xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/AArch64/horizontal.ll (revision 81d9ed605b126d163f2f1cc226c639f1dfdc5224)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -passes=slp-vectorizer -slp-threshold=-2 -S -pass-remarks-output=%t < %s | FileCheck %s
3; RUN: cat %t | FileCheck -check-prefix=YAML %s
4
5
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.
8
9target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
10target triple = "aarch64--linux"
11
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:            '-19'
19; YAML-NEXT:   - String:          ' and with tree size '
20; YAML-NEXT:   - TreeSize:        '8'
21
22define i32 @test_select(ptr noalias nocapture readonly %blk1, ptr 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_RDX:%.*]], [[FOR_BODY]] ]
32; CHECK-NEXT:    [[J_025:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
33; CHECK-NEXT:    [[P2_024:%.*]] = phi ptr [ [[BLK2:%.*]], [[FOR_BODY_LR_PH]] ], [ [[ADD_PTR29:%.*]], [[FOR_BODY]] ]
34; CHECK-NEXT:    [[P1_023:%.*]] = phi ptr [ [[BLK1:%.*]], [[FOR_BODY_LR_PH]] ], [ [[ADD_PTR:%.*]], [[FOR_BODY]] ]
35; CHECK-NEXT:    [[TMP0:%.*]] = load <4 x i32>, ptr [[P1_023]], align 4
36; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr [[P2_024]], align 4
37; CHECK-NEXT:    [[TMP2:%.*]] = sub nsw <4 x i32> [[TMP0]], [[TMP1]]
38; CHECK-NEXT:    [[TMP3:%.*]] = icmp slt <4 x i32> [[TMP2]], zeroinitializer
39; CHECK-NEXT:    [[TMP4:%.*]] = sub nsw <4 x i32> zeroinitializer, [[TMP2]]
40; CHECK-NEXT:    [[TMP5:%.*]] = select <4 x i1> [[TMP3]], <4 x i32> [[TMP4]], <4 x i32> [[TMP2]]
41; CHECK-NEXT:    [[TMP6:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP5]])
42; CHECK-NEXT:    [[OP_RDX]] = add i32 [[TMP6]], [[S_026]]
43; CHECK-NEXT:    [[ADD_PTR]] = getelementptr inbounds i32, ptr [[P1_023]], i64 [[IDX_EXT]]
44; CHECK-NEXT:    [[ADD_PTR29]] = getelementptr inbounds i32, ptr [[P2_024]], i64 [[IDX_EXT]]
45; CHECK-NEXT:    [[INC]] = add nuw nsw i32 [[J_025]], 1
46; CHECK-NEXT:    [[EXITCOND:%.*]] = icmp eq i32 [[INC]], [[H]]
47; CHECK-NEXT:    br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT:%.*]], label [[FOR_BODY]]
48; CHECK:       for.end.loopexit:
49; CHECK-NEXT:    br label [[FOR_END]]
50; CHECK:       for.end:
51; CHECK-NEXT:    [[S_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[OP_RDX]], [[FOR_END_LOOPEXIT]] ]
52; CHECK-NEXT:    ret i32 [[S_0_LCSSA]]
53;
54entry:
55  %cmp.22 = icmp sgt i32 %h, 0
56  br i1 %cmp.22, label %for.body.lr.ph, label %for.end
57
58for.body.lr.ph:                                   ; preds = %entry
59  %idx.ext = sext i32 %lx to i64
60  br label %for.body
61
62for.body:                                         ; preds = %for.body, %for.body.lr.ph
63  %s.026 = phi i32 [ 0, %for.body.lr.ph ], [ %add27, %for.body ]
64  %j.025 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
65  %p2.024 = phi ptr [ %blk2, %for.body.lr.ph ], [ %add.ptr29, %for.body ]
66  %p1.023 = phi ptr [ %blk1, %for.body.lr.ph ], [ %add.ptr, %for.body ]
67  %0 = load i32, ptr %p1.023, align 4
68  %1 = load i32, ptr %p2.024, align 4
69  %sub = sub nsw i32 %0, %1
70  %cmp2 = icmp slt i32 %sub, 0
71  %sub3 = sub nsw i32 0, %sub
72  %sub3.sub = select i1 %cmp2, i32 %sub3, i32 %sub
73  %add = add nsw i32 %sub3.sub, %s.026
74  %arrayidx4 = getelementptr inbounds i32, ptr %p1.023, i64 1
75  %2 = load i32, ptr %arrayidx4, align 4
76  %arrayidx5 = getelementptr inbounds i32, ptr %p2.024, i64 1
77  %3 = load i32, ptr %arrayidx5, align 4
78  %sub6 = sub nsw i32 %2, %3
79  %cmp7 = icmp slt i32 %sub6, 0
80  %sub9 = sub nsw i32 0, %sub6
81  %v.1 = select i1 %cmp7, i32 %sub9, i32 %sub6
82  %add11 = add nsw i32 %add, %v.1
83  %arrayidx12 = getelementptr inbounds i32, ptr %p1.023, i64 2
84  %4 = load i32, ptr %arrayidx12, align 4
85  %arrayidx13 = getelementptr inbounds i32, ptr %p2.024, i64 2
86  %5 = load i32, ptr %arrayidx13, align 4
87  %sub14 = sub nsw i32 %4, %5
88  %cmp15 = icmp slt i32 %sub14, 0
89  %sub17 = sub nsw i32 0, %sub14
90  %sub17.sub14 = select i1 %cmp15, i32 %sub17, i32 %sub14
91  %add19 = add nsw i32 %add11, %sub17.sub14
92  %arrayidx20 = getelementptr inbounds i32, ptr %p1.023, i64 3
93  %6 = load i32, ptr %arrayidx20, align 4
94  %arrayidx21 = getelementptr inbounds i32, ptr %p2.024, i64 3
95  %7 = load i32, ptr %arrayidx21, align 4
96  %sub22 = sub nsw i32 %6, %7
97  %cmp23 = icmp slt i32 %sub22, 0
98  %sub25 = sub nsw i32 0, %sub22
99  %v.3 = select i1 %cmp23, i32 %sub25, i32 %sub22
100  %add27 = add nsw i32 %add19, %v.3
101  %add.ptr = getelementptr inbounds i32, ptr %p1.023, i64 %idx.ext
102  %add.ptr29 = getelementptr inbounds i32, ptr %p2.024, i64 %idx.ext
103  %inc = add nuw nsw i32 %j.025, 1
104  %exitcond = icmp eq i32 %inc, %h
105  br i1 %exitcond, label %for.end.loopexit, label %for.body
106
107for.end.loopexit:                                 ; preds = %for.body
108  br label %for.end
109
110for.end:                                          ; preds = %for.end.loopexit, %entry
111  %s.0.lcssa = phi i32 [ 0, %entry ], [ %add27, %for.end.loopexit ]
112  ret i32 %s.0.lcssa
113}
114
115;; Check whether SLP can find a reduction phi whose incoming blocks are not
116;; the same as the block containing the phi.
117;;
118;; Came from code like,
119;;
120;; int s = 0;
121;; for (int j = 0; j < h; j++) {
122;;   s += p1ptr p2[0]
123;;   s += p1ptr p2[1];
124;;   s += p1ptr p2[2];
125;;   s += p1ptr p2[3];
126;;   if (s >= lim)
127;;      break;
128;;   p1 += lx;
129;;   p2 += lx;
130;; }
131define i32 @reduction_with_br(ptr noalias nocapture readonly %blk1, ptr noalias nocapture readonly %blk2, i32 %lx, i32 %h, i32 %lim) {
132; YAML:      --- !Passed
133; YAML-NEXT: Pass:            slp-vectorizer
134; YAML-NEXT: Name:            VectorizedHorizontalReduction
135; YAML-NEXT: Function:        reduction_with_br
136; YAML-NEXT: Args:
137; YAML-NEXT:   - String:          'Vectorized horizontal reduction with cost '
138; YAML-NEXT:   - Cost:            '-10'
139; YAML-NEXT:   - String:          ' and with tree size '
140; YAML-NEXT:   - TreeSize:        '3'
141; CHECK-LABEL: @reduction_with_br(
142; CHECK-NEXT:  entry:
143; CHECK-NEXT:    [[CMP_16:%.*]] = icmp sgt i32 [[H:%.*]], 0
144; CHECK-NEXT:    br i1 [[CMP_16]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_END:%.*]]
145; CHECK:       for.body.lr.ph:
146; CHECK-NEXT:    [[IDX_EXT:%.*]] = sext i32 [[LX:%.*]] to i64
147; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
148; CHECK:       for.body:
149; CHECK-NEXT:    [[S_020:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[OP_RDX:%.*]], [[IF_END:%.*]] ]
150; CHECK-NEXT:    [[J_019:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[INC:%.*]], [[IF_END]] ]
151; CHECK-NEXT:    [[P2_018:%.*]] = phi ptr [ [[BLK2:%.*]], [[FOR_BODY_LR_PH]] ], [ [[ADD_PTR16:%.*]], [[IF_END]] ]
152; CHECK-NEXT:    [[P1_017:%.*]] = phi ptr [ [[BLK1:%.*]], [[FOR_BODY_LR_PH]] ], [ [[ADD_PTR:%.*]], [[IF_END]] ]
153; CHECK-NEXT:    [[TMP0:%.*]] = load <4 x i32>, ptr [[P1_017]], align 4
154; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, ptr [[P2_018]], align 4
155; CHECK-NEXT:    [[TMP2:%.*]] = mul nsw <4 x i32> [[TMP1]], [[TMP0]]
156; CHECK-NEXT:    [[TMP3:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP2]])
157; CHECK-NEXT:    [[OP_RDX]] = add i32 [[TMP3]], [[S_020]]
158; CHECK-NEXT:    [[CMP14:%.*]] = icmp slt i32 [[OP_RDX]], [[LIM:%.*]]
159; CHECK-NEXT:    br i1 [[CMP14]], label [[IF_END]], label [[FOR_END_LOOPEXIT:%.*]]
160; CHECK:       if.end:
161; CHECK-NEXT:    [[ADD_PTR]] = getelementptr inbounds i32, ptr [[P1_017]], i64 [[IDX_EXT]]
162; CHECK-NEXT:    [[ADD_PTR16]] = getelementptr inbounds i32, ptr [[P2_018]], i64 [[IDX_EXT]]
163; CHECK-NEXT:    [[INC]] = add nuw nsw i32 [[J_019]], 1
164; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[INC]], [[H]]
165; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]]
166; CHECK:       for.end.loopexit:
167; CHECK-NEXT:    br label [[FOR_END]]
168; CHECK:       for.end:
169; CHECK-NEXT:    [[S_1:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[OP_RDX]], [[FOR_END_LOOPEXIT]] ]
170; CHECK-NEXT:    ret i32 [[S_1]]
171;
172entry:
173  %cmp.16 = icmp sgt i32 %h, 0
174  br i1 %cmp.16, label %for.body.lr.ph, label %for.end
175
176for.body.lr.ph:                                   ; preds = %entry
177  %idx.ext = sext i32 %lx to i64
178  br label %for.body
179
180for.body:                                         ; preds = %for.body.lr.ph, %if.end
181  %s.020 = phi i32 [ 0, %for.body.lr.ph ], [ %add13, %if.end ]
182  %j.019 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %if.end ]
183  %p2.018 = phi ptr [ %blk2, %for.body.lr.ph ], [ %add.ptr16, %if.end ]
184  %p1.017 = phi ptr [ %blk1, %for.body.lr.ph ], [ %add.ptr, %if.end ]
185  %0 = load i32, ptr %p1.017, align 4
186  %1 = load i32, ptr %p2.018, align 4
187  %mul = mul nsw i32 %1, %0
188  %add = add nsw i32 %mul, %s.020
189  %arrayidx2 = getelementptr inbounds i32, ptr %p1.017, i64 1
190  %2 = load i32, ptr %arrayidx2, align 4
191  %arrayidx3 = getelementptr inbounds i32, ptr %p2.018, i64 1
192  %3 = load i32, ptr %arrayidx3, align 4
193  %mul4 = mul nsw i32 %3, %2
194  %add5 = add nsw i32 %add, %mul4
195  %arrayidx6 = getelementptr inbounds i32, ptr %p1.017, i64 2
196  %4 = load i32, ptr %arrayidx6, align 4
197  %arrayidx7 = getelementptr inbounds i32, ptr %p2.018, i64 2
198  %5 = load i32, ptr %arrayidx7, align 4
199  %mul8 = mul nsw i32 %5, %4
200  %add9 = add nsw i32 %add5, %mul8
201  %arrayidx10 = getelementptr inbounds i32, ptr %p1.017, i64 3
202  %6 = load i32, ptr %arrayidx10, align 4
203  %arrayidx11 = getelementptr inbounds i32, ptr %p2.018, i64 3
204  %7 = load i32, ptr %arrayidx11, align 4
205  %mul12 = mul nsw i32 %7, %6
206  %add13 = add nsw i32 %add9, %mul12
207  %cmp14 = icmp slt i32 %add13, %lim
208  br i1 %cmp14, label %if.end, label %for.end.loopexit
209
210if.end:                                           ; preds = %for.body
211  %add.ptr = getelementptr inbounds i32, ptr %p1.017, i64 %idx.ext
212  %add.ptr16 = getelementptr inbounds i32, ptr %p2.018, i64 %idx.ext
213  %inc = add nuw nsw i32 %j.019, 1
214  %cmp = icmp slt i32 %inc, %h
215  br i1 %cmp, label %for.body, label %for.end.loopexit
216
217for.end.loopexit:                                 ; preds = %for.body, %if.end
218  br label %for.end
219
220for.end:                                          ; preds = %for.end.loopexit, %entry
221  %s.1 = phi i32 [ 0, %entry ], [ %add13, %for.end.loopexit ]
222  ret i32 %s.1
223}
224
225; YAML:      --- !Passed
226; YAML-NEXT: Pass:            slp-vectorizer
227; YAML-NEXT: Name:            VectorizedHorizontalReduction
228; YAML-NEXT: Function:        test_unrolled_select
229; YAML-NEXT: Args:
230; YAML-NEXT:   - String:          'Vectorized horizontal reduction with cost '
231; YAML-NEXT:   - Cost:            '-41'
232; YAML-NEXT:   - String:          ' and with tree size '
233; YAML-NEXT:   - TreeSize:        '10'
234
235define i32 @test_unrolled_select(ptr noalias nocapture readonly %blk1, ptr noalias nocapture readonly %blk2, i32 %lx, i32 %h, i32 %lim) #0 {
236; CHECK-LABEL: @test_unrolled_select(
237; CHECK-NEXT:  entry:
238; CHECK-NEXT:    [[CMP_43:%.*]] = icmp sgt i32 [[H:%.*]], 0
239; CHECK-NEXT:    br i1 [[CMP_43]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_END:%.*]]
240; CHECK:       for.body.lr.ph:
241; CHECK-NEXT:    [[IDX_EXT:%.*]] = sext i32 [[LX:%.*]] to i64
242; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
243; CHECK:       for.body:
244; CHECK-NEXT:    [[S_047:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[OP_RDX:%.*]], [[IF_END_86:%.*]] ]
245; CHECK-NEXT:    [[J_046:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[INC:%.*]], [[IF_END_86]] ]
246; CHECK-NEXT:    [[P2_045:%.*]] = phi ptr [ [[BLK2:%.*]], [[FOR_BODY_LR_PH]] ], [ [[ADD_PTR88:%.*]], [[IF_END_86]] ]
247; CHECK-NEXT:    [[P1_044:%.*]] = phi ptr [ [[BLK1:%.*]], [[FOR_BODY_LR_PH]] ], [ [[ADD_PTR:%.*]], [[IF_END_86]] ]
248; CHECK-NEXT:    [[TMP0:%.*]] = load <8 x i8>, ptr [[P1_044]], align 1
249; CHECK-NEXT:    [[TMP1:%.*]] = zext <8 x i8> [[TMP0]] to <8 x i16>
250; CHECK-NEXT:    [[TMP2:%.*]] = load <8 x i8>, ptr [[P2_045]], align 1
251; CHECK-NEXT:    [[TMP3:%.*]] = zext <8 x i8> [[TMP2]] to <8 x i16>
252; CHECK-NEXT:    [[TMP4:%.*]] = sub <8 x i16> [[TMP1]], [[TMP3]]
253; CHECK-NEXT:    [[TMP5:%.*]] = sext <8 x i16> [[TMP4]] to <8 x i32>
254; CHECK-NEXT:    [[TMP6:%.*]] = icmp slt <8 x i32> [[TMP5]], zeroinitializer
255; CHECK-NEXT:    [[TMP7:%.*]] = sub <8 x i16> zeroinitializer, [[TMP4]]
256; CHECK-NEXT:    [[TMP8:%.*]] = select <8 x i1> [[TMP6]], <8 x i16> [[TMP7]], <8 x i16> [[TMP4]]
257; CHECK-NEXT:    [[TMP9:%.*]] = sext <8 x i16> [[TMP8]] to <8 x i32>
258; CHECK-NEXT:    [[TMP10:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[TMP9]])
259; CHECK-NEXT:    [[OP_RDX]] = add i32 [[TMP10]], [[S_047]]
260; CHECK-NEXT:    [[CMP83:%.*]] = icmp slt i32 [[OP_RDX]], [[LIM:%.*]]
261; CHECK-NEXT:    br i1 [[CMP83]], label [[IF_END_86]], label [[FOR_END_LOOPEXIT:%.*]]
262; CHECK:       if.end.86:
263; CHECK-NEXT:    [[ADD_PTR]] = getelementptr inbounds i8, ptr [[P1_044]], i64 [[IDX_EXT]]
264; CHECK-NEXT:    [[ADD_PTR88]] = getelementptr inbounds i8, ptr [[P2_045]], i64 [[IDX_EXT]]
265; CHECK-NEXT:    [[INC]] = add nuw nsw i32 [[J_046]], 1
266; CHECK-NEXT:    [[CMP:%.*]] = icmp slt i32 [[INC]], [[H]]
267; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]]
268; CHECK:       for.end.loopexit:
269; CHECK-NEXT:    br label [[FOR_END]]
270; CHECK:       for.end:
271; CHECK-NEXT:    [[S_1:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[OP_RDX]], [[FOR_END_LOOPEXIT]] ]
272; CHECK-NEXT:    ret i32 [[S_1]]
273;
274entry:
275  %cmp.43 = icmp sgt i32 %h, 0
276  br i1 %cmp.43, label %for.body.lr.ph, label %for.end
277
278for.body.lr.ph:                                   ; preds = %entry
279  %idx.ext = sext i32 %lx to i64
280  br label %for.body
281
282for.body:                                         ; preds = %for.body.lr.ph, %if.end.86
283  %s.047 = phi i32 [ 0, %for.body.lr.ph ], [ %add82, %if.end.86 ]
284  %j.046 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %if.end.86 ]
285  %p2.045 = phi ptr [ %blk2, %for.body.lr.ph ], [ %add.ptr88, %if.end.86 ]
286  %p1.044 = phi ptr [ %blk1, %for.body.lr.ph ], [ %add.ptr, %if.end.86 ]
287  %0 = load i8, ptr %p1.044, align 1
288  %conv = zext i8 %0 to i32
289  %1 = load i8, ptr %p2.045, align 1
290  %conv2 = zext i8 %1 to i32
291  %sub = sub nsw i32 %conv, %conv2
292  %cmp3 = icmp slt i32 %sub, 0
293  %sub5 = sub nsw i32 0, %sub
294  %sub5.sub = select i1 %cmp3, i32 %sub5, i32 %sub
295  %add = add nsw i32 %sub5.sub, %s.047
296  %arrayidx6 = getelementptr inbounds i8, ptr %p1.044, i64 1
297  %2 = load i8, ptr %arrayidx6, align 1
298  %conv7 = zext i8 %2 to i32
299  %arrayidx8 = getelementptr inbounds i8, ptr %p2.045, i64 1
300  %3 = load i8, ptr %arrayidx8, align 1
301  %conv9 = zext i8 %3 to i32
302  %sub10 = sub nsw i32 %conv7, %conv9
303  %cmp11 = icmp slt i32 %sub10, 0
304  %sub14 = sub nsw i32 0, %sub10
305  %v.1 = select i1 %cmp11, i32 %sub14, i32 %sub10
306  %add16 = add nsw i32 %add, %v.1
307  %arrayidx17 = getelementptr inbounds i8, ptr %p1.044, i64 2
308  %4 = load i8, ptr %arrayidx17, align 1
309  %conv18 = zext i8 %4 to i32
310  %arrayidx19 = getelementptr inbounds i8, ptr %p2.045, i64 2
311  %5 = load i8, ptr %arrayidx19, align 1
312  %conv20 = zext i8 %5 to i32
313  %sub21 = sub nsw i32 %conv18, %conv20
314  %cmp22 = icmp slt i32 %sub21, 0
315  %sub25 = sub nsw i32 0, %sub21
316  %sub25.sub21 = select i1 %cmp22, i32 %sub25, i32 %sub21
317  %add27 = add nsw i32 %add16, %sub25.sub21
318  %arrayidx28 = getelementptr inbounds i8, ptr %p1.044, i64 3
319  %6 = load i8, ptr %arrayidx28, align 1
320  %conv29 = zext i8 %6 to i32
321  %arrayidx30 = getelementptr inbounds i8, ptr %p2.045, i64 3
322  %7 = load i8, ptr %arrayidx30, align 1
323  %conv31 = zext i8 %7 to i32
324  %sub32 = sub nsw i32 %conv29, %conv31
325  %cmp33 = icmp slt i32 %sub32, 0
326  %sub36 = sub nsw i32 0, %sub32
327  %v.3 = select i1 %cmp33, i32 %sub36, i32 %sub32
328  %add38 = add nsw i32 %add27, %v.3
329  %arrayidx39 = getelementptr inbounds i8, ptr %p1.044, i64 4
330  %8 = load i8, ptr %arrayidx39, align 1
331  %conv40 = zext i8 %8 to i32
332  %arrayidx41 = getelementptr inbounds i8, ptr %p2.045, i64 4
333  %9 = load i8, ptr %arrayidx41, align 1
334  %conv42 = zext i8 %9 to i32
335  %sub43 = sub nsw i32 %conv40, %conv42
336  %cmp44 = icmp slt i32 %sub43, 0
337  %sub47 = sub nsw i32 0, %sub43
338  %sub47.sub43 = select i1 %cmp44, i32 %sub47, i32 %sub43
339  %add49 = add nsw i32 %add38, %sub47.sub43
340  %arrayidx50 = getelementptr inbounds i8, ptr %p1.044, i64 5
341  %10 = load i8, ptr %arrayidx50, align 1
342  %conv51 = zext i8 %10 to i32
343  %arrayidx52 = getelementptr inbounds i8, ptr %p2.045, i64 5
344  %11 = load i8, ptr %arrayidx52, align 1
345  %conv53 = zext i8 %11 to i32
346  %sub54 = sub nsw i32 %conv51, %conv53
347  %cmp55 = icmp slt i32 %sub54, 0
348  %sub58 = sub nsw i32 0, %sub54
349  %v.5 = select i1 %cmp55, i32 %sub58, i32 %sub54
350  %add60 = add nsw i32 %add49, %v.5
351  %arrayidx61 = getelementptr inbounds i8, ptr %p1.044, i64 6
352  %12 = load i8, ptr %arrayidx61, align 1
353  %conv62 = zext i8 %12 to i32
354  %arrayidx63 = getelementptr inbounds i8, ptr %p2.045, i64 6
355  %13 = load i8, ptr %arrayidx63, align 1
356  %conv64 = zext i8 %13 to i32
357  %sub65 = sub nsw i32 %conv62, %conv64
358  %cmp66 = icmp slt i32 %sub65, 0
359  %sub69 = sub nsw i32 0, %sub65
360  %sub69.sub65 = select i1 %cmp66, i32 %sub69, i32 %sub65
361  %add71 = add nsw i32 %add60, %sub69.sub65
362  %arrayidx72 = getelementptr inbounds i8, ptr %p1.044, i64 7
363  %14 = load i8, ptr %arrayidx72, align 1
364  %conv73 = zext i8 %14 to i32
365  %arrayidx74 = getelementptr inbounds i8, ptr %p2.045, i64 7
366  %15 = load i8, ptr %arrayidx74, align 1
367  %conv75 = zext i8 %15 to i32
368  %sub76 = sub nsw i32 %conv73, %conv75
369  %cmp77 = icmp slt i32 %sub76, 0
370  %sub80 = sub nsw i32 0, %sub76
371  %v.7 = select i1 %cmp77, i32 %sub80, i32 %sub76
372  %add82 = add nsw i32 %add71, %v.7
373  %cmp83 = icmp slt i32 %add82, %lim
374  br i1 %cmp83, label %if.end.86, label %for.end.loopexit
375
376if.end.86:                                        ; preds = %for.body
377  %add.ptr = getelementptr inbounds i8, ptr %p1.044, i64 %idx.ext
378  %add.ptr88 = getelementptr inbounds i8, ptr %p2.045, i64 %idx.ext
379  %inc = add nuw nsw i32 %j.046, 1
380  %cmp = icmp slt i32 %inc, %h
381  br i1 %cmp, label %for.body, label %for.end.loopexit
382
383for.end.loopexit:                                 ; preds = %for.body, %if.end.86
384  br label %for.end
385
386for.end:                                          ; preds = %for.end.loopexit, %entry
387  %s.1 = phi i32 [ 0, %entry ], [ %add82, %for.end.loopexit ]
388  ret i32 %s.1
389}
390
391