1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -mattr=avx -force-vector-width=2 -force-vector-interleave=1 -loop-vectorize -simplifycfg -simplifycfg-require-and-preserve-domtree=1 -S | FileCheck %s
3; RUN: opt -mcpu=skylake-avx512 -S -force-vector-width=8 -force-vector-interleave=1 -loop-vectorize < %s | FileCheck %s --check-prefix=SINK-GATHER
4
5target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
6target triple = "x86_64-apple-macosx10.8.0"
7
8; This test ensures that we don't scalarize the predicated load. Since the load
9; can be vectorized with predication, scalarizing it would cause its pointer
10; operand to become non-uniform.
11;
12define i32 @predicated_sdiv_masked_load(i32* %a, i32* %b, i32 %x, i1 %c) {
13; CHECK-LABEL: @predicated_sdiv_masked_load(
14; CHECK-NEXT:  entry:
15; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i1> poison, i1 [[C:%.*]], i32 0
16; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i1> [[BROADCAST_SPLATINSERT]], <2 x i1> poison, <2 x i32> zeroinitializer
17; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
18; CHECK:       vector.body:
19; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[PRED_SDIV_CONTINUE2:%.*]] ]
20; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ zeroinitializer, [[ENTRY]] ], [ [[TMP19:%.*]], [[PRED_SDIV_CONTINUE2]] ]
21; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
22; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP0]]
23; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TMP1]], i32 0
24; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <2 x i32>*
25; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i32>, <2 x i32>* [[TMP3]], align 4
26; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr i32, i32* [[B:%.*]], i64 [[TMP0]]
27; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr i32, i32* [[TMP4]], i32 0
28; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i32* [[TMP5]] to <2 x i32>*
29; CHECK-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <2 x i32> @llvm.masked.load.v2i32.p0v2i32(<2 x i32>* [[TMP6]], i32 4, <2 x i1> [[BROADCAST_SPLAT]], <2 x i32> poison)
30; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 0
31; CHECK-NEXT:    br i1 [[TMP7]], label [[PRED_SDIV_IF:%.*]], label [[PRED_SDIV_CONTINUE:%.*]]
32; CHECK:       pred.sdiv.if:
33; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <2 x i32> [[WIDE_MASKED_LOAD]], i32 0
34; CHECK-NEXT:    [[TMP9:%.*]] = sdiv i32 [[TMP8]], [[X:%.*]]
35; CHECK-NEXT:    [[TMP10:%.*]] = insertelement <2 x i32> poison, i32 [[TMP9]], i32 0
36; CHECK-NEXT:    br label [[PRED_SDIV_CONTINUE]]
37; CHECK:       pred.sdiv.continue:
38; CHECK-NEXT:    [[TMP11:%.*]] = phi <2 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP10]], [[PRED_SDIV_IF]] ]
39; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 1
40; CHECK-NEXT:    br i1 [[TMP12]], label [[PRED_SDIV_IF1:%.*]], label [[PRED_SDIV_CONTINUE2]]
41; CHECK:       pred.sdiv.if1:
42; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x i32> [[WIDE_MASKED_LOAD]], i32 1
43; CHECK-NEXT:    [[TMP14:%.*]] = sdiv i32 [[TMP13]], [[X]]
44; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <2 x i32> [[TMP11]], i32 [[TMP14]], i32 1
45; CHECK-NEXT:    br label [[PRED_SDIV_CONTINUE2]]
46; CHECK:       pred.sdiv.continue2:
47; CHECK-NEXT:    [[TMP16:%.*]] = phi <2 x i32> [ [[TMP11]], [[PRED_SDIV_CONTINUE]] ], [ [[TMP15]], [[PRED_SDIV_IF1]] ]
48; CHECK-NEXT:    [[TMP17:%.*]] = add nsw <2 x i32> [[TMP16]], [[WIDE_LOAD]]
49; CHECK-NEXT:    [[TMP18:%.*]] = xor <2 x i1> [[BROADCAST_SPLAT]], <i1 true, i1 true>
50; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[BROADCAST_SPLAT]], <2 x i32> [[TMP17]], <2 x i32> [[WIDE_LOAD]]
51; CHECK-NEXT:    [[TMP19]] = add <2 x i32> [[VEC_PHI]], [[PREDPHI]]
52; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
53; CHECK-NEXT:    [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10000
54; CHECK-NEXT:    br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
55; CHECK:       middle.block:
56; CHECK-NEXT:    [[TMP21:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[TMP19]])
57; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 10000, 10000
58; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[FOR_BODY:%.*]]
59; CHECK:       for.body:
60; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_INC:%.*]] ], [ 10000, [[MIDDLE_BLOCK]] ]
61; CHECK-NEXT:    [[R:%.*]] = phi i32 [ [[T7:%.*]], [[FOR_INC]] ], [ [[TMP21]], [[MIDDLE_BLOCK]] ]
62; CHECK-NEXT:    [[T0:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[I]]
63; CHECK-NEXT:    [[T1:%.*]] = load i32, i32* [[T0]], align 4
64; CHECK-NEXT:    br i1 [[C]], label [[IF_THEN:%.*]], label [[FOR_INC]]
65; CHECK:       if.then:
66; CHECK-NEXT:    [[T2:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[I]]
67; CHECK-NEXT:    [[T3:%.*]] = load i32, i32* [[T2]], align 4
68; CHECK-NEXT:    [[T4:%.*]] = sdiv i32 [[T3]], [[X]]
69; CHECK-NEXT:    [[T5:%.*]] = add nsw i32 [[T4]], [[T1]]
70; CHECK-NEXT:    br label [[FOR_INC]]
71; CHECK:       for.inc:
72; CHECK-NEXT:    [[T6:%.*]] = phi i32 [ [[T1]], [[FOR_BODY]] ], [ [[T5]], [[IF_THEN]] ]
73; CHECK-NEXT:    [[T7]] = add i32 [[R]], [[T6]]
74; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
75; CHECK-NEXT:    [[COND:%.*]] = icmp eq i64 [[I_NEXT]], 10000
76; CHECK-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]]
77; CHECK:       for.end:
78; CHECK-NEXT:    [[T8:%.*]] = phi i32 [ [[T7]], [[FOR_INC]] ], [ [[TMP21]], [[MIDDLE_BLOCK]] ]
79; CHECK-NEXT:    ret i32 [[T8]]
80;
81; SINK-GATHER-LABEL: @predicated_sdiv_masked_load(
82; SINK-GATHER-NEXT:  entry:
83; SINK-GATHER-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
84; SINK-GATHER:       vector.ph:
85; SINK-GATHER-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i1> poison, i1 [[C:%.*]], i32 0
86; SINK-GATHER-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i1> [[BROADCAST_SPLATINSERT]], <8 x i1> poison, <8 x i32> zeroinitializer
87; SINK-GATHER-NEXT:    br label [[VECTOR_BODY:%.*]]
88; SINK-GATHER:       vector.body:
89; SINK-GATHER-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_SDIV_CONTINUE14:%.*]] ]
90; SINK-GATHER-NEXT:    [[VEC_PHI:%.*]] = phi <8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP49:%.*]], [[PRED_SDIV_CONTINUE14]] ]
91; SINK-GATHER-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
92; SINK-GATHER-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP0]]
93; SINK-GATHER-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, i32* [[TMP1]], i32 0
94; SINK-GATHER-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <8 x i32>*
95; SINK-GATHER-NEXT:    [[WIDE_LOAD:%.*]] = load <8 x i32>, <8 x i32>* [[TMP3]], align 4
96; SINK-GATHER-NEXT:    [[TMP4:%.*]] = getelementptr i32, i32* [[B:%.*]], i64 [[TMP0]]
97; SINK-GATHER-NEXT:    [[TMP5:%.*]] = getelementptr i32, i32* [[TMP4]], i32 0
98; SINK-GATHER-NEXT:    [[TMP6:%.*]] = bitcast i32* [[TMP5]] to <8 x i32>*
99; SINK-GATHER-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <8 x i32> @llvm.masked.load.v8i32.p0v8i32(<8 x i32>* [[TMP6]], i32 4, <8 x i1> [[BROADCAST_SPLAT]], <8 x i32> poison)
100; SINK-GATHER-NEXT:    [[TMP7:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 0
101; SINK-GATHER-NEXT:    br i1 [[TMP7]], label [[PRED_SDIV_IF:%.*]], label [[PRED_SDIV_CONTINUE:%.*]]
102; SINK-GATHER:       pred.sdiv.if:
103; SINK-GATHER-NEXT:    [[TMP8:%.*]] = extractelement <8 x i32> [[WIDE_MASKED_LOAD]], i32 0
104; SINK-GATHER-NEXT:    [[TMP9:%.*]] = sdiv i32 [[TMP8]], [[X:%.*]]
105; SINK-GATHER-NEXT:    [[TMP10:%.*]] = insertelement <8 x i32> poison, i32 [[TMP9]], i32 0
106; SINK-GATHER-NEXT:    br label [[PRED_SDIV_CONTINUE]]
107; SINK-GATHER:       pred.sdiv.continue:
108; SINK-GATHER-NEXT:    [[TMP11:%.*]] = phi <8 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP10]], [[PRED_SDIV_IF]] ]
109; SINK-GATHER-NEXT:    [[TMP12:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 1
110; SINK-GATHER-NEXT:    br i1 [[TMP12]], label [[PRED_SDIV_IF1:%.*]], label [[PRED_SDIV_CONTINUE2:%.*]]
111; SINK-GATHER:       pred.sdiv.if1:
112; SINK-GATHER-NEXT:    [[TMP13:%.*]] = extractelement <8 x i32> [[WIDE_MASKED_LOAD]], i32 1
113; SINK-GATHER-NEXT:    [[TMP14:%.*]] = sdiv i32 [[TMP13]], [[X]]
114; SINK-GATHER-NEXT:    [[TMP15:%.*]] = insertelement <8 x i32> [[TMP11]], i32 [[TMP14]], i32 1
115; SINK-GATHER-NEXT:    br label [[PRED_SDIV_CONTINUE2]]
116; SINK-GATHER:       pred.sdiv.continue2:
117; SINK-GATHER-NEXT:    [[TMP16:%.*]] = phi <8 x i32> [ [[TMP11]], [[PRED_SDIV_CONTINUE]] ], [ [[TMP15]], [[PRED_SDIV_IF1]] ]
118; SINK-GATHER-NEXT:    [[TMP17:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 2
119; SINK-GATHER-NEXT:    br i1 [[TMP17]], label [[PRED_SDIV_IF3:%.*]], label [[PRED_SDIV_CONTINUE4:%.*]]
120; SINK-GATHER:       pred.sdiv.if3:
121; SINK-GATHER-NEXT:    [[TMP18:%.*]] = extractelement <8 x i32> [[WIDE_MASKED_LOAD]], i32 2
122; SINK-GATHER-NEXT:    [[TMP19:%.*]] = sdiv i32 [[TMP18]], [[X]]
123; SINK-GATHER-NEXT:    [[TMP20:%.*]] = insertelement <8 x i32> [[TMP16]], i32 [[TMP19]], i32 2
124; SINK-GATHER-NEXT:    br label [[PRED_SDIV_CONTINUE4]]
125; SINK-GATHER:       pred.sdiv.continue4:
126; SINK-GATHER-NEXT:    [[TMP21:%.*]] = phi <8 x i32> [ [[TMP16]], [[PRED_SDIV_CONTINUE2]] ], [ [[TMP20]], [[PRED_SDIV_IF3]] ]
127; SINK-GATHER-NEXT:    [[TMP22:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 3
128; SINK-GATHER-NEXT:    br i1 [[TMP22]], label [[PRED_SDIV_IF5:%.*]], label [[PRED_SDIV_CONTINUE6:%.*]]
129; SINK-GATHER:       pred.sdiv.if5:
130; SINK-GATHER-NEXT:    [[TMP23:%.*]] = extractelement <8 x i32> [[WIDE_MASKED_LOAD]], i32 3
131; SINK-GATHER-NEXT:    [[TMP24:%.*]] = sdiv i32 [[TMP23]], [[X]]
132; SINK-GATHER-NEXT:    [[TMP25:%.*]] = insertelement <8 x i32> [[TMP21]], i32 [[TMP24]], i32 3
133; SINK-GATHER-NEXT:    br label [[PRED_SDIV_CONTINUE6]]
134; SINK-GATHER:       pred.sdiv.continue6:
135; SINK-GATHER-NEXT:    [[TMP26:%.*]] = phi <8 x i32> [ [[TMP21]], [[PRED_SDIV_CONTINUE4]] ], [ [[TMP25]], [[PRED_SDIV_IF5]] ]
136; SINK-GATHER-NEXT:    [[TMP27:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 4
137; SINK-GATHER-NEXT:    br i1 [[TMP27]], label [[PRED_SDIV_IF7:%.*]], label [[PRED_SDIV_CONTINUE8:%.*]]
138; SINK-GATHER:       pred.sdiv.if7:
139; SINK-GATHER-NEXT:    [[TMP28:%.*]] = extractelement <8 x i32> [[WIDE_MASKED_LOAD]], i32 4
140; SINK-GATHER-NEXT:    [[TMP29:%.*]] = sdiv i32 [[TMP28]], [[X]]
141; SINK-GATHER-NEXT:    [[TMP30:%.*]] = insertelement <8 x i32> [[TMP26]], i32 [[TMP29]], i32 4
142; SINK-GATHER-NEXT:    br label [[PRED_SDIV_CONTINUE8]]
143; SINK-GATHER:       pred.sdiv.continue8:
144; SINK-GATHER-NEXT:    [[TMP31:%.*]] = phi <8 x i32> [ [[TMP26]], [[PRED_SDIV_CONTINUE6]] ], [ [[TMP30]], [[PRED_SDIV_IF7]] ]
145; SINK-GATHER-NEXT:    [[TMP32:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 5
146; SINK-GATHER-NEXT:    br i1 [[TMP32]], label [[PRED_SDIV_IF9:%.*]], label [[PRED_SDIV_CONTINUE10:%.*]]
147; SINK-GATHER:       pred.sdiv.if9:
148; SINK-GATHER-NEXT:    [[TMP33:%.*]] = extractelement <8 x i32> [[WIDE_MASKED_LOAD]], i32 5
149; SINK-GATHER-NEXT:    [[TMP34:%.*]] = sdiv i32 [[TMP33]], [[X]]
150; SINK-GATHER-NEXT:    [[TMP35:%.*]] = insertelement <8 x i32> [[TMP31]], i32 [[TMP34]], i32 5
151; SINK-GATHER-NEXT:    br label [[PRED_SDIV_CONTINUE10]]
152; SINK-GATHER:       pred.sdiv.continue10:
153; SINK-GATHER-NEXT:    [[TMP36:%.*]] = phi <8 x i32> [ [[TMP31]], [[PRED_SDIV_CONTINUE8]] ], [ [[TMP35]], [[PRED_SDIV_IF9]] ]
154; SINK-GATHER-NEXT:    [[TMP37:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 6
155; SINK-GATHER-NEXT:    br i1 [[TMP37]], label [[PRED_SDIV_IF11:%.*]], label [[PRED_SDIV_CONTINUE12:%.*]]
156; SINK-GATHER:       pred.sdiv.if11:
157; SINK-GATHER-NEXT:    [[TMP38:%.*]] = extractelement <8 x i32> [[WIDE_MASKED_LOAD]], i32 6
158; SINK-GATHER-NEXT:    [[TMP39:%.*]] = sdiv i32 [[TMP38]], [[X]]
159; SINK-GATHER-NEXT:    [[TMP40:%.*]] = insertelement <8 x i32> [[TMP36]], i32 [[TMP39]], i32 6
160; SINK-GATHER-NEXT:    br label [[PRED_SDIV_CONTINUE12]]
161; SINK-GATHER:       pred.sdiv.continue12:
162; SINK-GATHER-NEXT:    [[TMP41:%.*]] = phi <8 x i32> [ [[TMP36]], [[PRED_SDIV_CONTINUE10]] ], [ [[TMP40]], [[PRED_SDIV_IF11]] ]
163; SINK-GATHER-NEXT:    [[TMP42:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 7
164; SINK-GATHER-NEXT:    br i1 [[TMP42]], label [[PRED_SDIV_IF13:%.*]], label [[PRED_SDIV_CONTINUE14]]
165; SINK-GATHER:       pred.sdiv.if13:
166; SINK-GATHER-NEXT:    [[TMP43:%.*]] = extractelement <8 x i32> [[WIDE_MASKED_LOAD]], i32 7
167; SINK-GATHER-NEXT:    [[TMP44:%.*]] = sdiv i32 [[TMP43]], [[X]]
168; SINK-GATHER-NEXT:    [[TMP45:%.*]] = insertelement <8 x i32> [[TMP41]], i32 [[TMP44]], i32 7
169; SINK-GATHER-NEXT:    br label [[PRED_SDIV_CONTINUE14]]
170; SINK-GATHER:       pred.sdiv.continue14:
171; SINK-GATHER-NEXT:    [[TMP46:%.*]] = phi <8 x i32> [ [[TMP41]], [[PRED_SDIV_CONTINUE12]] ], [ [[TMP45]], [[PRED_SDIV_IF13]] ]
172; SINK-GATHER-NEXT:    [[TMP47:%.*]] = add nsw <8 x i32> [[TMP46]], [[WIDE_LOAD]]
173; SINK-GATHER-NEXT:    [[TMP48:%.*]] = xor <8 x i1> [[BROADCAST_SPLAT]], <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>
174; SINK-GATHER-NEXT:    [[PREDPHI:%.*]] = select <8 x i1> [[BROADCAST_SPLAT]], <8 x i32> [[TMP47]], <8 x i32> [[WIDE_LOAD]]
175; SINK-GATHER-NEXT:    [[TMP49]] = add <8 x i32> [[VEC_PHI]], [[PREDPHI]]
176; SINK-GATHER-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
177; SINK-GATHER-NEXT:    [[TMP50:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10000
178; SINK-GATHER-NEXT:    br i1 [[TMP50]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
179; SINK-GATHER:       middle.block:
180; SINK-GATHER-NEXT:    [[TMP51:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[TMP49]])
181; SINK-GATHER-NEXT:    [[CMP_N:%.*]] = icmp eq i64 10000, 10000
182; SINK-GATHER-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
183; SINK-GATHER:       scalar.ph:
184; SINK-GATHER-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 10000, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
185; SINK-GATHER-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[TMP51]], [[MIDDLE_BLOCK]] ]
186; SINK-GATHER-NEXT:    br label [[FOR_BODY:%.*]]
187; SINK-GATHER:       for.body:
188; SINK-GATHER-NEXT:    [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_INC:%.*]] ]
189; SINK-GATHER-NEXT:    [[R:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[T7:%.*]], [[FOR_INC]] ]
190; SINK-GATHER-NEXT:    [[T0:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[I]]
191; SINK-GATHER-NEXT:    [[T1:%.*]] = load i32, i32* [[T0]], align 4
192; SINK-GATHER-NEXT:    br i1 [[C]], label [[IF_THEN:%.*]], label [[FOR_INC]]
193; SINK-GATHER:       if.then:
194; SINK-GATHER-NEXT:    [[T2:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[I]]
195; SINK-GATHER-NEXT:    [[T3:%.*]] = load i32, i32* [[T2]], align 4
196; SINK-GATHER-NEXT:    [[T4:%.*]] = sdiv i32 [[T3]], [[X]]
197; SINK-GATHER-NEXT:    [[T5:%.*]] = add nsw i32 [[T4]], [[T1]]
198; SINK-GATHER-NEXT:    br label [[FOR_INC]]
199; SINK-GATHER:       for.inc:
200; SINK-GATHER-NEXT:    [[T6:%.*]] = phi i32 [ [[T1]], [[FOR_BODY]] ], [ [[T5]], [[IF_THEN]] ]
201; SINK-GATHER-NEXT:    [[T7]] = add i32 [[R]], [[T6]]
202; SINK-GATHER-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
203; SINK-GATHER-NEXT:    [[COND:%.*]] = icmp eq i64 [[I_NEXT]], 10000
204; SINK-GATHER-NEXT:    br i1 [[COND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]]
205; SINK-GATHER:       for.end:
206; SINK-GATHER-NEXT:    [[T8:%.*]] = phi i32 [ [[T7]], [[FOR_INC]] ], [ [[TMP51]], [[MIDDLE_BLOCK]] ]
207; SINK-GATHER-NEXT:    ret i32 [[T8]]
208;
209entry:
210  br label %for.body
211
212for.body:
213  %i = phi i64 [ 0, %entry ], [ %i.next, %for.inc ]
214  %r = phi i32 [ 0, %entry ], [ %t7, %for.inc ]
215  %t0 = getelementptr inbounds i32, i32* %a, i64 %i
216  %t1 = load i32, i32* %t0, align 4
217  br i1 %c, label %if.then, label %for.inc
218
219if.then:
220  %t2 = getelementptr inbounds i32, i32* %b, i64 %i
221  %t3 = load i32, i32* %t2, align 4
222  %t4 = sdiv i32 %t3, %x
223  %t5 = add nsw i32 %t4, %t1
224  br label %for.inc
225
226for.inc:
227  %t6 = phi i32 [ %t1, %for.body ], [ %t5, %if.then]
228  %t7 = add i32 %r, %t6
229  %i.next = add nuw nsw i64 %i, 1
230  %cond = icmp eq i64 %i.next, 10000
231  br i1 %cond, label %for.end, label %for.body
232
233for.end:
234  %t8 = phi i32 [ %t7, %for.inc ]
235  ret i32 %t8
236}
237
238; This test ensures that a load, which would have been widened otherwise is
239; instead scalarized if Cost-Model so decided as part of its
240; sink-scalar-operands optimization for predicated instructions.
241define i32 @scalarize_and_sink_gather(i32* %a, i1 %c, i32 %x, i64 %n) {
242; CHECK-LABEL: @scalarize_and_sink_gather(
243; CHECK-NEXT:  entry:
244; CHECK-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
245; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 2
246; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
247; CHECK:       vector.ph:
248; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[SMAX]], 2
249; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[SMAX]], [[N_MOD_VF]]
250; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i1> poison, i1 [[C:%.*]], i32 0
251; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i1> [[BROADCAST_SPLATINSERT]], <2 x i1> poison, <2 x i32> zeroinitializer
252; CHECK-NEXT:    [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <2 x i32> poison, i32 [[X:%.*]], i32 0
253; CHECK-NEXT:    [[BROADCAST_SPLAT4:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT3]], <2 x i32> poison, <2 x i32> zeroinitializer
254; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
255; CHECK:       vector.body:
256; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UDIV_CONTINUE2:%.*]] ]
257; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_UDIV_CONTINUE2]] ]
258; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP18:%.*]], [[PRED_UDIV_CONTINUE2]] ]
259; CHECK-NEXT:    [[TMP0:%.*]] = mul <2 x i64> [[VEC_IND]], <i64 777, i64 777>
260; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 0
261; CHECK-NEXT:    br i1 [[TMP1]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]]
262; CHECK:       pred.udiv.if:
263; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x i64> [[TMP0]], i32 0
264; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP2]]
265; CHECK-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
266; CHECK-NEXT:    [[TMP5:%.*]] = udiv i32 [[TMP4]], [[X]]
267; CHECK-NEXT:    [[TMP6:%.*]] = insertelement <2 x i32> poison, i32 [[TMP5]], i32 0
268; CHECK-NEXT:    br label [[PRED_UDIV_CONTINUE]]
269; CHECK:       pred.udiv.continue:
270; CHECK-NEXT:    [[TMP7:%.*]] = phi i32 [ poison, [[VECTOR_BODY]] ], [ [[TMP4]], [[PRED_UDIV_IF]] ]
271; CHECK-NEXT:    [[TMP8:%.*]] = phi <2 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP6]], [[PRED_UDIV_IF]] ]
272; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x i1> [[BROADCAST_SPLAT]], i32 1
273; CHECK-NEXT:    br i1 [[TMP9]], label [[PRED_UDIV_IF1:%.*]], label [[PRED_UDIV_CONTINUE2]]
274; CHECK:       pred.udiv.if1:
275; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x i64> [[TMP0]], i32 1
276; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP10]]
277; CHECK-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
278; CHECK-NEXT:    [[TMP13:%.*]] = udiv i32 [[TMP12]], [[X]]
279; CHECK-NEXT:    [[TMP14:%.*]] = insertelement <2 x i32> [[TMP8]], i32 [[TMP13]], i32 1
280; CHECK-NEXT:    br label [[PRED_UDIV_CONTINUE2]]
281; CHECK:       pred.udiv.continue2:
282; CHECK-NEXT:    [[TMP15:%.*]] = phi i32 [ poison, [[PRED_UDIV_CONTINUE]] ], [ [[TMP12]], [[PRED_UDIV_IF1]] ]
283; CHECK-NEXT:    [[TMP16:%.*]] = phi <2 x i32> [ [[TMP8]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP14]], [[PRED_UDIV_IF1]] ]
284; CHECK-NEXT:    [[TMP17:%.*]] = xor <2 x i1> [[BROADCAST_SPLAT]], <i1 true, i1 true>
285; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[BROADCAST_SPLAT]], <2 x i32> [[TMP16]], <2 x i32> [[BROADCAST_SPLAT4]]
286; CHECK-NEXT:    [[TMP18]] = add <2 x i32> [[VEC_PHI]], [[PREDPHI]]
287; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
288; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
289; CHECK-NEXT:    [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
290; CHECK-NEXT:    br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
291; CHECK:       middle.block:
292; CHECK-NEXT:    [[TMP20:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[TMP18]])
293; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
294; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
295; CHECK:       scalar.ph:
296; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
297; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[TMP20]], [[MIDDLE_BLOCK]] ]
298; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
299; CHECK:       for.body:
300; CHECK-NEXT:    [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_INC:%.*]] ]
301; CHECK-NEXT:    [[R:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[T6:%.*]], [[FOR_INC]] ]
302; CHECK-NEXT:    [[I7:%.*]] = mul i64 [[I]], 777
303; CHECK-NEXT:    br i1 [[C]], label [[IF_THEN:%.*]], label [[FOR_INC]]
304; CHECK:       if.then:
305; CHECK-NEXT:    [[T0:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[I7]]
306; CHECK-NEXT:    [[T2:%.*]] = load i32, i32* [[T0]], align 4
307; CHECK-NEXT:    [[T4:%.*]] = udiv i32 [[T2]], [[X]]
308; CHECK-NEXT:    br label [[FOR_INC]]
309; CHECK:       for.inc:
310; CHECK-NEXT:    [[T5:%.*]] = phi i32 [ [[X]], [[FOR_BODY]] ], [ [[T4]], [[IF_THEN]] ]
311; CHECK-NEXT:    [[T6]] = add i32 [[R]], [[T5]]
312; CHECK-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
313; CHECK-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
314; CHECK-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP5:![0-9]+]]
315; CHECK:       for.end:
316; CHECK-NEXT:    [[T7:%.*]] = phi i32 [ [[T6]], [[FOR_INC]] ], [ [[TMP20]], [[MIDDLE_BLOCK]] ]
317; CHECK-NEXT:    ret i32 [[T7]]
318;
319; SINK-GATHER-LABEL: @scalarize_and_sink_gather(
320; SINK-GATHER-NEXT:  entry:
321; SINK-GATHER-NEXT:    [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[N:%.*]], i64 1)
322; SINK-GATHER-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 8
323; SINK-GATHER-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
324; SINK-GATHER:       vector.ph:
325; SINK-GATHER-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[SMAX]], 8
326; SINK-GATHER-NEXT:    [[N_VEC:%.*]] = sub i64 [[SMAX]], [[N_MOD_VF]]
327; SINK-GATHER-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i1> poison, i1 [[C:%.*]], i32 0
328; SINK-GATHER-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i1> [[BROADCAST_SPLATINSERT]], <8 x i1> poison, <8 x i32> zeroinitializer
329; SINK-GATHER-NEXT:    [[BROADCAST_SPLATINSERT15:%.*]] = insertelement <8 x i32> poison, i32 [[X:%.*]], i32 0
330; SINK-GATHER-NEXT:    [[BROADCAST_SPLAT16:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT15]], <8 x i32> poison, <8 x i32> zeroinitializer
331; SINK-GATHER-NEXT:    br label [[VECTOR_BODY:%.*]]
332; SINK-GATHER:       vector.body:
333; SINK-GATHER-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UDIV_CONTINUE14:%.*]] ]
334; SINK-GATHER-NEXT:    [[VEC_IND:%.*]] = phi <8 x i64> [ <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_UDIV_CONTINUE14]] ]
335; SINK-GATHER-NEXT:    [[VEC_PHI:%.*]] = phi <8 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP66:%.*]], [[PRED_UDIV_CONTINUE14]] ]
336; SINK-GATHER-NEXT:    [[TMP0:%.*]] = mul <8 x i64> [[VEC_IND]], <i64 777, i64 777, i64 777, i64 777, i64 777, i64 777, i64 777, i64 777>
337; SINK-GATHER-NEXT:    [[TMP1:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 0
338; SINK-GATHER-NEXT:    br i1 [[TMP1]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]]
339; SINK-GATHER:       pred.udiv.if:
340; SINK-GATHER-NEXT:    [[TMP2:%.*]] = extractelement <8 x i64> [[TMP0]], i32 0
341; SINK-GATHER-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, i32* [[A:%.*]], i64 [[TMP2]]
342; SINK-GATHER-NEXT:    [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
343; SINK-GATHER-NEXT:    [[TMP5:%.*]] = udiv i32 [[TMP4]], [[X]]
344; SINK-GATHER-NEXT:    [[TMP6:%.*]] = insertelement <8 x i32> poison, i32 [[TMP5]], i32 0
345; SINK-GATHER-NEXT:    br label [[PRED_UDIV_CONTINUE]]
346; SINK-GATHER:       pred.udiv.continue:
347; SINK-GATHER-NEXT:    [[TMP7:%.*]] = phi i32 [ poison, [[VECTOR_BODY]] ], [ [[TMP4]], [[PRED_UDIV_IF]] ]
348; SINK-GATHER-NEXT:    [[TMP8:%.*]] = phi <8 x i32> [ poison, [[VECTOR_BODY]] ], [ [[TMP6]], [[PRED_UDIV_IF]] ]
349; SINK-GATHER-NEXT:    [[TMP9:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 1
350; SINK-GATHER-NEXT:    br i1 [[TMP9]], label [[PRED_UDIV_IF1:%.*]], label [[PRED_UDIV_CONTINUE2:%.*]]
351; SINK-GATHER:       pred.udiv.if1:
352; SINK-GATHER-NEXT:    [[TMP10:%.*]] = extractelement <8 x i64> [[TMP0]], i32 1
353; SINK-GATHER-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP10]]
354; SINK-GATHER-NEXT:    [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
355; SINK-GATHER-NEXT:    [[TMP13:%.*]] = udiv i32 [[TMP12]], [[X]]
356; SINK-GATHER-NEXT:    [[TMP14:%.*]] = insertelement <8 x i32> [[TMP8]], i32 [[TMP13]], i32 1
357; SINK-GATHER-NEXT:    br label [[PRED_UDIV_CONTINUE2]]
358; SINK-GATHER:       pred.udiv.continue2:
359; SINK-GATHER-NEXT:    [[TMP15:%.*]] = phi i32 [ poison, [[PRED_UDIV_CONTINUE]] ], [ [[TMP12]], [[PRED_UDIV_IF1]] ]
360; SINK-GATHER-NEXT:    [[TMP16:%.*]] = phi <8 x i32> [ [[TMP8]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP14]], [[PRED_UDIV_IF1]] ]
361; SINK-GATHER-NEXT:    [[TMP17:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 2
362; SINK-GATHER-NEXT:    br i1 [[TMP17]], label [[PRED_UDIV_IF3:%.*]], label [[PRED_UDIV_CONTINUE4:%.*]]
363; SINK-GATHER:       pred.udiv.if3:
364; SINK-GATHER-NEXT:    [[TMP18:%.*]] = extractelement <8 x i64> [[TMP0]], i32 2
365; SINK-GATHER-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP18]]
366; SINK-GATHER-NEXT:    [[TMP20:%.*]] = load i32, i32* [[TMP19]], align 4
367; SINK-GATHER-NEXT:    [[TMP21:%.*]] = udiv i32 [[TMP20]], [[X]]
368; SINK-GATHER-NEXT:    [[TMP22:%.*]] = insertelement <8 x i32> [[TMP16]], i32 [[TMP21]], i32 2
369; SINK-GATHER-NEXT:    br label [[PRED_UDIV_CONTINUE4]]
370; SINK-GATHER:       pred.udiv.continue4:
371; SINK-GATHER-NEXT:    [[TMP23:%.*]] = phi i32 [ poison, [[PRED_UDIV_CONTINUE2]] ], [ [[TMP20]], [[PRED_UDIV_IF3]] ]
372; SINK-GATHER-NEXT:    [[TMP24:%.*]] = phi <8 x i32> [ [[TMP16]], [[PRED_UDIV_CONTINUE2]] ], [ [[TMP22]], [[PRED_UDIV_IF3]] ]
373; SINK-GATHER-NEXT:    [[TMP25:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 3
374; SINK-GATHER-NEXT:    br i1 [[TMP25]], label [[PRED_UDIV_IF5:%.*]], label [[PRED_UDIV_CONTINUE6:%.*]]
375; SINK-GATHER:       pred.udiv.if5:
376; SINK-GATHER-NEXT:    [[TMP26:%.*]] = extractelement <8 x i64> [[TMP0]], i32 3
377; SINK-GATHER-NEXT:    [[TMP27:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP26]]
378; SINK-GATHER-NEXT:    [[TMP28:%.*]] = load i32, i32* [[TMP27]], align 4
379; SINK-GATHER-NEXT:    [[TMP29:%.*]] = udiv i32 [[TMP28]], [[X]]
380; SINK-GATHER-NEXT:    [[TMP30:%.*]] = insertelement <8 x i32> [[TMP24]], i32 [[TMP29]], i32 3
381; SINK-GATHER-NEXT:    br label [[PRED_UDIV_CONTINUE6]]
382; SINK-GATHER:       pred.udiv.continue6:
383; SINK-GATHER-NEXT:    [[TMP31:%.*]] = phi i32 [ poison, [[PRED_UDIV_CONTINUE4]] ], [ [[TMP28]], [[PRED_UDIV_IF5]] ]
384; SINK-GATHER-NEXT:    [[TMP32:%.*]] = phi <8 x i32> [ [[TMP24]], [[PRED_UDIV_CONTINUE4]] ], [ [[TMP30]], [[PRED_UDIV_IF5]] ]
385; SINK-GATHER-NEXT:    [[TMP33:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 4
386; SINK-GATHER-NEXT:    br i1 [[TMP33]], label [[PRED_UDIV_IF7:%.*]], label [[PRED_UDIV_CONTINUE8:%.*]]
387; SINK-GATHER:       pred.udiv.if7:
388; SINK-GATHER-NEXT:    [[TMP34:%.*]] = extractelement <8 x i64> [[TMP0]], i32 4
389; SINK-GATHER-NEXT:    [[TMP35:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP34]]
390; SINK-GATHER-NEXT:    [[TMP36:%.*]] = load i32, i32* [[TMP35]], align 4
391; SINK-GATHER-NEXT:    [[TMP37:%.*]] = udiv i32 [[TMP36]], [[X]]
392; SINK-GATHER-NEXT:    [[TMP38:%.*]] = insertelement <8 x i32> [[TMP32]], i32 [[TMP37]], i32 4
393; SINK-GATHER-NEXT:    br label [[PRED_UDIV_CONTINUE8]]
394; SINK-GATHER:       pred.udiv.continue8:
395; SINK-GATHER-NEXT:    [[TMP39:%.*]] = phi i32 [ poison, [[PRED_UDIV_CONTINUE6]] ], [ [[TMP36]], [[PRED_UDIV_IF7]] ]
396; SINK-GATHER-NEXT:    [[TMP40:%.*]] = phi <8 x i32> [ [[TMP32]], [[PRED_UDIV_CONTINUE6]] ], [ [[TMP38]], [[PRED_UDIV_IF7]] ]
397; SINK-GATHER-NEXT:    [[TMP41:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 5
398; SINK-GATHER-NEXT:    br i1 [[TMP41]], label [[PRED_UDIV_IF9:%.*]], label [[PRED_UDIV_CONTINUE10:%.*]]
399; SINK-GATHER:       pred.udiv.if9:
400; SINK-GATHER-NEXT:    [[TMP42:%.*]] = extractelement <8 x i64> [[TMP0]], i32 5
401; SINK-GATHER-NEXT:    [[TMP43:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP42]]
402; SINK-GATHER-NEXT:    [[TMP44:%.*]] = load i32, i32* [[TMP43]], align 4
403; SINK-GATHER-NEXT:    [[TMP45:%.*]] = udiv i32 [[TMP44]], [[X]]
404; SINK-GATHER-NEXT:    [[TMP46:%.*]] = insertelement <8 x i32> [[TMP40]], i32 [[TMP45]], i32 5
405; SINK-GATHER-NEXT:    br label [[PRED_UDIV_CONTINUE10]]
406; SINK-GATHER:       pred.udiv.continue10:
407; SINK-GATHER-NEXT:    [[TMP47:%.*]] = phi i32 [ poison, [[PRED_UDIV_CONTINUE8]] ], [ [[TMP44]], [[PRED_UDIV_IF9]] ]
408; SINK-GATHER-NEXT:    [[TMP48:%.*]] = phi <8 x i32> [ [[TMP40]], [[PRED_UDIV_CONTINUE8]] ], [ [[TMP46]], [[PRED_UDIV_IF9]] ]
409; SINK-GATHER-NEXT:    [[TMP49:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 6
410; SINK-GATHER-NEXT:    br i1 [[TMP49]], label [[PRED_UDIV_IF11:%.*]], label [[PRED_UDIV_CONTINUE12:%.*]]
411; SINK-GATHER:       pred.udiv.if11:
412; SINK-GATHER-NEXT:    [[TMP50:%.*]] = extractelement <8 x i64> [[TMP0]], i32 6
413; SINK-GATHER-NEXT:    [[TMP51:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP50]]
414; SINK-GATHER-NEXT:    [[TMP52:%.*]] = load i32, i32* [[TMP51]], align 4
415; SINK-GATHER-NEXT:    [[TMP53:%.*]] = udiv i32 [[TMP52]], [[X]]
416; SINK-GATHER-NEXT:    [[TMP54:%.*]] = insertelement <8 x i32> [[TMP48]], i32 [[TMP53]], i32 6
417; SINK-GATHER-NEXT:    br label [[PRED_UDIV_CONTINUE12]]
418; SINK-GATHER:       pred.udiv.continue12:
419; SINK-GATHER-NEXT:    [[TMP55:%.*]] = phi i32 [ poison, [[PRED_UDIV_CONTINUE10]] ], [ [[TMP52]], [[PRED_UDIV_IF11]] ]
420; SINK-GATHER-NEXT:    [[TMP56:%.*]] = phi <8 x i32> [ [[TMP48]], [[PRED_UDIV_CONTINUE10]] ], [ [[TMP54]], [[PRED_UDIV_IF11]] ]
421; SINK-GATHER-NEXT:    [[TMP57:%.*]] = extractelement <8 x i1> [[BROADCAST_SPLAT]], i32 7
422; SINK-GATHER-NEXT:    br i1 [[TMP57]], label [[PRED_UDIV_IF13:%.*]], label [[PRED_UDIV_CONTINUE14]]
423; SINK-GATHER:       pred.udiv.if13:
424; SINK-GATHER-NEXT:    [[TMP58:%.*]] = extractelement <8 x i64> [[TMP0]], i32 7
425; SINK-GATHER-NEXT:    [[TMP59:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[TMP58]]
426; SINK-GATHER-NEXT:    [[TMP60:%.*]] = load i32, i32* [[TMP59]], align 4
427; SINK-GATHER-NEXT:    [[TMP61:%.*]] = udiv i32 [[TMP60]], [[X]]
428; SINK-GATHER-NEXT:    [[TMP62:%.*]] = insertelement <8 x i32> [[TMP56]], i32 [[TMP61]], i32 7
429; SINK-GATHER-NEXT:    br label [[PRED_UDIV_CONTINUE14]]
430; SINK-GATHER:       pred.udiv.continue14:
431; SINK-GATHER-NEXT:    [[TMP63:%.*]] = phi i32 [ poison, [[PRED_UDIV_CONTINUE12]] ], [ [[TMP60]], [[PRED_UDIV_IF13]] ]
432; SINK-GATHER-NEXT:    [[TMP64:%.*]] = phi <8 x i32> [ [[TMP56]], [[PRED_UDIV_CONTINUE12]] ], [ [[TMP62]], [[PRED_UDIV_IF13]] ]
433; SINK-GATHER-NEXT:    [[TMP65:%.*]] = xor <8 x i1> [[BROADCAST_SPLAT]], <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>
434; SINK-GATHER-NEXT:    [[PREDPHI:%.*]] = select <8 x i1> [[BROADCAST_SPLAT]], <8 x i32> [[TMP64]], <8 x i32> [[BROADCAST_SPLAT16]]
435; SINK-GATHER-NEXT:    [[TMP66]] = add <8 x i32> [[VEC_PHI]], [[PREDPHI]]
436; SINK-GATHER-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
437; SINK-GATHER-NEXT:    [[VEC_IND_NEXT]] = add <8 x i64> [[VEC_IND]], <i64 8, i64 8, i64 8, i64 8, i64 8, i64 8, i64 8, i64 8>
438; SINK-GATHER-NEXT:    [[TMP67:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
439; SINK-GATHER-NEXT:    br i1 [[TMP67]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
440; SINK-GATHER:       middle.block:
441; SINK-GATHER-NEXT:    [[TMP68:%.*]] = call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[TMP66]])
442; SINK-GATHER-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
443; SINK-GATHER-NEXT:    br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]
444; SINK-GATHER:       scalar.ph:
445; SINK-GATHER-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
446; SINK-GATHER-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[TMP68]], [[MIDDLE_BLOCK]] ]
447; SINK-GATHER-NEXT:    br label [[FOR_BODY:%.*]]
448; SINK-GATHER:       for.body:
449; SINK-GATHER-NEXT:    [[I:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[I_NEXT:%.*]], [[FOR_INC:%.*]] ]
450; SINK-GATHER-NEXT:    [[R:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[T6:%.*]], [[FOR_INC]] ]
451; SINK-GATHER-NEXT:    [[I7:%.*]] = mul i64 [[I]], 777
452; SINK-GATHER-NEXT:    br i1 [[C]], label [[IF_THEN:%.*]], label [[FOR_INC]]
453; SINK-GATHER:       if.then:
454; SINK-GATHER-NEXT:    [[T0:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[I7]]
455; SINK-GATHER-NEXT:    [[T2:%.*]] = load i32, i32* [[T0]], align 4
456; SINK-GATHER-NEXT:    [[T4:%.*]] = udiv i32 [[T2]], [[X]]
457; SINK-GATHER-NEXT:    br label [[FOR_INC]]
458; SINK-GATHER:       for.inc:
459; SINK-GATHER-NEXT:    [[T5:%.*]] = phi i32 [ [[X]], [[FOR_BODY]] ], [ [[T4]], [[IF_THEN]] ]
460; SINK-GATHER-NEXT:    [[T6]] = add i32 [[R]], [[T5]]
461; SINK-GATHER-NEXT:    [[I_NEXT]] = add nuw nsw i64 [[I]], 1
462; SINK-GATHER-NEXT:    [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
463; SINK-GATHER-NEXT:    br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop [[LOOP5:![0-9]+]]
464; SINK-GATHER:       for.end:
465; SINK-GATHER-NEXT:    [[T7:%.*]] = phi i32 [ [[T6]], [[FOR_INC]] ], [ [[TMP68]], [[MIDDLE_BLOCK]] ]
466; SINK-GATHER-NEXT:    ret i32 [[T7]]
467;
468entry:
469  br label %for.body
470
471for.body:
472  %i = phi i64 [ 0, %entry ], [ %i.next, %for.inc ]
473  %r = phi i32 [ 0, %entry ], [ %t6, %for.inc ]
474  %i7 = mul i64 %i, 777
475  br i1 %c, label %if.then, label %for.inc
476
477if.then:
478  %t0 = getelementptr inbounds i32, i32* %a, i64 %i7
479  %t2 = load i32, i32* %t0, align 4
480  %t4 = udiv i32 %t2, %x
481  br label %for.inc
482
483for.inc:
484  %t5 = phi i32 [ %x, %for.body ], [ %t4, %if.then]
485  %t6 = add i32 %r, %t5
486  %i.next = add nuw nsw i64 %i, 1
487  %cond = icmp slt i64 %i.next, %n
488  br i1 %cond, label %for.body, label %for.end
489
490for.end:
491  %t7 = phi i32 [ %t6, %for.inc ]
492  ret i32 %t7
493}
494