1; RUN: llc -O2 -o - %s | FileCheck %s 2target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" 3target triple = "x86_64-unknown-linux-gnu" 4 5; Function Attrs: nounwind uwtable 6; CHECK-LABEL: tail_dup_merge_loops 7; CHECK: # %entry 8; CHECK-NOT: # %{{[a-zA-Z_]+}} 9; CHECK: # %inner_loop_exit 10; CHECK-NOT: # %{{[a-zA-Z_]+}} 11; CHECK: # %inner_loop_latch 12; CHECK-NOT: # %{{[a-zA-Z_]+}} 13; CHECK: # %inner_loop_top 14; CHECK-NOT: # %{{[a-zA-Z_]+}} 15; CHECK: # %exit 16define void @tail_dup_merge_loops(i32 %a, i8* %b, i8* %c) local_unnamed_addr #0 { 17entry: 18 %notlhs674.i = icmp eq i32 %a, 0 19 br label %outer_loop_top 20 21outer_loop_top: ; preds = %inner_loop_exit, %entry 22 %dst.0.ph.i = phi i8* [ %b, %entry ], [ %scevgep679.i, %inner_loop_exit ] 23 br i1 %notlhs674.i, label %exit, label %inner_loop_preheader 24 25inner_loop_preheader: ; preds = %outer_loop_top 26 br label %inner_loop_top 27 28inner_loop_top: ; preds = %inner_loop_latch, %inner_loop_preheader 29 %dst.0.i = phi i8* [ %inc, %inner_loop_latch ], [ %dst.0.ph.i, %inner_loop_preheader ] 30 %var = load i8, i8* %dst.0.i 31 %tobool1.i = icmp slt i8 %var, 0 32 br label %inner_loop_test 33 34inner_loop_test: ; preds = %inner_loop_top 35 br i1 %tobool1.i, label %inner_loop_exit, label %inner_loop_latch 36 37inner_loop_exit: ; preds = %inner_loop_test 38 %scevgep.i = getelementptr i8, i8* %dst.0.i, i64 1 39 %scevgep679.i = getelementptr i8, i8* %scevgep.i, i64 0 40 br label %outer_loop_top 41 42inner_loop_latch: ; preds = %inner_loop_test 43 %cmp75.i = icmp ult i8* %dst.0.i, %c 44 %inc = getelementptr i8, i8* %dst.0.i, i64 2 45 br label %inner_loop_top 46 47exit: ; preds = %outer_loop_top 48 ret void 49} 50 51@.str.6 = external unnamed_addr constant [23 x i8], align 1 52 53; There is an erroneus check in LoopBase::addBasicBlockToLoop(), where it 54; assumes that the header block for a loop is unique. 55; For most of compilation this assumption is true, but during layout we allow 56; this assumption to be violated. The following code will trigger the bug: 57 58; The loops in question is eventually headed by the block shared_loop_header 59; 60; During layout The block labeled outer_loop_header gets tail-duplicated into 61; outer_loop_latch, and into shared_preheader, and then removed. This leaves 62; shared_loop_header as the header of both loops. The end result 63; is that there are 2 valid loops, and that they share a header. If we re-ran 64; the loop analysis, it would classify this as a single loop. 65; So far this is fine as far as layout is concerned. 66; After layout we tail merge blocks merge_other and merge_predecessor_split. 67; We do this even though they share only a single instruction, because 68; merge_predecessor_split falls through to their shared successor: 69; outer_loop_latch. 70; The rest of the blocks in the function are noise unfortunately. Bugpoint 71; couldn't shrink the test any further. 72 73; CHECK-LABEL: loop_shared_header 74; CHECK: # %entry 75; CHECK: # %shared_preheader 76; CHECK: # %shared_loop_header 77; CHECK: # %inner_loop_body 78; CHECK: # %outer_loop_latch 79; CHECK: # %merge_predecessor_split 80; CHECK: # %outer_loop_latch 81; CHECK: # %cleanup 82define i32 @loop_shared_header(i8* %exe, i32 %exesz, i32 %headsize, i32 %min, i32 %wwprva, i32 %e_lfanew, i8* readonly %wwp, i32 %wwpsz, i16 zeroext %sects) local_unnamed_addr #0 { 83entry: 84 %0 = load i32, i32* undef, align 4 85 %mul = shl nsw i32 %0, 2 86 br i1 undef, label %if.end19, label %cleanup 87 88if.end19: ; preds = %entry 89 %conv = zext i32 %mul to i64 90 %call = tail call i8* @cli_calloc(i64 %conv, i64 1) 91 %1 = icmp eq i32 %exesz, 0 92 %notrhs = icmp eq i32 %0, 0 93 %or.cond117.not = or i1 %1, %notrhs 94 %or.cond202 = or i1 %or.cond117.not, undef 95 %cmp35 = icmp ult i8* undef, %exe 96 %or.cond203 = or i1 %or.cond202, %cmp35 97 br i1 %or.cond203, label %cleanup, label %if.end50 98 99if.end50: ; preds = %if.end19 100 tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull %call, i8* undef, i64 %conv, i1 false) 101 %cmp1.i.i = icmp ugt i32 %mul, 3 102 br i1 %cmp1.i.i, label %shared_preheader, label %wunpsect.exit.thread.loopexit391 103 104shared_preheader: ; preds = %if.end50 105 br label %outer_loop_header 106 107outer_loop_header: ; preds = %outer_loop_latch, %shared_preheader 108 %bits.1.i = phi i8 [ 32, %shared_preheader ], [ %bits.43.i, %outer_loop_latch ] 109 %dst.0.ph.i = phi i8* [ undef, %shared_preheader ], [ %scevgep679.i, %outer_loop_latch ] 110 %2 = icmp eq i32 undef, 0 111 br i1 %2, label %while.cond.us1412.i, label %shared_loop_header 112 113while.cond.us1412.i: ; preds = %outer_loop_header 114 %.pre.i = add i8 %bits.1.i, -1 115 %tobool2.us1420.i = icmp eq i8 %.pre.i, 0 116 %or.cond.us1421.i = or i1 undef, %tobool2.us1420.i 117 br i1 %or.cond.us1421.i, label %if.end41.us1436.i, label %cleanup 118 119if.end41.us1436.i: ; preds = %while.cond.us1412.i 120 unreachable 121 122shared_loop_header: ; preds = %dup_early2, %dup_early1 123 %dst.0.i = phi i8* [ undef, %inner_loop_body ], [ %dst.0.ph.i, %outer_loop_header ], [ undef, %dead_block ] 124 %cmp3.i1172.i = icmp ult i8* undef, %call 125 br i1 %cmp3.i1172.i, label %wunpsect.exit.thread.loopexit389, label %inner_loop_body 126 127inner_loop_body: ; preds = %shared_loop_header 128 %3 = icmp slt i32 undef, 0 129 br i1 %3, label %if.end96.i, label %shared_loop_header 130 131dead_block: ; preds = %inner_loop_body 132 %cmp75.i = icmp ult i8* %dst.0.i, undef 133 br label %shared_loop_header 134 135if.end96.i: ; preds = %inner_loop_body 136 %cmp97.i = icmp ugt i32 undef, 2 137 br i1 %cmp97.i, label %if.then99.i, label %if.end287.i 138 139if.then99.i: ; preds = %if.end96.i 140 tail call void (i8*, ...) @cli_dbgmsg(i8* getelementptr inbounds ([23 x i8], [23 x i8]* @.str.6, i64 0, i64 0), i32 undef) 141 br label %cleanup 142 143if.end287.i: ; preds = %if.end96.i 144 %cmp291.i = icmp ne i32 undef, 1 145 %conv294.i = select i1 %cmp291.i, i16 4, i16 3 146 br i1 undef, label %if.end308.i, label %outer_loop_latch 147 148if.end308.i: ; preds = %if.end287.i 149 br i1 undef, label %if.end335.i, label %merge_predecessor_split 150 151merge_predecessor_split: ; preds = %if.end308.i 152 %4 = bitcast i8* undef to i32* 153 br label %outer_loop_latch 154 155if.end335.i: ; preds = %if.end308.i 156 br i1 undef, label %outer_loop_latch, label %merge_other 157 158merge_other: ; preds = %if.end335.i 159 br label %outer_loop_latch 160 161outer_loop_latch: ; preds = %merge_other, %if.end335.i, %merge_predecessor_split, %if.end287.i 162 %bits.43.i = phi i8 [ undef, %if.end287.i ], [ undef, %merge_other ], [ 32, %merge_predecessor_split ], [ 0, %if.end335.i ] 163 %backsize.0.i = phi i16 [ %conv294.i, %if.end287.i ], [ 0, %merge_other ], [ 0, %merge_predecessor_split ], [ 0, %if.end335.i ] 164 %5 = add i16 %backsize.0.i, -1 165 %6 = zext i16 %5 to i64 166 %scevgep.i = getelementptr i8, i8* %dst.0.ph.i, i64 1 167 %scevgep679.i = getelementptr i8, i8* %scevgep.i, i64 %6 168 br label %outer_loop_header 169 170wunpsect.exit.thread.loopexit389: ; preds = %shared_loop_header 171 unreachable 172 173wunpsect.exit.thread.loopexit391: ; preds = %if.end50 174 unreachable 175 176cleanup: ; preds = %if.then99.i, %while.cond.us1412.i, %if.end19, %entry 177 %retval.0 = phi i32 [ 0, %if.then99.i ], [ 1, %entry ], [ 1, %if.end19 ], [ 1, %while.cond.us1412.i ] 178 ret i32 %retval.0 179} 180 181; Function Attrs: nounwind 182declare void @cli_dbgmsg(i8*, ...) local_unnamed_addr #0 183 184; Function Attrs: nounwind 185declare i8* @cli_calloc(i64, i64) local_unnamed_addr #0 186 187; Function Attrs: argmemonly nounwind 188declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i1) #1 189attributes #0 = { nounwind } 190attributes #1 = { argmemonly nounwind } 191