xref: /llvm-project/llvm/test/Transforms/LoopVectorize/interleaved-accesses.ll (revision 7f3428d3ed71d87a2088b77b6cab9f3d86544234)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -S -passes=loop-vectorize,instcombine -force-vector-width=4 -force-vector-interleave=1 -enable-interleaved-mem-accesses=true -runtime-memory-check-threshold=24 < %s | FileCheck %s
3
4target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
5
6; Check vectorization on an interleaved load group of factor 2 and an interleaved
7; store group of factor 2.
8
9; int AB[1024];
10; int CD[1024];
11;  void test_array_load2_store2(int C, int D) {
12;   for (int i = 0; i < 1024; i+=2) {
13;     int A = AB[i];
14;     int B = AB[i+1];
15;     CD[i] = A + C;
16;     CD[i+1] = B * D;
17;   }
18; }
19
20
21@AB = common global [1024 x i32] zeroinitializer, align 4
22@CD = common global [1024 x i32] zeroinitializer, align 4
23
24define void @test_array_load2_store2(i32 %C, i32 %D) {
25; CHECK-LABEL: @test_array_load2_store2(
26; CHECK-NEXT:  entry:
27; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
28; CHECK:       vector.ph:
29; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[C:%.*]], i64 0
30; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
31; CHECK-NEXT:    [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <4 x i32> poison, i32 [[D:%.*]], i64 0
32; CHECK-NEXT:    [[BROADCAST_SPLAT3:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT2]], <4 x i32> poison, <4 x i32> zeroinitializer
33; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
34; CHECK:       vector.body:
35; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
36; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 1
37; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [1024 x i32], ptr @AB, i64 0, i64 [[OFFSET_IDX]]
38; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP0]], align 4
39; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
40; CHECK-NEXT:    [[STRIDED_VEC1:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
41; CHECK-NEXT:    [[TMP2:%.*]] = add nsw <4 x i32> [[STRIDED_VEC]], [[BROADCAST_SPLAT]]
42; CHECK-NEXT:    [[TMP3:%.*]] = mul nsw <4 x i32> [[STRIDED_VEC1]], [[BROADCAST_SPLAT3]]
43; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [1024 x i32], ptr @CD, i64 0, i64 [[OFFSET_IDX]]
44; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> [[TMP3]], <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>
45; CHECK-NEXT:    store <8 x i32> [[INTERLEAVED_VEC]], ptr [[TMP4]], align 4
46; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
47; CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 512
48; CHECK-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
49; CHECK:       middle.block:
50; CHECK-NEXT:    br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]]
51; CHECK:       scalar.ph:
52; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
53; CHECK:       for.body:
54; CHECK-NEXT:    br i1 poison, label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP3:![0-9]+]]
55; CHECK:       for.end:
56; CHECK-NEXT:    ret void
57;
58entry:
59  br label %for.body
60
61for.body:                                         ; preds = %for.body, %entry
62  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
63  %arrayidx0 = getelementptr inbounds [1024 x i32], ptr @AB, i64 0, i64 %indvars.iv
64  %tmp = load i32, ptr %arrayidx0, align 4
65  %tmp1 = or disjoint i64 %indvars.iv, 1
66  %arrayidx1 = getelementptr inbounds [1024 x i32], ptr @AB, i64 0, i64 %tmp1
67  %tmp2 = load i32, ptr %arrayidx1, align 4
68  %add = add nsw i32 %tmp, %C
69  %mul = mul nsw i32 %tmp2, %D
70  %arrayidx2 = getelementptr inbounds [1024 x i32], ptr @CD, i64 0, i64 %indvars.iv
71  store i32 %add, ptr %arrayidx2, align 4
72  %arrayidx3 = getelementptr inbounds [1024 x i32], ptr @CD, i64 0, i64 %tmp1
73  store i32 %mul, ptr %arrayidx3, align 4
74  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 2
75  %cmp = icmp slt i64 %indvars.iv.next, 1024
76  br i1 %cmp, label %for.body, label %for.end
77
78for.end:                                          ; preds = %for.body
79  ret void
80}
81
82; int A[3072];
83; struct ST S[1024];
84; void test_struct_st3() {
85;   int *ptr = A;
86;   for (int i = 0; i < 1024; i++) {
87;     int X1 = *ptr++;
88;     int X2 = *ptr++;
89;     int X3 = *ptr++;
90;     T[i].x = X1 + 1;
91;     T[i].y = X2 + 2;
92;     T[i].z = X3 + 3;
93;   }
94; }
95
96
97%struct.ST3 = type { i32, i32, i32 }
98@A = common global [3072 x i32] zeroinitializer, align 4
99@S = common global [1024 x %struct.ST3] zeroinitializer, align 4
100
101define void @test_struct_array_load3_store3() {
102; CHECK-LABEL: @test_struct_array_load3_store3(
103; CHECK-NEXT:  entry:
104; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
105; CHECK:       vector.ph:
106; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
107; CHECK:       vector.body:
108; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
109; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 12
110; CHECK-NEXT:    [[NEXT_GEP:%.*]] = getelementptr i8, ptr @A, i64 [[OFFSET_IDX]]
111; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <12 x i32>, ptr [[NEXT_GEP]], align 4
112; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <12 x i32> [[WIDE_VEC]], <12 x i32> poison, <4 x i32> <i32 0, i32 3, i32 6, i32 9>
113; CHECK-NEXT:    [[STRIDED_VEC2:%.*]] = shufflevector <12 x i32> [[WIDE_VEC]], <12 x i32> poison, <4 x i32> <i32 1, i32 4, i32 7, i32 10>
114; CHECK-NEXT:    [[STRIDED_VEC3:%.*]] = shufflevector <12 x i32> [[WIDE_VEC]], <12 x i32> poison, <4 x i32> <i32 2, i32 5, i32 8, i32 11>
115; CHECK-NEXT:    [[TMP0:%.*]] = add nsw <4 x i32> [[STRIDED_VEC]], splat (i32 1)
116; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [1024 x %struct.ST3], ptr @S, i64 0, i64 [[INDEX]], i32 0
117; CHECK-NEXT:    [[TMP1:%.*]] = add nsw <4 x i32> [[STRIDED_VEC2]], splat (i32 2)
118; CHECK-NEXT:    [[TMP2:%.*]] = add nsw <4 x i32> [[STRIDED_VEC3]], splat (i32 3)
119; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <4 x i32> [[TMP0]], <4 x i32> [[TMP1]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
120; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison>
121; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <8 x i32> [[TMP4]], <8 x i32> [[TMP5]], <12 x i32> <i32 0, i32 4, i32 8, i32 1, i32 5, i32 9, i32 2, i32 6, i32 10, i32 3, i32 7, i32 11>
122; CHECK-NEXT:    store <12 x i32> [[INTERLEAVED_VEC]], ptr [[TMP3]], align 4
123; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
124; CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
125; CHECK-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
126; CHECK:       middle.block:
127; CHECK-NEXT:    br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]]
128; CHECK:       scalar.ph:
129; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
130; CHECK:       for.body:
131; CHECK-NEXT:    br i1 poison, label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
132; CHECK:       for.end:
133; CHECK-NEXT:    ret void
134;
135entry:
136  br label %for.body
137
138for.body:                                         ; preds = %for.body, %entry
139  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
140  %ptr.016 = phi ptr [ @A, %entry ], [ %incdec.ptr2, %for.body ]
141  %incdec.ptr = getelementptr inbounds i32, ptr %ptr.016, i64 1
142  %tmp = load i32, ptr %ptr.016, align 4
143  %incdec.ptr1 = getelementptr inbounds i32, ptr %ptr.016, i64 2
144  %tmp1 = load i32, ptr %incdec.ptr, align 4
145  %incdec.ptr2 = getelementptr inbounds i32, ptr %ptr.016, i64 3
146  %tmp2 = load i32, ptr %incdec.ptr1, align 4
147  %add = add nsw i32 %tmp, 1
148  %x = getelementptr inbounds [1024 x %struct.ST3], ptr @S, i64 0, i64 %indvars.iv, i32 0
149  store i32 %add, ptr %x, align 4
150  %add3 = add nsw i32 %tmp1, 2
151  %y = getelementptr inbounds [1024 x %struct.ST3], ptr @S, i64 0, i64 %indvars.iv, i32 1
152  store i32 %add3, ptr %y, align 4
153  %add6 = add nsw i32 %tmp2, 3
154  %z = getelementptr inbounds [1024 x %struct.ST3], ptr @S, i64 0, i64 %indvars.iv, i32 2
155  store i32 %add6, ptr %z, align 4
156  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
157  %exitcond = icmp eq i64 %indvars.iv.next, 1024
158  br i1 %exitcond, label %for.end, label %for.body
159
160for.end:                                          ; preds = %for.body
161  ret void
162}
163
164; Check vectorization on an interleaved load group of factor 4.
165
166; struct ST4{
167;   int x;
168;   int y;
169;   int z;
170;   int w;
171; };
172; int test_struct_load4(struct ST4 *S) {
173;   int r = 0;
174;   for (int i = 0; i < 1024; i++) {
175;      r += S[i].x;
176;      r -= S[i].y;
177;      r += S[i].z;
178;      r -= S[i].w;
179;   }
180;   return r;
181; }
182
183%struct.ST4 = type { i32, i32, i32, i32 }
184
185define i32 @test_struct_load4(ptr nocapture readonly %S) {
186;
187; CHECK-LABEL: @test_struct_load4(
188; CHECK-NEXT:  entry:
189; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
190; CHECK:       vector.ph:
191; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
192; CHECK:       vector.body:
193; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
194; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP4:%.*]], [[VECTOR_BODY]] ]
195; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [[STRUCT_ST4:%.*]], ptr [[S:%.*]], i64 [[INDEX]], i32 0
196; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <16 x i32>, ptr [[TMP0]], align 4
197; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <16 x i32> [[WIDE_VEC]], <16 x i32> poison, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
198; CHECK-NEXT:    [[STRIDED_VEC1:%.*]] = shufflevector <16 x i32> [[WIDE_VEC]], <16 x i32> poison, <4 x i32> <i32 1, i32 5, i32 9, i32 13>
199; CHECK-NEXT:    [[STRIDED_VEC2:%.*]] = shufflevector <16 x i32> [[WIDE_VEC]], <16 x i32> poison, <4 x i32> <i32 2, i32 6, i32 10, i32 14>
200; CHECK-NEXT:    [[STRIDED_VEC3:%.*]] = shufflevector <16 x i32> [[WIDE_VEC]], <16 x i32> poison, <4 x i32> <i32 3, i32 7, i32 11, i32 15>
201; CHECK-NEXT:    [[TMP1:%.*]] = add <4 x i32> [[STRIDED_VEC]], [[VEC_PHI]]
202; CHECK-NEXT:    [[TMP2:%.*]] = add <4 x i32> [[TMP1]], [[STRIDED_VEC2]]
203; CHECK-NEXT:    [[TMP3:%.*]] = add <4 x i32> [[STRIDED_VEC1]], [[STRIDED_VEC3]]
204; CHECK-NEXT:    [[TMP4]] = sub <4 x i32> [[TMP2]], [[TMP3]]
205; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
206; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
207; CHECK-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
208; CHECK:       middle.block:
209; CHECK-NEXT:    [[TMP6:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP4]])
210; CHECK-NEXT:    br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]]
211; CHECK:       scalar.ph:
212; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
213; CHECK:       for.body:
214; CHECK-NEXT:    br i1 poison, label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]]
215; CHECK:       for.end:
216; CHECK-NEXT:    [[SUB8_LCSSA:%.*]] = phi i32 [ poison, [[FOR_BODY]] ], [ [[TMP6]], [[MIDDLE_BLOCK]] ]
217; CHECK-NEXT:    ret i32 [[SUB8_LCSSA]]
218;
219entry:
220  br label %for.body
221
222for.body:                                         ; preds = %for.body, %entry
223  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
224  %r.022 = phi i32 [ 0, %entry ], [ %sub8, %for.body ]
225  %x = getelementptr inbounds %struct.ST4, ptr %S, i64 %indvars.iv, i32 0
226  %tmp = load i32, ptr %x, align 4
227  %add = add nsw i32 %tmp, %r.022
228  %y = getelementptr inbounds %struct.ST4, ptr %S, i64 %indvars.iv, i32 1
229  %tmp1 = load i32, ptr %y, align 4
230  %sub = sub i32 %add, %tmp1
231  %z = getelementptr inbounds %struct.ST4, ptr %S, i64 %indvars.iv, i32 2
232  %tmp2 = load i32, ptr %z, align 4
233  %add5 = add nsw i32 %sub, %tmp2
234  %w = getelementptr inbounds %struct.ST4, ptr %S, i64 %indvars.iv, i32 3
235  %tmp3 = load i32, ptr %w, align 4
236  %sub8 = sub i32 %add5, %tmp3
237  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
238  %exitcond = icmp eq i64 %indvars.iv.next, 1024
239  br i1 %exitcond, label %for.end, label %for.body
240
241for.end:                                          ; preds = %for.body
242  ret i32 %sub8
243}
244
245; Check vectorization on an interleaved store group of factor 4.
246
247; void test_struct_store4(int *A, struct ST4 *B) {
248;   int *ptr = A;
249;   for (int i = 0; i < 1024; i++) {
250;     int X = *ptr++;
251;     B[i].x = X + 1;
252;     B[i].y = X * 2;
253;     B[i].z = X + 3;
254;     B[i].w = X + 4;
255;   }
256; }
257
258
259define void @test_struct_store4(ptr noalias nocapture readonly %A, ptr noalias nocapture %B) {
260; CHECK-LABEL: @test_struct_store4(
261; CHECK-NEXT:  entry:
262; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
263; CHECK:       vector.ph:
264; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
265; CHECK:       vector.body:
266; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
267; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 2
268; CHECK-NEXT:    [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[OFFSET_IDX]]
269; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[NEXT_GEP]], align 4
270; CHECK-NEXT:    [[TMP0:%.*]] = add nsw <4 x i32> [[WIDE_LOAD]], splat (i32 1)
271; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_ST4:%.*]], ptr [[B:%.*]], i64 [[INDEX]], i32 0
272; CHECK-NEXT:    [[TMP1:%.*]] = shl nsw <4 x i32> [[WIDE_LOAD]], splat (i32 1)
273; CHECK-NEXT:    [[TMP2:%.*]] = add nsw <4 x i32> [[WIDE_LOAD]], splat (i32 3)
274; CHECK-NEXT:    [[TMP3:%.*]] = add nsw <4 x i32> [[WIDE_LOAD]], splat (i32 4)
275; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <4 x i32> [[TMP0]], <4 x i32> [[TMP1]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
276; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> [[TMP3]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
277; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <8 x i32> [[TMP5]], <8 x i32> [[TMP6]], <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 1, i32 5, i32 9, i32 13, i32 2, i32 6, i32 10, i32 14, i32 3, i32 7, i32 11, i32 15>
278; CHECK-NEXT:    store <16 x i32> [[INTERLEAVED_VEC]], ptr [[TMP4]], align 4
279; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
280; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
281; CHECK-NEXT:    br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
282; CHECK:       middle.block:
283; CHECK-NEXT:    br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
284; CHECK:       scalar.ph:
285; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
286; CHECK:       for.cond.cleanup:
287; CHECK-NEXT:    ret void
288; CHECK:       for.body:
289; CHECK-NEXT:    br i1 poison, label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
290;
291entry:
292  br label %for.body
293
294for.cond.cleanup:                                 ; preds = %for.body
295  ret void
296
297for.body:                                         ; preds = %for.body, %entry
298  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
299  %ptr.024 = phi ptr [ %A, %entry ], [ %incdec.ptr, %for.body ]
300  %incdec.ptr = getelementptr inbounds i32, ptr %ptr.024, i64 1
301  %tmp = load i32, ptr %ptr.024, align 4
302  %add = add nsw i32 %tmp, 1
303  %x = getelementptr inbounds %struct.ST4, ptr %B, i64 %indvars.iv, i32 0
304  store i32 %add, ptr %x, align 4
305  %mul = shl nsw i32 %tmp, 1
306  %y = getelementptr inbounds %struct.ST4, ptr %B, i64 %indvars.iv, i32 1
307  store i32 %mul, ptr %y, align 4
308  %add3 = add nsw i32 %tmp, 3
309  %z = getelementptr inbounds %struct.ST4, ptr %B, i64 %indvars.iv, i32 2
310  store i32 %add3, ptr %z, align 4
311  %add6 = add nsw i32 %tmp, 4
312  %w = getelementptr inbounds %struct.ST4, ptr %B, i64 %indvars.iv, i32 3
313  store i32 %add6, ptr %w, align 4
314  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
315  %exitcond = icmp eq i64 %indvars.iv.next, 1024
316  br i1 %exitcond, label %for.cond.cleanup, label %for.body
317}
318
319; Check vectorization on a reverse interleaved load group of factor 2 and
320; a reverse interleaved store group of factor 2.
321
322; struct ST2 {
323;  int x;
324;  int y;
325; };
326;
327; void test_reversed_load2_store2(struct ST2 *A, struct ST2 *B) {
328;   for (int i = 1023; i >= 0; i--) {
329;     int a = A[i].x + i;  // interleaved load of index 0
330;     int b = A[i].y - i;  // interleaved load of index 1
331;     B[i].x = a;          // interleaved store of index 0
332;     B[i].y = b;          // interleaved store of index 1
333;   }
334; }
335
336
337%struct.ST2 = type { i32, i32 }
338
339define void @test_reversed_load2_store2(ptr noalias nocapture readonly %A, ptr noalias nocapture %B) {
340; CHECK-LABEL: @test_reversed_load2_store2(
341; CHECK-NEXT:  entry:
342; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
343; CHECK:       vector.ph:
344; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
345; CHECK:       vector.body:
346; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
347; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 1023, i32 1022, i32 1021, i32 1020>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
348; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = sub i64 1023, [[INDEX]]
349; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [[STRUCT_ST2:%.*]], ptr [[A:%.*]], i64 [[OFFSET_IDX]], i32 0
350; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[TMP0]], i64 -24
351; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP1]], align 4
352; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
353; CHECK-NEXT:    [[REVERSE:%.*]] = shufflevector <4 x i32> [[STRIDED_VEC]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
354; CHECK-NEXT:    [[STRIDED_VEC1:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
355; CHECK-NEXT:    [[REVERSE2:%.*]] = shufflevector <4 x i32> [[STRIDED_VEC1]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
356; CHECK-NEXT:    [[TMP2:%.*]] = add nsw <4 x i32> [[REVERSE]], [[VEC_IND]]
357; CHECK-NEXT:    [[TMP3:%.*]] = sub nsw <4 x i32> [[REVERSE2]], [[VEC_IND]]
358; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[STRUCT_ST2]], ptr [[B:%.*]], i64 [[OFFSET_IDX]], i32 0
359; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i64 -24
360; CHECK-NEXT:    [[REVERSE3:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
361; CHECK-NEXT:    [[REVERSE4:%.*]] = shufflevector <4 x i32> [[TMP3]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
362; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <4 x i32> [[REVERSE3]], <4 x i32> [[REVERSE4]], <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>
363; CHECK-NEXT:    store <8 x i32> [[INTERLEAVED_VEC]], ptr [[TMP5]], align 4
364; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
365; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], splat (i32 -4)
366; CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
367; CHECK-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
368; CHECK:       middle.block:
369; CHECK-NEXT:    br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
370; CHECK:       scalar.ph:
371; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
372; CHECK:       for.cond.cleanup:
373; CHECK-NEXT:    ret void
374; CHECK:       for.body:
375; CHECK-NEXT:    br i1 poison, label [[FOR_BODY]], label [[FOR_COND_CLEANUP]], !llvm.loop [[LOOP11:![0-9]+]]
376;
377entry:
378  br label %for.body
379
380for.cond.cleanup:                                 ; preds = %for.body
381  ret void
382
383for.body:                                         ; preds = %for.body, %entry
384  %indvars.iv = phi i64 [ 1023, %entry ], [ %indvars.iv.next, %for.body ]
385  %x = getelementptr inbounds %struct.ST2, ptr %A, i64 %indvars.iv, i32 0
386  %tmp = load i32, ptr %x, align 4
387  %tmp1 = trunc i64 %indvars.iv to i32
388  %add = add nsw i32 %tmp, %tmp1
389  %y = getelementptr inbounds %struct.ST2, ptr %A, i64 %indvars.iv, i32 1
390  %tmp2 = load i32, ptr %y, align 4
391  %sub = sub nsw i32 %tmp2, %tmp1
392  %x5 = getelementptr inbounds %struct.ST2, ptr %B, i64 %indvars.iv, i32 0
393  store i32 %add, ptr %x5, align 4
394  %y8 = getelementptr inbounds %struct.ST2, ptr %B, i64 %indvars.iv, i32 1
395  store i32 %sub, ptr %y8, align 4
396  %indvars.iv.next = add nsw i64 %indvars.iv, -1
397  %cmp = icmp sgt i64 %indvars.iv, 0
398  br i1 %cmp, label %for.body, label %for.cond.cleanup
399}
400
401; Check vectorization on an interleaved load group of factor 2 with 1 gap
402; (missing the load of odd elements). Because the vectorized loop would
403; speculatively access memory out-of-bounds, we must execute at least one
404; iteration of the scalar loop.
405
406; void even_load_static_tc(int *A, int *B) {
407;  for (unsigned i = 0; i < 1024; i+=2)
408;     B[i/2] = A[i] * 2;
409; }
410
411
412define void @even_load_static_tc(ptr noalias nocapture readonly %A, ptr noalias nocapture %B) {
413; CHECK-LABEL: @even_load_static_tc(
414; CHECK-NEXT:  entry:
415; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
416; CHECK:       vector.ph:
417; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
418; CHECK:       vector.body:
419; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
420; CHECK-NEXT:    [[DOTIDX:%.*]] = shl i64 [[INDEX]], 3
421; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds i8, ptr [[A:%.*]], i64 [[DOTIDX]]
422; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP0]], align 4
423; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
424; CHECK-NEXT:    [[TMP1:%.*]] = shl nsw <4 x i32> [[STRIDED_VEC]], splat (i32 1)
425; CHECK-NEXT:    [[TMP2:%.*]] = and i64 [[INDEX]], 9223372036854775804
426; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds nuw i32, ptr [[B:%.*]], i64 [[TMP2]]
427; CHECK-NEXT:    store <4 x i32> [[TMP1]], ptr [[TMP3]], align 4
428; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
429; CHECK-NEXT:    [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], 508
430; CHECK-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
431; CHECK:       middle.block:
432; CHECK-NEXT:    br label [[SCALAR_PH]]
433; CHECK:       scalar.ph:
434; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
435; CHECK:       for.cond.cleanup:
436; CHECK-NEXT:    ret void
437; CHECK:       for.body:
438; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ 1016, [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
439; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[A]], i64 [[INDVARS_IV]]
440; CHECK-NEXT:    [[TMP:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
441; CHECK-NEXT:    [[MUL:%.*]] = shl nsw i32 [[TMP]], 1
442; CHECK-NEXT:    [[TMP1:%.*]] = lshr exact i64 [[INDVARS_IV]], 1
443; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds nuw i32, ptr [[B]], i64 [[TMP1]]
444; CHECK-NEXT:    store i32 [[MUL]], ptr [[ARRAYIDX2]], align 4
445; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 2
446; CHECK-NEXT:    [[CMP:%.*]] = icmp samesign ult i64 [[INDVARS_IV]], 1022
447; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND_CLEANUP:%.*]], !llvm.loop [[LOOP13:![0-9]+]]
448;
449entry:
450  br label %for.body
451
452for.cond.cleanup:                                 ; preds = %for.body
453  ret void
454
455for.body:                                         ; preds = %for.body, %entry
456  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
457  %arrayidx = getelementptr inbounds i32, ptr %A, i64 %indvars.iv
458  %tmp = load i32, ptr %arrayidx, align 4
459  %mul = shl nsw i32 %tmp, 1
460  %tmp1 = lshr exact i64 %indvars.iv, 1
461  %arrayidx2 = getelementptr inbounds i32, ptr %B, i64 %tmp1
462  store i32 %mul, ptr %arrayidx2, align 4
463  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 2
464  %cmp = icmp ult i64 %indvars.iv.next, 1024
465  br i1 %cmp, label %for.body, label %for.cond.cleanup
466}
467
468; Check vectorization on an interleaved load group of factor 2 with 1 gap
469; (missing the load of odd elements). Because the vectorized loop would
470; speculatively access memory out-of-bounds, we must execute at least one
471; iteration of the scalar loop.
472
473; void even_load_dynamic_tc(int *A, int *B, unsigned N) {
474;  for (unsigned i = 0; i < N; i+=2)
475;     B[i/2] = A[i] * 2;
476; }
477
478
479define void @even_load_dynamic_tc(ptr noalias nocapture readonly %A, ptr noalias nocapture %B, i64 %N) {
480; CHECK-LABEL: @even_load_dynamic_tc(
481; CHECK-NEXT:  entry:
482; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 9
483; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
484; CHECK:       vector.ph:
485; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[N]], -1
486; CHECK-NEXT:    [[TMP1:%.*]] = lshr i64 [[TMP0]], 1
487; CHECK-NEXT:    [[TMP2:%.*]] = add nuw i64 [[TMP1]], 1
488; CHECK-NEXT:    [[N_MOD_VF:%.*]] = and i64 [[TMP2]], 3
489; CHECK-NEXT:    [[TMP3:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
490; CHECK-NEXT:    [[TMP4:%.*]] = select i1 [[TMP3]], i64 4, i64 [[N_MOD_VF]]
491; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP2]], [[TMP4]]
492; CHECK-NEXT:    [[IND_END:%.*]] = shl i64 [[N_VEC]], 1
493; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
494; CHECK:       vector.body:
495; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
496; CHECK-NEXT:    [[DOTIDX:%.*]] = shl i64 [[INDEX]], 3
497; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[A:%.*]], i64 [[DOTIDX]]
498; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP5]], align 4
499; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
500; CHECK-NEXT:    [[TMP6:%.*]] = shl nsw <4 x i32> [[STRIDED_VEC]], splat (i32 1)
501; CHECK-NEXT:    [[TMP7:%.*]] = and i64 [[INDEX]], 9223372036854775804
502; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds nuw i32, ptr [[B:%.*]], i64 [[TMP7]]
503; CHECK-NEXT:    store <4 x i32> [[TMP6]], ptr [[TMP8]], align 4
504; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
505; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
506; CHECK-NEXT:    br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]]
507; CHECK:       middle.block:
508; CHECK-NEXT:    br label [[SCALAR_PH]]
509; CHECK:       scalar.ph:
510; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
511; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
512; CHECK:       for.cond.cleanup:
513; CHECK-NEXT:    ret void
514; CHECK:       for.body:
515; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
516; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV]]
517; CHECK-NEXT:    [[TMP:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
518; CHECK-NEXT:    [[MUL:%.*]] = shl nsw i32 [[TMP]], 1
519; CHECK-NEXT:    [[TMP1:%.*]] = lshr exact i64 [[INDVARS_IV]], 1
520; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds nuw i32, ptr [[B]], i64 [[TMP1]]
521; CHECK-NEXT:    store i32 [[MUL]], ptr [[ARRAYIDX2]], align 4
522; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 2
523; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i64 [[INDVARS_IV_NEXT]], [[N]]
524; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND_CLEANUP:%.*]], !llvm.loop [[LOOP15:![0-9]+]]
525;
526entry:
527  br label %for.body
528
529for.cond.cleanup:                                 ; preds = %for.body
530  ret void
531
532for.body:                                         ; preds = %for.body, %entry
533  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
534  %arrayidx = getelementptr inbounds i32, ptr %A, i64 %indvars.iv
535  %tmp = load i32, ptr %arrayidx, align 4
536  %mul = shl nsw i32 %tmp, 1
537  %tmp1 = lshr exact i64 %indvars.iv, 1
538  %arrayidx2 = getelementptr inbounds i32, ptr %B, i64 %tmp1
539  store i32 %mul, ptr %arrayidx2, align 4
540  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 2
541  %cmp = icmp ult i64 %indvars.iv.next, %N
542  br i1 %cmp, label %for.body, label %for.cond.cleanup
543}
544
545; Check vectorization on a reverse interleaved load group of factor 2 with 1
546; gap and a reverse interleaved store group of factor 2. The interleaved load
547; group should be removed since it has a gap and is reverse.
548
549; struct pair {
550;  int x;
551;  int y;
552; };
553;
554; void load_gap_reverse(struct pair *P1, struct pair *P2, int X) {
555;   for (int i = 1023; i >= 0; i--) {
556;     int a = X + i;
557;     int b = B[i].y - i;
558;     A[i].x = a;
559;     B[i].y = b;
560;   }
561; }
562
563
564%pair = type { i64, i64 }
565define void @load_gap_reverse(ptr noalias nocapture %P1, ptr noalias nocapture %P2, i64 %X) {
566; CHECK-LABEL: @load_gap_reverse(
567; CHECK-NEXT:  entry:
568; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
569; CHECK:       vector.ph:
570; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[X:%.*]], i64 0
571; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
572; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
573; CHECK:       vector.body:
574; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
575; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 1023, i64 1022, i64 1021, i64 1020>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
576; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = sub i64 1023, [[INDEX]]
577; CHECK-NEXT:    [[TMP0:%.*]] = sub i64 1022, [[INDEX]]
578; CHECK-NEXT:    [[TMP1:%.*]] = sub i64 1021, [[INDEX]]
579; CHECK-NEXT:    [[TMP2:%.*]] = sub i64 1020, [[INDEX]]
580; CHECK-NEXT:    [[TMP3:%.*]] = add nsw <4 x i64> [[BROADCAST_SPLAT]], [[VEC_IND]]
581; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[PAIR:%.*]], ptr [[P1:%.*]], i64 [[OFFSET_IDX]], i32 0
582; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[PAIR]], ptr [[P1]], i64 [[TMP0]], i32 0
583; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[PAIR]], ptr [[P1]], i64 [[TMP1]], i32 0
584; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[PAIR]], ptr [[P1]], i64 [[TMP2]], i32 0
585; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[PAIR]], ptr [[P2:%.*]], i64 [[OFFSET_IDX]], i32 1
586; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[PAIR]], ptr [[P2]], i64 [[TMP0]], i32 1
587; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[PAIR]], ptr [[P2]], i64 [[TMP1]], i32 1
588; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[PAIR]], ptr [[P2]], i64 [[TMP2]], i32 1
589; CHECK-NEXT:    [[TMP12:%.*]] = load i64, ptr [[TMP8]], align 8
590; CHECK-NEXT:    [[TMP13:%.*]] = load i64, ptr [[TMP9]], align 8
591; CHECK-NEXT:    [[TMP14:%.*]] = load i64, ptr [[TMP10]], align 8
592; CHECK-NEXT:    [[TMP15:%.*]] = load i64, ptr [[TMP11]], align 8
593; CHECK-NEXT:    [[TMP16:%.*]] = insertelement <4 x i64> poison, i64 [[TMP12]], i64 0
594; CHECK-NEXT:    [[TMP17:%.*]] = insertelement <4 x i64> [[TMP16]], i64 [[TMP13]], i64 1
595; CHECK-NEXT:    [[TMP18:%.*]] = insertelement <4 x i64> [[TMP17]], i64 [[TMP14]], i64 2
596; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <4 x i64> [[TMP18]], i64 [[TMP15]], i64 3
597; CHECK-NEXT:    [[TMP20:%.*]] = sub nsw <4 x i64> [[TMP19]], [[VEC_IND]]
598; CHECK-NEXT:    [[TMP21:%.*]] = extractelement <4 x i64> [[TMP3]], i64 0
599; CHECK-NEXT:    store i64 [[TMP21]], ptr [[TMP4]], align 8
600; CHECK-NEXT:    [[TMP22:%.*]] = extractelement <4 x i64> [[TMP3]], i64 1
601; CHECK-NEXT:    store i64 [[TMP22]], ptr [[TMP5]], align 8
602; CHECK-NEXT:    [[TMP23:%.*]] = extractelement <4 x i64> [[TMP3]], i64 2
603; CHECK-NEXT:    store i64 [[TMP23]], ptr [[TMP6]], align 8
604; CHECK-NEXT:    [[TMP24:%.*]] = extractelement <4 x i64> [[TMP3]], i64 3
605; CHECK-NEXT:    store i64 [[TMP24]], ptr [[TMP7]], align 8
606; CHECK-NEXT:    [[TMP25:%.*]] = extractelement <4 x i64> [[TMP20]], i64 0
607; CHECK-NEXT:    store i64 [[TMP25]], ptr [[TMP8]], align 8
608; CHECK-NEXT:    [[TMP26:%.*]] = extractelement <4 x i64> [[TMP20]], i64 1
609; CHECK-NEXT:    store i64 [[TMP26]], ptr [[TMP9]], align 8
610; CHECK-NEXT:    [[TMP27:%.*]] = extractelement <4 x i64> [[TMP20]], i64 2
611; CHECK-NEXT:    store i64 [[TMP27]], ptr [[TMP10]], align 8
612; CHECK-NEXT:    [[TMP28:%.*]] = extractelement <4 x i64> [[TMP20]], i64 3
613; CHECK-NEXT:    store i64 [[TMP28]], ptr [[TMP11]], align 8
614; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
615; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 -4)
616; CHECK-NEXT:    [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
617; CHECK-NEXT:    br i1 [[TMP29]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
618; CHECK:       middle.block:
619; CHECK-NEXT:    br i1 true, label [[FOR_EXIT:%.*]], label [[SCALAR_PH]]
620; CHECK:       scalar.ph:
621; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
622; CHECK:       for.body:
623; CHECK-NEXT:    br i1 poison, label [[FOR_BODY]], label [[FOR_EXIT]], !llvm.loop [[LOOP17:![0-9]+]]
624; CHECK:       for.exit:
625; CHECK-NEXT:    ret void
626;
627entry:
628  br label %for.body
629
630for.body:
631  %i = phi i64 [ 1023, %entry ], [ %i.next, %for.body ]
632  %0 = add nsw i64 %X, %i
633  %1 = getelementptr inbounds %pair, ptr %P1, i64 %i, i32 0
634  %2 = getelementptr inbounds %pair, ptr %P2, i64 %i, i32 1
635  %3 = load i64, ptr %2, align 8
636  %4 = sub nsw i64 %3, %i
637  store i64 %0, ptr %1, align 8
638  store i64 %4, ptr %2, align 8
639  %i.next = add nsw i64 %i, -1
640  %cond = icmp sgt i64 %i, 0
641  br i1 %cond, label %for.body, label %for.exit
642
643for.exit:
644  ret void
645}
646
647; Check vectorization on interleaved access groups identified from mixed
648; loads/stores.
649; void mixed_load2_store2(int *A, int *B) {
650;   for (unsigned i = 0; i < 1024; i+=2)  {
651;     B[i] = A[i] * A[i+1];
652;     B[i+1] = A[i] + A[i+1];
653;   }
654; }
655
656
657define void @mixed_load2_store2(ptr noalias nocapture readonly %A, ptr noalias nocapture %B) {
658; CHECK-LABEL: @mixed_load2_store2(
659; CHECK-NEXT:  entry:
660; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
661; CHECK:       vector.ph:
662; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
663; CHECK:       vector.body:
664; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
665; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 1
666; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[OFFSET_IDX]]
667; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP0]], align 4
668; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
669; CHECK-NEXT:    [[STRIDED_VEC1:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
670; CHECK-NEXT:    [[TMP2:%.*]] = mul nsw <4 x i32> [[STRIDED_VEC1]], [[STRIDED_VEC]]
671; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[OFFSET_IDX]]
672; CHECK-NEXT:    [[STRIDED_VEC3:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
673; CHECK-NEXT:    [[STRIDED_VEC4:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
674; CHECK-NEXT:    [[TMP3:%.*]] = add nsw <4 x i32> [[STRIDED_VEC4]], [[STRIDED_VEC3]]
675; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> [[TMP3]], <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>
676; CHECK-NEXT:    store <8 x i32> [[INTERLEAVED_VEC]], ptr [[TMP4]], align 4
677; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
678; CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 512
679; CHECK-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]]
680; CHECK:       middle.block:
681; CHECK-NEXT:    br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
682; CHECK:       scalar.ph:
683; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
684; CHECK:       for.cond.cleanup:
685; CHECK-NEXT:    ret void
686; CHECK:       for.body:
687; CHECK-NEXT:    br i1 poison, label [[FOR_BODY]], label [[FOR_COND_CLEANUP]], !llvm.loop [[LOOP19:![0-9]+]]
688;
689entry:
690  br label %for.body
691
692for.cond.cleanup:                                 ; preds = %for.body
693  ret void
694
695for.body:                                         ; preds = %for.body, %entry
696  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
697  %arrayidx = getelementptr inbounds i32, ptr %A, i64 %indvars.iv
698  %tmp = load i32, ptr %arrayidx, align 4
699  %tmp1 = or disjoint i64 %indvars.iv, 1
700  %arrayidx2 = getelementptr inbounds i32, ptr %A, i64 %tmp1
701  %tmp2 = load i32, ptr %arrayidx2, align 4
702  %mul = mul nsw i32 %tmp2, %tmp
703  %arrayidx4 = getelementptr inbounds i32, ptr %B, i64 %indvars.iv
704  store i32 %mul, ptr %arrayidx4, align 4
705  %tmp3 = load i32, ptr %arrayidx, align 4
706  %tmp4 = load i32, ptr %arrayidx2, align 4
707  %add10 = add nsw i32 %tmp4, %tmp3
708  %arrayidx13 = getelementptr inbounds i32, ptr %B, i64 %tmp1
709  store i32 %add10, ptr %arrayidx13, align 4
710  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 2
711  %cmp = icmp ult i64 %indvars.iv.next, 1024
712  br i1 %cmp, label %for.body, label %for.cond.cleanup
713}
714
715; Check vectorization on interleaved access groups identified from mixed
716; loads/stores.
717; void mixed_load3_store3(int *A) {
718;   for (unsigned i = 0; i < 1024; i++)  {
719;     *A++ += i;
720;     *A++ += i;
721;     *A++ += i;
722;   }
723; }
724
725
726define void @mixed_load3_store3(ptr nocapture %A) {
727; CHECK-LABEL: @mixed_load3_store3(
728; CHECK-NEXT:  entry:
729; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
730; CHECK:       vector.ph:
731; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
732; CHECK:       vector.body:
733; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
734; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
735; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 12
736; CHECK-NEXT:    [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 [[OFFSET_IDX]]
737; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <12 x i32>, ptr [[NEXT_GEP]], align 4
738; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <12 x i32> [[WIDE_VEC]], <12 x i32> poison, <4 x i32> <i32 0, i32 3, i32 6, i32 9>
739; CHECK-NEXT:    [[STRIDED_VEC2:%.*]] = shufflevector <12 x i32> [[WIDE_VEC]], <12 x i32> poison, <4 x i32> <i32 1, i32 4, i32 7, i32 10>
740; CHECK-NEXT:    [[STRIDED_VEC3:%.*]] = shufflevector <12 x i32> [[WIDE_VEC]], <12 x i32> poison, <4 x i32> <i32 2, i32 5, i32 8, i32 11>
741; CHECK-NEXT:    [[TMP0:%.*]] = add <4 x i32> [[STRIDED_VEC]], [[VEC_IND]]
742; CHECK-NEXT:    [[TMP1:%.*]] = add <4 x i32> [[STRIDED_VEC2]], [[VEC_IND]]
743; CHECK-NEXT:    [[TMP2:%.*]] = add <4 x i32> [[STRIDED_VEC3]], [[VEC_IND]]
744; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x i32> [[TMP0]], <4 x i32> [[TMP1]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
745; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison>
746; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <8 x i32> [[TMP3]], <8 x i32> [[TMP4]], <12 x i32> <i32 0, i32 4, i32 8, i32 1, i32 5, i32 9, i32 2, i32 6, i32 10, i32 3, i32 7, i32 11>
747; CHECK-NEXT:    store <12 x i32> [[INTERLEAVED_VEC]], ptr [[NEXT_GEP]], align 4
748; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
749; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i32> [[VEC_IND]], splat (i32 4)
750; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
751; CHECK-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]]
752; CHECK:       middle.block:
753; CHECK-NEXT:    br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
754; CHECK:       scalar.ph:
755; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
756; CHECK:       for.cond.cleanup:
757; CHECK-NEXT:    ret void
758; CHECK:       for.body:
759; CHECK-NEXT:    br i1 poison, label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP21:![0-9]+]]
760;
761entry:
762  br label %for.body
763
764for.cond.cleanup:                                 ; preds = %for.body
765  ret void
766
767for.body:                                         ; preds = %for.body, %entry
768  %i.013 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
769  %A.addr.012 = phi ptr [ %A, %entry ], [ %incdec.ptr3, %for.body ]
770  %incdec.ptr = getelementptr inbounds i32, ptr %A.addr.012, i64 1
771  %tmp = load i32, ptr %A.addr.012, align 4
772  %add = add i32 %tmp, %i.013
773  store i32 %add, ptr %A.addr.012, align 4
774  %incdec.ptr1 = getelementptr inbounds i32, ptr %A.addr.012, i64 2
775  %tmp1 = load i32, ptr %incdec.ptr, align 4
776  %add2 = add i32 %tmp1, %i.013
777  store i32 %add2, ptr %incdec.ptr, align 4
778  %incdec.ptr3 = getelementptr inbounds i32, ptr %A.addr.012, i64 3
779  %tmp2 = load i32, ptr %incdec.ptr1, align 4
780  %add4 = add i32 %tmp2, %i.013
781  store i32 %add4, ptr %incdec.ptr1, align 4
782  %inc = add nuw nsw i32 %i.013, 1
783  %exitcond = icmp eq i32 %inc, 1024
784  br i1 %exitcond, label %for.cond.cleanup, label %for.body
785}
786
787; Check vectorization on interleaved access groups with members having different
788; kinds of type.
789
790; struct IntFloat {
791;   int a;
792;   float b;
793; };
794;
795; int SA;
796; float SB;
797;
798; void int_float_struct(struct IntFloat *A) {
799;   int SumA;
800;   float SumB;
801;   for (unsigned i = 0; i < 1024; i++)  {
802;     SumA += A[i].a;
803;     SumB += A[i].b;
804;   }
805;   SA = SumA;
806;   SB = SumB;
807; }
808
809
810%struct.IntFloat = type { i32, float }
811
812@SA = common global i32 0, align 4
813@SB = common global float 0.000000e+00, align 4
814
815define void @int_float_struct(ptr nocapture readonly %A) #0 {
816; CHECK-LABEL: @int_float_struct(
817; CHECK-NEXT:  entry:
818; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
819; CHECK:       vector.ph:
820; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
821; CHECK:       vector.body:
822; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
823; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <4 x float> [ <float undef, float 0.000000e+00, float 0.000000e+00, float 0.000000e+00>, [[VECTOR_PH]] ], [ [[TMP3:%.*]], [[VECTOR_BODY]] ]
824; CHECK-NEXT:    [[VEC_PHI1:%.*]] = phi <4 x i32> [ <i32 undef, i32 0, i32 0, i32 0>, [[VECTOR_PH]] ], [ [[TMP2:%.*]], [[VECTOR_BODY]] ]
825; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds [[STRUCT_INTFLOAT:%.*]], ptr [[A:%.*]], i64 [[INDEX]], i32 0
826; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP0]], align 4
827; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
828; CHECK-NEXT:    [[STRIDED_VEC2:%.*]] = shufflevector <8 x i32> [[WIDE_VEC]], <8 x i32> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
829; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i32> [[STRIDED_VEC2]] to <4 x float>
830; CHECK-NEXT:    [[TMP2]] = add <4 x i32> [[STRIDED_VEC]], [[VEC_PHI1]]
831; CHECK-NEXT:    [[TMP3]] = fadd fast <4 x float> [[VEC_PHI]], [[TMP1]]
832; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
833; CHECK-NEXT:    [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024
834; CHECK-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]]
835; CHECK:       middle.block:
836; CHECK-NEXT:    [[TMP5:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float 0.000000e+00, <4 x float> [[TMP3]])
837; CHECK-NEXT:    [[TMP6:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP2]])
838; CHECK-NEXT:    br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
839; CHECK:       scalar.ph:
840; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
841; CHECK:       for.cond.cleanup:
842; CHECK-NEXT:    [[ADD_LCSSA:%.*]] = phi i32 [ poison, [[FOR_BODY]] ], [ [[TMP6]], [[MIDDLE_BLOCK]] ]
843; CHECK-NEXT:    [[ADD3_LCSSA:%.*]] = phi float [ poison, [[FOR_BODY]] ], [ [[TMP5]], [[MIDDLE_BLOCK]] ]
844; CHECK-NEXT:    store i32 [[ADD_LCSSA]], ptr @SA, align 4
845; CHECK-NEXT:    store float [[ADD3_LCSSA]], ptr @SB, align 4
846; CHECK-NEXT:    ret void
847; CHECK:       for.body:
848; CHECK-NEXT:    br i1 poison, label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]]
849;
850entry:
851  br label %for.body
852
853for.cond.cleanup:                                 ; preds = %for.body
854  store i32 %add, ptr @SA, align 4
855  store float %add3, ptr @SB, align 4
856  ret void
857
858for.body:                                         ; preds = %for.body, %entry
859  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
860  %SumB.014 = phi float [ undef, %entry ], [ %add3, %for.body ]
861  %SumA.013 = phi i32 [ undef, %entry ], [ %add, %for.body ]
862  %a = getelementptr inbounds %struct.IntFloat, ptr %A, i64 %indvars.iv, i32 0
863  %tmp = load i32, ptr %a, align 4
864  %add = add nsw i32 %tmp, %SumA.013
865  %b = getelementptr inbounds %struct.IntFloat, ptr %A, i64 %indvars.iv, i32 1
866  %tmp1 = load float, ptr %b, align 4
867  %add3 = fadd fast float %SumB.014, %tmp1
868  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
869  %exitcond = icmp eq i64 %indvars.iv.next, 1024
870  br i1 %exitcond, label %for.cond.cleanup, label %for.body
871}
872
873; Check vectorization of interleaved access groups in the presence of
874; dependences (PR27626). The following tests check that we don't reorder
875; dependent loads and stores when generating code for interleaved access
876; groups. Stores should be scalarized because the required code motion would
877; break dependences, and the remaining interleaved load groups should have
878; gaps.
879
880; PR27626_0: Ensure a strided store is not moved after a dependent (zero
881;            distance) strided load.
882
883; void PR27626_0(struct pair *p, int z, int n) {
884;   for (int i = 0; i < n; i++) {
885;     p[i].x = z;
886;     p[i].y = p[i].x;
887;   }
888; }
889
890
891%pair.i32 = type { i32, i32 }
892define void @PR27626_0(ptr %p, i32 %z, i64 %n) {
893; CHECK-LABEL: @PR27626_0(
894; CHECK-NEXT:  entry:
895; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N:%.*]], 5
896; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
897; CHECK:       vector.ph:
898; CHECK-NEXT:    [[N_MOD_VF:%.*]] = and i64 [[N]], 3
899; CHECK-NEXT:    [[TMP0:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
900; CHECK-NEXT:    [[TMP1:%.*]] = select i1 [[TMP0]], i64 4, i64 [[N_MOD_VF]]
901; CHECK-NEXT:    [[N_VEC:%.*]] = sub nsw i64 [[N]], [[TMP1]]
902; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
903; CHECK:       vector.body:
904; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
905; CHECK-NEXT:    [[TMP2:%.*]] = or disjoint i64 [[INDEX]], 1
906; CHECK-NEXT:    [[TMP3:%.*]] = or disjoint i64 [[INDEX]], 2
907; CHECK-NEXT:    [[TMP4:%.*]] = or disjoint i64 [[INDEX]], 3
908; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], i64 [[INDEX]], i32 0
909; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP2]], i32 0
910; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP3]], i32 0
911; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP4]], i32 0
912; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[INDEX]], i32 1
913; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP2]], i32 1
914; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP3]], i32 1
915; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP4]], i32 1
916; CHECK-NEXT:    store i32 [[Z:%.*]], ptr [[TMP5]], align 4
917; CHECK-NEXT:    store i32 [[Z]], ptr [[TMP6]], align 4
918; CHECK-NEXT:    store i32 [[Z]], ptr [[TMP7]], align 4
919; CHECK-NEXT:    store i32 [[Z]], ptr [[TMP8]], align 4
920; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP5]], align 4
921; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 0
922; CHECK-NEXT:    store i32 [[TMP13]], ptr [[TMP9]], align 4
923; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 2
924; CHECK-NEXT:    store i32 [[TMP14]], ptr [[TMP10]], align 4
925; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 4
926; CHECK-NEXT:    store i32 [[TMP15]], ptr [[TMP11]], align 4
927; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 6
928; CHECK-NEXT:    store i32 [[TMP16]], ptr [[TMP12]], align 4
929; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
930; CHECK-NEXT:    [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
931; CHECK-NEXT:    br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]]
932; CHECK:       middle.block:
933; CHECK-NEXT:    br label [[SCALAR_PH]]
934; CHECK:       scalar.ph:
935; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
936; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
937; CHECK:       for.body:
938; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
939; CHECK-NEXT:    [[P_I_X:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 0
940; CHECK-NEXT:    [[P_I_Y:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1
941; CHECK-NEXT:    store i32 [[Z]], ptr [[P_I_X]], align 4
942; CHECK-NEXT:    store i32 [[Z]], ptr [[P_I_Y]], align 4
943; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
944; CHECK-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
945; CHECK-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END:%.*]], !llvm.loop [[LOOP25:![0-9]+]]
946; CHECK:       for.end:
947; CHECK-NEXT:    ret void
948;
949entry:
950  br label %for.body
951
952for.body:
953  %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ]
954  %p_i.x = getelementptr inbounds %pair.i32, ptr %p, i64 %i, i32 0
955  %p_i.y = getelementptr inbounds %pair.i32, ptr %p, i64 %i, i32 1
956  store i32 %z, ptr %p_i.x, align 4
957  %0 = load i32, ptr %p_i.x, align 4
958  store i32 %0, ptr %p_i.y, align 4
959  %i.next = add nuw nsw i64 %i, 1
960  %cond = icmp slt i64 %i.next, %n
961  br i1 %cond, label %for.body, label %for.end
962
963for.end:
964  ret void
965}
966
967; PR27626_1: Ensure a strided load is not moved before a dependent (zero
968;            distance) strided store.
969
970; void PR27626_1(struct pair *p, int n) {
971;   int s = 0;
972;   for (int i = 0; i < n; i++) {
973;     p[i].y = p[i].x;
974;     s += p[i].y
975;   }
976; }
977
978
979define i32 @PR27626_1(ptr %p, i64 %n) {
980; CHECK-LABEL: @PR27626_1(
981; CHECK-NEXT:  entry:
982; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N:%.*]], 5
983; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
984; CHECK:       vector.ph:
985; CHECK-NEXT:    [[N_MOD_VF:%.*]] = and i64 [[N]], 3
986; CHECK-NEXT:    [[TMP0:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
987; CHECK-NEXT:    [[TMP1:%.*]] = select i1 [[TMP0]], i64 4, i64 [[N_MOD_VF]]
988; CHECK-NEXT:    [[N_VEC:%.*]] = sub nsw i64 [[N]], [[TMP1]]
989; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
990; CHECK:       vector.body:
991; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
992; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP14:%.*]], [[VECTOR_BODY]] ]
993; CHECK-NEXT:    [[TMP2:%.*]] = or disjoint i64 [[INDEX]], 1
994; CHECK-NEXT:    [[TMP3:%.*]] = or disjoint i64 [[INDEX]], 2
995; CHECK-NEXT:    [[TMP4:%.*]] = or disjoint i64 [[INDEX]], 3
996; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], i64 [[INDEX]], i32 0
997; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[INDEX]], i32 1
998; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP2]], i32 1
999; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP3]], i32 1
1000; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP4]], i32 1
1001; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP5]], align 4
1002; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 0
1003; CHECK-NEXT:    store i32 [[TMP10]], ptr [[TMP6]], align 4
1004; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 2
1005; CHECK-NEXT:    store i32 [[TMP11]], ptr [[TMP7]], align 4
1006; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 4
1007; CHECK-NEXT:    store i32 [[TMP12]], ptr [[TMP8]], align 4
1008; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 6
1009; CHECK-NEXT:    store i32 [[TMP13]], ptr [[TMP9]], align 4
1010; CHECK-NEXT:    [[WIDE_VEC1:%.*]] = load <8 x i32>, ptr [[TMP6]], align 4
1011; CHECK-NEXT:    [[STRIDED_VEC2:%.*]] = shufflevector <8 x i32> [[WIDE_VEC1]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
1012; CHECK-NEXT:    [[TMP14]] = add <4 x i32> [[STRIDED_VEC2]], [[VEC_PHI]]
1013; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
1014; CHECK-NEXT:    [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1015; CHECK-NEXT:    br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]]
1016; CHECK:       middle.block:
1017; CHECK-NEXT:    [[TMP16:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP14]])
1018; CHECK-NEXT:    br label [[SCALAR_PH]]
1019; CHECK:       scalar.ph:
1020; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1021; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP16]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
1022; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
1023; CHECK:       for.body:
1024; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1025; CHECK-NEXT:    [[S:%.*]] = phi i32 [ [[TMP18:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
1026; CHECK-NEXT:    [[P_I_X:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 0
1027; CHECK-NEXT:    [[P_I_Y:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1
1028; CHECK-NEXT:    [[TMP17:%.*]] = load i32, ptr [[P_I_X]], align 4
1029; CHECK-NEXT:    store i32 [[TMP17]], ptr [[P_I_Y]], align 4
1030; CHECK-NEXT:    [[TMP18]] = add nsw i32 [[TMP17]], [[S]]
1031; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
1032; CHECK-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1033; CHECK-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END:%.*]], !llvm.loop [[LOOP27:![0-9]+]]
1034; CHECK:       for.end:
1035; CHECK-NEXT:    ret i32 [[TMP18]]
1036;
1037entry:
1038  br label %for.body
1039
1040for.body:
1041  %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ]
1042  %s = phi i32 [ %2, %for.body ], [ 0, %entry ]
1043  %p_i.x = getelementptr inbounds %pair.i32, ptr %p, i64 %i, i32 0
1044  %p_i.y = getelementptr inbounds %pair.i32, ptr %p, i64 %i, i32 1
1045  %0 = load i32, ptr %p_i.x, align 4
1046  store i32 %0, ptr %p_i.y, align 4
1047  %1 = load i32, ptr %p_i.y, align 4
1048  %2 = add nsw i32 %1, %s
1049  %i.next = add nuw nsw i64 %i, 1
1050  %cond = icmp slt i64 %i.next, %n
1051  br i1 %cond, label %for.body, label %for.end
1052
1053for.end:
1054  %3 = phi i32 [ %2, %for.body ]
1055  ret i32 %3
1056}
1057
1058; PR27626_2: Ensure a strided store is not moved after a dependent (negative
1059;            distance) strided load.
1060
1061; void PR27626_2(struct pair *p, int z, int n) {
1062;   for (int i = 0; i < n; i++) {
1063;     p[i].x = z;
1064;     p[i].y = p[i - 1].x;
1065;   }
1066; }
1067
1068
1069define void @PR27626_2(ptr %p, i64 %n, i32 %z) {
1070; CHECK-LABEL: @PR27626_2(
1071; CHECK-NEXT:  entry:
1072; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N:%.*]], 5
1073; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
1074; CHECK:       vector.ph:
1075; CHECK-NEXT:    [[N_MOD_VF:%.*]] = and i64 [[N]], 3
1076; CHECK-NEXT:    [[TMP0:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
1077; CHECK-NEXT:    [[TMP1:%.*]] = select i1 [[TMP0]], i64 4, i64 [[N_MOD_VF]]
1078; CHECK-NEXT:    [[N_VEC:%.*]] = sub nsw i64 [[N]], [[TMP1]]
1079; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
1080; CHECK:       vector.body:
1081; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1082; CHECK-NEXT:    [[TMP2:%.*]] = or disjoint i64 [[INDEX]], 1
1083; CHECK-NEXT:    [[TMP3:%.*]] = or disjoint i64 [[INDEX]], 2
1084; CHECK-NEXT:    [[TMP4:%.*]] = or disjoint i64 [[INDEX]], 3
1085; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], i64 [[INDEX]], i32 0
1086; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP2]], i32 0
1087; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP3]], i32 0
1088; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP4]], i32 0
1089; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i8, ptr [[P]], i64 -8
1090; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[INDEX]], i32 1
1091; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP2]], i32 1
1092; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP3]], i32 1
1093; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP4]], i32 1
1094; CHECK-NEXT:    store i32 [[Z:%.*]], ptr [[TMP5]], align 4
1095; CHECK-NEXT:    store i32 [[Z]], ptr [[TMP6]], align 4
1096; CHECK-NEXT:    store i32 [[Z]], ptr [[TMP7]], align 4
1097; CHECK-NEXT:    store i32 [[Z]], ptr [[TMP8]], align 4
1098; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP9]], align 4
1099; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 0
1100; CHECK-NEXT:    store i32 [[TMP14]], ptr [[TMP10]], align 4
1101; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 2
1102; CHECK-NEXT:    store i32 [[TMP15]], ptr [[TMP11]], align 4
1103; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 4
1104; CHECK-NEXT:    store i32 [[TMP16]], ptr [[TMP12]], align 4
1105; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 6
1106; CHECK-NEXT:    store i32 [[TMP17]], ptr [[TMP13]], align 4
1107; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
1108; CHECK-NEXT:    [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1109; CHECK-NEXT:    br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]]
1110; CHECK:       middle.block:
1111; CHECK-NEXT:    br label [[SCALAR_PH]]
1112; CHECK:       scalar.ph:
1113; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1114; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
1115; CHECK:       for.body:
1116; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1117; CHECK-NEXT:    [[P_I_X:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 0
1118; CHECK-NEXT:    [[P_I_MINUS_1_X:%.*]] = getelementptr inbounds i8, ptr [[P]], i64 -8
1119; CHECK-NEXT:    [[P_I_Y:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1
1120; CHECK-NEXT:    store i32 [[Z]], ptr [[P_I_X]], align 4
1121; CHECK-NEXT:    [[TMP19:%.*]] = load i32, ptr [[P_I_MINUS_1_X]], align 4
1122; CHECK-NEXT:    store i32 [[TMP19]], ptr [[P_I_Y]], align 4
1123; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
1124; CHECK-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1125; CHECK-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END:%.*]], !llvm.loop [[LOOP29:![0-9]+]]
1126; CHECK:       for.end:
1127; CHECK-NEXT:    ret void
1128;
1129entry:
1130  br label %for.body
1131
1132for.body:
1133  %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ]
1134  %i_minus_1 = add nuw nsw i64 %i, -1
1135  %p_i.x = getelementptr inbounds %pair.i32, ptr %p, i64 %i, i32 0
1136  %p_i_minus_1.x = getelementptr inbounds %pair.i32, ptr %p, i64 %i_minus_1, i32 0
1137  %p_i.y = getelementptr inbounds %pair.i32, ptr %p, i64 %i, i32 1
1138  store i32 %z, ptr %p_i.x, align 4
1139  %0 = load i32, ptr %p_i_minus_1.x, align 4
1140  store i32 %0, ptr %p_i.y, align 4
1141  %i.next = add nuw nsw i64 %i, 1
1142  %cond = icmp slt i64 %i.next, %n
1143  br i1 %cond, label %for.body, label %for.end
1144
1145for.end:
1146  ret void
1147}
1148
1149; PR27626_3: Ensure a strided load is not moved before a dependent (negative
1150;            distance) strided store.
1151
1152; void PR27626_3(struct pair *p, int z, int n) {
1153;   for (int i = 0; i < n; i++) {
1154;     p[i + 1].y = p[i].x;
1155;     s += p[i].y;
1156;   }
1157; }
1158
1159
1160define i32 @PR27626_3(ptr %p, i64 %n, i32 %z) {
1161; CHECK-LABEL: @PR27626_3(
1162; CHECK-NEXT:  entry:
1163; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N:%.*]], 5
1164; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
1165; CHECK:       vector.ph:
1166; CHECK-NEXT:    [[N_MOD_VF:%.*]] = and i64 [[N]], 3
1167; CHECK-NEXT:    [[TMP0:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
1168; CHECK-NEXT:    [[TMP1:%.*]] = select i1 [[TMP0]], i64 4, i64 [[N_MOD_VF]]
1169; CHECK-NEXT:    [[N_VEC:%.*]] = sub nsw i64 [[N]], [[TMP1]]
1170; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
1171; CHECK:       vector.body:
1172; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1173; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
1174; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP17:%.*]], [[VECTOR_BODY]] ]
1175; CHECK-NEXT:    [[TMP2:%.*]] = add nuw nsw <4 x i64> [[VEC_IND]], splat (i64 1)
1176; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [[PAIR_I32:%.*]], ptr [[P:%.*]], i64 [[INDEX]], i32 0
1177; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[INDEX]], i32 1
1178; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x i64> [[TMP2]], i64 0
1179; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP5]], i32 1
1180; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x i64> [[TMP2]], i64 1
1181; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP7]], i32 1
1182; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <4 x i64> [[TMP2]], i64 2
1183; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP9]], i32 1
1184; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <4 x i64> [[TMP2]], i64 3
1185; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[TMP11]], i32 1
1186; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i32>, ptr [[TMP3]], align 4
1187; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 0
1188; CHECK-NEXT:    store i32 [[TMP13]], ptr [[TMP6]], align 4
1189; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 2
1190; CHECK-NEXT:    store i32 [[TMP14]], ptr [[TMP8]], align 4
1191; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 4
1192; CHECK-NEXT:    store i32 [[TMP15]], ptr [[TMP10]], align 4
1193; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <8 x i32> [[WIDE_VEC]], i64 6
1194; CHECK-NEXT:    store i32 [[TMP16]], ptr [[TMP12]], align 4
1195; CHECK-NEXT:    [[WIDE_VEC1:%.*]] = load <8 x i32>, ptr [[TMP4]], align 4
1196; CHECK-NEXT:    [[STRIDED_VEC2:%.*]] = shufflevector <8 x i32> [[WIDE_VEC1]], <8 x i32> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
1197; CHECK-NEXT:    [[TMP17]] = add <4 x i32> [[STRIDED_VEC2]], [[VEC_PHI]]
1198; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
1199; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 4)
1200; CHECK-NEXT:    [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1201; CHECK-NEXT:    br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP30:![0-9]+]]
1202; CHECK:       middle.block:
1203; CHECK-NEXT:    [[TMP19:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP17]])
1204; CHECK-NEXT:    br label [[SCALAR_PH]]
1205; CHECK:       scalar.ph:
1206; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1207; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP19]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
1208; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
1209; CHECK:       for.body:
1210; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1211; CHECK-NEXT:    [[S:%.*]] = phi i32 [ [[TMP22:%.*]], [[FOR_BODY]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ]
1212; CHECK-NEXT:    [[I_PLUS_1:%.*]] = add nuw nsw i64 [[I]], 1
1213; CHECK-NEXT:    [[P_I_X:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 0
1214; CHECK-NEXT:    [[P_I_Y:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I]], i32 1
1215; CHECK-NEXT:    [[P_I_PLUS_1_Y:%.*]] = getelementptr inbounds [[PAIR_I32]], ptr [[P]], i64 [[I_PLUS_1]], i32 1
1216; CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr [[P_I_X]], align 4
1217; CHECK-NEXT:    store i32 [[TMP20]], ptr [[P_I_PLUS_1_Y]], align 4
1218; CHECK-NEXT:    [[TMP21:%.*]] = load i32, ptr [[P_I_Y]], align 4
1219; CHECK-NEXT:    [[TMP22]] = add nsw i32 [[TMP21]], [[S]]
1220; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
1221; CHECK-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1222; CHECK-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END:%.*]], !llvm.loop [[LOOP31:![0-9]+]]
1223; CHECK:       for.end:
1224; CHECK-NEXT:    ret i32 [[TMP22]]
1225;
1226entry:
1227  br label %for.body
1228
1229for.body:
1230  %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ]
1231  %s = phi i32 [ %2, %for.body ], [ 0, %entry ]
1232  %i_plus_1 = add nuw nsw i64 %i, 1
1233  %p_i.x = getelementptr inbounds %pair.i32, ptr %p, i64 %i, i32 0
1234  %p_i.y = getelementptr inbounds %pair.i32, ptr %p, i64 %i, i32 1
1235  %p_i_plus_1.y = getelementptr inbounds %pair.i32, ptr %p, i64 %i_plus_1, i32 1
1236  %0 = load i32, ptr %p_i.x, align 4
1237  store i32 %0, ptr %p_i_plus_1.y, align 4
1238  %1 = load i32, ptr %p_i.y, align 4
1239  %2 = add nsw i32 %1, %s
1240  %i.next = add nuw nsw i64 %i, 1
1241  %cond = icmp slt i64 %i.next, %n
1242  br i1 %cond, label %for.body, label %for.end
1243
1244for.end:
1245  %3 = phi i32 [ %2, %for.body ]
1246  ret i32 %3
1247}
1248
1249; PR27626_4: Ensure we form an interleaved group for strided stores in the
1250;            presence of a write-after-write dependence. We create a group for
1251;            (2) and (3) while excluding (1).
1252
1253; void PR27626_4(int *a, int x, int y, int z, int n) {
1254;   for (int i = 0; i < n; i += 2) {
1255;     a[i] = x;      // (1)
1256;     a[i] = y;      // (2)
1257;     a[i + 1] = z;  // (3)
1258;   }
1259; }
1260
1261
1262define void @PR27626_4(ptr %a, i32 %x, i32 %y, i32 %z, i64 %n) {
1263; CHECK-LABEL: @PR27626_4(
1264; CHECK-NEXT:  entry:
1265; CHECK-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 2)
1266; CHECK-NEXT:    [[TMP0:%.*]] = add nsw i64 [[SMAX]], -1
1267; CHECK-NEXT:    [[TMP1:%.*]] = lshr i64 [[TMP0]], 1
1268; CHECK-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
1269; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N]], 7
1270; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
1271; CHECK:       vector.ph:
1272; CHECK-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 9223372036854775804
1273; CHECK-NEXT:    [[IND_END:%.*]] = shl nuw i64 [[N_VEC]], 1
1274; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[Y:%.*]], i64 0
1275; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
1276; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i32> poison, i32 [[Z:%.*]], i64 0
1277; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT1]], <4 x i32> poison, <4 x i32> zeroinitializer
1278; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
1279; CHECK:       vector.body:
1280; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1281; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 1
1282; CHECK-NEXT:    [[TMP3:%.*]] = or disjoint i64 [[OFFSET_IDX]], 2
1283; CHECK-NEXT:    [[TMP4:%.*]] = or disjoint i64 [[OFFSET_IDX]], 4
1284; CHECK-NEXT:    [[TMP5:%.*]] = or disjoint i64 [[OFFSET_IDX]], 6
1285; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[OFFSET_IDX]]
1286; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP3]]
1287; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP4]]
1288; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP5]]
1289; CHECK-NEXT:    store i32 [[X:%.*]], ptr [[TMP7]], align 4
1290; CHECK-NEXT:    store i32 [[X]], ptr [[TMP8]], align 4
1291; CHECK-NEXT:    store i32 [[X]], ptr [[TMP9]], align 4
1292; CHECK-NEXT:    store i32 [[X]], ptr [[TMP10]], align 4
1293; CHECK-NEXT:    [[INTERLEAVED_VEC:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLAT]], <4 x i32> [[BROADCAST_SPLAT2]], <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>
1294; CHECK-NEXT:    store <8 x i32> [[INTERLEAVED_VEC]], ptr [[TMP7]], align 4
1295; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
1296; CHECK-NEXT:    [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1297; CHECK-NEXT:    br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP32:![0-9]+]]
1298; CHECK:       middle.block:
1299; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
1300; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
1301; CHECK:       scalar.ph:
1302; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
1303; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
1304; CHECK:       for.body:
1305; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1306; CHECK-NEXT:    [[I_PLUS_1:%.*]] = or disjoint i64 [[I]], 1
1307; CHECK-NEXT:    [[A_I:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[I]]
1308; CHECK-NEXT:    [[A_I_PLUS_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[I_PLUS_1]]
1309; CHECK-NEXT:    store i32 [[Y]], ptr [[A_I]], align 4
1310; CHECK-NEXT:    store i32 [[Z]], ptr [[A_I_PLUS_1]], align 4
1311; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 2
1312; CHECK-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1313; CHECK-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP33:![0-9]+]]
1314; CHECK:       for.end:
1315; CHECK-NEXT:    ret void
1316;
1317entry:
1318  br label %for.body
1319
1320for.body:
1321  %i = phi i64 [ %i.next, %for.body ], [ 0, %entry ]
1322  %i_plus_1 = add i64 %i, 1
1323  %a_i = getelementptr inbounds i32, ptr %a, i64 %i
1324  %a_i_plus_1 = getelementptr inbounds i32, ptr %a, i64 %i_plus_1
1325  store i32 %x, ptr %a_i, align 4
1326  store i32 %y, ptr %a_i, align 4
1327  store i32 %z, ptr %a_i_plus_1, align 4
1328  %i.next = add nuw nsw i64 %i, 2
1329  %cond = icmp slt i64 %i.next, %n
1330  br i1 %cond, label %for.body, label %for.end
1331
1332for.end:
1333  ret void
1334}
1335
1336; PR27626_5: Ensure we do not form an interleaved group for strided stores in
1337;            the presence of a write-after-write dependence.
1338
1339; void PR27626_5(int *a, int x, int y, int z, int n) {
1340;   for (int i = 3; i < n; i += 2) {
1341;     a[i - 1] = x;
1342;     a[i - 3] = y;
1343;     a[i] = z;
1344;   }
1345; }
1346
1347
1348define void @PR27626_5(ptr %a, i32 %x, i32 %y, i32 %z, i64 %n) {
1349; CHECK-LABEL: @PR27626_5(
1350; CHECK-NEXT:  entry:
1351; CHECK-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 5)
1352; CHECK-NEXT:    [[TMP0:%.*]] = add nsw i64 [[SMAX]], -4
1353; CHECK-NEXT:    [[TMP1:%.*]] = lshr i64 [[TMP0]], 1
1354; CHECK-NEXT:    [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
1355; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp slt i64 [[N]], 10
1356; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
1357; CHECK:       vector.ph:
1358; CHECK-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP2]], 9223372036854775804
1359; CHECK-NEXT:    [[TMP3:%.*]] = shl nuw i64 [[N_VEC]], 1
1360; CHECK-NEXT:    [[IND_END:%.*]] = or disjoint i64 [[TMP3]], 3
1361; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
1362; CHECK:       vector.body:
1363; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1364; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 3, i64 5, i64 7, i64 9>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
1365; CHECK-NEXT:    [[TMP4:%.*]] = shl i64 [[INDEX]], 1
1366; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = or disjoint i64 [[TMP4]], 3
1367; CHECK-NEXT:    [[TMP5:%.*]] = or disjoint i64 [[TMP4]], 5
1368; CHECK-NEXT:    [[TMP6:%.*]] = or disjoint i64 [[TMP4]], 7
1369; CHECK-NEXT:    [[TMP7:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 -1)
1370; CHECK-NEXT:    [[TMP8:%.*]] = add <4 x i64> [[VEC_IND]], splat (i64 -3)
1371; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[OFFSET_IDX]]
1372; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP5]]
1373; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP6]]
1374; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr i32, ptr [[A]], i64 [[TMP4]]
1375; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr i8, ptr [[TMP12]], i64 36
1376; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <4 x i64> [[TMP7]], i64 0
1377; CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP14]]
1378; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <4 x i64> [[TMP7]], i64 1
1379; CHECK-NEXT:    [[TMP17:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP16]]
1380; CHECK-NEXT:    [[TMP18:%.*]] = extractelement <4 x i64> [[TMP7]], i64 2
1381; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP18]]
1382; CHECK-NEXT:    [[TMP20:%.*]] = extractelement <4 x i64> [[TMP7]], i64 3
1383; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP20]]
1384; CHECK-NEXT:    [[TMP22:%.*]] = extractelement <4 x i64> [[TMP8]], i64 0
1385; CHECK-NEXT:    [[TMP23:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP22]]
1386; CHECK-NEXT:    [[TMP24:%.*]] = extractelement <4 x i64> [[TMP8]], i64 1
1387; CHECK-NEXT:    [[TMP25:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP24]]
1388; CHECK-NEXT:    [[TMP26:%.*]] = extractelement <4 x i64> [[TMP8]], i64 2
1389; CHECK-NEXT:    [[TMP27:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP26]]
1390; CHECK-NEXT:    [[TMP28:%.*]] = extractelement <4 x i64> [[TMP8]], i64 3
1391; CHECK-NEXT:    [[TMP29:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP28]]
1392; CHECK-NEXT:    store i32 [[X:%.*]], ptr [[TMP15]], align 4
1393; CHECK-NEXT:    store i32 [[X]], ptr [[TMP17]], align 4
1394; CHECK-NEXT:    store i32 [[X]], ptr [[TMP19]], align 4
1395; CHECK-NEXT:    store i32 [[X]], ptr [[TMP21]], align 4
1396; CHECK-NEXT:    store i32 [[Y:%.*]], ptr [[TMP23]], align 4
1397; CHECK-NEXT:    store i32 [[Y]], ptr [[TMP25]], align 4
1398; CHECK-NEXT:    store i32 [[Y]], ptr [[TMP27]], align 4
1399; CHECK-NEXT:    store i32 [[Y]], ptr [[TMP29]], align 4
1400; CHECK-NEXT:    store i32 [[Z:%.*]], ptr [[TMP9]], align 4
1401; CHECK-NEXT:    store i32 [[Z]], ptr [[TMP10]], align 4
1402; CHECK-NEXT:    store i32 [[Z]], ptr [[TMP11]], align 4
1403; CHECK-NEXT:    store i32 [[Z]], ptr [[TMP13]], align 4
1404; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
1405; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 8)
1406; CHECK-NEXT:    [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1407; CHECK-NEXT:    br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP34:![0-9]+]]
1408; CHECK:       middle.block:
1409; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
1410; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
1411; CHECK:       scalar.ph:
1412; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 3, [[ENTRY:%.*]] ]
1413; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
1414; CHECK:       for.body:
1415; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
1416; CHECK-NEXT:    [[A_I:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[I]]
1417; CHECK-NEXT:    [[TMP31:%.*]] = getelementptr i32, ptr [[A]], i64 [[I]]
1418; CHECK-NEXT:    [[A_I_MINUS_1:%.*]] = getelementptr i8, ptr [[TMP31]], i64 -4
1419; CHECK-NEXT:    [[TMP32:%.*]] = getelementptr i32, ptr [[A]], i64 [[I]]
1420; CHECK-NEXT:    [[A_I_MINUS_3:%.*]] = getelementptr i8, ptr [[TMP32]], i64 -12
1421; CHECK-NEXT:    store i32 [[X]], ptr [[A_I_MINUS_1]], align 4
1422; CHECK-NEXT:    store i32 [[Y]], ptr [[A_I_MINUS_3]], align 4
1423; CHECK-NEXT:    store i32 [[Z]], ptr [[A_I]], align 4
1424; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 2
1425; CHECK-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
1426; CHECK-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP35:![0-9]+]]
1427; CHECK:       for.end:
1428; CHECK-NEXT:    ret void
1429;
1430entry:
1431  br label %for.body
1432
1433for.body:
1434  %i = phi i64 [ %i.next, %for.body ], [ 3, %entry ]
1435  %i_minus_1 = sub i64 %i, 1
1436  %i_minus_3 = sub i64 %i_minus_1, 2
1437  %a_i = getelementptr inbounds i32, ptr %a, i64 %i
1438  %a_i_minus_1 = getelementptr inbounds i32, ptr %a, i64 %i_minus_1
1439  %a_i_minus_3 = getelementptr inbounds i32, ptr %a, i64 %i_minus_3
1440  store i32 %x, ptr %a_i_minus_1, align 4
1441  store i32 %y, ptr %a_i_minus_3, align 4
1442  store i32 %z, ptr %a_i, align 4
1443  %i.next = add nuw nsw i64 %i, 2
1444  %cond = icmp slt i64 %i.next, %n
1445  br i1 %cond, label %for.body, label %for.end
1446
1447for.end:
1448  ret void
1449}
1450
1451; PR34743: Ensure that a cast which needs to sink after a load that belongs to
1452; an interleaved group, indeeded gets sunk.
1453
1454; void PR34743(short *a, int *b, int n) {
1455;   for (int i = 0, iv = 0; iv < n; i++, iv += 2) {
1456;     b[i] = a[iv] * a[iv+1] * a[iv+2];
1457;   }
1458; }
1459
1460
1461define void @PR34743(ptr %a, ptr %b, i64 %n) {
1462; CHECK-LABEL: @PR34743(
1463; CHECK-NEXT:  entry:
1464; CHECK-NEXT:    [[DOTPRE:%.*]] = load i16, ptr [[A:%.*]], align 2
1465; CHECK-NEXT:    [[TMP0:%.*]] = lshr i64 [[N:%.*]], 1
1466; CHECK-NEXT:    [[TMP1:%.*]] = add nuw i64 [[TMP0]], 1
1467; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 6
1468; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
1469; CHECK:       vector.memcheck:
1470; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[N]], 1
1471; CHECK-NEXT:    [[TMP3:%.*]] = and i64 [[TMP2]], -4
1472; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr i8, ptr [[B:%.*]], i64 [[TMP3]]
1473; CHECK-NEXT:    [[SCEVGEP:%.*]] = getelementptr i8, ptr [[TMP4]], i64 4
1474; CHECK-NEXT:    [[SCEVGEP1:%.*]] = getelementptr nuw i8, ptr [[A]], i64 2
1475; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP3]]
1476; CHECK-NEXT:    [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[TMP5]], i64 6
1477; CHECK-NEXT:    [[BOUND0:%.*]] = icmp ult ptr [[B]], [[SCEVGEP2]]
1478; CHECK-NEXT:    [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP1]], [[SCEVGEP]]
1479; CHECK-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
1480; CHECK-NEXT:    br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
1481; CHECK:       vector.ph:
1482; CHECK-NEXT:    [[N_VEC:%.*]] = and i64 [[TMP1]], -4
1483; CHECK-NEXT:    [[IND_END:%.*]] = shl i64 [[N_VEC]], 1
1484; CHECK-NEXT:    [[VECTOR_RECUR_INIT:%.*]] = insertelement <4 x i16> poison, i16 [[DOTPRE]], i64 3
1485; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
1486; CHECK:       vector.body:
1487; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
1488; CHECK-NEXT:    [[VECTOR_RECUR:%.*]] = phi <4 x i16> [ [[VECTOR_RECUR_INIT]], [[VECTOR_PH]] ], [ [[STRIDED_VEC4:%.*]], [[VECTOR_BODY]] ]
1489; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 1
1490; CHECK-NEXT:    [[TMP6:%.*]] = or disjoint i64 [[OFFSET_IDX]], 1
1491; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i16, ptr [[A]], i64 [[TMP6]]
1492; CHECK-NEXT:    [[WIDE_VEC:%.*]] = load <8 x i16>, ptr [[TMP7]], align 4
1493; CHECK-NEXT:    [[STRIDED_VEC:%.*]] = shufflevector <8 x i16> [[WIDE_VEC]], <8 x i16> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
1494; CHECK-NEXT:    [[STRIDED_VEC4]] = shufflevector <8 x i16> [[WIDE_VEC]], <8 x i16> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
1495; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <4 x i16> [[VECTOR_RECUR]], <4 x i16> [[STRIDED_VEC4]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
1496; CHECK-NEXT:    [[TMP9:%.*]] = sext <4 x i16> [[TMP8]] to <4 x i32>
1497; CHECK-NEXT:    [[TMP10:%.*]] = sext <4 x i16> [[STRIDED_VEC]] to <4 x i32>
1498; CHECK-NEXT:    [[TMP11:%.*]] = sext <4 x i16> [[STRIDED_VEC4]] to <4 x i32>
1499; CHECK-NEXT:    [[TMP12:%.*]] = mul nsw <4 x i32> [[TMP9]], [[TMP10]]
1500; CHECK-NEXT:    [[TMP13:%.*]] = mul nsw <4 x i32> [[TMP12]], [[TMP11]]
1501; CHECK-NEXT:    [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDEX]]
1502; CHECK-NEXT:    store <4 x i32> [[TMP13]], ptr [[TMP14]], align 4, !alias.scope [[META36:![0-9]+]], !noalias [[META39:![0-9]+]]
1503; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
1504; CHECK-NEXT:    [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
1505; CHECK-NEXT:    br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP41:![0-9]+]]
1506; CHECK:       middle.block:
1507; CHECK-NEXT:    [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <8 x i16> [[WIDE_VEC]], i64 7
1508; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP1]], [[N_VEC]]
1509; CHECK-NEXT:    br i1 [[CMP_N]], label [[END:%.*]], label [[SCALAR_PH]]
1510; CHECK:       scalar.ph:
1511; CHECK-NEXT:    [[SCALAR_RECUR_INIT:%.*]] = phi i16 [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ], [ [[DOTPRE]], [[VECTOR_MEMCHECK]] ], [ [[DOTPRE]], [[ENTRY:%.*]] ]
1512; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ENTRY]] ]
1513; CHECK-NEXT:    [[BC_RESUME_VAL3:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[ENTRY]] ]
1514; CHECK-NEXT:    br label [[LOOP:%.*]]
1515; CHECK:       loop:
1516; CHECK-NEXT:    [[TMP16:%.*]] = phi i16 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[LOAD2:%.*]], [[LOOP]] ]
1517; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV2:%.*]], [[LOOP]] ]
1518; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ], [ [[I1:%.*]], [[LOOP]] ]
1519; CHECK-NEXT:    [[CONV:%.*]] = sext i16 [[TMP16]] to i32
1520; CHECK-NEXT:    [[I1]] = add nuw nsw i64 [[I]], 1
1521; CHECK-NEXT:    [[IV1:%.*]] = or disjoint i64 [[IV]], 1
1522; CHECK-NEXT:    [[IV2]] = add nuw nsw i64 [[IV]], 2
1523; CHECK-NEXT:    [[GEP1:%.*]] = getelementptr inbounds i16, ptr [[A]], i64 [[IV1]]
1524; CHECK-NEXT:    [[LOAD1:%.*]] = load i16, ptr [[GEP1]], align 4
1525; CHECK-NEXT:    [[CONV1:%.*]] = sext i16 [[LOAD1]] to i32
1526; CHECK-NEXT:    [[GEP2:%.*]] = getelementptr inbounds i16, ptr [[A]], i64 [[IV2]]
1527; CHECK-NEXT:    [[LOAD2]] = load i16, ptr [[GEP2]], align 4
1528; CHECK-NEXT:    [[CONV2:%.*]] = sext i16 [[LOAD2]] to i32
1529; CHECK-NEXT:    [[MUL01:%.*]] = mul nsw i32 [[CONV]], [[CONV1]]
1530; CHECK-NEXT:    [[MUL012:%.*]] = mul nsw i32 [[MUL01]], [[CONV2]]
1531; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[I]]
1532; CHECK-NEXT:    store i32 [[MUL012]], ptr [[ARRAYIDX5]], align 4
1533; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i64 [[IV]], [[N]]
1534; CHECK-NEXT:    br i1 [[EXITCOND]], label [[END]], label [[LOOP]], !llvm.loop [[LOOP42:![0-9]+]]
1535; CHECK:       end:
1536; CHECK-NEXT:    ret void
1537;
1538entry:
1539  %.pre = load i16, ptr %a
1540  br label %loop
1541
1542loop:
1543  %0 = phi i16 [ %.pre, %entry ], [ %load2, %loop ]
1544  %iv = phi i64 [ 0, %entry ], [ %iv2, %loop ]
1545  %i = phi i64 [ 0, %entry ], [ %i1, %loop ]
1546  %conv = sext i16 %0 to i32
1547  %i1 = add nuw nsw i64 %i, 1
1548  %iv1 = add nuw nsw i64 %iv, 1
1549  %iv2 = add nuw nsw i64 %iv, 2
1550  %gep1 = getelementptr inbounds i16, ptr %a, i64 %iv1
1551  %load1 = load i16, ptr %gep1, align 4
1552  %conv1 = sext i16 %load1 to i32
1553  %gep2 = getelementptr inbounds i16, ptr %a, i64 %iv2
1554  %load2 = load i16, ptr %gep2, align 4
1555  %conv2 = sext i16 %load2 to i32
1556  %mul01 = mul nsw i32 %conv, %conv1
1557  %mul012 = mul nsw i32 %mul01, %conv2
1558  %arrayidx5 = getelementptr inbounds i32, ptr %b, i64 %i
1559  store i32 %mul012, ptr %arrayidx5
1560  %exitcond = icmp eq i64 %iv, %n
1561  br i1 %exitcond, label %end, label %loop
1562
1563end:
1564  ret void
1565}
1566
1567attributes #0 = { "unsafe-fp-math"="true" }
1568