xref: /llvm-project/llvm/test/Transforms/LoopVectorize/X86/pr55096-scalarize-add.ll (revision 7f7f540a48982d7901412502d045d7863d951ffe)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
2; RUN: opt -mtriple=x86_64-apple-macosx -passes=loop-vectorize,simplifycfg,dce -force-vector-width=2 -force-vector-interleave=1 -S %s | FileCheck %s
3
4define void @test_pr55096(i64 %c, ptr %p) {
5; CHECK-LABEL: define void @test_pr55096(
6; CHECK-SAME: i64 [[C:%.*]], ptr [[P:%.*]]) {
7; CHECK-NEXT:  entry:
8; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[C]], i64 0
9; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
10; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
11; CHECK:       vector.body:
12; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH:%.*]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE2:%.*]] ]
13; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 122, i64 123>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE2]] ]
14; CHECK-NEXT:    [[DOTCAST:%.*]] = trunc i64 [[INDEX]] to i16
15; CHECK-NEXT:    [[TMP0:%.*]] = mul i16 [[DOTCAST]], 2008
16; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = add i16 6229, [[TMP0]]
17; CHECK-NEXT:    [[TMP1:%.*]] = icmp ult <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
18; CHECK-NEXT:    [[TMP2:%.*]] = xor <2 x i1> [[TMP1]], splat (i1 true)
19; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x i1> [[TMP2]], i32 0
20; CHECK-NEXT:    br i1 [[TMP3]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
21; CHECK:       pred.store.if:
22; CHECK-NEXT:    [[TMP4:%.*]] = add i16 [[OFFSET_IDX]], 0
23; CHECK-NEXT:    [[TMP5:%.*]] = add i16 [[TMP4]], 2008
24; CHECK-NEXT:    [[TMP6:%.*]] = udiv i16 4943, [[TMP5]]
25; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i16, ptr [[P]], i16 [[TMP6]]
26; CHECK-NEXT:    store i16 0, ptr [[TMP7]], align 2
27; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE]]
28; CHECK:       pred.store.continue:
29; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x i1> [[TMP2]], i32 1
30; CHECK-NEXT:    br i1 [[TMP9]], label [[PRED_STORE_IF1:%.*]], label [[PRED_STORE_CONTINUE2]]
31; CHECK:       pred.store.if1:
32; CHECK-NEXT:    [[TMP10:%.*]] = add i16 [[OFFSET_IDX]], 2008
33; CHECK-NEXT:    [[TMP11:%.*]] = add i16 [[TMP10]], 2008
34; CHECK-NEXT:    [[TMP12:%.*]] = udiv i16 4943, [[TMP11]]
35; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds i16, ptr [[P]], i16 [[TMP12]]
36; CHECK-NEXT:    store i16 0, ptr [[TMP13]], align 2
37; CHECK-NEXT:    br label [[PRED_STORE_CONTINUE2]]
38; CHECK:       pred.store.continue2:
39; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
40; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], splat (i64 2)
41; CHECK-NEXT:    [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], 340
42; CHECK-NEXT:    br i1 [[TMP15]], label [[EXIT:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
43; CHECK:       exit:
44; CHECK-NEXT:    ret void
45;
46entry:
47  br label %loop.header
48
49loop.header:
50  %iv.1 = phi i64 [ 122, %entry ], [ %iv.1.next, %loop.latch ]
51  %iv.2 = phi i16 [ 6229, %entry ], [ %iv.2.next, %loop.latch ]
52  %iv.2.next = add i16 %iv.2, 2008
53  %cmp = icmp ult i64 %iv.1, %c
54  br i1 %cmp, label %loop.latch, label %loop.then
55
56loop.then:
57  %div = udiv i16 4943, %iv.2.next
58  %gep = getelementptr inbounds i16, ptr %p, i16 %div
59  store i16 0, ptr %gep, align 2
60  br label %loop.latch
61
62loop.latch:
63  %iv.1.next = add nuw nsw i64 %iv.1, 1
64  %exitcond.not = icmp eq i64 %iv.1.next, 462
65  br i1 %exitcond.not, label %exit, label %loop.header
66
67exit:
68  ret void
69}
70;.
71; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
72; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
73; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
74;.
75