32(<4 x i32>)
388declare <4 x i32> @llvm.masked.gather.v4i32.v4p0i32(<4 x i32*>, i32, <4 x i1>, <4 x i32>)
389declare <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32, i32)
390declare <4 x i32> @llvm.masked.load.v4i32.p0v4i32(<4 x i32>*, i32, <4 x i1>, <4 x i32>)
391declare void @llvm.masked.scatter.v4i32.v4p0i32(<4 x i32>, <4 x i32*>, i32, <4 x i1>)
392