1 ; RUN: opt < %s -loop-unroll -S | FileCheck %s 2 ; 3 ; Verify that the unrolling pass removes existing unroll count metadata 4 ; and adds a disable unrolling node after unrolling is complete. 5 6 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" 7 target triple = "x86_64-unknown-linux-gnu" 8 9 ; #pragma clang loop vectorize(enable) unroll_count(4) vectorize_width(8) 10 ; 11 ; Unroll count metadata should be replaced with unroll(disable). Vectorize 12 ; metadata should be untouched. 13 ; 14 ; CHECK-LABEL: @unroll_count_4( 15 ; CHECK: br i1 {{.*}}, label {{.*}}, label {{.*}}, !llvm.loop ![[LOOP_1:.*]] 16 define void @unroll_count_4(i32* nocapture %a) { 17 entry: 18 br label %for.body 19 20 for.body: ; preds = %for.body, %entry 21 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 22 %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv 23 %0 = load i32, i32* %arrayidx, align 4 24 %inc = add nsw i32 %0, 1 25 store i32 %inc, i32* %arrayidx, align 4 26 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 27 %exitcond = icmp eq i64 %indvars.iv.next, 64 28 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !1 29 30 for.end: ; preds = %for.body 31 ret void 32 } 33 !1 = !{!1, !2, !3, !4} 34 !2 = !{!"llvm.loop.vectorize.enable", i1 true} 35 !3 = !{!"llvm.loop.unroll.count", i32 4} 36 !4 = !{!"llvm.loop.vectorize.width", i32 8} 37 38 ; #pragma clang loop unroll(full) 39 ; 40 ; An unroll disable metadata node is only added for the unroll count case. 41 ; In this case, the loop has a full unroll metadata but can't be fully unrolled 42 ; because the trip count is dynamic. The full unroll metadata should remain 43 ; after unrolling. 44 ; 45 ; CHECK-LABEL: @unroll_full( 46 ; CHECK: br i1 {{.*}}, label {{.*}}, label {{.*}}, !llvm.loop ![[LOOP_2:.*]] 47 define void @unroll_full(i32* nocapture %a, i32 %b) { 48 entry: 49 %cmp3 = icmp sgt i32 %b, 0 50 br i1 %cmp3, label %for.body, label %for.end, !llvm.loop !5 51 52 for.body: ; preds = %entry, %for.body 53 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ] 54 %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv 55 %0 = load i32, i32* %arrayidx, align 4 56 %inc = add nsw i32 %0, 1 57 store i32 %inc, i32* %arrayidx, align 4 58 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 59 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 60 %exitcond = icmp eq i32 %lftr.wideiv, %b 61 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !5 62 63 for.end: ; preds = %for.body, %entry 64 ret void 65 } 66 !5 = !{!5, !6} 67 !6 = !{!"llvm.loop.unroll.full"} 68 69 ; #pragma clang loop unroll(disable) 70 ; 71 ; Unroll metadata should not change. 72 ; 73 ; CHECK-LABEL: @unroll_disable( 74 ; CHECK: br i1 {{.*}}, label {{.*}}, label {{.*}}, !llvm.loop ![[LOOP_3:.*]] 75 define void @unroll_disable(i32* nocapture %a) { 76 entry: 77 br label %for.body 78 79 for.body: ; preds = %for.body, %entry 80 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 81 %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv 82 %0 = load i32, i32* %arrayidx, align 4 83 %inc = add nsw i32 %0, 1 84 store i32 %inc, i32* %arrayidx, align 4 85 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 86 %exitcond = icmp eq i64 %indvars.iv.next, 64 87 br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !7 88 89 for.end: ; preds = %for.body 90 ret void 91 } 92 !7 = !{!7, !8} 93 !8 = !{!"llvm.loop.unroll.disable"} 94 95 ; This function contains two loops which share the same llvm.loop metadata node 96 ; with an llvm.loop.unroll.count 2 hint. Both loops should be unrolled. This 97 ; verifies that adding disable metadata to a loop after unrolling doesn't affect 98 ; other loops which previously shared the same llvm.loop metadata. 99 ; 100 ; CHECK-LABEL: @shared_metadata( 101 ; CHECK: store i32 102 ; CHECK: store i32 103 ; CHECK: br i1 {{.*}}, label {{.*}}, label {{.*}}, !llvm.loop ![[LOOP_4:.*]] 104 ; CHECK: store i32 105 ; CHECK: store i32 106 ; CHECK: br i1 {{.*}}, label {{.*}}, label {{.*}}, !llvm.loop ![[LOOP_5:.*]] 107 define void @shared_metadata(i32* nocapture %List) #0 { 108 entry: 109 br label %for.body3 110 111 for.body3: ; preds = %for.body3, %entry 112 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body3 ] 113 %arrayidx = getelementptr inbounds i32, i32* %List, i64 %indvars.iv 114 %0 = load i32, i32* %arrayidx, align 4 115 %add4 = add nsw i32 %0, 10 116 store i32 %add4, i32* %arrayidx, align 4 117 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 118 %exitcond = icmp eq i64 %indvars.iv.next, 4 119 br i1 %exitcond, label %for.body3.1.preheader, label %for.body3, !llvm.loop !9 120 121 for.body3.1.preheader: ; preds = %for.body3 122 br label %for.body3.1 123 124 for.body3.1: ; preds = %for.body3.1.preheader, %for.body3.1 125 %indvars.iv.1 = phi i64 [ %1, %for.body3.1 ], [ 0, %for.body3.1.preheader ] 126 %1 = add nsw i64 %indvars.iv.1, 1 127 %arrayidx.1 = getelementptr inbounds i32, i32* %List, i64 %1 128 %2 = load i32, i32* %arrayidx.1, align 4 129 %add4.1 = add nsw i32 %2, 10 130 store i32 %add4.1, i32* %arrayidx.1, align 4 131 %exitcond.1 = icmp eq i64 %1, 4 132 br i1 %exitcond.1, label %for.inc5.1, label %for.body3.1, !llvm.loop !9 133 134 for.inc5.1: ; preds = %for.body3.1 135 ret void 136 } 137 !9 = !{!9, !10} 138 !10 = !{!"llvm.loop.unroll.count", i32 2} 139 140 141 ; CHECK: ![[LOOP_1]] = distinct !{![[LOOP_1]], ![[VEC_ENABLE:.*]], ![[WIDTH_8:.*]], ![[UNROLL_DISABLE:.*]]} 142 ; CHECK: ![[VEC_ENABLE]] = !{!"llvm.loop.vectorize.enable", i1 true} 143 ; CHECK: ![[WIDTH_8]] = !{!"llvm.loop.vectorize.width", i32 8} 144 ; CHECK: ![[UNROLL_DISABLE]] = !{!"llvm.loop.unroll.disable"} 145 ; CHECK: ![[LOOP_2]] = distinct !{![[LOOP_2]], ![[UNROLL_FULL:.*]]} 146 ; CHECK: ![[UNROLL_FULL]] = !{!"llvm.loop.unroll.full"} 147 ; CHECK: ![[LOOP_3]] = distinct !{![[LOOP_3]], ![[UNROLL_DISABLE:.*]]} 148 ; CHECK: ![[LOOP_4]] = distinct !{![[LOOP_4]], ![[UNROLL_DISABLE:.*]]} 149 ; CHECK: ![[LOOP_5]] = distinct !{![[LOOP_5]], ![[UNROLL_DISABLE:.*]]} 150
