1 ; RUN: opt < %s -loop-vectorize -force-vector-unroll=1 -force-vector-width=2 -S | FileCheck %s
3 target 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"
5 ; Make sure that we can handle multiple integer induction variables.
6 ; CHECK-LABEL: @multi_int_induction(
8 ; CHECK: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
9 ; CHECK: %normalized.idx = sub i64 %index, 0
10 ; CHECK: %[[VAR:.*]] = trunc i64 %normalized.idx to i32
11 ; CHECK: %offset.idx = add i32 190, %[[VAR]]
12 define void @multi_int_induction(i32* %A, i32 %N) {
17 %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ]
18 %count.09 = phi i32 [ 190, %for.body.lr.ph ], [ %inc, %for.body ]
19 %arrayidx2 = getelementptr inbounds i32* %A, i64 %indvars.iv
20 store i32 %count.09, i32* %arrayidx2, align 4
21 %inc = add nsw i32 %count.09, 1
22 %indvars.iv.next = add i64 %indvars.iv, 1
23 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
24 %exitcond = icmp ne i32 %lftr.wideiv, %N
25 br i1 %exitcond, label %for.body, label %for.end
31 ; RUN: opt < %s -loop-vectorize -force-vector-unroll=1 -force-vector-width=2 -instcombine -S | FileCheck %s --check-prefix=IND
33 ; Make sure we remove unneeded vectorization of induction variables.
34 ; In order for instcombine to cleanup the vectorized induction variables that we
35 ; create in the loop vectorizer we need to perform some form of redundancy
36 ; elimination to get rid of multiple uses.
38 ; IND-LABEL: scalar_use
40 ; IND: br label %vector.body
42 ; Vectorized induction variable.
43 ; IND-NOT: insertelement <2 x i64>
44 ; IND-NOT: shufflevector <2 x i64>
45 ; IND: br {{.*}}, label %vector.body
47 define void @scalar_use(float* %a, float %b, i64 %offset, i64 %offset2, i64 %n) {
52 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
53 %ind.sum = add i64 %iv, %offset
54 %arr.idx = getelementptr inbounds float* %a, i64 %ind.sum
55 %l1 = load float* %arr.idx, align 4
56 %ind.sum2 = add i64 %iv, %offset2
57 %arr.idx2 = getelementptr inbounds float* %a, i64 %ind.sum2
58 %l2 = load float* %arr.idx2, align 4
59 %m = fmul fast float %b, %l2
60 %ad = fadd fast float %l1, %m
61 store float %ad, float* %arr.idx, align 4
62 %iv.next = add nuw nsw i64 %iv, 1
63 %exitcond = icmp eq i64 %iv.next, %n
64 br i1 %exitcond, label %loopexit, label %for.body
71 ; Make sure that the loop exit count computation does not overflow for i8 and
72 ; i16. The exit count of these loops is i8/i16 max + 1. If we don't cast the
73 ; induction variable to a bigger type the exit count computation will overflow
77 ; CHECK-LABEL: i8_loop
78 ; CHECK: icmp eq i32 {{.*}}, 256
79 define i32 @i8_loop() nounwind readnone ssp uwtable {
82 ; <label>:1 ; preds = %1, %0
83 %a.0 = phi i32 [ 1, %0 ], [ %2, %1 ]
84 %b.0 = phi i8 [ 0, %0 ], [ %3, %1 ]
88 br i1 %4, label %5, label %1
90 ; <label>:5 ; preds = %1
94 ; CHECK-LABEL: i16_loop
95 ; CHECK: icmp eq i32 {{.*}}, 65536
97 define i32 @i16_loop() nounwind readnone ssp uwtable {
100 ; <label>:1 ; preds = %1, %0
101 %a.0 = phi i32 [ 1, %0 ], [ %2, %1 ]
102 %b.0 = phi i16 [ 0, %0 ], [ %3, %1 ]
104 %3 = add i16 %b.0, -1
105 %4 = icmp eq i16 %3, 0
106 br i1 %4, label %5, label %1
108 ; <label>:5 ; preds = %1