1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; REQUIRES: asserts
3; RUN: opt < %s -loop-vectorize -disable-output -debug-only=loop-vectorize 2>&1 | FileCheck %s --check-prefix=COST
4; RUN: opt < %s -loop-vectorize -force-vector-width=2 -instcombine -simplifycfg -simplifycfg-require-and-preserve-domtree=1 -S | FileCheck %s
5
6target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
7target triple = "aarch64--linux-gnu"
8
9; This test checks that we correctly compute the scalarized operands for a
10; user-specified vectorization factor when interleaving is disabled. We use the
11; "optsize" attribute to disable all interleaving calculations.  A cost of 4
12; for %var4 indicates that we would scalarize it's operand (%var3), giving
13; %var4 a lower scalarization overhead.
14;
15; COST-LABEL:  predicated_udiv_scalarized_operand
16; COST:        LV: Found an estimated cost of 4 for VF 2 For instruction: %var4 = udiv i64 %var2, %var3
17;
18;
19define i64 @predicated_udiv_scalarized_operand(i64* %a, i64 %x) optsize {
20; CHECK-LABEL: @predicated_udiv_scalarized_operand(
21; CHECK-NEXT:  entry:
22; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
23; CHECK:       vector.body:
24; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[PRED_UDIV_CONTINUE2:%.*]] ]
25; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[ENTRY]] ], [ [[TMP17:%.*]], [[PRED_UDIV_CONTINUE2]] ]
26; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr inbounds i64, i64* [[A:%.*]], i64 [[INDEX]]
27; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i64* [[TMP0]] to <2 x i64>*
28; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i64>, <2 x i64>* [[TMP1]], align 4
29; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt <2 x i64> [[WIDE_LOAD]], zeroinitializer
30; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x i1> [[TMP2]], i64 0
31; CHECK-NEXT:    br i1 [[TMP3]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]]
32; CHECK:       pred.udiv.if:
33; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i64> [[WIDE_LOAD]], i64 0
34; CHECK-NEXT:    [[TMP5:%.*]] = add nsw i64 [[TMP4]], [[X:%.*]]
35; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i64> [[WIDE_LOAD]], i64 0
36; CHECK-NEXT:    [[TMP7:%.*]] = udiv i64 [[TMP6]], [[TMP5]]
37; CHECK-NEXT:    [[TMP8:%.*]] = insertelement <2 x i64> poison, i64 [[TMP7]], i64 0
38; CHECK-NEXT:    br label [[PRED_UDIV_CONTINUE]]
39; CHECK:       pred.udiv.continue:
40; CHECK-NEXT:    [[TMP9:%.*]] = phi <2 x i64> [ poison, [[VECTOR_BODY]] ], [ [[TMP8]], [[PRED_UDIV_IF]] ]
41; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x i1> [[TMP2]], i64 1
42; CHECK-NEXT:    br i1 [[TMP10]], label [[PRED_UDIV_IF1:%.*]], label [[PRED_UDIV_CONTINUE2]]
43; CHECK:       pred.udiv.if1:
44; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <2 x i64> [[WIDE_LOAD]], i64 1
45; CHECK-NEXT:    [[TMP12:%.*]] = add nsw i64 [[TMP11]], [[X]]
46; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x i64> [[WIDE_LOAD]], i64 1
47; CHECK-NEXT:    [[TMP14:%.*]] = udiv i64 [[TMP13]], [[TMP12]]
48; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <2 x i64> [[TMP9]], i64 [[TMP14]], i64 1
49; CHECK-NEXT:    br label [[PRED_UDIV_CONTINUE2]]
50; CHECK:       pred.udiv.continue2:
51; CHECK-NEXT:    [[TMP16:%.*]] = phi <2 x i64> [ [[TMP9]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP15]], [[PRED_UDIV_IF1]] ]
52; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP2]], <2 x i64> [[TMP16]], <2 x i64> [[WIDE_LOAD]]
53; CHECK-NEXT:    [[TMP17]] = add <2 x i64> [[VEC_PHI]], [[PREDPHI]]
54; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
55; CHECK-NEXT:    [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
56; CHECK-NEXT:    br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
57; CHECK:       middle.block:
58; CHECK-NEXT:    [[TMP19:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[TMP17]])
59; CHECK-NEXT:    ret i64 [[TMP19]]
60;
61entry:
62  br label %for.body
63
64for.body:
65  %i = phi i64 [ 0, %entry ], [ %i.next, %for.inc ]
66  %r = phi i64 [ 0, %entry ], [ %var6, %for.inc ]
67  %var0 = getelementptr inbounds i64, i64* %a, i64 %i
68  %var2 = load i64, i64* %var0, align 4
69  %cond0 = icmp sgt i64 %var2, 0
70  br i1 %cond0, label %if.then, label %for.inc
71
72if.then:
73  %var3 = add nsw i64 %var2, %x
74  %var4 = udiv i64 %var2, %var3
75  br label %for.inc
76
77for.inc:
78  %var5 = phi i64 [ %var2, %for.body ], [ %var4, %if.then]
79  %var6 = add i64 %r, %var5
80  %i.next = add nuw nsw i64 %i, 1
81  %cond1 = icmp slt i64 %i.next, 100
82  br i1 %cond1, label %for.body, label %for.end
83
84for.end:
85  %var7 = phi i64 [ %var6, %for.inc ]
86  ret i64 %var7
87}
88