1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -S -slp-vectorizer -slp-threshold=-18 -dce -instcombine -pass-remarks-output=%t < %s | FileCheck %s
3; RUN: cat %t | FileCheck -check-prefix=YAML %s
4; RUN: opt -S -passes='slp-vectorizer,dce,instcombine' -slp-threshold=-18 -pass-remarks-output=%t < %s | FileCheck %s
5; RUN: cat %t | FileCheck -check-prefix=YAML %s
6
7
8target datalayout = "e-m:e-i32:64-i128:128-n32:64-S128"
9target triple = "aarch64--linux-gnu"
10
11; These tests check that we remove from consideration pairs of seed
12; getelementptrs when they are known to have a constant difference. Such pairs
13; are likely not good candidates for vectorization since one can be computed
14; from the other. We use an unprofitable threshold to force vectorization.
15;
16; int getelementptr(int *g, int n, int w, int x, int y, int z) {
17;   int sum = 0;
18;   for (int i = 0; i < n ; ++i) {
19;     sum += g[2*i + w]; sum += g[2*i + x];
20;     sum += g[2*i + y]; sum += g[2*i + z];
21;   }
22;   return sum;
23; }
24;
25
26; YAML:      --- !Passed
27; YAML-NEXT: Pass:            slp-vectorizer
28; YAML-NEXT: Name:            VectorizedList
29; YAML-NEXT: Function:        getelementptr_4x32
30; YAML-NEXT: Args:
31; YAML-NEXT:   - String:          'SLP vectorized with cost '
32; YAML-NEXT:   - Cost:            '11'
33; YAML-NEXT:   - String:          ' and with tree size '
34; YAML-NEXT:   - TreeSize:        '5'
35
36; YAML:      --- !Passed
37; YAML-NEXT: Pass:            slp-vectorizer
38; YAML-NEXT: Name:            VectorizedList
39; YAML-NEXT: Function:        getelementptr_4x32
40; YAML-NEXT: Args:
41; YAML-NEXT:   - String:          'SLP vectorized with cost '
42; YAML-NEXT:   - Cost:            '16'
43; YAML-NEXT:   - String:          ' and with tree size '
44; YAML-NEXT:   - TreeSize:        '3'
45
46define i32 @getelementptr_4x32(i32* nocapture readonly %g, i32 %n, i32 %x, i32 %y, i32 %z) {
47; CHECK-LABEL: @getelementptr_4x32(
48; CHECK-NEXT:  entry:
49; CHECK-NEXT:    [[CMP31:%.*]] = icmp sgt i32 [[N:%.*]], 0
50; CHECK-NEXT:    br i1 [[CMP31]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
51; CHECK:       for.body.preheader:
52; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>, i32 [[X:%.*]], i32 1
53; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x i32> [[TMP0]], i32 [[Y:%.*]], i32 2
54; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x i32> [[TMP1]], i32 [[Z:%.*]], i32 3
55; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
56; CHECK:       for.cond.cleanup.loopexit:
57; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x i32> [[TMP21:%.*]], i32 1
58; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]
59; CHECK:       for.cond.cleanup:
60; CHECK-NEXT:    [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[TMP3]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ]
61; CHECK-NEXT:    ret i32 [[SUM_0_LCSSA]]
62; CHECK:       for.body:
63; CHECK-NEXT:    [[TMP4:%.*]] = phi <2 x i32> [ zeroinitializer, [[FOR_BODY_PREHEADER]] ], [ [[TMP21]], [[FOR_BODY]] ]
64; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x i32> [[TMP4]], i32 0
65; CHECK-NEXT:    [[T4:%.*]] = shl nsw i32 [[TMP5]], 1
66; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <4 x i32> undef, i32 [[T4]], i32 0
67; CHECK-NEXT:    [[TMP7:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> undef, <4 x i32> zeroinitializer
68; CHECK-NEXT:    [[TMP8:%.*]] = add nsw <4 x i32> [[TMP7]], [[TMP2]]
69; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <4 x i32> [[TMP8]], i32 0
70; CHECK-NEXT:    [[TMP10:%.*]] = sext i32 [[TMP9]] to i64
71; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[G:%.*]], i64 [[TMP10]]
72; CHECK-NEXT:    [[T6:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
73; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <2 x i32> [[TMP4]], i32 1
74; CHECK-NEXT:    [[ADD1:%.*]] = add nsw i32 [[T6]], [[TMP11]]
75; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <4 x i32> [[TMP8]], i32 1
76; CHECK-NEXT:    [[TMP13:%.*]] = sext i32 [[TMP12]] to i64
77; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP13]]
78; CHECK-NEXT:    [[T8:%.*]] = load i32, i32* [[ARRAYIDX5]], align 4
79; CHECK-NEXT:    [[ADD6:%.*]] = add nsw i32 [[ADD1]], [[T8]]
80; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <4 x i32> [[TMP8]], i32 2
81; CHECK-NEXT:    [[TMP15:%.*]] = sext i32 [[TMP14]] to i64
82; CHECK-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP15]]
83; CHECK-NEXT:    [[T10:%.*]] = load i32, i32* [[ARRAYIDX10]], align 4
84; CHECK-NEXT:    [[ADD11:%.*]] = add nsw i32 [[ADD6]], [[T10]]
85; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <4 x i32> [[TMP8]], i32 3
86; CHECK-NEXT:    [[TMP17:%.*]] = sext i32 [[TMP16]] to i64
87; CHECK-NEXT:    [[ARRAYIDX15:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP17]]
88; CHECK-NEXT:    [[T12:%.*]] = load i32, i32* [[ARRAYIDX15]], align 4
89; CHECK-NEXT:    [[TMP18:%.*]] = insertelement <2 x i32> undef, i32 [[TMP5]], i32 0
90; CHECK-NEXT:    [[TMP19:%.*]] = insertelement <2 x i32> [[TMP18]], i32 [[ADD11]], i32 1
91; CHECK-NEXT:    [[TMP20:%.*]] = insertelement <2 x i32> <i32 1, i32 undef>, i32 [[T12]], i32 1
92; CHECK-NEXT:    [[TMP21]] = add nsw <2 x i32> [[TMP19]], [[TMP20]]
93; CHECK-NEXT:    [[TMP22:%.*]] = extractelement <2 x i32> [[TMP21]], i32 0
94; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[TMP22]], [[N]]
95; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]]
96;
97entry:
98  %cmp31 = icmp sgt i32 %n, 0
99  br i1 %cmp31, label %for.body.preheader, label %for.cond.cleanup
100
101for.body.preheader:
102  br label %for.body
103
104for.cond.cleanup.loopexit:
105  br label %for.cond.cleanup
106
107for.cond.cleanup:
108  %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add16, %for.cond.cleanup.loopexit ]
109  ret i32 %sum.0.lcssa
110
111for.body:
112  %indvars.iv = phi i32 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
113  %sum.032 = phi i32 [ 0, %for.body.preheader ], [ %add16, %for.body ]
114  %t4 = shl nsw i32 %indvars.iv, 1
115  %t5 = add nsw i32 %t4, 0
116  %arrayidx = getelementptr inbounds i32, i32* %g, i32 %t5
117  %t6 = load i32, i32* %arrayidx, align 4
118  %add1 = add nsw i32 %t6, %sum.032
119  %t7 = add nsw i32 %t4, %x
120  %arrayidx5 = getelementptr inbounds i32, i32* %g, i32 %t7
121  %t8 = load i32, i32* %arrayidx5, align 4
122  %add6 = add nsw i32 %add1, %t8
123  %t9 = add nsw i32 %t4, %y
124  %arrayidx10 = getelementptr inbounds i32, i32* %g, i32 %t9
125  %t10 = load i32, i32* %arrayidx10, align 4
126  %add11 = add nsw i32 %add6, %t10
127  %t11 = add nsw i32 %t4, %z
128  %arrayidx15 = getelementptr inbounds i32, i32* %g, i32 %t11
129  %t12 = load i32, i32* %arrayidx15, align 4
130  %add16 = add nsw i32 %add11, %t12
131  %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1
132  %exitcond = icmp eq i32 %indvars.iv.next , %n
133  br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body
134}
135
136; YAML:      --- !Passed
137; YAML-NEXT: Pass:            slp-vectorizer
138; YAML-NEXT: Name:            VectorizedList
139; YAML-NEXT: Function:        getelementptr_2x32
140; YAML-NEXT: Args:
141; YAML-NEXT:   - String:          'SLP vectorized with cost '
142; YAML-NEXT:   - Cost:            '11'
143; YAML-NEXT:   - String:          ' and with tree size '
144; YAML-NEXT:   - TreeSize:        '5'
145
146; YAML:      --- !Passed
147; YAML-NEXT: Pass:            slp-vectorizer
148; YAML-NEXT: Name:            VectorizedList
149; YAML-NEXT: Function:        getelementptr_2x32
150; YAML-NEXT: Args:
151; YAML-NEXT:   - String:          'SLP vectorized with cost '
152; YAML-NEXT:   - Cost:            '6'
153; YAML-NEXT:   - String:          ' and with tree size '
154; YAML-NEXT:   - TreeSize:        '3'
155
156define i32 @getelementptr_2x32(i32* nocapture readonly %g, i32 %n, i32 %x, i32 %y, i32 %z) {
157; CHECK-LABEL: @getelementptr_2x32(
158; CHECK-NEXT:  entry:
159; CHECK-NEXT:    [[CMP31:%.*]] = icmp sgt i32 [[N:%.*]], 0
160; CHECK-NEXT:    br i1 [[CMP31]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
161; CHECK:       for.body.preheader:
162; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <2 x i32> undef, i32 [[Y:%.*]], i32 0
163; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> [[TMP0]], i32 [[Z:%.*]], i32 1
164; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
165; CHECK:       for.cond.cleanup.loopexit:
166; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x i32> [[TMP18:%.*]], i32 1
167; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]
168; CHECK:       for.cond.cleanup:
169; CHECK-NEXT:    [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[TMP2]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ]
170; CHECK-NEXT:    ret i32 [[SUM_0_LCSSA]]
171; CHECK:       for.body:
172; CHECK-NEXT:    [[TMP3:%.*]] = phi <2 x i32> [ zeroinitializer, [[FOR_BODY_PREHEADER]] ], [ [[TMP18]], [[FOR_BODY]] ]
173; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i32> [[TMP3]], i32 0
174; CHECK-NEXT:    [[T4:%.*]] = shl nsw i32 [[TMP4]], 1
175; CHECK-NEXT:    [[TMP5:%.*]] = sext i32 [[T4]] to i64
176; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[G:%.*]], i64 [[TMP5]]
177; CHECK-NEXT:    [[T6:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
178; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i32> [[TMP3]], i32 1
179; CHECK-NEXT:    [[ADD1:%.*]] = add nsw i32 [[T6]], [[TMP6]]
180; CHECK-NEXT:    [[T7:%.*]] = or i32 [[T4]], 1
181; CHECK-NEXT:    [[TMP7:%.*]] = sext i32 [[T7]] to i64
182; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP7]]
183; CHECK-NEXT:    [[T8:%.*]] = load i32, i32* [[ARRAYIDX5]], align 4
184; CHECK-NEXT:    [[ADD6:%.*]] = add nsw i32 [[ADD1]], [[T8]]
185; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <2 x i32> undef, i32 [[T4]], i32 0
186; CHECK-NEXT:    [[TMP9:%.*]] = shufflevector <2 x i32> [[TMP8]], <2 x i32> undef, <2 x i32> zeroinitializer
187; CHECK-NEXT:    [[TMP10:%.*]] = add nsw <2 x i32> [[TMP9]], [[TMP1]]
188; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <2 x i32> [[TMP10]], i32 0
189; CHECK-NEXT:    [[TMP12:%.*]] = sext i32 [[TMP11]] to i64
190; CHECK-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP12]]
191; CHECK-NEXT:    [[T10:%.*]] = load i32, i32* [[ARRAYIDX10]], align 4
192; CHECK-NEXT:    [[ADD11:%.*]] = add nsw i32 [[ADD6]], [[T10]]
193; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x i32> [[TMP10]], i32 1
194; CHECK-NEXT:    [[TMP14:%.*]] = sext i32 [[TMP13]] to i64
195; CHECK-NEXT:    [[ARRAYIDX15:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP14]]
196; CHECK-NEXT:    [[T12:%.*]] = load i32, i32* [[ARRAYIDX15]], align 4
197; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <2 x i32> undef, i32 [[TMP4]], i32 0
198; CHECK-NEXT:    [[TMP16:%.*]] = insertelement <2 x i32> [[TMP15]], i32 [[ADD11]], i32 1
199; CHECK-NEXT:    [[TMP17:%.*]] = insertelement <2 x i32> <i32 1, i32 undef>, i32 [[T12]], i32 1
200; CHECK-NEXT:    [[TMP18]] = add nsw <2 x i32> [[TMP16]], [[TMP17]]
201; CHECK-NEXT:    [[TMP19:%.*]] = extractelement <2 x i32> [[TMP18]], i32 0
202; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[TMP19]], [[N]]
203; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]]
204;
205entry:
206  %cmp31 = icmp sgt i32 %n, 0
207  br i1 %cmp31, label %for.body.preheader, label %for.cond.cleanup
208
209for.body.preheader:
210  br label %for.body
211
212for.cond.cleanup.loopexit:
213  br label %for.cond.cleanup
214
215for.cond.cleanup:
216  %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add16, %for.cond.cleanup.loopexit ]
217  ret i32 %sum.0.lcssa
218
219for.body:
220  %indvars.iv = phi i32 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
221  %sum.032 = phi i32 [ 0, %for.body.preheader ], [ %add16, %for.body ]
222  %t4 = shl nsw i32 %indvars.iv, 1
223  %t5 = add nsw i32 %t4, 0
224  %arrayidx = getelementptr inbounds i32, i32* %g, i32 %t5
225  %t6 = load i32, i32* %arrayidx, align 4
226  %add1 = add nsw i32 %t6, %sum.032
227  %t7 = add nsw i32 %t4, 1
228  %arrayidx5 = getelementptr inbounds i32, i32* %g, i32 %t7
229  %t8 = load i32, i32* %arrayidx5, align 4
230  %add6 = add nsw i32 %add1, %t8
231  %t9 = add nsw i32 %t4, %y
232  %arrayidx10 = getelementptr inbounds i32, i32* %g, i32 %t9
233  %t10 = load i32, i32* %arrayidx10, align 4
234  %add11 = add nsw i32 %add6, %t10
235  %t11 = add nsw i32 %t4, %z
236  %arrayidx15 = getelementptr inbounds i32, i32* %g, i32 %t11
237  %t12 = load i32, i32* %arrayidx15, align 4
238  %add16 = add nsw i32 %add11, %t12
239  %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1
240  %exitcond = icmp eq i32 %indvars.iv.next , %n
241  br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body
242}
243