2 ; RUN: opt < %s -loop-vectorize -disable-output -debug-only=loop-vectorize 2>&1 | FileCheck %s --check-prefix=COST
3 ; RUN: opt < %s -loop-vectorize -force-vector-width=2 -instcombine -simplifycfg -S | FileCheck %s
5 target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
6 target triple = "aarch64--linux-gnu"
8 ; This test checks that we correctly compute the scalarized operands for a
9 ; user-specified vectorization factor when interleaving is disabled. We use the
10 ; "optsize" attribute to disable all interleaving calculations. A cost of 4
11 ; for %tmp4 indicates that we would scalarize it's operand (%tmp3), giving
12 ; %tmp4 a lower scalarization overhead.
14 ; COST-LABEL: predicated_udiv_scalarized_operand
15 ; COST: LV: Found an estimated cost of 4 for VF 2 For instruction: %tmp4 = udiv i64 %tmp2, %tmp3
17 ; CHECK-LABEL: @predicated_udiv_scalarized_operand(
19 ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %entry ], [ [[INDEX_NEXT:%.*]], %[[PRED_UDIV_CONTINUE2:.*]] ]
20 ; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, %entry ], [ [[TMP17:%.*]], %[[PRED_UDIV_CONTINUE2]] ]
21 ; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds i64, i64* %a, i64 [[INDEX]]
22 ; CHECK-NEXT: [[TMP1:%.*]] = bitcast i64* [[TMP0]] to <2 x i64>*
23 ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i64>, <2 x i64>* [[TMP1]], align 4
24 ; CHECK-NEXT: [[TMP2:%.*]] = icmp sgt <2 x i64> [[WIDE_LOAD]], zeroinitializer
25 ; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i1> [[TMP2]], i32 0
26 ; CHECK-NEXT: br i1 [[TMP3]], label %[[PRED_UDIV_IF:.*]], label %[[PRED_UDIV_CONTINUE:.*]]
27 ; CHECK: [[PRED_UDIV_IF]]:
28 ; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i64> [[WIDE_LOAD]], i32 0
29 ; CHECK-NEXT: [[TMP5:%.*]] = add nsw i64 [[TMP4]], %x
30 ; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i64> [[WIDE_LOAD]], i32 0
31 ; CHECK-NEXT: [[TMP7:%.*]] = udiv i64 [[TMP6]], [[TMP5]]
32 ; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x i64> undef, i64 [[TMP7]], i32 0
33 ; CHECK-NEXT: br label %[[PRED_UDIV_CONTINUE]]
34 ; CHECK: [[PRED_UDIV_CONTINUE]]:
35 ; CHECK-NEXT: [[TMP9:%.*]] = phi <2 x i64> [ undef, %vector.body ], [ [[TMP8]], %[[PRED_UDIV_IF]] ]
36 ; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x i1> [[TMP2]], i32 1
37 ; CHECK-NEXT: br i1 [[TMP10]], label %[[PRED_UDIV_IF1:.*]], label %[[PRED_UDIV_CONTINUE2]]
38 ; CHECK: [[PRED_UDIV_IF1]]:
39 ; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i64> [[WIDE_LOAD]], i32 1
40 ; CHECK-NEXT: [[TMP12:%.*]] = add nsw i64 [[TMP11]], %x
41 ; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i64> [[WIDE_LOAD]], i32 1
42 ; CHECK-NEXT: [[TMP14:%.*]] = udiv i64 [[TMP13]], [[TMP12]]
43 ; CHECK-NEXT: [[TMP15:%.*]] = insertelement <2 x i64> [[TMP9]], i64 [[TMP14]], i32 1
44 ; CHECK-NEXT: br label %[[PRED_UDIV_CONTINUE2]]
45 ; CHECK: [[PRED_UDIV_CONTINUE2]]:
46 ; CHECK-NEXT: [[TMP16:%.*]] = phi <2 x i64> [ [[TMP9]], %[[PRED_UDIV_CONTINUE]] ], [ [[TMP15]], %[[PRED_UDIV_IF1]] ]
47 ; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP2]], <2 x i64> [[TMP16]], <2 x i64> [[WIDE_LOAD]]
48 ; CHECK-NEXT: [[TMP17]] = add <2 x i64> [[VEC_PHI]], [[PREDPHI]]
49 ; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2
50 ; CHECK: br i1 {{.*}}, label %middle.block, label %vector.body
52 define i64 @predicated_udiv_scalarized_operand(i64* %a, i64 %x) optsize {
57 %i = phi i64 [ 0, %entry ], [ %i.next, %for.inc ]
58 %r = phi i64 [ 0, %entry ], [ %tmp6, %for.inc ]
59 %tmp0 = getelementptr inbounds i64, i64* %a, i64 %i
60 %tmp2 = load i64, i64* %tmp0, align 4
61 %cond0 = icmp sgt i64 %tmp2, 0
62 br i1 %cond0, label %if.then, label %for.inc
65 %tmp3 = add nsw i64 %tmp2, %x
66 %tmp4 = udiv i64 %tmp2, %tmp3
70 %tmp5 = phi i64 [ %tmp2, %for.body ], [ %tmp4, %if.then]
71 %tmp6 = add i64 %r, %tmp5
72 %i.next = add nuw nsw i64 %i, 1
73 %cond1 = icmp slt i64 %i.next, 100
74 br i1 %cond1, label %for.body, label %for.end
77 %tmp7 = phi i64 [ %tmp6, %for.inc ]