1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt -S -passes=slp-vectorizer,dce,instcombine < %s | FileCheck %s --check-prefix=GENERIC
3 ; RUN: opt -S -mcpu=kryo -passes=slp-vectorizer,dce,instcombine < %s | FileCheck %s --check-prefix=KRYO
5 target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
6 target triple = "aarch64--linux-gnu"
8 ; These tests check that we vectorize the index calculations in the
9 ; gather-reduce pattern shown below. We check cases having i32 and i64
12 ; int gather_reduce_8x16(short *a, short *b, short *g, int n) {
14 ; for (int i = 0; i < n ; ++i) {
15 ; sum += g[*a++ - b[0]]; sum += g[*a++ - b[1]];
16 ; sum += g[*a++ - b[2]]; sum += g[*a++ - b[3]];
17 ; sum += g[*a++ - b[4]]; sum += g[*a++ - b[5]];
18 ; sum += g[*a++ - b[6]]; sum += g[*a++ - b[7]];
23 define i32 @gather_reduce_8x16_i32(ptr nocapture readonly %a, ptr nocapture readonly %b, ptr nocapture readonly %g, i32 %n) {
24 ; GENERIC-LABEL: @gather_reduce_8x16_i32(
25 ; GENERIC-NEXT: entry:
26 ; GENERIC-NEXT: [[CMP_99:%.*]] = icmp sgt i32 [[N:%.*]], 0
27 ; GENERIC-NEXT: br i1 [[CMP_99]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
28 ; GENERIC: for.body.preheader:
29 ; GENERIC-NEXT: br label [[FOR_BODY:%.*]]
30 ; GENERIC: for.cond.cleanup.loopexit:
31 ; GENERIC-NEXT: br label [[FOR_COND_CLEANUP]]
32 ; GENERIC: for.cond.cleanup:
33 ; GENERIC-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[ADD66:%.*]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ]
34 ; GENERIC-NEXT: ret i32 [[SUM_0_LCSSA]]
36 ; GENERIC-NEXT: [[I_0103:%.*]] = phi i32 [ [[INC:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
37 ; GENERIC-NEXT: [[SUM_0102:%.*]] = phi i32 [ [[ADD66]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
38 ; GENERIC-NEXT: [[A_ADDR_0101:%.*]] = phi ptr [ [[INCDEC_PTR58:%.*]], [[FOR_BODY]] ], [ [[A:%.*]], [[FOR_BODY_PREHEADER]] ]
39 ; GENERIC-NEXT: [[INCDEC_PTR58]] = getelementptr inbounds i8, ptr [[A_ADDR_0101]], i64 16
40 ; GENERIC-NEXT: [[TMP0:%.*]] = load <8 x i16>, ptr [[A_ADDR_0101]], align 2
41 ; GENERIC-NEXT: [[TMP1:%.*]] = zext <8 x i16> [[TMP0]] to <8 x i32>
42 ; GENERIC-NEXT: [[TMP2:%.*]] = load <8 x i16>, ptr [[B:%.*]], align 2
43 ; GENERIC-NEXT: [[TMP3:%.*]] = zext <8 x i16> [[TMP2]] to <8 x i32>
44 ; GENERIC-NEXT: [[TMP4:%.*]] = sub nsw <8 x i32> [[TMP1]], [[TMP3]]
45 ; GENERIC-NEXT: [[TMP5:%.*]] = extractelement <8 x i32> [[TMP4]], i64 0
46 ; GENERIC-NEXT: [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
47 ; GENERIC-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[G:%.*]], i64 [[TMP6]]
48 ; GENERIC-NEXT: [[TMP7:%.*]] = load i16, ptr [[ARRAYIDX]], align 2
49 ; GENERIC-NEXT: [[CONV3:%.*]] = zext i16 [[TMP7]] to i32
50 ; GENERIC-NEXT: [[ADD:%.*]] = add nsw i32 [[SUM_0102]], [[CONV3]]
51 ; GENERIC-NEXT: [[TMP8:%.*]] = extractelement <8 x i32> [[TMP4]], i64 1
52 ; GENERIC-NEXT: [[TMP9:%.*]] = sext i32 [[TMP8]] to i64
53 ; GENERIC-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP9]]
54 ; GENERIC-NEXT: [[TMP10:%.*]] = load i16, ptr [[ARRAYIDX10]], align 2
55 ; GENERIC-NEXT: [[CONV11:%.*]] = zext i16 [[TMP10]] to i32
56 ; GENERIC-NEXT: [[ADD12:%.*]] = add nsw i32 [[ADD]], [[CONV11]]
57 ; GENERIC-NEXT: [[TMP11:%.*]] = extractelement <8 x i32> [[TMP4]], i64 2
58 ; GENERIC-NEXT: [[TMP12:%.*]] = sext i32 [[TMP11]] to i64
59 ; GENERIC-NEXT: [[ARRAYIDX19:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP12]]
60 ; GENERIC-NEXT: [[TMP13:%.*]] = load i16, ptr [[ARRAYIDX19]], align 2
61 ; GENERIC-NEXT: [[CONV20:%.*]] = zext i16 [[TMP13]] to i32
62 ; GENERIC-NEXT: [[ADD21:%.*]] = add nsw i32 [[ADD12]], [[CONV20]]
63 ; GENERIC-NEXT: [[TMP14:%.*]] = extractelement <8 x i32> [[TMP4]], i64 3
64 ; GENERIC-NEXT: [[TMP15:%.*]] = sext i32 [[TMP14]] to i64
65 ; GENERIC-NEXT: [[ARRAYIDX28:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP15]]
66 ; GENERIC-NEXT: [[TMP16:%.*]] = load i16, ptr [[ARRAYIDX28]], align 2
67 ; GENERIC-NEXT: [[CONV29:%.*]] = zext i16 [[TMP16]] to i32
68 ; GENERIC-NEXT: [[ADD30:%.*]] = add nsw i32 [[ADD21]], [[CONV29]]
69 ; GENERIC-NEXT: [[TMP17:%.*]] = extractelement <8 x i32> [[TMP4]], i64 4
70 ; GENERIC-NEXT: [[TMP18:%.*]] = sext i32 [[TMP17]] to i64
71 ; GENERIC-NEXT: [[ARRAYIDX37:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP18]]
72 ; GENERIC-NEXT: [[TMP19:%.*]] = load i16, ptr [[ARRAYIDX37]], align 2
73 ; GENERIC-NEXT: [[CONV38:%.*]] = zext i16 [[TMP19]] to i32
74 ; GENERIC-NEXT: [[ADD39:%.*]] = add nsw i32 [[ADD30]], [[CONV38]]
75 ; GENERIC-NEXT: [[TMP20:%.*]] = extractelement <8 x i32> [[TMP4]], i64 5
76 ; GENERIC-NEXT: [[TMP21:%.*]] = sext i32 [[TMP20]] to i64
77 ; GENERIC-NEXT: [[ARRAYIDX46:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP21]]
78 ; GENERIC-NEXT: [[TMP22:%.*]] = load i16, ptr [[ARRAYIDX46]], align 2
79 ; GENERIC-NEXT: [[CONV47:%.*]] = zext i16 [[TMP22]] to i32
80 ; GENERIC-NEXT: [[ADD48:%.*]] = add nsw i32 [[ADD39]], [[CONV47]]
81 ; GENERIC-NEXT: [[TMP23:%.*]] = extractelement <8 x i32> [[TMP4]], i64 6
82 ; GENERIC-NEXT: [[TMP24:%.*]] = sext i32 [[TMP23]] to i64
83 ; GENERIC-NEXT: [[ARRAYIDX55:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP24]]
84 ; GENERIC-NEXT: [[TMP25:%.*]] = load i16, ptr [[ARRAYIDX55]], align 2
85 ; GENERIC-NEXT: [[CONV56:%.*]] = zext i16 [[TMP25]] to i32
86 ; GENERIC-NEXT: [[ADD57:%.*]] = add nsw i32 [[ADD48]], [[CONV56]]
87 ; GENERIC-NEXT: [[TMP26:%.*]] = extractelement <8 x i32> [[TMP4]], i64 7
88 ; GENERIC-NEXT: [[TMP27:%.*]] = sext i32 [[TMP26]] to i64
89 ; GENERIC-NEXT: [[ARRAYIDX64:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP27]]
90 ; GENERIC-NEXT: [[TMP28:%.*]] = load i16, ptr [[ARRAYIDX64]], align 2
91 ; GENERIC-NEXT: [[CONV65:%.*]] = zext i16 [[TMP28]] to i32
92 ; GENERIC-NEXT: [[ADD66]] = add nsw i32 [[ADD57]], [[CONV65]]
93 ; GENERIC-NEXT: [[INC]] = add nuw nsw i32 [[I_0103]], 1
94 ; GENERIC-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[INC]], [[N]]
95 ; GENERIC-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]]
97 ; KRYO-LABEL: @gather_reduce_8x16_i32(
99 ; KRYO-NEXT: [[CMP_99:%.*]] = icmp sgt i32 [[N:%.*]], 0
100 ; KRYO-NEXT: br i1 [[CMP_99]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
101 ; KRYO: for.body.preheader:
102 ; KRYO-NEXT: br label [[FOR_BODY:%.*]]
103 ; KRYO: for.cond.cleanup.loopexit:
104 ; KRYO-NEXT: br label [[FOR_COND_CLEANUP]]
105 ; KRYO: for.cond.cleanup:
106 ; KRYO-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[ADD66:%.*]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ]
107 ; KRYO-NEXT: ret i32 [[SUM_0_LCSSA]]
109 ; KRYO-NEXT: [[I_0103:%.*]] = phi i32 [ [[INC:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
110 ; KRYO-NEXT: [[SUM_0102:%.*]] = phi i32 [ [[ADD66]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
111 ; KRYO-NEXT: [[A_ADDR_0101:%.*]] = phi ptr [ [[INCDEC_PTR58:%.*]], [[FOR_BODY]] ], [ [[A:%.*]], [[FOR_BODY_PREHEADER]] ]
112 ; KRYO-NEXT: [[INCDEC_PTR58]] = getelementptr inbounds i8, ptr [[A_ADDR_0101]], i64 16
113 ; KRYO-NEXT: [[TMP0:%.*]] = load <8 x i16>, ptr [[A_ADDR_0101]], align 2
114 ; KRYO-NEXT: [[TMP1:%.*]] = zext <8 x i16> [[TMP0]] to <8 x i32>
115 ; KRYO-NEXT: [[TMP2:%.*]] = load <8 x i16>, ptr [[B:%.*]], align 2
116 ; KRYO-NEXT: [[TMP3:%.*]] = zext <8 x i16> [[TMP2]] to <8 x i32>
117 ; KRYO-NEXT: [[TMP4:%.*]] = sub nsw <8 x i32> [[TMP1]], [[TMP3]]
118 ; KRYO-NEXT: [[TMP5:%.*]] = extractelement <8 x i32> [[TMP4]], i64 0
119 ; KRYO-NEXT: [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
120 ; KRYO-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[G:%.*]], i64 [[TMP6]]
121 ; KRYO-NEXT: [[TMP7:%.*]] = load i16, ptr [[ARRAYIDX]], align 2
122 ; KRYO-NEXT: [[CONV3:%.*]] = zext i16 [[TMP7]] to i32
123 ; KRYO-NEXT: [[ADD:%.*]] = add nsw i32 [[SUM_0102]], [[CONV3]]
124 ; KRYO-NEXT: [[TMP8:%.*]] = extractelement <8 x i32> [[TMP4]], i64 1
125 ; KRYO-NEXT: [[TMP9:%.*]] = sext i32 [[TMP8]] to i64
126 ; KRYO-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP9]]
127 ; KRYO-NEXT: [[TMP10:%.*]] = load i16, ptr [[ARRAYIDX10]], align 2
128 ; KRYO-NEXT: [[CONV11:%.*]] = zext i16 [[TMP10]] to i32
129 ; KRYO-NEXT: [[ADD12:%.*]] = add nsw i32 [[ADD]], [[CONV11]]
130 ; KRYO-NEXT: [[TMP11:%.*]] = extractelement <8 x i32> [[TMP4]], i64 2
131 ; KRYO-NEXT: [[TMP12:%.*]] = sext i32 [[TMP11]] to i64
132 ; KRYO-NEXT: [[ARRAYIDX19:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP12]]
133 ; KRYO-NEXT: [[TMP13:%.*]] = load i16, ptr [[ARRAYIDX19]], align 2
134 ; KRYO-NEXT: [[CONV20:%.*]] = zext i16 [[TMP13]] to i32
135 ; KRYO-NEXT: [[ADD21:%.*]] = add nsw i32 [[ADD12]], [[CONV20]]
136 ; KRYO-NEXT: [[TMP14:%.*]] = extractelement <8 x i32> [[TMP4]], i64 3
137 ; KRYO-NEXT: [[TMP15:%.*]] = sext i32 [[TMP14]] to i64
138 ; KRYO-NEXT: [[ARRAYIDX28:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP15]]
139 ; KRYO-NEXT: [[TMP16:%.*]] = load i16, ptr [[ARRAYIDX28]], align 2
140 ; KRYO-NEXT: [[CONV29:%.*]] = zext i16 [[TMP16]] to i32
141 ; KRYO-NEXT: [[ADD30:%.*]] = add nsw i32 [[ADD21]], [[CONV29]]
142 ; KRYO-NEXT: [[TMP17:%.*]] = extractelement <8 x i32> [[TMP4]], i64 4
143 ; KRYO-NEXT: [[TMP18:%.*]] = sext i32 [[TMP17]] to i64
144 ; KRYO-NEXT: [[ARRAYIDX37:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP18]]
145 ; KRYO-NEXT: [[TMP19:%.*]] = load i16, ptr [[ARRAYIDX37]], align 2
146 ; KRYO-NEXT: [[CONV38:%.*]] = zext i16 [[TMP19]] to i32
147 ; KRYO-NEXT: [[ADD39:%.*]] = add nsw i32 [[ADD30]], [[CONV38]]
148 ; KRYO-NEXT: [[TMP20:%.*]] = extractelement <8 x i32> [[TMP4]], i64 5
149 ; KRYO-NEXT: [[TMP21:%.*]] = sext i32 [[TMP20]] to i64
150 ; KRYO-NEXT: [[ARRAYIDX46:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP21]]
151 ; KRYO-NEXT: [[TMP22:%.*]] = load i16, ptr [[ARRAYIDX46]], align 2
152 ; KRYO-NEXT: [[CONV47:%.*]] = zext i16 [[TMP22]] to i32
153 ; KRYO-NEXT: [[ADD48:%.*]] = add nsw i32 [[ADD39]], [[CONV47]]
154 ; KRYO-NEXT: [[TMP23:%.*]] = extractelement <8 x i32> [[TMP4]], i64 6
155 ; KRYO-NEXT: [[TMP24:%.*]] = sext i32 [[TMP23]] to i64
156 ; KRYO-NEXT: [[ARRAYIDX55:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP24]]
157 ; KRYO-NEXT: [[TMP25:%.*]] = load i16, ptr [[ARRAYIDX55]], align 2
158 ; KRYO-NEXT: [[CONV56:%.*]] = zext i16 [[TMP25]] to i32
159 ; KRYO-NEXT: [[ADD57:%.*]] = add nsw i32 [[ADD48]], [[CONV56]]
160 ; KRYO-NEXT: [[TMP26:%.*]] = extractelement <8 x i32> [[TMP4]], i64 7
161 ; KRYO-NEXT: [[TMP27:%.*]] = sext i32 [[TMP26]] to i64
162 ; KRYO-NEXT: [[ARRAYIDX64:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP27]]
163 ; KRYO-NEXT: [[TMP28:%.*]] = load i16, ptr [[ARRAYIDX64]], align 2
164 ; KRYO-NEXT: [[CONV65:%.*]] = zext i16 [[TMP28]] to i32
165 ; KRYO-NEXT: [[ADD66]] = add nsw i32 [[ADD57]], [[CONV65]]
166 ; KRYO-NEXT: [[INC]] = add nuw nsw i32 [[I_0103]], 1
167 ; KRYO-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[INC]], [[N]]
168 ; KRYO-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]]
171 %cmp.99 = icmp sgt i32 %n, 0
172 br i1 %cmp.99, label %for.body.preheader, label %for.cond.cleanup
177 for.cond.cleanup.loopexit:
178 br label %for.cond.cleanup
181 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add66, %for.cond.cleanup.loopexit ]
185 %i.0103 = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]
186 %sum.0102 = phi i32 [ %add66, %for.body ], [ 0, %for.body.preheader ]
187 %a.addr.0101 = phi ptr [ %incdec.ptr58, %for.body ], [ %a, %for.body.preheader ]
188 %incdec.ptr = getelementptr inbounds i16, ptr %a.addr.0101, i64 1
189 %0 = load i16, ptr %a.addr.0101, align 2
190 %conv = zext i16 %0 to i32
191 %incdec.ptr1 = getelementptr inbounds i16, ptr %b, i64 1
192 %1 = load i16, ptr %b, align 2
193 %conv2 = zext i16 %1 to i32
194 %sub = sub nsw i32 %conv, %conv2
195 %arrayidx = getelementptr inbounds i16, ptr %g, i32 %sub
196 %2 = load i16, ptr %arrayidx, align 2
197 %conv3 = zext i16 %2 to i32
198 %add = add nsw i32 %conv3, %sum.0102
199 %incdec.ptr4 = getelementptr inbounds i16, ptr %a.addr.0101, i64 2
200 %3 = load i16, ptr %incdec.ptr, align 2
201 %conv5 = zext i16 %3 to i32
202 %incdec.ptr6 = getelementptr inbounds i16, ptr %b, i64 2
203 %4 = load i16, ptr %incdec.ptr1, align 2
204 %conv7 = zext i16 %4 to i32
205 %sub8 = sub nsw i32 %conv5, %conv7
206 %arrayidx10 = getelementptr inbounds i16, ptr %g, i32 %sub8
207 %5 = load i16, ptr %arrayidx10, align 2
208 %conv11 = zext i16 %5 to i32
209 %add12 = add nsw i32 %add, %conv11
210 %incdec.ptr13 = getelementptr inbounds i16, ptr %a.addr.0101, i64 3
211 %6 = load i16, ptr %incdec.ptr4, align 2
212 %conv14 = zext i16 %6 to i32
213 %incdec.ptr15 = getelementptr inbounds i16, ptr %b, i64 3
214 %7 = load i16, ptr %incdec.ptr6, align 2
215 %conv16 = zext i16 %7 to i32
216 %sub17 = sub nsw i32 %conv14, %conv16
217 %arrayidx19 = getelementptr inbounds i16, ptr %g, i32 %sub17
218 %8 = load i16, ptr %arrayidx19, align 2
219 %conv20 = zext i16 %8 to i32
220 %add21 = add nsw i32 %add12, %conv20
221 %incdec.ptr22 = getelementptr inbounds i16, ptr %a.addr.0101, i64 4
222 %9 = load i16, ptr %incdec.ptr13, align 2
223 %conv23 = zext i16 %9 to i32
224 %incdec.ptr24 = getelementptr inbounds i16, ptr %b, i64 4
225 %10 = load i16, ptr %incdec.ptr15, align 2
226 %conv25 = zext i16 %10 to i32
227 %sub26 = sub nsw i32 %conv23, %conv25
228 %arrayidx28 = getelementptr inbounds i16, ptr %g, i32 %sub26
229 %11 = load i16, ptr %arrayidx28, align 2
230 %conv29 = zext i16 %11 to i32
231 %add30 = add nsw i32 %add21, %conv29
232 %incdec.ptr31 = getelementptr inbounds i16, ptr %a.addr.0101, i64 5
233 %12 = load i16, ptr %incdec.ptr22, align 2
234 %conv32 = zext i16 %12 to i32
235 %incdec.ptr33 = getelementptr inbounds i16, ptr %b, i64 5
236 %13 = load i16, ptr %incdec.ptr24, align 2
237 %conv34 = zext i16 %13 to i32
238 %sub35 = sub nsw i32 %conv32, %conv34
239 %arrayidx37 = getelementptr inbounds i16, ptr %g, i32 %sub35
240 %14 = load i16, ptr %arrayidx37, align 2
241 %conv38 = zext i16 %14 to i32
242 %add39 = add nsw i32 %add30, %conv38
243 %incdec.ptr40 = getelementptr inbounds i16, ptr %a.addr.0101, i64 6
244 %15 = load i16, ptr %incdec.ptr31, align 2
245 %conv41 = zext i16 %15 to i32
246 %incdec.ptr42 = getelementptr inbounds i16, ptr %b, i64 6
247 %16 = load i16, ptr %incdec.ptr33, align 2
248 %conv43 = zext i16 %16 to i32
249 %sub44 = sub nsw i32 %conv41, %conv43
250 %arrayidx46 = getelementptr inbounds i16, ptr %g, i32 %sub44
251 %17 = load i16, ptr %arrayidx46, align 2
252 %conv47 = zext i16 %17 to i32
253 %add48 = add nsw i32 %add39, %conv47
254 %incdec.ptr49 = getelementptr inbounds i16, ptr %a.addr.0101, i64 7
255 %18 = load i16, ptr %incdec.ptr40, align 2
256 %conv50 = zext i16 %18 to i32
257 %incdec.ptr51 = getelementptr inbounds i16, ptr %b, i64 7
258 %19 = load i16, ptr %incdec.ptr42, align 2
259 %conv52 = zext i16 %19 to i32
260 %sub53 = sub nsw i32 %conv50, %conv52
261 %arrayidx55 = getelementptr inbounds i16, ptr %g, i32 %sub53
262 %20 = load i16, ptr %arrayidx55, align 2
263 %conv56 = zext i16 %20 to i32
264 %add57 = add nsw i32 %add48, %conv56
265 %incdec.ptr58 = getelementptr inbounds i16, ptr %a.addr.0101, i64 8
266 %21 = load i16, ptr %incdec.ptr49, align 2
267 %conv59 = zext i16 %21 to i32
268 %22 = load i16, ptr %incdec.ptr51, align 2
269 %conv61 = zext i16 %22 to i32
270 %sub62 = sub nsw i32 %conv59, %conv61
271 %arrayidx64 = getelementptr inbounds i16, ptr %g, i32 %sub62
272 %23 = load i16, ptr %arrayidx64, align 2
273 %conv65 = zext i16 %23 to i32
274 %add66 = add nsw i32 %add57, %conv65
275 %inc = add nuw nsw i32 %i.0103, 1
276 %exitcond = icmp eq i32 %inc, %n
277 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body
280 define i32 @gather_reduce_8x16_i64(ptr nocapture readonly %a, ptr nocapture readonly %b, ptr nocapture readonly %g, i32 %n) {
281 ; GENERIC-LABEL: @gather_reduce_8x16_i64(
282 ; GENERIC-NEXT: entry:
283 ; GENERIC-NEXT: [[CMP_99:%.*]] = icmp sgt i32 [[N:%.*]], 0
284 ; GENERIC-NEXT: br i1 [[CMP_99]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
285 ; GENERIC: for.body.preheader:
286 ; GENERIC-NEXT: br label [[FOR_BODY:%.*]]
287 ; GENERIC: for.cond.cleanup.loopexit:
288 ; GENERIC-NEXT: br label [[FOR_COND_CLEANUP]]
289 ; GENERIC: for.cond.cleanup:
290 ; GENERIC-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[ADD66:%.*]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ]
291 ; GENERIC-NEXT: ret i32 [[SUM_0_LCSSA]]
293 ; GENERIC-NEXT: [[I_0103:%.*]] = phi i32 [ [[INC:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
294 ; GENERIC-NEXT: [[SUM_0102:%.*]] = phi i32 [ [[ADD66]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
295 ; GENERIC-NEXT: [[A_ADDR_0101:%.*]] = phi ptr [ [[INCDEC_PTR58:%.*]], [[FOR_BODY]] ], [ [[A:%.*]], [[FOR_BODY_PREHEADER]] ]
296 ; GENERIC-NEXT: [[INCDEC_PTR58]] = getelementptr inbounds i8, ptr [[A_ADDR_0101]], i64 16
297 ; GENERIC-NEXT: [[TMP0:%.*]] = load <8 x i16>, ptr [[A_ADDR_0101]], align 2
298 ; GENERIC-NEXT: [[TMP1:%.*]] = zext <8 x i16> [[TMP0]] to <8 x i32>
299 ; GENERIC-NEXT: [[TMP2:%.*]] = load <8 x i16>, ptr [[B:%.*]], align 2
300 ; GENERIC-NEXT: [[TMP3:%.*]] = zext <8 x i16> [[TMP2]] to <8 x i32>
301 ; GENERIC-NEXT: [[TMP4:%.*]] = sub nsw <8 x i32> [[TMP1]], [[TMP3]]
302 ; GENERIC-NEXT: [[TMP5:%.*]] = extractelement <8 x i32> [[TMP4]], i64 0
303 ; GENERIC-NEXT: [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
304 ; GENERIC-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[G:%.*]], i64 [[TMP6]]
305 ; GENERIC-NEXT: [[TMP7:%.*]] = load i16, ptr [[ARRAYIDX]], align 2
306 ; GENERIC-NEXT: [[CONV3:%.*]] = zext i16 [[TMP7]] to i32
307 ; GENERIC-NEXT: [[ADD:%.*]] = add nsw i32 [[SUM_0102]], [[CONV3]]
308 ; GENERIC-NEXT: [[TMP8:%.*]] = extractelement <8 x i32> [[TMP4]], i64 1
309 ; GENERIC-NEXT: [[TMP9:%.*]] = sext i32 [[TMP8]] to i64
310 ; GENERIC-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP9]]
311 ; GENERIC-NEXT: [[TMP10:%.*]] = load i16, ptr [[ARRAYIDX10]], align 2
312 ; GENERIC-NEXT: [[CONV11:%.*]] = zext i16 [[TMP10]] to i32
313 ; GENERIC-NEXT: [[ADD12:%.*]] = add nsw i32 [[ADD]], [[CONV11]]
314 ; GENERIC-NEXT: [[TMP11:%.*]] = extractelement <8 x i32> [[TMP4]], i64 2
315 ; GENERIC-NEXT: [[TMP12:%.*]] = sext i32 [[TMP11]] to i64
316 ; GENERIC-NEXT: [[ARRAYIDX19:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP12]]
317 ; GENERIC-NEXT: [[TMP13:%.*]] = load i16, ptr [[ARRAYIDX19]], align 2
318 ; GENERIC-NEXT: [[CONV20:%.*]] = zext i16 [[TMP13]] to i32
319 ; GENERIC-NEXT: [[ADD21:%.*]] = add nsw i32 [[ADD12]], [[CONV20]]
320 ; GENERIC-NEXT: [[TMP14:%.*]] = extractelement <8 x i32> [[TMP4]], i64 3
321 ; GENERIC-NEXT: [[TMP15:%.*]] = sext i32 [[TMP14]] to i64
322 ; GENERIC-NEXT: [[ARRAYIDX28:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP15]]
323 ; GENERIC-NEXT: [[TMP16:%.*]] = load i16, ptr [[ARRAYIDX28]], align 2
324 ; GENERIC-NEXT: [[CONV29:%.*]] = zext i16 [[TMP16]] to i32
325 ; GENERIC-NEXT: [[ADD30:%.*]] = add nsw i32 [[ADD21]], [[CONV29]]
326 ; GENERIC-NEXT: [[TMP17:%.*]] = extractelement <8 x i32> [[TMP4]], i64 4
327 ; GENERIC-NEXT: [[TMP18:%.*]] = sext i32 [[TMP17]] to i64
328 ; GENERIC-NEXT: [[ARRAYIDX37:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP18]]
329 ; GENERIC-NEXT: [[TMP19:%.*]] = load i16, ptr [[ARRAYIDX37]], align 2
330 ; GENERIC-NEXT: [[CONV38:%.*]] = zext i16 [[TMP19]] to i32
331 ; GENERIC-NEXT: [[ADD39:%.*]] = add nsw i32 [[ADD30]], [[CONV38]]
332 ; GENERIC-NEXT: [[TMP20:%.*]] = extractelement <8 x i32> [[TMP4]], i64 5
333 ; GENERIC-NEXT: [[TMP21:%.*]] = sext i32 [[TMP20]] to i64
334 ; GENERIC-NEXT: [[ARRAYIDX46:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP21]]
335 ; GENERIC-NEXT: [[TMP22:%.*]] = load i16, ptr [[ARRAYIDX46]], align 2
336 ; GENERIC-NEXT: [[CONV47:%.*]] = zext i16 [[TMP22]] to i32
337 ; GENERIC-NEXT: [[ADD48:%.*]] = add nsw i32 [[ADD39]], [[CONV47]]
338 ; GENERIC-NEXT: [[TMP23:%.*]] = extractelement <8 x i32> [[TMP4]], i64 6
339 ; GENERIC-NEXT: [[TMP24:%.*]] = sext i32 [[TMP23]] to i64
340 ; GENERIC-NEXT: [[ARRAYIDX55:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP24]]
341 ; GENERIC-NEXT: [[TMP25:%.*]] = load i16, ptr [[ARRAYIDX55]], align 2
342 ; GENERIC-NEXT: [[CONV56:%.*]] = zext i16 [[TMP25]] to i32
343 ; GENERIC-NEXT: [[ADD57:%.*]] = add nsw i32 [[ADD48]], [[CONV56]]
344 ; GENERIC-NEXT: [[TMP26:%.*]] = extractelement <8 x i32> [[TMP4]], i64 7
345 ; GENERIC-NEXT: [[TMP27:%.*]] = sext i32 [[TMP26]] to i64
346 ; GENERIC-NEXT: [[ARRAYIDX64:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP27]]
347 ; GENERIC-NEXT: [[TMP28:%.*]] = load i16, ptr [[ARRAYIDX64]], align 2
348 ; GENERIC-NEXT: [[CONV65:%.*]] = zext i16 [[TMP28]] to i32
349 ; GENERIC-NEXT: [[ADD66]] = add nsw i32 [[ADD57]], [[CONV65]]
350 ; GENERIC-NEXT: [[INC]] = add nuw nsw i32 [[I_0103]], 1
351 ; GENERIC-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[INC]], [[N]]
352 ; GENERIC-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]]
354 ; KRYO-LABEL: @gather_reduce_8x16_i64(
356 ; KRYO-NEXT: [[CMP_99:%.*]] = icmp sgt i32 [[N:%.*]], 0
357 ; KRYO-NEXT: br i1 [[CMP_99]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
358 ; KRYO: for.body.preheader:
359 ; KRYO-NEXT: br label [[FOR_BODY:%.*]]
360 ; KRYO: for.cond.cleanup.loopexit:
361 ; KRYO-NEXT: br label [[FOR_COND_CLEANUP]]
362 ; KRYO: for.cond.cleanup:
363 ; KRYO-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[ADD66:%.*]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ]
364 ; KRYO-NEXT: ret i32 [[SUM_0_LCSSA]]
366 ; KRYO-NEXT: [[I_0103:%.*]] = phi i32 [ [[INC:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
367 ; KRYO-NEXT: [[SUM_0102:%.*]] = phi i32 [ [[ADD66]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
368 ; KRYO-NEXT: [[A_ADDR_0101:%.*]] = phi ptr [ [[INCDEC_PTR58:%.*]], [[FOR_BODY]] ], [ [[A:%.*]], [[FOR_BODY_PREHEADER]] ]
369 ; KRYO-NEXT: [[INCDEC_PTR58]] = getelementptr inbounds i8, ptr [[A_ADDR_0101]], i64 16
370 ; KRYO-NEXT: [[TMP0:%.*]] = load <8 x i16>, ptr [[A_ADDR_0101]], align 2
371 ; KRYO-NEXT: [[TMP1:%.*]] = zext <8 x i16> [[TMP0]] to <8 x i32>
372 ; KRYO-NEXT: [[TMP2:%.*]] = load <8 x i16>, ptr [[B:%.*]], align 2
373 ; KRYO-NEXT: [[TMP3:%.*]] = zext <8 x i16> [[TMP2]] to <8 x i32>
374 ; KRYO-NEXT: [[TMP4:%.*]] = sub nsw <8 x i32> [[TMP1]], [[TMP3]]
375 ; KRYO-NEXT: [[TMP5:%.*]] = extractelement <8 x i32> [[TMP4]], i64 0
376 ; KRYO-NEXT: [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
377 ; KRYO-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[G:%.*]], i64 [[TMP6]]
378 ; KRYO-NEXT: [[TMP7:%.*]] = load i16, ptr [[ARRAYIDX]], align 2
379 ; KRYO-NEXT: [[CONV3:%.*]] = zext i16 [[TMP7]] to i32
380 ; KRYO-NEXT: [[ADD:%.*]] = add nsw i32 [[SUM_0102]], [[CONV3]]
381 ; KRYO-NEXT: [[TMP8:%.*]] = extractelement <8 x i32> [[TMP4]], i64 1
382 ; KRYO-NEXT: [[TMP9:%.*]] = sext i32 [[TMP8]] to i64
383 ; KRYO-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP9]]
384 ; KRYO-NEXT: [[TMP10:%.*]] = load i16, ptr [[ARRAYIDX10]], align 2
385 ; KRYO-NEXT: [[CONV11:%.*]] = zext i16 [[TMP10]] to i32
386 ; KRYO-NEXT: [[ADD12:%.*]] = add nsw i32 [[ADD]], [[CONV11]]
387 ; KRYO-NEXT: [[TMP11:%.*]] = extractelement <8 x i32> [[TMP4]], i64 2
388 ; KRYO-NEXT: [[TMP12:%.*]] = sext i32 [[TMP11]] to i64
389 ; KRYO-NEXT: [[ARRAYIDX19:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP12]]
390 ; KRYO-NEXT: [[TMP13:%.*]] = load i16, ptr [[ARRAYIDX19]], align 2
391 ; KRYO-NEXT: [[CONV20:%.*]] = zext i16 [[TMP13]] to i32
392 ; KRYO-NEXT: [[ADD21:%.*]] = add nsw i32 [[ADD12]], [[CONV20]]
393 ; KRYO-NEXT: [[TMP14:%.*]] = extractelement <8 x i32> [[TMP4]], i64 3
394 ; KRYO-NEXT: [[TMP15:%.*]] = sext i32 [[TMP14]] to i64
395 ; KRYO-NEXT: [[ARRAYIDX28:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP15]]
396 ; KRYO-NEXT: [[TMP16:%.*]] = load i16, ptr [[ARRAYIDX28]], align 2
397 ; KRYO-NEXT: [[CONV29:%.*]] = zext i16 [[TMP16]] to i32
398 ; KRYO-NEXT: [[ADD30:%.*]] = add nsw i32 [[ADD21]], [[CONV29]]
399 ; KRYO-NEXT: [[TMP17:%.*]] = extractelement <8 x i32> [[TMP4]], i64 4
400 ; KRYO-NEXT: [[TMP18:%.*]] = sext i32 [[TMP17]] to i64
401 ; KRYO-NEXT: [[ARRAYIDX37:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP18]]
402 ; KRYO-NEXT: [[TMP19:%.*]] = load i16, ptr [[ARRAYIDX37]], align 2
403 ; KRYO-NEXT: [[CONV38:%.*]] = zext i16 [[TMP19]] to i32
404 ; KRYO-NEXT: [[ADD39:%.*]] = add nsw i32 [[ADD30]], [[CONV38]]
405 ; KRYO-NEXT: [[TMP20:%.*]] = extractelement <8 x i32> [[TMP4]], i64 5
406 ; KRYO-NEXT: [[TMP21:%.*]] = sext i32 [[TMP20]] to i64
407 ; KRYO-NEXT: [[ARRAYIDX46:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP21]]
408 ; KRYO-NEXT: [[TMP22:%.*]] = load i16, ptr [[ARRAYIDX46]], align 2
409 ; KRYO-NEXT: [[CONV47:%.*]] = zext i16 [[TMP22]] to i32
410 ; KRYO-NEXT: [[ADD48:%.*]] = add nsw i32 [[ADD39]], [[CONV47]]
411 ; KRYO-NEXT: [[TMP23:%.*]] = extractelement <8 x i32> [[TMP4]], i64 6
412 ; KRYO-NEXT: [[TMP24:%.*]] = sext i32 [[TMP23]] to i64
413 ; KRYO-NEXT: [[ARRAYIDX55:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP24]]
414 ; KRYO-NEXT: [[TMP25:%.*]] = load i16, ptr [[ARRAYIDX55]], align 2
415 ; KRYO-NEXT: [[CONV56:%.*]] = zext i16 [[TMP25]] to i32
416 ; KRYO-NEXT: [[ADD57:%.*]] = add nsw i32 [[ADD48]], [[CONV56]]
417 ; KRYO-NEXT: [[TMP26:%.*]] = extractelement <8 x i32> [[TMP4]], i64 7
418 ; KRYO-NEXT: [[TMP27:%.*]] = sext i32 [[TMP26]] to i64
419 ; KRYO-NEXT: [[ARRAYIDX64:%.*]] = getelementptr inbounds i16, ptr [[G]], i64 [[TMP27]]
420 ; KRYO-NEXT: [[TMP28:%.*]] = load i16, ptr [[ARRAYIDX64]], align 2
421 ; KRYO-NEXT: [[CONV65:%.*]] = zext i16 [[TMP28]] to i32
422 ; KRYO-NEXT: [[ADD66]] = add nsw i32 [[ADD57]], [[CONV65]]
423 ; KRYO-NEXT: [[INC]] = add nuw nsw i32 [[I_0103]], 1
424 ; KRYO-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[INC]], [[N]]
425 ; KRYO-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]]
428 %cmp.99 = icmp sgt i32 %n, 0
429 br i1 %cmp.99, label %for.body.preheader, label %for.cond.cleanup
434 for.cond.cleanup.loopexit:
435 br label %for.cond.cleanup
438 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add66, %for.cond.cleanup.loopexit ]
442 %i.0103 = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]
443 %sum.0102 = phi i32 [ %add66, %for.body ], [ 0, %for.body.preheader ]
444 %a.addr.0101 = phi ptr [ %incdec.ptr58, %for.body ], [ %a, %for.body.preheader ]
445 %incdec.ptr = getelementptr inbounds i16, ptr %a.addr.0101, i64 1
446 %0 = load i16, ptr %a.addr.0101, align 2
447 %conv = zext i16 %0 to i64
448 %incdec.ptr1 = getelementptr inbounds i16, ptr %b, i64 1
449 %1 = load i16, ptr %b, align 2
450 %conv2 = zext i16 %1 to i64
451 %sub = sub nsw i64 %conv, %conv2
452 %arrayidx = getelementptr inbounds i16, ptr %g, i64 %sub
453 %2 = load i16, ptr %arrayidx, align 2
454 %conv3 = zext i16 %2 to i32
455 %add = add nsw i32 %conv3, %sum.0102
456 %incdec.ptr4 = getelementptr inbounds i16, ptr %a.addr.0101, i64 2
457 %3 = load i16, ptr %incdec.ptr, align 2
458 %conv5 = zext i16 %3 to i64
459 %incdec.ptr6 = getelementptr inbounds i16, ptr %b, i64 2
460 %4 = load i16, ptr %incdec.ptr1, align 2
461 %conv7 = zext i16 %4 to i64
462 %sub8 = sub nsw i64 %conv5, %conv7
463 %arrayidx10 = getelementptr inbounds i16, ptr %g, i64 %sub8
464 %5 = load i16, ptr %arrayidx10, align 2
465 %conv11 = zext i16 %5 to i32
466 %add12 = add nsw i32 %add, %conv11
467 %incdec.ptr13 = getelementptr inbounds i16, ptr %a.addr.0101, i64 3
468 %6 = load i16, ptr %incdec.ptr4, align 2
469 %conv14 = zext i16 %6 to i64
470 %incdec.ptr15 = getelementptr inbounds i16, ptr %b, i64 3
471 %7 = load i16, ptr %incdec.ptr6, align 2
472 %conv16 = zext i16 %7 to i64
473 %sub17 = sub nsw i64 %conv14, %conv16
474 %arrayidx19 = getelementptr inbounds i16, ptr %g, i64 %sub17
475 %8 = load i16, ptr %arrayidx19, align 2
476 %conv20 = zext i16 %8 to i32
477 %add21 = add nsw i32 %add12, %conv20
478 %incdec.ptr22 = getelementptr inbounds i16, ptr %a.addr.0101, i64 4
479 %9 = load i16, ptr %incdec.ptr13, align 2
480 %conv23 = zext i16 %9 to i64
481 %incdec.ptr24 = getelementptr inbounds i16, ptr %b, i64 4
482 %10 = load i16, ptr %incdec.ptr15, align 2
483 %conv25 = zext i16 %10 to i64
484 %sub26 = sub nsw i64 %conv23, %conv25
485 %arrayidx28 = getelementptr inbounds i16, ptr %g, i64 %sub26
486 %11 = load i16, ptr %arrayidx28, align 2
487 %conv29 = zext i16 %11 to i32
488 %add30 = add nsw i32 %add21, %conv29
489 %incdec.ptr31 = getelementptr inbounds i16, ptr %a.addr.0101, i64 5
490 %12 = load i16, ptr %incdec.ptr22, align 2
491 %conv32 = zext i16 %12 to i64
492 %incdec.ptr33 = getelementptr inbounds i16, ptr %b, i64 5
493 %13 = load i16, ptr %incdec.ptr24, align 2
494 %conv34 = zext i16 %13 to i64
495 %sub35 = sub nsw i64 %conv32, %conv34
496 %arrayidx37 = getelementptr inbounds i16, ptr %g, i64 %sub35
497 %14 = load i16, ptr %arrayidx37, align 2
498 %conv38 = zext i16 %14 to i32
499 %add39 = add nsw i32 %add30, %conv38
500 %incdec.ptr40 = getelementptr inbounds i16, ptr %a.addr.0101, i64 6
501 %15 = load i16, ptr %incdec.ptr31, align 2
502 %conv41 = zext i16 %15 to i64
503 %incdec.ptr42 = getelementptr inbounds i16, ptr %b, i64 6
504 %16 = load i16, ptr %incdec.ptr33, align 2
505 %conv43 = zext i16 %16 to i64
506 %sub44 = sub nsw i64 %conv41, %conv43
507 %arrayidx46 = getelementptr inbounds i16, ptr %g, i64 %sub44
508 %17 = load i16, ptr %arrayidx46, align 2
509 %conv47 = zext i16 %17 to i32
510 %add48 = add nsw i32 %add39, %conv47
511 %incdec.ptr49 = getelementptr inbounds i16, ptr %a.addr.0101, i64 7
512 %18 = load i16, ptr %incdec.ptr40, align 2
513 %conv50 = zext i16 %18 to i64
514 %incdec.ptr51 = getelementptr inbounds i16, ptr %b, i64 7
515 %19 = load i16, ptr %incdec.ptr42, align 2
516 %conv52 = zext i16 %19 to i64
517 %sub53 = sub nsw i64 %conv50, %conv52
518 %arrayidx55 = getelementptr inbounds i16, ptr %g, i64 %sub53
519 %20 = load i16, ptr %arrayidx55, align 2
520 %conv56 = zext i16 %20 to i32
521 %add57 = add nsw i32 %add48, %conv56
522 %incdec.ptr58 = getelementptr inbounds i16, ptr %a.addr.0101, i64 8
523 %21 = load i16, ptr %incdec.ptr49, align 2
524 %conv59 = zext i16 %21 to i64
525 %22 = load i16, ptr %incdec.ptr51, align 2
526 %conv61 = zext i16 %22 to i64
527 %sub62 = sub nsw i64 %conv59, %conv61
528 %arrayidx64 = getelementptr inbounds i16, ptr %g, i64 %sub62
529 %23 = load i16, ptr %arrayidx64, align 2
530 %conv65 = zext i16 %23 to i32
531 %add66 = add nsw i32 %add57, %conv65
532 %inc = add nuw nsw i32 %i.0103, 1
533 %exitcond = icmp eq i32 %inc, %n
534 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body