| blob | 7e4b95d7234994bc0e5f4eba068f31b7618af5b8 |
1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt -slp-vectorizer -slp-threshold=-6 -S -pass-remarks-output=%t < %s | FileCheck %s
3 ; RUN: cat %t | FileCheck -check-prefix=YAML %s
6 ; FIXME: The threshold is changed to keep this test case a bit smaller.
7 ; The AArch64 cost model should not give such high costs to select statements.
9 target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
10 target triple = "aarch64--linux"
12 ; YAML: --- !Passed
13 ; YAML-NEXT: Pass: slp-vectorizer
14 ; YAML-NEXT: Name: VectorizedHorizontalReduction
15 ; YAML-NEXT: Function: test_select
16 ; YAML-NEXT: Args:
17 ; YAML-NEXT: - String: 'Vectorized horizontal reduction with cost '
18 ; YAML-NEXT: - Cost: '-8'
19 ; YAML-NEXT: - String: ' and with tree size '
20 ; YAML-NEXT: - TreeSize: '8'
22 define i32 @test_select(i32* noalias nocapture readonly %blk1, i32* noalias nocapture readonly %blk2, i32 %lx, i32 %h) {
23 ; CHECK-LABEL: @test_select(
24 ; CHECK-NEXT: entry:
25 ; CHECK-NEXT: [[CMP_22:%.*]] = icmp sgt i32 [[H:%.*]], 0
26 ; CHECK-NEXT: br i1 [[CMP_22]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_END:%.*]]
27 ; CHECK: for.body.lr.ph:
28 ; CHECK-NEXT: [[IDX_EXT:%.*]] = sext i32 [[LX:%.*]] to i64
29 ; CHECK-NEXT: br label [[FOR_BODY:%.*]]
30 ; CHECK: for.body:
31 ; CHECK-NEXT: [[S_026:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[OP_EXTRA:%.*]], [[FOR_BODY]] ]
32 ; CHECK-NEXT: [[J_025:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
33 ; CHECK-NEXT: [[P2_024:%.*]] = phi i32* [ [[BLK2:%.*]], [[FOR_BODY_LR_PH]] ], [ [[ADD_PTR29:%.*]], [[FOR_BODY]] ]
34 ; CHECK-NEXT: [[P1_023:%.*]] = phi i32* [ [[BLK1:%.*]], [[FOR_BODY_LR_PH]] ], [ [[ADD_PTR:%.*]], [[FOR_BODY]] ]
35 ; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, i32* [[P1_023]], i64 1
36 ; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, i32* [[P2_024]], i64 1
37 ; CHECK-NEXT: [[ARRAYIDX12:%.*]] = getelementptr inbounds i32, i32* [[P1_023]], i64 2
38 ; CHECK-NEXT: [[ARRAYIDX13:%.*]] = getelementptr inbounds i32, i32* [[P2_024]], i64 2
39 ; CHECK-NEXT: [[ARRAYIDX20:%.*]] = getelementptr inbounds i32, i32* [[P1_023]], i64 3
40 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[P1_023]] to <4 x i32>*
41 ; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
42 ; CHECK-NEXT: [[ARRAYIDX21:%.*]] = getelementptr inbounds i32, i32* [[P2_024]], i64 3
43 ; CHECK-NEXT: [[TMP2:%.*]] = bitcast i32* [[P2_024]] to <4 x i32>*
44 ; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, <4 x i32>* [[TMP2]], align 4
45 ; CHECK-NEXT: [[TMP4:%.*]] = sub nsw <4 x i32> [[TMP1]], [[TMP3]]
46 ; CHECK-NEXT: [[TMP5:%.*]] = icmp slt <4 x i32> [[TMP4]], zeroinitializer
47 ; CHECK-NEXT: [[TMP6:%.*]] = sub nsw <4 x i32> zeroinitializer, [[TMP4]]
48 ; CHECK-NEXT: [[TMP7:%.*]] = select <4 x i1> [[TMP5]], <4 x i32> [[TMP6]], <4 x i32> [[TMP4]]
49 ; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 undef, [[S_026]]
50 ; CHECK-NEXT: [[ADD11:%.*]] = add nsw i32 [[ADD]], undef
51 ; CHECK-NEXT: [[ADD19:%.*]] = add nsw i32 [[ADD11]], undef
52 ; CHECK-NEXT: [[TMP8:%.*]] = call i32 @llvm.experimental.vector.reduce.add.v4i32(<4 x i32> [[TMP7]])
53 ; CHECK-NEXT: [[OP_EXTRA]] = add nsw i32 [[TMP8]], [[S_026]]
54 ; CHECK-NEXT: [[ADD27:%.*]] = add nsw i32 [[ADD19]], undef
55 ; CHECK-NEXT: [[ADD_PTR]] = getelementptr inbounds i32, i32* [[P1_023]], i64 [[IDX_EXT]]
56 ; CHECK-NEXT: [[ADD_PTR29]] = getelementptr inbounds i32, i32* [[P2_024]], i64 [[IDX_EXT]]
57 ; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[J_025]], 1
58 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[INC]], [[H]]
59 ; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT:%.*]], label [[FOR_BODY]]
60 ; CHECK: for.end.loopexit:
61 ; CHECK-NEXT: br label [[FOR_END]]
62 ; CHECK: for.end:
63 ; CHECK-NEXT: [[S_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[OP_EXTRA]], [[FOR_END_LOOPEXIT]] ]
64 ; CHECK-NEXT: ret i32 [[S_0_LCSSA]]
65 ;
66 entry:
67 %cmp.22 = icmp sgt i32 %h, 0
68 br i1 %cmp.22, label %for.body.lr.ph, label %for.end
70 for.body.lr.ph: ; preds = %entry
71 %idx.ext = sext i32 %lx to i64
72 br label %for.body
74 for.body: ; preds = %for.body, %for.body.lr.ph
75 %s.026 = phi i32 [ 0, %for.body.lr.ph ], [ %add27, %for.body ]
76 %j.025 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
77 %p2.024 = phi i32* [ %blk2, %for.body.lr.ph ], [ %add.ptr29, %for.body ]
78 %p1.023 = phi i32* [ %blk1, %for.body.lr.ph ], [ %add.ptr, %for.body ]
79 %0 = load i32, i32* %p1.023, align 4
80 %1 = load i32, i32* %p2.024, align 4
81 %sub = sub nsw i32 %0, %1
82 %cmp2 = icmp slt i32 %sub, 0
83 %sub3 = sub nsw i32 0, %sub
84 %sub3.sub = select i1 %cmp2, i32 %sub3, i32 %sub
85 %add = add nsw i32 %sub3.sub, %s.026
86 %arrayidx4 = getelementptr inbounds i32, i32* %p1.023, i64 1
87 %2 = load i32, i32* %arrayidx4, align 4
88 %arrayidx5 = getelementptr inbounds i32, i32* %p2.024, i64 1
89 %3 = load i32, i32* %arrayidx5, align 4
90 %sub6 = sub nsw i32 %2, %3
91 %cmp7 = icmp slt i32 %sub6, 0
92 %sub9 = sub nsw i32 0, %sub6
93 %v.1 = select i1 %cmp7, i32 %sub9, i32 %sub6
94 %add11 = add nsw i32 %add, %v.1
95 %arrayidx12 = getelementptr inbounds i32, i32* %p1.023, i64 2
96 %4 = load i32, i32* %arrayidx12, align 4
97 %arrayidx13 = getelementptr inbounds i32, i32* %p2.024, i64 2
98 %5 = load i32, i32* %arrayidx13, align 4
99 %sub14 = sub nsw i32 %4, %5
100 %cmp15 = icmp slt i32 %sub14, 0
101 %sub17 = sub nsw i32 0, %sub14
102 %sub17.sub14 = select i1 %cmp15, i32 %sub17, i32 %sub14
103 %add19 = add nsw i32 %add11, %sub17.sub14
104 %arrayidx20 = getelementptr inbounds i32, i32* %p1.023, i64 3
105 %6 = load i32, i32* %arrayidx20, align 4
106 %arrayidx21 = getelementptr inbounds i32, i32* %p2.024, i64 3
107 %7 = load i32, i32* %arrayidx21, align 4
108 %sub22 = sub nsw i32 %6, %7
109 %cmp23 = icmp slt i32 %sub22, 0
110 %sub25 = sub nsw i32 0, %sub22
111 %v.3 = select i1 %cmp23, i32 %sub25, i32 %sub22
112 %add27 = add nsw i32 %add19, %v.3
113 %add.ptr = getelementptr inbounds i32, i32* %p1.023, i64 %idx.ext
114 %add.ptr29 = getelementptr inbounds i32, i32* %p2.024, i64 %idx.ext
115 %inc = add nuw nsw i32 %j.025, 1
116 %exitcond = icmp eq i32 %inc, %h
117 br i1 %exitcond, label %for.end.loopexit, label %for.body
119 for.end.loopexit: ; preds = %for.body
120 br label %for.end
122 for.end: ; preds = %for.end.loopexit, %entry
123 %s.0.lcssa = phi i32 [ 0, %entry ], [ %add27, %for.end.loopexit ]
124 ret i32 %s.0.lcssa
125 }
127 ;; Check whether SLP can find a reduction phi whose incoming blocks are not
128 ;; the same as the block containing the phi.
129 ;;
130 ;; Came from code like,
131 ;;
132 ;; int s = 0;
133 ;; for (int j = 0; j < h; j++) {
134 ;; s += p1[0] * p2[0]
135 ;; s += p1[1] * p2[1];
136 ;; s += p1[2] * p2[2];
137 ;; s += p1[3] * p2[3];
138 ;; if (s >= lim)
139 ;; break;
140 ;; p1 += lx;
141 ;; p2 += lx;
142 ;; }
143 define i32 @reduction_with_br(i32* noalias nocapture readonly %blk1, i32* noalias nocapture readonly %blk2, i32 %lx, i32 %h, i32 %lim) {
144 ; YAML: --- !Passed
145 ; YAML-NEXT: Pass: slp-vectorizer
146 ; YAML-NEXT: Name: VectorizedHorizontalReduction
147 ; YAML-NEXT: Function: reduction_with_br
148 ; YAML-NEXT: Args:
149 ; YAML-NEXT: - String: 'Vectorized horizontal reduction with cost '
150 ; YAML-NEXT: - Cost: '-11'
151 ; YAML-NEXT: - String: ' and with tree size '
152 ; YAML-NEXT: - TreeSize: '3'
153 ; CHECK-LABEL: @reduction_with_br(
154 ; CHECK-NEXT: entry:
155 ; CHECK-NEXT: [[CMP_16:%.*]] = icmp sgt i32 [[H:%.*]], 0
156 ; CHECK-NEXT: br i1 [[CMP_16]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_END:%.*]]
157 ; CHECK: for.body.lr.ph:
158 ; CHECK-NEXT: [[IDX_EXT:%.*]] = sext i32 [[LX:%.*]] to i64
159 ; CHECK-NEXT: br label [[FOR_BODY:%.*]]
160 ; CHECK: for.body:
161 ; CHECK-NEXT: [[S_020:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[OP_EXTRA:%.*]], [[IF_END:%.*]] ]
162 ; CHECK-NEXT: [[J_019:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[INC:%.*]], [[IF_END]] ]
163 ; CHECK-NEXT: [[P2_018:%.*]] = phi i32* [ [[BLK2:%.*]], [[FOR_BODY_LR_PH]] ], [ [[ADD_PTR16:%.*]], [[IF_END]] ]
164 ; CHECK-NEXT: [[P1_017:%.*]] = phi i32* [ [[BLK1:%.*]], [[FOR_BODY_LR_PH]] ], [ [[ADD_PTR:%.*]], [[IF_END]] ]
165 ; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, i32* [[P1_017]], i64 1
166 ; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds i32, i32* [[P2_018]], i64 1
167 ; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds i32, i32* [[P1_017]], i64 2
168 ; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds i32, i32* [[P2_018]], i64 2
169 ; CHECK-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds i32, i32* [[P1_017]], i64 3
170 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[P1_017]] to <4 x i32>*
171 ; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
172 ; CHECK-NEXT: [[ARRAYIDX11:%.*]] = getelementptr inbounds i32, i32* [[P2_018]], i64 3
173 ; CHECK-NEXT: [[TMP2:%.*]] = bitcast i32* [[P2_018]] to <4 x i32>*
174 ; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, <4 x i32>* [[TMP2]], align 4
175 ; CHECK-NEXT: [[TMP4:%.*]] = mul nsw <4 x i32> [[TMP3]], [[TMP1]]
176 ; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 undef, [[S_020]]
177 ; CHECK-NEXT: [[ADD5:%.*]] = add nsw i32 [[ADD]], undef
178 ; CHECK-NEXT: [[ADD9:%.*]] = add nsw i32 [[ADD5]], undef
179 ; CHECK-NEXT: [[TMP5:%.*]] = call i32 @llvm.experimental.vector.reduce.add.v4i32(<4 x i32> [[TMP4]])
180 ; CHECK-NEXT: [[OP_EXTRA]] = add nsw i32 [[TMP5]], [[S_020]]
181 ; CHECK-NEXT: [[ADD13:%.*]] = add nsw i32 [[ADD9]], undef
182 ; CHECK-NEXT: [[CMP14:%.*]] = icmp slt i32 [[OP_EXTRA]], [[LIM:%.*]]
183 ; CHECK-NEXT: br i1 [[CMP14]], label [[IF_END]], label [[FOR_END_LOOPEXIT:%.*]]
184 ; CHECK: if.end:
185 ; CHECK-NEXT: [[ADD_PTR]] = getelementptr inbounds i32, i32* [[P1_017]], i64 [[IDX_EXT]]
186 ; CHECK-NEXT: [[ADD_PTR16]] = getelementptr inbounds i32, i32* [[P2_018]], i64 [[IDX_EXT]]
187 ; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[J_019]], 1
188 ; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[INC]], [[H]]
189 ; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]]
190 ; CHECK: for.end.loopexit:
191 ; CHECK-NEXT: br label [[FOR_END]]
192 ; CHECK: for.end:
193 ; CHECK-NEXT: [[S_1:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[OP_EXTRA]], [[FOR_END_LOOPEXIT]] ]
194 ; CHECK-NEXT: ret i32 [[S_1]]
195 ;
196 entry:
197 %cmp.16 = icmp sgt i32 %h, 0
198 br i1 %cmp.16, label %for.body.lr.ph, label %for.end
200 for.body.lr.ph: ; preds = %entry
201 %idx.ext = sext i32 %lx to i64
202 br label %for.body
204 for.body: ; preds = %for.body.lr.ph, %if.end
205 %s.020 = phi i32 [ 0, %for.body.lr.ph ], [ %add13, %if.end ]
206 %j.019 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %if.end ]
207 %p2.018 = phi i32* [ %blk2, %for.body.lr.ph ], [ %add.ptr16, %if.end ]
208 %p1.017 = phi i32* [ %blk1, %for.body.lr.ph ], [ %add.ptr, %if.end ]
209 %0 = load i32, i32* %p1.017, align 4
210 %1 = load i32, i32* %p2.018, align 4
211 %mul = mul nsw i32 %1, %0
212 %add = add nsw i32 %mul, %s.020
213 %arrayidx2 = getelementptr inbounds i32, i32* %p1.017, i64 1
214 %2 = load i32, i32* %arrayidx2, align 4
215 %arrayidx3 = getelementptr inbounds i32, i32* %p2.018, i64 1
216 %3 = load i32, i32* %arrayidx3, align 4
217 %mul4 = mul nsw i32 %3, %2
218 %add5 = add nsw i32 %add, %mul4
219 %arrayidx6 = getelementptr inbounds i32, i32* %p1.017, i64 2
220 %4 = load i32, i32* %arrayidx6, align 4
221 %arrayidx7 = getelementptr inbounds i32, i32* %p2.018, i64 2
222 %5 = load i32, i32* %arrayidx7, align 4
223 %mul8 = mul nsw i32 %5, %4
224 %add9 = add nsw i32 %add5, %mul8
225 %arrayidx10 = getelementptr inbounds i32, i32* %p1.017, i64 3
226 %6 = load i32, i32* %arrayidx10, align 4
227 %arrayidx11 = getelementptr inbounds i32, i32* %p2.018, i64 3
228 %7 = load i32, i32* %arrayidx11, align 4
229 %mul12 = mul nsw i32 %7, %6
230 %add13 = add nsw i32 %add9, %mul12
231 %cmp14 = icmp slt i32 %add13, %lim
232 br i1 %cmp14, label %if.end, label %for.end.loopexit
234 if.end: ; preds = %for.body
235 %add.ptr = getelementptr inbounds i32, i32* %p1.017, i64 %idx.ext
236 %add.ptr16 = getelementptr inbounds i32, i32* %p2.018, i64 %idx.ext
237 %inc = add nuw nsw i32 %j.019, 1
238 %cmp = icmp slt i32 %inc, %h
239 br i1 %cmp, label %for.body, label %for.end.loopexit
241 for.end.loopexit: ; preds = %for.body, %if.end
242 br label %for.end
244 for.end: ; preds = %for.end.loopexit, %entry
245 %s.1 = phi i32 [ 0, %entry ], [ %add13, %for.end.loopexit ]
246 ret i32 %s.1
247 }
249 ; YAML: --- !Passed
250 ; YAML-NEXT: Pass: slp-vectorizer
251 ; YAML-NEXT: Name: VectorizedHorizontalReduction
252 ; YAML-NEXT: Function: test_unrolled_select
253 ; YAML-NEXT: Args:
254 ; YAML-NEXT: - String: 'Vectorized horizontal reduction with cost '
255 ; YAML-NEXT: - Cost: '-47'
256 ; YAML-NEXT: - String: ' and with tree size '
257 ; YAML-NEXT: - TreeSize: '10'
259 define i32 @test_unrolled_select(i8* noalias nocapture readonly %blk1, i8* noalias nocapture readonly %blk2, i32 %lx, i32 %h, i32 %lim) #0 {
260 ; CHECK-LABEL: @test_unrolled_select(
261 ; CHECK-NEXT: entry:
262 ; CHECK-NEXT: [[CMP_43:%.*]] = icmp sgt i32 [[H:%.*]], 0
263 ; CHECK-NEXT: br i1 [[CMP_43]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_END:%.*]]
264 ; CHECK: for.body.lr.ph:
265 ; CHECK-NEXT: [[IDX_EXT:%.*]] = sext i32 [[LX:%.*]] to i64
266 ; CHECK-NEXT: br label [[FOR_BODY:%.*]]
267 ; CHECK: for.body:
268 ; CHECK-NEXT: [[S_047:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[OP_EXTRA:%.*]], [[IF_END_86:%.*]] ]
269 ; CHECK-NEXT: [[J_046:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[INC:%.*]], [[IF_END_86]] ]
270 ; CHECK-NEXT: [[P2_045:%.*]] = phi i8* [ [[BLK2:%.*]], [[FOR_BODY_LR_PH]] ], [ [[ADD_PTR88:%.*]], [[IF_END_86]] ]
271 ; CHECK-NEXT: [[P1_044:%.*]] = phi i8* [ [[BLK1:%.*]], [[FOR_BODY_LR_PH]] ], [ [[ADD_PTR:%.*]], [[IF_END_86]] ]
272 ; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds i8, i8* [[P1_044]], i64 1
273 ; CHECK-NEXT: [[ARRAYIDX8:%.*]] = getelementptr inbounds i8, i8* [[P2_045]], i64 1
274 ; CHECK-NEXT: [[ARRAYIDX17:%.*]] = getelementptr inbounds i8, i8* [[P1_044]], i64 2
275 ; CHECK-NEXT: [[ARRAYIDX19:%.*]] = getelementptr inbounds i8, i8* [[P2_045]], i64 2
276 ; CHECK-NEXT: [[ARRAYIDX28:%.*]] = getelementptr inbounds i8, i8* [[P1_044]], i64 3
277 ; CHECK-NEXT: [[ARRAYIDX30:%.*]] = getelementptr inbounds i8, i8* [[P2_045]], i64 3
278 ; CHECK-NEXT: [[ARRAYIDX39:%.*]] = getelementptr inbounds i8, i8* [[P1_044]], i64 4
279 ; CHECK-NEXT: [[ARRAYIDX41:%.*]] = getelementptr inbounds i8, i8* [[P2_045]], i64 4
280 ; CHECK-NEXT: [[ARRAYIDX50:%.*]] = getelementptr inbounds i8, i8* [[P1_044]], i64 5
281 ; CHECK-NEXT: [[ARRAYIDX52:%.*]] = getelementptr inbounds i8, i8* [[P2_045]], i64 5
282 ; CHECK-NEXT: [[ARRAYIDX61:%.*]] = getelementptr inbounds i8, i8* [[P1_044]], i64 6
283 ; CHECK-NEXT: [[ARRAYIDX63:%.*]] = getelementptr inbounds i8, i8* [[P2_045]], i64 6
284 ; CHECK-NEXT: [[ARRAYIDX72:%.*]] = getelementptr inbounds i8, i8* [[P1_044]], i64 7
285 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast i8* [[P1_044]] to <8 x i8>*
286 ; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i8>, <8 x i8>* [[TMP0]], align 1
287 ; CHECK-NEXT: [[TMP2:%.*]] = zext <8 x i8> [[TMP1]] to <8 x i32>
288 ; CHECK-NEXT: [[ARRAYIDX74:%.*]] = getelementptr inbounds i8, i8* [[P2_045]], i64 7
289 ; CHECK-NEXT: [[TMP3:%.*]] = bitcast i8* [[P2_045]] to <8 x i8>*
290 ; CHECK-NEXT: [[TMP4:%.*]] = load <8 x i8>, <8 x i8>* [[TMP3]], align 1
291 ; CHECK-NEXT: [[TMP5:%.*]] = zext <8 x i8> [[TMP4]] to <8 x i32>
292 ; CHECK-NEXT: [[TMP6:%.*]] = sub nsw <8 x i32> [[TMP2]], [[TMP5]]
293 ; CHECK-NEXT: [[TMP7:%.*]] = icmp slt <8 x i32> [[TMP6]], zeroinitializer
294 ; CHECK-NEXT: [[TMP8:%.*]] = sub nsw <8 x i32> zeroinitializer, [[TMP6]]
295 ; CHECK-NEXT: [[TMP9:%.*]] = select <8 x i1> [[TMP7]], <8 x i32> [[TMP8]], <8 x i32> [[TMP6]]
296 ; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 undef, [[S_047]]
297 ; CHECK-NEXT: [[ADD16:%.*]] = add nsw i32 [[ADD]], undef
298 ; CHECK-NEXT: [[ADD27:%.*]] = add nsw i32 [[ADD16]], undef
299 ; CHECK-NEXT: [[ADD38:%.*]] = add nsw i32 [[ADD27]], undef
300 ; CHECK-NEXT: [[ADD49:%.*]] = add nsw i32 [[ADD38]], undef
301 ; CHECK-NEXT: [[ADD60:%.*]] = add nsw i32 [[ADD49]], undef
302 ; CHECK-NEXT: [[ADD71:%.*]] = add nsw i32 [[ADD60]], undef
303 ; CHECK-NEXT: [[TMP10:%.*]] = call i32 @llvm.experimental.vector.reduce.add.v8i32(<8 x i32> [[TMP9]])
304 ; CHECK-NEXT: [[OP_EXTRA]] = add nsw i32 [[TMP10]], [[S_047]]
305 ; CHECK-NEXT: [[ADD82:%.*]] = add nsw i32 [[ADD71]], undef
306 ; CHECK-NEXT: [[CMP83:%.*]] = icmp slt i32 [[OP_EXTRA]], [[LIM:%.*]]
307 ; CHECK-NEXT: br i1 [[CMP83]], label [[IF_END_86]], label [[FOR_END_LOOPEXIT:%.*]]
308 ; CHECK: if.end.86:
309 ; CHECK-NEXT: [[ADD_PTR]] = getelementptr inbounds i8, i8* [[P1_044]], i64 [[IDX_EXT]]
310 ; CHECK-NEXT: [[ADD_PTR88]] = getelementptr inbounds i8, i8* [[P2_045]], i64 [[IDX_EXT]]
311 ; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[J_046]], 1
312 ; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[INC]], [[H]]
313 ; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]]
314 ; CHECK: for.end.loopexit:
315 ; CHECK-NEXT: br label [[FOR_END]]
316 ; CHECK: for.end:
317 ; CHECK-NEXT: [[S_1:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[OP_EXTRA]], [[FOR_END_LOOPEXIT]] ]
318 ; CHECK-NEXT: ret i32 [[S_1]]
319 ;
320 entry:
321 %cmp.43 = icmp sgt i32 %h, 0
322 br i1 %cmp.43, label %for.body.lr.ph, label %for.end
324 for.body.lr.ph: ; preds = %entry
325 %idx.ext = sext i32 %lx to i64
326 br label %for.body
328 for.body: ; preds = %for.body.lr.ph, %if.end.86
329 %s.047 = phi i32 [ 0, %for.body.lr.ph ], [ %add82, %if.end.86 ]
330 %j.046 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %if.end.86 ]
331 %p2.045 = phi i8* [ %blk2, %for.body.lr.ph ], [ %add.ptr88, %if.end.86 ]
332 %p1.044 = phi i8* [ %blk1, %for.body.lr.ph ], [ %add.ptr, %if.end.86 ]
333 %0 = load i8, i8* %p1.044, align 1
334 %conv = zext i8 %0 to i32
335 %1 = load i8, i8* %p2.045, align 1
336 %conv2 = zext i8 %1 to i32
337 %sub = sub nsw i32 %conv, %conv2
338 %cmp3 = icmp slt i32 %sub, 0
339 %sub5 = sub nsw i32 0, %sub
340 %sub5.sub = select i1 %cmp3, i32 %sub5, i32 %sub
341 %add = add nsw i32 %sub5.sub, %s.047
342 %arrayidx6 = getelementptr inbounds i8, i8* %p1.044, i64 1
343 %2 = load i8, i8* %arrayidx6, align 1
344 %conv7 = zext i8 %2 to i32
345 %arrayidx8 = getelementptr inbounds i8, i8* %p2.045, i64 1
346 %3 = load i8, i8* %arrayidx8, align 1
347 %conv9 = zext i8 %3 to i32
348 %sub10 = sub nsw i32 %conv7, %conv9
349 %cmp11 = icmp slt i32 %sub10, 0
350 %sub14 = sub nsw i32 0, %sub10
351 %v.1 = select i1 %cmp11, i32 %sub14, i32 %sub10
352 %add16 = add nsw i32 %add, %v.1
353 %arrayidx17 = getelementptr inbounds i8, i8* %p1.044, i64 2
354 %4 = load i8, i8* %arrayidx17, align 1
355 %conv18 = zext i8 %4 to i32
356 %arrayidx19 = getelementptr inbounds i8, i8* %p2.045, i64 2
357 %5 = load i8, i8* %arrayidx19, align 1
358 %conv20 = zext i8 %5 to i32
359 %sub21 = sub nsw i32 %conv18, %conv20
360 %cmp22 = icmp slt i32 %sub21, 0
361 %sub25 = sub nsw i32 0, %sub21
362 %sub25.sub21 = select i1 %cmp22, i32 %sub25, i32 %sub21
363 %add27 = add nsw i32 %add16, %sub25.sub21
364 %arrayidx28 = getelementptr inbounds i8, i8* %p1.044, i64 3
365 %6 = load i8, i8* %arrayidx28, align 1
366 %conv29 = zext i8 %6 to i32
367 %arrayidx30 = getelementptr inbounds i8, i8* %p2.045, i64 3
368 %7 = load i8, i8* %arrayidx30, align 1
369 %conv31 = zext i8 %7 to i32
370 %sub32 = sub nsw i32 %conv29, %conv31
371 %cmp33 = icmp slt i32 %sub32, 0
372 %sub36 = sub nsw i32 0, %sub32
373 %v.3 = select i1 %cmp33, i32 %sub36, i32 %sub32
374 %add38 = add nsw i32 %add27, %v.3
375 %arrayidx39 = getelementptr inbounds i8, i8* %p1.044, i64 4
376 %8 = load i8, i8* %arrayidx39, align 1
377 %conv40 = zext i8 %8 to i32
378 %arrayidx41 = getelementptr inbounds i8, i8* %p2.045, i64 4
379 %9 = load i8, i8* %arrayidx41, align 1
380 %conv42 = zext i8 %9 to i32
381 %sub43 = sub nsw i32 %conv40, %conv42
382 %cmp44 = icmp slt i32 %sub43, 0
383 %sub47 = sub nsw i32 0, %sub43
384 %sub47.sub43 = select i1 %cmp44, i32 %sub47, i32 %sub43
385 %add49 = add nsw i32 %add38, %sub47.sub43
386 %arrayidx50 = getelementptr inbounds i8, i8* %p1.044, i64 5
387 %10 = load i8, i8* %arrayidx50, align 1
388 %conv51 = zext i8 %10 to i32
389 %arrayidx52 = getelementptr inbounds i8, i8* %p2.045, i64 5
390 %11 = load i8, i8* %arrayidx52, align 1
391 %conv53 = zext i8 %11 to i32
392 %sub54 = sub nsw i32 %conv51, %conv53
393 %cmp55 = icmp slt i32 %sub54, 0
394 %sub58 = sub nsw i32 0, %sub54
395 %v.5 = select i1 %cmp55, i32 %sub58, i32 %sub54
396 %add60 = add nsw i32 %add49, %v.5
397 %arrayidx61 = getelementptr inbounds i8, i8* %p1.044, i64 6
398 %12 = load i8, i8* %arrayidx61, align 1
399 %conv62 = zext i8 %12 to i32
400 %arrayidx63 = getelementptr inbounds i8, i8* %p2.045, i64 6
401 %13 = load i8, i8* %arrayidx63, align 1
402 %conv64 = zext i8 %13 to i32
403 %sub65 = sub nsw i32 %conv62, %conv64
404 %cmp66 = icmp slt i32 %sub65, 0
405 %sub69 = sub nsw i32 0, %sub65
406 %sub69.sub65 = select i1 %cmp66, i32 %sub69, i32 %sub65
407 %add71 = add nsw i32 %add60, %sub69.sub65
408 %arrayidx72 = getelementptr inbounds i8, i8* %p1.044, i64 7
409 %14 = load i8, i8* %arrayidx72, align 1
410 %conv73 = zext i8 %14 to i32
411 %arrayidx74 = getelementptr inbounds i8, i8* %p2.045, i64 7
412 %15 = load i8, i8* %arrayidx74, align 1
413 %conv75 = zext i8 %15 to i32
414 %sub76 = sub nsw i32 %conv73, %conv75
415 %cmp77 = icmp slt i32 %sub76, 0
416 %sub80 = sub nsw i32 0, %sub76
417 %v.7 = select i1 %cmp77, i32 %sub80, i32 %sub76
418 %add82 = add nsw i32 %add71, %v.7
419 %cmp83 = icmp slt i32 %add82, %lim
420 br i1 %cmp83, label %if.end.86, label %for.end.loopexit
422 if.end.86: ; preds = %for.body
423 %add.ptr = getelementptr inbounds i8, i8* %p1.044, i64 %idx.ext
424 %add.ptr88 = getelementptr inbounds i8, i8* %p2.045, i64 %idx.ext
425 %inc = add nuw nsw i32 %j.046, 1
426 %cmp = icmp slt i32 %inc, %h
427 br i1 %cmp, label %for.body, label %for.end.loopexit
429 for.end.loopexit: ; preds = %for.body, %if.end.86
430 br label %for.end
432 for.end: ; preds = %for.end.loopexit, %entry
433 %s.1 = phi i32 [ 0, %entry ], [ %add82, %for.end.loopexit ]
434 ret i32 %s.1
435 }