xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/AArch64/getelementptr.ll (revision 462cb3cd6cecd0511ecaf0e3ebcaba455ece587d)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -S -mtriple=aarch64--linux-gnu -passes=slp-vectorizer,dce,instcombine -slp-threshold=-7 -pass-remarks-output=%t < %s | FileCheck %s
3; RUN: cat %t | FileCheck -check-prefix=YAML %s
4; RUN: opt -S -mtriple=aarch64--linux-gnu -passes='slp-vectorizer,dce,instcombine' -slp-threshold=-7 -pass-remarks-output=%t < %s | FileCheck %s
5; RUN: cat %t | FileCheck -check-prefix=YAML %s
6
7; These tests check that we remove from consideration pairs of seed
8; getelementptrs when they are known to have a constant difference. Such pairs
9; are likely not good candidates for vectorization since one can be computed
10; from the other. We use an unprofitable threshold to force vectorization.
11;
12; int getelementptr(int *g, int n, int w, int x, int y, int z) {
13;   int sum = 0;
14;   for (int i = 0; i < n ; ++i) {
15;     sum += g[2*i + w]; sum += g[2*i + x];
16;     sum += g[2*i + y]; sum += g[2*i + z];
17;   }
18;   return sum;
19; }
20;
21
22; YAML-LABEL: Function:        getelementptr_4x32
23; YAML:      --- !Passed
24; YAML-NEXT: Pass:            slp-vectorizer
25; YAML-NEXT: Name:            VectorizedList
26; YAML-NEXT: Function:        getelementptr_4x32
27; YAML-NEXT: Args:
28; YAML-NEXT:   - String:          'SLP vectorized with cost '
29; YAML-NEXT:   - Cost:            '4'
30; YAML-NEXT:   - String:          ' and with tree size '
31; YAML-NEXT:   - TreeSize:        '3'
32
33; YAML:      --- !Passed
34; YAML-NEXT: Pass:            slp-vectorizer
35; YAML-NEXT: Name:            VectorizedList
36; YAML-NEXT: Function:        getelementptr_4x32
37; YAML-NEXT: Args:
38; YAML-NEXT:   - String:          'SLP vectorized with cost '
39; YAML-NEXT:   - Cost:            '6'
40; YAML-NEXT:   - String:          ' and with tree size '
41; YAML-NEXT:   - TreeSize:        '3'
42
43define i32 @getelementptr_4x32(ptr nocapture readonly %g, i32 %n, i32 %x, i32 %y, i32 %z) {
44; CHECK-LABEL: @getelementptr_4x32(
45; CHECK-NEXT:  entry:
46; CHECK-NEXT:    [[CMP31:%.*]] = icmp sgt i32 [[N:%.*]], 0
47; CHECK-NEXT:    br i1 [[CMP31]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
48; CHECK:       for.body.preheader:
49; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <2 x i32> <i32 0, i32 poison>, i32 [[X:%.*]], i64 1
50; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> poison, i32 [[Y:%.*]], i64 0
51; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[Z:%.*]], i64 1
52; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
53; CHECK:       for.cond.cleanup.loopexit:
54; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]
55; CHECK:       for.cond.cleanup:
56; CHECK-NEXT:    [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[ADD16:%.*]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ]
57; CHECK-NEXT:    ret i32 [[SUM_0_LCSSA]]
58; CHECK:       for.body:
59; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
60; CHECK-NEXT:    [[SUM_032:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[ADD16]], [[FOR_BODY]] ]
61; CHECK-NEXT:    [[T4:%.*]] = shl nuw nsw i32 [[INDVARS_IV]], 1
62; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x i32> poison, i32 [[T4]], i64 0
63; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <2 x i32> [[TMP3]], <2 x i32> poison, <2 x i32> zeroinitializer
64; CHECK-NEXT:    [[TMP5:%.*]] = add nsw <2 x i32> [[TMP4]], [[TMP0]]
65; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i32> [[TMP5]], i64 0
66; CHECK-NEXT:    [[TMP7:%.*]] = zext nneg i32 [[TMP6]] to i64
67; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[G:%.*]], i64 [[TMP7]]
68; CHECK-NEXT:    [[T6:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
69; CHECK-NEXT:    [[ADD1:%.*]] = add nsw i32 [[T6]], [[SUM_032]]
70; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <2 x i32> [[TMP5]], i64 1
71; CHECK-NEXT:    [[TMP9:%.*]] = sext i32 [[TMP8]] to i64
72; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, ptr [[G]], i64 [[TMP9]]
73; CHECK-NEXT:    [[T8:%.*]] = load i32, ptr [[ARRAYIDX5]], align 4
74; CHECK-NEXT:    [[ADD6:%.*]] = add nsw i32 [[ADD1]], [[T8]]
75; CHECK-NEXT:    [[TMP10:%.*]] = add nsw <2 x i32> [[TMP4]], [[TMP2]]
76; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <2 x i32> [[TMP10]], i64 0
77; CHECK-NEXT:    [[TMP12:%.*]] = sext i32 [[TMP11]] to i64
78; CHECK-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds i32, ptr [[G]], i64 [[TMP12]]
79; CHECK-NEXT:    [[T10:%.*]] = load i32, ptr [[ARRAYIDX10]], align 4
80; CHECK-NEXT:    [[ADD11:%.*]] = add nsw i32 [[ADD6]], [[T10]]
81; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x i32> [[TMP10]], i64 1
82; CHECK-NEXT:    [[TMP14:%.*]] = sext i32 [[TMP13]] to i64
83; CHECK-NEXT:    [[ARRAYIDX15:%.*]] = getelementptr inbounds i32, ptr [[G]], i64 [[TMP14]]
84; CHECK-NEXT:    [[T12:%.*]] = load i32, ptr [[ARRAYIDX15]], align 4
85; CHECK-NEXT:    [[ADD16]] = add nsw i32 [[ADD11]], [[T12]]
86; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
87; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[N]]
88; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]]
89;
90entry:
91  %cmp31 = icmp sgt i32 %n, 0
92  br i1 %cmp31, label %for.body.preheader, label %for.cond.cleanup
93
94for.body.preheader:
95  br label %for.body
96
97for.cond.cleanup.loopexit:
98  br label %for.cond.cleanup
99
100for.cond.cleanup:
101  %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add16, %for.cond.cleanup.loopexit ]
102  ret i32 %sum.0.lcssa
103
104for.body:
105  %indvars.iv = phi i32 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
106  %sum.032 = phi i32 [ 0, %for.body.preheader ], [ %add16, %for.body ]
107  %t4 = shl nsw i32 %indvars.iv, 1
108  %t5 = add nsw i32 %t4, 0
109  %arrayidx = getelementptr inbounds i32, ptr %g, i32 %t5
110  %t6 = load i32, ptr %arrayidx, align 4
111  %add1 = add nsw i32 %t6, %sum.032
112  %t7 = add nsw i32 %t4, %x
113  %arrayidx5 = getelementptr inbounds i32, ptr %g, i32 %t7
114  %t8 = load i32, ptr %arrayidx5, align 4
115  %add6 = add nsw i32 %add1, %t8
116  %t9 = add nsw i32 %t4, %y
117  %arrayidx10 = getelementptr inbounds i32, ptr %g, i32 %t9
118  %t10 = load i32, ptr %arrayidx10, align 4
119  %add11 = add nsw i32 %add6, %t10
120  %t11 = add nsw i32 %t4, %z
121  %arrayidx15 = getelementptr inbounds i32, ptr %g, i32 %t11
122  %t12 = load i32, ptr %arrayidx15, align 4
123  %add16 = add nsw i32 %add11, %t12
124  %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1
125  %exitcond = icmp eq i32 %indvars.iv.next , %n
126  br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body
127}
128
129; YAML-LABEL: Function:        getelementptr_2x32
130; YAML:      --- !Passed
131; YAML:      Pass:            slp-vectorizer
132; YAML:      Name:            VectorizedList
133; YAML:      Function:        getelementptr_2x32
134; YAML:     Args:
135; YAML:        - String:          'SLP vectorized with cost '
136; YAML:        - Cost:            '4'
137; YAML-NEXT:   - String:          ' and with tree size '
138; YAML-NEXT:   - TreeSize:        '3'
139
140define i32 @getelementptr_2x32(ptr nocapture readonly %g, i32 %n, i32 %x, i32 %y, i32 %z) {
141; CHECK-LABEL: @getelementptr_2x32(
142; CHECK-NEXT:  entry:
143; CHECK-NEXT:    [[CMP31:%.*]] = icmp sgt i32 [[N:%.*]], 0
144; CHECK-NEXT:    br i1 [[CMP31]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
145; CHECK:       for.body.preheader:
146; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <2 x i32> <i32 0, i32 poison>, i32 [[Y:%.*]], i64 1
147; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
148; CHECK:       for.cond.cleanup.loopexit:
149; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]
150; CHECK:       for.cond.cleanup:
151; CHECK-NEXT:    [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[OP_RDX:%.*]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ]
152; CHECK-NEXT:    ret i32 [[SUM_0_LCSSA]]
153; CHECK:       for.body:
154; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
155; CHECK-NEXT:    [[SUM_032:%.*]] = phi i32 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[OP_RDX]], [[FOR_BODY]] ]
156; CHECK-NEXT:    [[T4:%.*]] = shl nuw nsw i32 [[INDVARS_IV]], 1
157; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x i32> poison, i32 [[T4]], i64 0
158; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> poison, <2 x i32> zeroinitializer
159; CHECK-NEXT:    [[TMP3:%.*]] = add nsw <2 x i32> [[TMP2]], [[TMP0]]
160; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i32> [[TMP3]], i64 0
161; CHECK-NEXT:    [[TMP5:%.*]] = zext nneg i32 [[TMP4]] to i64
162; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw i32, ptr [[G:%.*]], i64 [[TMP5]]
163; CHECK-NEXT:    [[TMP6:%.*]] = load <2 x i32>, ptr [[ARRAYIDX]], align 4
164; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i32> [[TMP3]], i64 1
165; CHECK-NEXT:    [[TMP8:%.*]] = sext i32 [[TMP7]] to i64
166; CHECK-NEXT:    [[ARRAYIDX10:%.*]] = getelementptr inbounds i32, ptr [[G]], i64 [[TMP8]]
167; CHECK-NEXT:    [[T10:%.*]] = load i32, ptr [[ARRAYIDX10]], align 4
168; CHECK-NEXT:    [[T11:%.*]] = add nsw i32 [[T4]], [[Z:%.*]]
169; CHECK-NEXT:    [[TMP9:%.*]] = sext i32 [[T11]] to i64
170; CHECK-NEXT:    [[ARRAYIDX15:%.*]] = getelementptr inbounds i32, ptr [[G]], i64 [[TMP9]]
171; CHECK-NEXT:    [[T12:%.*]] = load i32, ptr [[ARRAYIDX15]], align 4
172; CHECK-NEXT:    [[TMP10:%.*]] = insertelement <4 x i32> poison, i32 [[T10]], i64 2
173; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <4 x i32> [[TMP10]], i32 [[T12]], i64 3
174; CHECK-NEXT:    [[TMP12:%.*]] = shufflevector <2 x i32> [[TMP6]], <2 x i32> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison>
175; CHECK-NEXT:    [[TMP13:%.*]] = shufflevector <4 x i32> [[TMP12]], <4 x i32> [[TMP11]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>
176; CHECK-NEXT:    [[TMP14:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP13]])
177; CHECK-NEXT:    [[OP_RDX]] = add i32 [[TMP14]], [[SUM_032]]
178; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i32 [[INDVARS_IV]], 1
179; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INDVARS_IV_NEXT]], [[N]]
180; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]]
181;
182entry:
183  %cmp31 = icmp sgt i32 %n, 0
184  br i1 %cmp31, label %for.body.preheader, label %for.cond.cleanup
185
186for.body.preheader:
187  br label %for.body
188
189for.cond.cleanup.loopexit:
190  br label %for.cond.cleanup
191
192for.cond.cleanup:
193  %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add16, %for.cond.cleanup.loopexit ]
194  ret i32 %sum.0.lcssa
195
196for.body:
197  %indvars.iv = phi i32 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
198  %sum.032 = phi i32 [ 0, %for.body.preheader ], [ %add16, %for.body ]
199  %t4 = shl nsw i32 %indvars.iv, 1
200  %t5 = add nsw i32 %t4, 0
201  %arrayidx = getelementptr inbounds i32, ptr %g, i32 %t5
202  %t6 = load i32, ptr %arrayidx, align 4
203  %add1 = add nsw i32 %t6, %sum.032
204  %t7 = add nsw i32 %t4, 1
205  %arrayidx5 = getelementptr inbounds i32, ptr %g, i32 %t7
206  %t8 = load i32, ptr %arrayidx5, align 4
207  %add6 = add nsw i32 %add1, %t8
208  %t9 = add nsw i32 %t4, %y
209  %arrayidx10 = getelementptr inbounds i32, ptr %g, i32 %t9
210  %t10 = load i32, ptr %arrayidx10, align 4
211  %add11 = add nsw i32 %add6, %t10
212  %t11 = add nsw i32 %t4, %z
213  %arrayidx15 = getelementptr inbounds i32, ptr %g, i32 %t11
214  %t12 = load i32, ptr %arrayidx15, align 4
215  %add16 = add nsw i32 %add11, %t12
216  %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1
217  %exitcond = icmp eq i32 %indvars.iv.next , %n
218  br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body
219}
220