1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu | FileCheck %s 3 4; Function Attrs: nounwind uwtable 5define void @tail_dup_merge_loops(i32 %a, ptr %b, ptr %c) local_unnamed_addr #0 { 6; CHECK-LABEL: tail_dup_merge_loops: 7; CHECK: # %bb.0: # %entry 8; CHECK-NEXT: jmp .LBB0_1 9; CHECK-NEXT: .p2align 4 10; CHECK-NEXT: .LBB0_3: # %inner_loop_exit 11; CHECK-NEXT: # in Loop: Header=BB0_1 Depth=1 12; CHECK-NEXT: incq %rsi 13; CHECK-NEXT: .LBB0_1: # %outer_loop_top 14; CHECK-NEXT: # =>This Loop Header: Depth=1 15; CHECK-NEXT: # Child Loop BB0_4 Depth 2 16; CHECK-NEXT: testl %edi, %edi 17; CHECK-NEXT: je .LBB0_5 18; CHECK-NEXT: .p2align 4 19; CHECK-NEXT: # %bb.2: # %inner_loop_top 20; CHECK-NEXT: # in Loop: Header=BB0_1 Depth=1 21; CHECK-NEXT: cmpb $0, (%rsi) 22; CHECK-NEXT: js .LBB0_3 23; CHECK-NEXT: .LBB0_4: # %inner_loop_latch 24; CHECK-NEXT: # Parent Loop BB0_1 Depth=1 25; CHECK-NEXT: # => This Inner Loop Header: Depth=2 26; CHECK-NEXT: addq $2, %rsi 27; CHECK-NEXT: cmpb $0, (%rsi) 28; CHECK-NEXT: jns .LBB0_4 29; CHECK-NEXT: jmp .LBB0_3 30; CHECK-NEXT: .LBB0_5: # %exit 31; CHECK-NEXT: retq 32entry: 33 %notlhs674.i = icmp eq i32 %a, 0 34 br label %outer_loop_top 35 36outer_loop_top: ; preds = %inner_loop_exit, %entry 37 %dst.0.ph.i = phi ptr [ %b, %entry ], [ %scevgep679.i, %inner_loop_exit ] 38 br i1 %notlhs674.i, label %exit, label %inner_loop_preheader 39 40inner_loop_preheader: ; preds = %outer_loop_top 41 br label %inner_loop_top 42 43inner_loop_top: ; preds = %inner_loop_latch, %inner_loop_preheader 44 %dst.0.i = phi ptr [ %inc, %inner_loop_latch ], [ %dst.0.ph.i, %inner_loop_preheader ] 45 %var = load i8, ptr %dst.0.i 46 %tobool1.i = icmp slt i8 %var, 0 47 br label %inner_loop_test 48 49inner_loop_test: ; preds = %inner_loop_top 50 br i1 %tobool1.i, label %inner_loop_exit, label %inner_loop_latch 51 52inner_loop_exit: ; preds = %inner_loop_test 53 %scevgep.i = getelementptr i8, ptr %dst.0.i, i64 1 54 %scevgep679.i = getelementptr i8, ptr %scevgep.i, i64 0 55 br label %outer_loop_top 56 57inner_loop_latch: ; preds = %inner_loop_test 58 %cmp75.i = icmp ult ptr %dst.0.i, %c 59 %inc = getelementptr i8, ptr %dst.0.i, i64 2 60 br label %inner_loop_top 61 62exit: ; preds = %outer_loop_top 63 ret void 64} 65 66@.str.6 = external unnamed_addr constant [23 x i8], align 1 67 68; There is an erroneus check in LoopBase::addBasicBlockToLoop(), where it 69; assumes that the header block for a loop is unique. 70; For most of compilation this assumption is true, but during layout we allow 71; this assumption to be violated. The following code will trigger the bug: 72 73; The loops in question is eventually headed by the block shared_loop_header 74; 75; During layout The block labeled outer_loop_header gets tail-duplicated into 76; outer_loop_latch, and into shared_preheader, and then removed. This leaves 77; shared_loop_header as the header of both loops. The end result 78; is that there are 2 valid loops, and that they share a header. If we re-ran 79; the loop analysis, it would classify this as a single loop. 80; So far this is fine as far as layout is concerned. 81; After layout we tail merge blocks merge_other and merge_predecessor_split. 82; We do this even though they share only a single instruction, because 83; merge_predecessor_split falls through to their shared successor: 84; outer_loop_latch. 85; The rest of the blocks in the function are noise unfortunately. Bugpoint 86; couldn't shrink the test any further. 87 88define i32 @loop_shared_header(ptr %exe, i32 %exesz, i32 %headsize, i32 %min, i32 %wwprva, i32 %e_lfanew, ptr readonly %wwp, i32 %wwpsz, i16 zeroext %sects) local_unnamed_addr #0 { 89; CHECK-LABEL: loop_shared_header: 90; CHECK: # %bb.0: # %entry 91; CHECK-NEXT: pushq %rbp 92; CHECK-NEXT: pushq %r15 93; CHECK-NEXT: pushq %r14 94; CHECK-NEXT: pushq %r12 95; CHECK-NEXT: pushq %rbx 96; CHECK-NEXT: movl $1, %ebx 97; CHECK-NEXT: xorl %eax, %eax 98; CHECK-NEXT: testb %al, %al 99; CHECK-NEXT: jne .LBB1_12 100; CHECK-NEXT: # %bb.1: # %if.end19 101; CHECK-NEXT: movl (%rax), %r12d 102; CHECK-NEXT: leal (,%r12,4), %ebp 103; CHECK-NEXT: movl %ebp, %r15d 104; CHECK-NEXT: movl $1, %esi 105; CHECK-NEXT: movq %r15, %rdi 106; CHECK-NEXT: callq cli_calloc@PLT 107; CHECK-NEXT: movq %rax, %r14 108; CHECK-NEXT: movb $1, %al 109; CHECK-NEXT: testb %al, %al 110; CHECK-NEXT: jne .LBB1_12 111; CHECK-NEXT: # %bb.2: # %if.end50 112; CHECK-NEXT: movq %r14, %rdi 113; CHECK-NEXT: movq %r15, %rdx 114; CHECK-NEXT: callq memcpy@PLT 115; CHECK-NEXT: cmpl $4, %ebp 116; CHECK-NEXT: jb .LBB1_19 117; CHECK-NEXT: # %bb.3: # %shared_preheader 118; CHECK-NEXT: movb $32, %cl 119; CHECK-NEXT: xorl %eax, %eax 120; CHECK-NEXT: jmp .LBB1_4 121; CHECK-NEXT: .p2align 4 122; CHECK-NEXT: .LBB1_15: # %merge_predecessor_split 123; CHECK-NEXT: # in Loop: Header=BB1_4 Depth=1 124; CHECK-NEXT: movb $32, %cl 125; CHECK-NEXT: .LBB1_4: # %outer_loop_header 126; CHECK-NEXT: # =>This Loop Header: Depth=1 127; CHECK-NEXT: # Child Loop BB1_8 Depth 2 128; CHECK-NEXT: testl %r12d, %r12d 129; CHECK-NEXT: je .LBB1_5 130; CHECK-NEXT: .p2align 4 131; CHECK-NEXT: .LBB1_8: # %shared_loop_header 132; CHECK-NEXT: # Parent Loop BB1_4 Depth=1 133; CHECK-NEXT: # => This Inner Loop Header: Depth=2 134; CHECK-NEXT: testq %r14, %r14 135; CHECK-NEXT: jne .LBB1_18 136; CHECK-NEXT: # %bb.9: # %inner_loop_body 137; CHECK-NEXT: # in Loop: Header=BB1_8 Depth=2 138; CHECK-NEXT: testb %al, %al 139; CHECK-NEXT: je .LBB1_8 140; CHECK-NEXT: # %bb.10: # %if.end96.i 141; CHECK-NEXT: # in Loop: Header=BB1_4 Depth=1 142; CHECK-NEXT: cmpl $3, %r12d 143; CHECK-NEXT: jae .LBB1_11 144; CHECK-NEXT: # %bb.13: # %if.end287.i 145; CHECK-NEXT: # in Loop: Header=BB1_4 Depth=1 146; CHECK-NEXT: testb %al, %al 147; CHECK-NEXT: # implicit-def: $cl 148; CHECK-NEXT: jne .LBB1_4 149; CHECK-NEXT: # %bb.14: # %if.end308.i 150; CHECK-NEXT: # in Loop: Header=BB1_4 Depth=1 151; CHECK-NEXT: testb %al, %al 152; CHECK-NEXT: je .LBB1_15 153; CHECK-NEXT: # %bb.16: # %if.end335.i 154; CHECK-NEXT: # in Loop: Header=BB1_4 Depth=1 155; CHECK-NEXT: xorl %ecx, %ecx 156; CHECK-NEXT: testb %cl, %cl 157; CHECK-NEXT: jne .LBB1_4 158; CHECK-NEXT: # %bb.17: # %merge_other 159; CHECK-NEXT: # in Loop: Header=BB1_4 Depth=1 160; CHECK-NEXT: # implicit-def: $cl 161; CHECK-NEXT: jmp .LBB1_4 162; CHECK-NEXT: .LBB1_5: # %while.cond.us1412.i 163; CHECK-NEXT: xorl %eax, %eax 164; CHECK-NEXT: testb %al, %al 165; CHECK-NEXT: jne .LBB1_7 166; CHECK-NEXT: # %bb.6: # %while.cond.us1412.i 167; CHECK-NEXT: decb %cl 168; CHECK-NEXT: jne .LBB1_12 169; CHECK-NEXT: .LBB1_7: # %if.end41.us1436.i 170; CHECK-NEXT: .LBB1_11: # %if.then99.i 171; CHECK-NEXT: movq .str.6@GOTPCREL(%rip), %rdi 172; CHECK-NEXT: xorl %ebx, %ebx 173; CHECK-NEXT: xorl %eax, %eax 174; CHECK-NEXT: callq cli_dbgmsg@PLT 175; CHECK-NEXT: .LBB1_12: # %cleanup 176; CHECK-NEXT: movl %ebx, %eax 177; CHECK-NEXT: popq %rbx 178; CHECK-NEXT: popq %r12 179; CHECK-NEXT: popq %r14 180; CHECK-NEXT: popq %r15 181; CHECK-NEXT: popq %rbp 182; CHECK-NEXT: retq 183; CHECK-NEXT: .LBB1_18: # %wunpsect.exit.thread.loopexit389 184; CHECK-NEXT: .LBB1_19: # %wunpsect.exit.thread.loopexit391 185entry: 186 %0 = load i32, ptr undef, align 4 187 %mul = shl nsw i32 %0, 2 188 br i1 undef, label %if.end19, label %cleanup 189 190if.end19: ; preds = %entry 191 %conv = zext i32 %mul to i64 192 %call = tail call ptr @cli_calloc(i64 %conv, i64 1) 193 %1 = icmp eq i32 %exesz, 0 194 %notrhs = icmp eq i32 %0, 0 195 %or.cond117.not = or i1 %1, %notrhs 196 %or.cond202 = or i1 %or.cond117.not, undef 197 %cmp35 = icmp ult ptr %call, %exe 198 %or.cond203 = or i1 %or.cond202, %cmp35 199 br i1 %or.cond203, label %cleanup, label %if.end50 200 201if.end50: ; preds = %if.end19 202 tail call void @llvm.memcpy.p0.p0.i64(ptr nonnull %call, ptr undef, i64 %conv, i1 false) 203 %cmp1.i.i = icmp ugt i32 %mul, 3 204 br i1 %cmp1.i.i, label %shared_preheader, label %wunpsect.exit.thread.loopexit391 205 206shared_preheader: ; preds = %if.end50 207 br label %outer_loop_header 208 209outer_loop_header: ; preds = %outer_loop_latch, %shared_preheader 210 %bits.1.i = phi i8 [ 32, %shared_preheader ], [ %bits.43.i, %outer_loop_latch ] 211 %dst.0.ph.i = phi ptr [ undef, %shared_preheader ], [ %scevgep679.i, %outer_loop_latch ] 212 %2 = icmp eq i32 %0, 0 213 br i1 %2, label %while.cond.us1412.i, label %shared_loop_header 214 215while.cond.us1412.i: ; preds = %outer_loop_header 216 %.pre.i = add i8 %bits.1.i, -1 217 %tobool2.us1420.i = icmp eq i8 %.pre.i, 0 218 %or.cond.us1421.i = or i1 undef, %tobool2.us1420.i 219 br i1 %or.cond.us1421.i, label %if.end41.us1436.i, label %cleanup 220 221if.end41.us1436.i: ; preds = %while.cond.us1412.i 222 unreachable 223 224shared_loop_header: ; preds = %dup_early2, %dup_early1 225 %dst.0.i = phi ptr [ undef, %inner_loop_body ], [ %dst.0.ph.i, %outer_loop_header ], [ undef, %dead_block ] 226 %cmp3.i1172.i = icmp ult ptr null, %call 227 br i1 %cmp3.i1172.i, label %wunpsect.exit.thread.loopexit389, label %inner_loop_body 228 229inner_loop_body: ; preds = %shared_loop_header 230 %3 = icmp slt i32 undef, 0 231 br i1 %3, label %if.end96.i, label %shared_loop_header 232 233dead_block: ; preds = %inner_loop_body 234 %cmp75.i = icmp ult ptr %dst.0.i, null 235 br label %shared_loop_header 236 237if.end96.i: ; preds = %inner_loop_body 238 %cmp97.i = icmp ugt i32 %0, 2 239 br i1 %cmp97.i, label %if.then99.i, label %if.end287.i 240 241if.then99.i: ; preds = %if.end96.i 242 tail call void (ptr, ...) @cli_dbgmsg(ptr @.str.6, i32 undef) 243 br label %cleanup 244 245if.end287.i: ; preds = %if.end96.i 246 %cmp291.i = icmp ne i32 %0, 1 247 %conv294.i = select i1 %cmp291.i, i16 4, i16 3 248 br i1 undef, label %if.end308.i, label %outer_loop_latch 249 250if.end308.i: ; preds = %if.end287.i 251 br i1 undef, label %if.end335.i, label %merge_predecessor_split 252 253merge_predecessor_split: ; preds = %if.end308.i 254 %4 = bitcast ptr undef to ptr 255 br label %outer_loop_latch 256 257if.end335.i: ; preds = %if.end308.i 258 br i1 undef, label %outer_loop_latch, label %merge_other 259 260merge_other: ; preds = %if.end335.i 261 br label %outer_loop_latch 262 263outer_loop_latch: ; preds = %merge_other, %if.end335.i, %merge_predecessor_split, %if.end287.i 264 %bits.43.i = phi i8 [ undef, %if.end287.i ], [ undef, %merge_other ], [ 32, %merge_predecessor_split ], [ 0, %if.end335.i ] 265 %backsize.0.i = phi i16 [ %conv294.i, %if.end287.i ], [ 0, %merge_other ], [ 0, %merge_predecessor_split ], [ 0, %if.end335.i ] 266 %5 = add i16 %backsize.0.i, -1 267 %6 = zext i16 %5 to i64 268 %scevgep.i = getelementptr i8, ptr %dst.0.ph.i, i64 1 269 %scevgep679.i = getelementptr i8, ptr %scevgep.i, i64 %6 270 br label %outer_loop_header 271 272wunpsect.exit.thread.loopexit389: ; preds = %shared_loop_header 273 unreachable 274 275wunpsect.exit.thread.loopexit391: ; preds = %if.end50 276 unreachable 277 278cleanup: ; preds = %if.then99.i, %while.cond.us1412.i, %if.end19, %entry 279 %retval.0 = phi i32 [ 0, %if.then99.i ], [ 1, %entry ], [ 1, %if.end19 ], [ 1, %while.cond.us1412.i ] 280 ret i32 %retval.0 281} 282 283; Function Attrs: nounwind 284declare void @cli_dbgmsg(ptr, ...) local_unnamed_addr #0 285 286; Function Attrs: nounwind 287declare ptr @cli_calloc(i64, i64) local_unnamed_addr #0 288 289; Function Attrs: argmemonly nounwind 290declare void @llvm.memcpy.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i1) #1 291attributes #0 = { nounwind } 292attributes #1 = { argmemonly nounwind } 293