1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
3 ; RUN: opt < %s -passes=loop-vectorize -disable-output -debug-only=loop-vectorize 2>&1 | FileCheck %s --check-prefix=COST
4 ; RUN: opt < %s -passes=loop-vectorize,instcombine,simplifycfg -force-vector-width=2 -simplifycfg-require-and-preserve-domtree=1 -S | FileCheck %s
6 target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
7 target triple = "aarch64--linux-gnu"
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.
15 ; COST-LABEL: predicated_udiv_scalarized_operand
16 ; COST: LV: Found an estimated cost of 5 for VF 2 For instruction: %var4 = udiv i64 %var2, %var3
19 define i64 @predicated_udiv_scalarized_operand(ptr %a, i64 %x) optsize {
20 ; CHECK-LABEL: @predicated_udiv_scalarized_operand(
22 ; CHECK-NEXT: br label [[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, ptr [[A:%.*]], i64 [[INDEX]]
27 ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, ptr [[TMP0]], align 4
28 ; CHECK-NEXT: [[TMP2:%.*]] = icmp sgt <2 x i64> [[WIDE_LOAD]], zeroinitializer
29 ; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i1> [[TMP2]], i64 0
30 ; CHECK-NEXT: br i1 [[TMP3]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]]
31 ; CHECK: pred.udiv.if:
32 ; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i64> [[WIDE_LOAD]], i64 0
33 ; CHECK-NEXT: [[TMP5:%.*]] = add nsw i64 [[TMP4]], [[X:%.*]]
34 ; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i64> [[WIDE_LOAD]], i64 0
35 ; CHECK-NEXT: [[TMP7:%.*]] = udiv i64 [[TMP6]], [[TMP5]]
36 ; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x i64> poison, i64 [[TMP7]], i64 0
37 ; CHECK-NEXT: br label [[PRED_UDIV_CONTINUE]]
38 ; CHECK: pred.udiv.continue:
39 ; CHECK-NEXT: [[TMP9:%.*]] = phi <2 x i64> [ poison, [[VECTOR_BODY]] ], [ [[TMP8]], [[PRED_UDIV_IF]] ]
40 ; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x i1> [[TMP2]], i64 1
41 ; CHECK-NEXT: br i1 [[TMP10]], label [[PRED_UDIV_IF1:%.*]], label [[PRED_UDIV_CONTINUE2]]
42 ; CHECK: pred.udiv.if1:
43 ; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i64> [[WIDE_LOAD]], i64 1
44 ; CHECK-NEXT: [[TMP12:%.*]] = add nsw i64 [[TMP11]], [[X]]
45 ; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i64> [[WIDE_LOAD]], i64 1
46 ; CHECK-NEXT: [[TMP14:%.*]] = udiv i64 [[TMP13]], [[TMP12]]
47 ; CHECK-NEXT: [[TMP15:%.*]] = insertelement <2 x i64> [[TMP9]], i64 [[TMP14]], i64 1
48 ; CHECK-NEXT: br label [[PRED_UDIV_CONTINUE2]]
49 ; CHECK: pred.udiv.continue2:
50 ; CHECK-NEXT: [[TMP16:%.*]] = phi <2 x i64> [ [[TMP9]], [[PRED_UDIV_CONTINUE]] ], [ [[TMP15]], [[PRED_UDIV_IF1]] ]
51 ; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP2]], <2 x i64> [[TMP16]], <2 x i64> [[WIDE_LOAD]]
52 ; CHECK-NEXT: [[TMP17]] = add <2 x i64> [[VEC_PHI]], [[PREDPHI]]
53 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
54 ; CHECK-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
55 ; CHECK-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
56 ; CHECK: middle.block:
57 ; CHECK-NEXT: [[TMP19:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[TMP17]])
58 ; CHECK-NEXT: ret i64 [[TMP19]]
64 %i = phi i64 [ 0, %entry ], [ %i.next, %for.inc ]
65 %r = phi i64 [ 0, %entry ], [ %var6, %for.inc ]
66 %var0 = getelementptr inbounds i64, ptr %a, i64 %i
67 %var2 = load i64, ptr %var0, align 4
68 %cond0 = icmp sgt i64 %var2, 0
69 br i1 %cond0, label %if.then, label %for.inc
72 %var3 = add nsw i64 %var2, %x
73 %var4 = udiv i64 %var2, %var3
77 %var5 = phi i64 [ %var2, %for.body ], [ %var4, %if.then]
78 %var6 = add i64 %r, %var5
79 %i.next = add nuw nsw i64 %i, 1
80 %cond1 = icmp slt i64 %i.next, 100
81 br i1 %cond1, label %for.body, label %for.end
84 %var7 = phi i64 [ %var6, %for.inc ]