| Adam Nemet | d2fa414 | 2016-04-27 05:28:18 +0000 | [diff] [blame] | 1 | ; RUN: opt -basicaa -loop-distribute -enable-loop-distribute=0 -S < %s | FileCheck %s --check-prefix=CHECK --check-prefix=EXPLICIT --check-prefix=DEFAULT_OFF |
| 2 | ; RUN: opt -basicaa -loop-distribute -enable-loop-distribute=1 -S < %s | FileCheck %s --check-prefix=CHECK --check-prefix=EXPLICIT --check-prefix=DEFAULT_ON |
| 3 | |
| 4 | ; Same loop as in basic.ll. Check that distribution is enabled/disabled |
| 5 | ; properly according to -enable-loop-distribute=0/1 and the |
| 6 | ; llvm.loop.distribute.enable metadata. |
| 7 | |
| 8 | target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" |
| 9 | target triple = "x86_64-apple-macosx10.10.0" |
| 10 | |
| 11 | ; CHECK-LABEL: @explicit_on( |
| 12 | define void @explicit_on(i32* noalias %a, |
| 13 | i32* noalias %b, |
| 14 | i32* noalias %c, |
| 15 | i32* noalias %d, |
| 16 | i32* noalias %e) { |
| 17 | entry: |
| 18 | br label %for.body |
| 19 | |
| 20 | ; EXPLICIT: for.body.ldist1: |
| 21 | |
| 22 | for.body: ; preds = %for.body, %entry |
| 23 | %ind = phi i64 [ 0, %entry ], [ %add, %for.body ] |
| 24 | |
| 25 | %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %ind |
| 26 | %loadA = load i32, i32* %arrayidxA, align 4 |
| 27 | |
| 28 | %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %ind |
| 29 | %loadB = load i32, i32* %arrayidxB, align 4 |
| 30 | |
| 31 | %mulA = mul i32 %loadB, %loadA |
| 32 | |
| 33 | %add = add nuw nsw i64 %ind, 1 |
| 34 | %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add |
| 35 | store i32 %mulA, i32* %arrayidxA_plus_4, align 4 |
| 36 | |
| 37 | %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %ind |
| 38 | %loadD = load i32, i32* %arrayidxD, align 4 |
| 39 | |
| 40 | %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %ind |
| 41 | %loadE = load i32, i32* %arrayidxE, align 4 |
| 42 | |
| 43 | %mulC = mul i32 %loadD, %loadE |
| 44 | |
| 45 | %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %ind |
| 46 | store i32 %mulC, i32* %arrayidxC, align 4 |
| 47 | |
| 48 | %exitcond = icmp eq i64 %add, 20 |
| 49 | br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !0 |
| 50 | |
| 51 | for.end: ; preds = %for.body |
| 52 | ret void |
| 53 | } |
| 54 | |
| 55 | ; CHECK-LABEL: @explicit_off( |
| 56 | define void @explicit_off(i32* noalias %a, |
| 57 | i32* noalias %b, |
| 58 | i32* noalias %c, |
| 59 | i32* noalias %d, |
| 60 | i32* noalias %e) { |
| 61 | entry: |
| 62 | br label %for.body |
| 63 | |
| 64 | ; EXPLICIT-NOT: for.body.ldist1: |
| 65 | |
| 66 | for.body: ; preds = %for.body, %entry |
| 67 | %ind = phi i64 [ 0, %entry ], [ %add, %for.body ] |
| 68 | |
| 69 | %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %ind |
| 70 | %loadA = load i32, i32* %arrayidxA, align 4 |
| 71 | |
| 72 | %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %ind |
| 73 | %loadB = load i32, i32* %arrayidxB, align 4 |
| 74 | |
| 75 | %mulA = mul i32 %loadB, %loadA |
| 76 | |
| 77 | %add = add nuw nsw i64 %ind, 1 |
| 78 | %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add |
| 79 | store i32 %mulA, i32* %arrayidxA_plus_4, align 4 |
| 80 | |
| 81 | %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %ind |
| 82 | %loadD = load i32, i32* %arrayidxD, align 4 |
| 83 | |
| 84 | %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %ind |
| 85 | %loadE = load i32, i32* %arrayidxE, align 4 |
| 86 | |
| 87 | %mulC = mul i32 %loadD, %loadE |
| 88 | |
| 89 | %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %ind |
| 90 | store i32 %mulC, i32* %arrayidxC, align 4 |
| 91 | |
| 92 | %exitcond = icmp eq i64 %add, 20 |
| 93 | br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !2 |
| 94 | |
| 95 | for.end: ; preds = %for.body |
| 96 | ret void |
| 97 | } |
| 98 | |
| 99 | ; CHECK-LABEL: @default_distribute( |
| 100 | define void @default_distribute(i32* noalias %a, |
| 101 | i32* noalias %b, |
| 102 | i32* noalias %c, |
| 103 | i32* noalias %d, |
| 104 | i32* noalias %e) { |
| 105 | entry: |
| 106 | br label %for.body |
| 107 | |
| 108 | ; Verify the two distributed loops. |
| 109 | |
| 110 | ; DEFAULT_ON: for.body.ldist1: |
| 111 | ; DEFAULT_OFF-NOT: for.body.ldist1: |
| 112 | |
| 113 | for.body: ; preds = %for.body, %entry |
| 114 | %ind = phi i64 [ 0, %entry ], [ %add, %for.body ] |
| 115 | |
| 116 | %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %ind |
| 117 | %loadA = load i32, i32* %arrayidxA, align 4 |
| 118 | |
| 119 | %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %ind |
| 120 | %loadB = load i32, i32* %arrayidxB, align 4 |
| 121 | |
| 122 | %mulA = mul i32 %loadB, %loadA |
| 123 | |
| 124 | %add = add nuw nsw i64 %ind, 1 |
| 125 | %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add |
| 126 | store i32 %mulA, i32* %arrayidxA_plus_4, align 4 |
| 127 | |
| 128 | %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %ind |
| 129 | %loadD = load i32, i32* %arrayidxD, align 4 |
| 130 | |
| 131 | %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %ind |
| 132 | %loadE = load i32, i32* %arrayidxE, align 4 |
| 133 | |
| 134 | %mulC = mul i32 %loadD, %loadE |
| 135 | |
| 136 | %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %ind |
| 137 | store i32 %mulC, i32* %arrayidxC, align 4 |
| 138 | |
| 139 | %exitcond = icmp eq i64 %add, 20 |
| 140 | br i1 %exitcond, label %for.end, label %for.body |
| 141 | |
| 142 | for.end: ; preds = %for.body |
| 143 | ret void |
| 144 | } |
| 145 | |
| 146 | !0 = distinct !{!0, !1} |
| 147 | !1 = !{!"llvm.loop.distribute.enable", i1 true} |
| 148 | !2 = distinct !{!2, !3} |
| 149 | !3 = !{!"llvm.loop.distribute.enable", i1 false} |