Home | History | Annotate | Download | only in LoopVectorize
      1 ; RUN: opt -S -vectorize-num-stores-pred=1 -force-vector-width=1 -force-vector-interleave=2 -loop-vectorize -simplifycfg < %s | FileCheck %s --check-prefix=UNROLL
      2 ; RUN: opt -S -vectorize-num-stores-pred=1 -force-vector-width=1 -force-vector-interleave=2 -loop-vectorize < %s | FileCheck %s --check-prefix=UNROLL-NOSIMPLIFY
      3 ; RUN: opt -S -vectorize-num-stores-pred=1 -force-vector-width=2 -force-vector-interleave=1 -loop-vectorize -enable-cond-stores-vec -simplifycfg < %s | FileCheck %s --check-prefix=VEC
      4 ; RUN: opt -S -vectorize-num-stores-pred=1 -force-vector-width=2 -force-vector-interleave=1 -loop-vectorize -enable-cond-stores-vec -simplifycfg -instcombine < %s | FileCheck %s --check-prefix=VEC-IC
      5 
      6 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
      7 target triple = "x86_64-apple-macosx10.9.0"
      8 
      9 ; Test predication of stores.
     10 define i32 @test(i32* nocapture %f) #0 {
     11 entry:
     12   br label %for.body
     13 
     14 ; VEC-LABEL: test
     15 ; VEC:   %[[v8:.+]] = icmp sgt <2 x i32> %{{.*}}, <i32 100, i32 100>
     16 ; VEC:   %[[v9:.+]] = add nsw <2 x i32> %{{.*}}, <i32 20, i32 20>
     17 ; VEC:   %[[v10:.+]] = and <2 x i1> %[[v8]], <i1 true, i1 true>
     18 ; VEC:   %[[v11:.+]] = extractelement <2 x i1> %[[v10]], i32 0
     19 ; VEC:   %[[v12:.+]] = icmp eq i1 %[[v11]], true
     20 ; VEC:   %[[v13:.+]] = extractelement <2 x i32> %[[v9]], i32 0
     21 ; VEC:   %[[v14:.+]] = extractelement <2 x i32*> %{{.*}}, i32 0
     22 ; VEC:   br i1 %[[v12]], label %[[cond:.+]], label %[[else:.+]]
     23 ;
     24 ; VEC: [[cond]]:
     25 ; VEC:   store i32 %[[v13]], i32* %[[v14]], align 4
     26 ; VEC:   br label %[[else:.+]]
     27 ;
     28 ; VEC: [[else]]:
     29 ; VEC:   %[[v15:.+]] = extractelement <2 x i1> %[[v10]], i32 1
     30 ; VEC:   %[[v16:.+]] = icmp eq i1 %[[v15]], true
     31 ; VEC:   %[[v17:.+]] = extractelement <2 x i32> %[[v9]], i32 1
     32 ; VEC:   %[[v18:.+]] = extractelement <2 x i32*> %{{.+}} i32 1
     33 ; VEC:   br i1 %[[v16]], label %[[cond2:.+]], label %[[else2:.+]]
     34 ;
     35 ; VEC: [[cond2]]:
     36 ; VEC:   store i32 %[[v17]], i32* %[[v18]], align 4
     37 ; VEC:   br label %[[else2:.+]]
     38 ;
     39 ; VEC: [[else2]]:
     40 
     41 ; VEC-IC-LABEL: test
     42 ; VEC-IC:   %[[v1:.+]] = icmp sgt <2 x i32> %{{.*}}, <i32 100, i32 100>
     43 ; VEC-IC:   %[[v2:.+]] = add nsw <2 x i32> %{{.*}}, <i32 20, i32 20>
     44 ; VEC-IC:   %[[v3:.+]] = extractelement <2 x i1> %[[v1]], i32 0
     45 ; VEC-IC:   br i1 %[[v3]], label %[[cond:.+]], label %[[else:.+]]
     46 ;
     47 ; VEC-IC: [[cond]]:
     48 ; VEC-IC:   %[[v4:.+]] = extractelement <2 x i32> %[[v2]], i32 0
     49 ; VEC-IC:   store i32 %[[v4]], i32* %{{.*}}, align 4
     50 ; VEC-IC:   br label %[[else:.+]]
     51 ;
     52 ; VEC-IC: [[else]]:
     53 ; VEC-IC:   %[[v5:.+]] = extractelement <2 x i1> %[[v1]], i32 1
     54 ; VEC-IC:   br i1 %[[v5]], label %[[cond2:.+]], label %[[else2:.+]]
     55 ;
     56 ; VEC-IC: [[cond2]]:
     57 ; VEC-IC:   %[[v6:.+]] = extractelement <2 x i32> %[[v2]], i32 1
     58 ; VEC-IC:   store i32 %[[v6]], i32* %{{.*}}, align 4
     59 ; VEC-IC:   br label %[[else2:.+]]
     60 ;
     61 ; VEC-IC: [[else2]]:
     62 
     63 ; UNROLL-LABEL: test
     64 ; UNROLL: vector.body:
     65 ; UNROLL:   %[[IND:[a-zA-Z0-9]+]] = add i64 %{{.*}}, 0
     66 ; UNROLL:   %[[IND1:[a-zA-Z0-9]+]] = add i64 %{{.*}}, 1
     67 ; UNROLL:   %[[v0:[a-zA-Z0-9]+]] = getelementptr inbounds i32, i32* %f, i64 %[[IND]]
     68 ; UNROLL:   %[[v1:[a-zA-Z0-9]+]] = getelementptr inbounds i32, i32* %f, i64 %[[IND1]]
     69 ; UNROLL:   %[[v2:[a-zA-Z0-9]+]] = load i32, i32* %[[v0]], align 4
     70 ; UNROLL:   %[[v3:[a-zA-Z0-9]+]] = load i32, i32* %[[v1]], align 4
     71 ; UNROLL:   %[[v4:[a-zA-Z0-9]+]] = icmp sgt i32 %[[v2]], 100
     72 ; UNROLL:   %[[v5:[a-zA-Z0-9]+]] = icmp sgt i32 %[[v3]], 100
     73 ; UNROLL:   %[[v6:[a-zA-Z0-9]+]] = add nsw i32 %[[v2]], 20
     74 ; UNROLL:   %[[v7:[a-zA-Z0-9]+]] = add nsw i32 %[[v3]], 20
     75 ; UNROLL:   %[[v8:[a-zA-Z0-9]+]] = icmp eq i1 %[[v4]], true
     76 ; UNROLL:   br i1 %[[v8]], label %[[cond:[a-zA-Z0-9.]+]], label %[[else:[a-zA-Z0-9.]+]]
     77 ;
     78 ; UNROLL: [[cond]]:
     79 ; UNROLL:   store i32 %[[v6]], i32* %[[v0]], align 4
     80 ; UNROLL:   br label %[[else]]
     81 ;
     82 ; UNROLL: [[else]]:
     83 ; UNROLL:   %[[v9:[a-zA-Z0-9]+]] = icmp eq i1 %[[v5]], true
     84 ; UNROLL:   br i1 %[[v9]], label %[[cond2:[a-zA-Z0-9.]+]], label %[[else2:[a-zA-Z0-9.]+]]
     85 ;
     86 ; UNROLL: [[cond2]]:
     87 ; UNROLL:   store i32 %[[v7]], i32* %[[v1]], align 4
     88 ; UNROLL:   br label %[[else2]]
     89 ;
     90 ; UNROLL: [[else2]]:
     91 
     92 for.body:
     93   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.inc ]
     94   %arrayidx = getelementptr inbounds i32, i32* %f, i64 %indvars.iv
     95   %0 = load i32, i32* %arrayidx, align 4
     96   %cmp1 = icmp sgt i32 %0, 100
     97   br i1 %cmp1, label %if.then, label %for.inc
     98 
     99 if.then:
    100   %add = add nsw i32 %0, 20
    101   store i32 %add, i32* %arrayidx, align 4
    102   br label %for.inc
    103 
    104 for.inc:
    105   %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
    106   %exitcond = icmp eq i64 %indvars.iv.next, 128
    107   br i1 %exitcond, label %for.end, label %for.body
    108 
    109 for.end:
    110   ret i32 0
    111 }
    112 
    113 ; Track basic blocks when unrolling conditional blocks. This code used to assert
    114 ; because we did not update the phi nodes with the proper predecessor in the
    115 ; vectorized loop body.
    116 ; PR18724
    117 
    118 ; UNROLL-NOSIMPLIFY-LABEL: bug18724
    119 ; UNROLL-NOSIMPLIFY: store i32
    120 ; UNROLL-NOSIMPLIFY: store i32
    121 
    122 define void @bug18724() {
    123 entry:
    124   br label %for.body9
    125 
    126 for.body9:
    127   br i1 undef, label %for.inc26, label %for.body14
    128 
    129 for.body14:
    130   %indvars.iv3 = phi i64 [ %indvars.iv.next4, %for.inc23 ], [ undef, %for.body9 ]
    131   %iNewChunks.120 = phi i32 [ %iNewChunks.2, %for.inc23 ], [ undef, %for.body9 ]
    132   %arrayidx16 = getelementptr inbounds [768 x i32], [768 x i32]* undef, i64 0, i64 %indvars.iv3
    133   %tmp = load i32, i32* %arrayidx16, align 4
    134   br i1 undef, label %if.then18, label %for.inc23
    135 
    136 if.then18:
    137   store i32 2, i32* %arrayidx16, align 4
    138   %inc21 = add nsw i32 %iNewChunks.120, 1
    139   br label %for.inc23
    140 
    141 for.inc23:
    142   %iNewChunks.2 = phi i32 [ %inc21, %if.then18 ], [ %iNewChunks.120, %for.body14 ]
    143   %indvars.iv.next4 = add nsw i64 %indvars.iv3, 1
    144   %tmp1 = trunc i64 %indvars.iv3 to i32
    145   %cmp13 = icmp slt i32 %tmp1, 0
    146   br i1 %cmp13, label %for.body14, label %for.inc26
    147 
    148 for.inc26:
    149   %iNewChunks.1.lcssa = phi i32 [ undef, %for.body9 ], [ %iNewChunks.2, %for.inc23 ]
    150   unreachable
    151 }
    152