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
8 target datalayout = "e-m:e-i32:64-i128:128-n32:64-S128"
9 target triple = "aarch64--linux-gnu"
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.
16 ; int getelementptr(int *g, int n, int w, int x, int y, int z) {
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];
26 ; YAML-LABEL: Function: getelementptr_4x32
28 ; YAML-NEXT: - String: 'SLP vectorized with cost '
29 ; YAML-NEXT: - Cost: '11'
30 ; YAML-NEXT: - String: ' and with tree size '
31 ; YAML-NEXT: - TreeSize: '5'
34 ; YAML-NEXT: Pass: slp-vectorizer
35 ; YAML-NEXT: Name: VectorizedList
36 ; YAML-NEXT: Function: getelementptr_4x32
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'
43 define i32 @getelementptr_4x32(i32* nocapture readonly %g, i32 %n, i32 %x, i32 %y, i32 %z) {
44 ; CHECK-LABEL: @getelementptr_4x32(
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 undef>, i32 [[X:%.*]], i32 1
50 ; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i32> undef, i32 [[Y:%.*]], i32 0
51 ; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[Z:%.*]], i32 1
52 ; CHECK-NEXT: br label [[FOR_BODY:%.*]]
53 ; CHECK: for.cond.cleanup.loopexit:
54 ; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[TMP22:%.*]], i32 1
55 ; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]
56 ; CHECK: for.cond.cleanup:
57 ; CHECK-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[TMP3]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ]
58 ; CHECK-NEXT: ret i32 [[SUM_0_LCSSA]]
60 ; CHECK-NEXT: [[TMP4:%.*]] = phi <2 x i32> [ zeroinitializer, [[FOR_BODY_PREHEADER]] ], [ [[TMP22]], [[FOR_BODY]] ]
61 ; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i32> [[TMP4]], i32 0
62 ; CHECK-NEXT: [[T4:%.*]] = shl nsw i32 [[TMP5]], 1
63 ; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i32> undef, i32 [[T4]], i32 0
64 ; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <2 x i32> [[TMP6]], <2 x i32> undef, <2 x i32> zeroinitializer
65 ; CHECK-NEXT: [[TMP8:%.*]] = add nsw <2 x i32> [[TMP7]], [[TMP0]]
66 ; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x i32> [[TMP8]], i32 0
67 ; CHECK-NEXT: [[TMP10:%.*]] = sext i32 [[TMP9]] to i64
68 ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[G:%.*]], i64 [[TMP10]]
69 ; CHECK-NEXT: [[T6:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
70 ; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i32> [[TMP4]], i32 1
71 ; CHECK-NEXT: [[ADD1:%.*]] = add nsw i32 [[T6]], [[TMP11]]
72 ; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x i32> [[TMP8]], i32 1
73 ; CHECK-NEXT: [[TMP13:%.*]] = sext i32 [[TMP12]] to i64
74 ; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP13]]
75 ; CHECK-NEXT: [[T8:%.*]] = load i32, i32* [[ARRAYIDX5]], align 4
76 ; CHECK-NEXT: [[ADD6:%.*]] = add nsw i32 [[ADD1]], [[T8]]
77 ; CHECK-NEXT: [[TMP14:%.*]] = add nsw <2 x i32> [[TMP7]], [[TMP2]]
78 ; CHECK-NEXT: [[TMP15:%.*]] = extractelement <2 x i32> [[TMP14]], i32 0
79 ; CHECK-NEXT: [[TMP16:%.*]] = sext i32 [[TMP15]] to i64
80 ; CHECK-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP16]]
81 ; CHECK-NEXT: [[T10:%.*]] = load i32, i32* [[ARRAYIDX10]], align 4
82 ; CHECK-NEXT: [[ADD11:%.*]] = add nsw i32 [[ADD6]], [[T10]]
83 ; CHECK-NEXT: [[TMP17:%.*]] = extractelement <2 x i32> [[TMP14]], i32 1
84 ; CHECK-NEXT: [[TMP18:%.*]] = sext i32 [[TMP17]] to i64
85 ; CHECK-NEXT: [[ARRAYIDX15:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP18]]
86 ; CHECK-NEXT: [[T12:%.*]] = load i32, i32* [[ARRAYIDX15]], align 4
87 ; CHECK-NEXT: [[TMP19:%.*]] = insertelement <2 x i32> undef, i32 [[TMP5]], i32 0
88 ; CHECK-NEXT: [[TMP20:%.*]] = insertelement <2 x i32> [[TMP19]], i32 [[ADD11]], i32 1
89 ; CHECK-NEXT: [[TMP21:%.*]] = insertelement <2 x i32> <i32 1, i32 undef>, i32 [[T12]], i32 1
90 ; CHECK-NEXT: [[TMP22]] = add nsw <2 x i32> [[TMP20]], [[TMP21]]
91 ; CHECK-NEXT: [[TMP23:%.*]] = extractelement <2 x i32> [[TMP22]], i32 0
92 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[TMP23]], [[N]]
93 ; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]]
96 %cmp31 = icmp sgt i32 %n, 0
97 br i1 %cmp31, label %for.body.preheader, label %for.cond.cleanup
102 for.cond.cleanup.loopexit:
103 br label %for.cond.cleanup
106 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add16, %for.cond.cleanup.loopexit ]
110 %indvars.iv = phi i32 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
111 %sum.032 = phi i32 [ 0, %for.body.preheader ], [ %add16, %for.body ]
112 %t4 = shl nsw i32 %indvars.iv, 1
113 %t5 = add nsw i32 %t4, 0
114 %arrayidx = getelementptr inbounds i32, i32* %g, i32 %t5
115 %t6 = load i32, i32* %arrayidx, align 4
116 %add1 = add nsw i32 %t6, %sum.032
117 %t7 = add nsw i32 %t4, %x
118 %arrayidx5 = getelementptr inbounds i32, i32* %g, i32 %t7
119 %t8 = load i32, i32* %arrayidx5, align 4
120 %add6 = add nsw i32 %add1, %t8
121 %t9 = add nsw i32 %t4, %y
122 %arrayidx10 = getelementptr inbounds i32, i32* %g, i32 %t9
123 %t10 = load i32, i32* %arrayidx10, align 4
124 %add11 = add nsw i32 %add6, %t10
125 %t11 = add nsw i32 %t4, %z
126 %arrayidx15 = getelementptr inbounds i32, i32* %g, i32 %t11
127 %t12 = load i32, i32* %arrayidx15, align 4
128 %add16 = add nsw i32 %add11, %t12
129 %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1
130 %exitcond = icmp eq i32 %indvars.iv.next , %n
131 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body
134 ; YAML-LABEL: Function: getelementptr_2x32
136 ; YAML-NEXT: - String: 'SLP vectorized with cost '
137 ; YAML-NEXT: - Cost: '11'
138 ; YAML-NEXT: - String: ' and with tree size '
139 ; YAML-NEXT: - TreeSize: '5'
142 ; YAML-NEXT: Pass: slp-vectorizer
143 ; YAML-NEXT: Name: VectorizedList
144 ; YAML-NEXT: Function: getelementptr_2x32
146 ; YAML-NEXT: - String: 'SLP vectorized with cost '
147 ; YAML-NEXT: - Cost: '6'
148 ; YAML-NEXT: - String: ' and with tree size '
149 ; YAML-NEXT: - TreeSize: '3'
151 define i32 @getelementptr_2x32(i32* nocapture readonly %g, i32 %n, i32 %x, i32 %y, i32 %z) {
152 ; CHECK-LABEL: @getelementptr_2x32(
154 ; CHECK-NEXT: [[CMP31:%.*]] = icmp sgt i32 [[N:%.*]], 0
155 ; CHECK-NEXT: br i1 [[CMP31]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
156 ; CHECK: for.body.preheader:
157 ; CHECK-NEXT: [[TMP0:%.*]] = insertelement <2 x i32> undef, i32 [[Y:%.*]], i32 0
158 ; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i32> [[TMP0]], i32 [[Z:%.*]], i32 1
159 ; CHECK-NEXT: br label [[FOR_BODY:%.*]]
160 ; CHECK: for.cond.cleanup.loopexit:
161 ; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i32> [[TMP18:%.*]], i32 1
162 ; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]
163 ; CHECK: for.cond.cleanup:
164 ; CHECK-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[TMP2]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ]
165 ; CHECK-NEXT: ret i32 [[SUM_0_LCSSA]]
167 ; CHECK-NEXT: [[TMP3:%.*]] = phi <2 x i32> [ zeroinitializer, [[FOR_BODY_PREHEADER]] ], [ [[TMP18]], [[FOR_BODY]] ]
168 ; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i32> [[TMP3]], i32 0
169 ; CHECK-NEXT: [[T4:%.*]] = shl nsw i32 [[TMP4]], 1
170 ; CHECK-NEXT: [[TMP5:%.*]] = sext i32 [[T4]] to i64
171 ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[G:%.*]], i64 [[TMP5]]
172 ; CHECK-NEXT: [[T6:%.*]] = load i32, i32* [[ARRAYIDX]], align 4
173 ; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i32> [[TMP3]], i32 1
174 ; CHECK-NEXT: [[ADD1:%.*]] = add nsw i32 [[T6]], [[TMP6]]
175 ; CHECK-NEXT: [[T7:%.*]] = or i32 [[T4]], 1
176 ; CHECK-NEXT: [[TMP7:%.*]] = sext i32 [[T7]] to i64
177 ; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP7]]
178 ; CHECK-NEXT: [[T8:%.*]] = load i32, i32* [[ARRAYIDX5]], align 4
179 ; CHECK-NEXT: [[ADD6:%.*]] = add nsw i32 [[ADD1]], [[T8]]
180 ; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x i32> undef, i32 [[T4]], i32 0
181 ; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <2 x i32> [[TMP8]], <2 x i32> undef, <2 x i32> zeroinitializer
182 ; CHECK-NEXT: [[TMP10:%.*]] = add nsw <2 x i32> [[TMP9]], [[TMP1]]
183 ; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i32> [[TMP10]], i32 0
184 ; CHECK-NEXT: [[TMP12:%.*]] = sext i32 [[TMP11]] to i64
185 ; CHECK-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP12]]
186 ; CHECK-NEXT: [[T10:%.*]] = load i32, i32* [[ARRAYIDX10]], align 4
187 ; CHECK-NEXT: [[ADD11:%.*]] = add nsw i32 [[ADD6]], [[T10]]
188 ; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i32> [[TMP10]], i32 1
189 ; CHECK-NEXT: [[TMP14:%.*]] = sext i32 [[TMP13]] to i64
190 ; CHECK-NEXT: [[ARRAYIDX15:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP14]]
191 ; CHECK-NEXT: [[T12:%.*]] = load i32, i32* [[ARRAYIDX15]], align 4
192 ; CHECK-NEXT: [[TMP15:%.*]] = insertelement <2 x i32> undef, i32 [[TMP4]], i32 0
193 ; CHECK-NEXT: [[TMP16:%.*]] = insertelement <2 x i32> [[TMP15]], i32 [[ADD11]], i32 1
194 ; CHECK-NEXT: [[TMP17:%.*]] = insertelement <2 x i32> <i32 1, i32 undef>, i32 [[T12]], i32 1
195 ; CHECK-NEXT: [[TMP18]] = add nsw <2 x i32> [[TMP16]], [[TMP17]]
196 ; CHECK-NEXT: [[TMP19:%.*]] = extractelement <2 x i32> [[TMP18]], i32 0
197 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[TMP19]], [[N]]
198 ; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]]
201 %cmp31 = icmp sgt i32 %n, 0
202 br i1 %cmp31, label %for.body.preheader, label %for.cond.cleanup
207 for.cond.cleanup.loopexit:
208 br label %for.cond.cleanup
211 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add16, %for.cond.cleanup.loopexit ]
215 %indvars.iv = phi i32 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
216 %sum.032 = phi i32 [ 0, %for.body.preheader ], [ %add16, %for.body ]
217 %t4 = shl nsw i32 %indvars.iv, 1
218 %t5 = add nsw i32 %t4, 0
219 %arrayidx = getelementptr inbounds i32, i32* %g, i32 %t5
220 %t6 = load i32, i32* %arrayidx, align 4
221 %add1 = add nsw i32 %t6, %sum.032
222 %t7 = add nsw i32 %t4, 1
223 %arrayidx5 = getelementptr inbounds i32, i32* %g, i32 %t7
224 %t8 = load i32, i32* %arrayidx5, align 4
225 %add6 = add nsw i32 %add1, %t8
226 %t9 = add nsw i32 %t4, %y
227 %arrayidx10 = getelementptr inbounds i32, i32* %g, i32 %t9
228 %t10 = load i32, i32* %arrayidx10, align 4
229 %add11 = add nsw i32 %add6, %t10
230 %t11 = add nsw i32 %t4, %z
231 %arrayidx15 = getelementptr inbounds i32, i32* %g, i32 %t11
232 %t12 = load i32, i32* %arrayidx15, align 4
233 %add16 = add nsw i32 %add11, %t12
234 %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1
235 %exitcond = icmp eq i32 %indvars.iv.next , %n
236 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body