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