xref: /llvm-project/llvm/test/CodeGen/X86/tail-dup-merge-loop-headers.ll (revision 0aef747b846586c29ed3285bbed20a3d607576fa)
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, i8* %b, i8* %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, 0x90
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, 0x90
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 i8* [ %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 i8* [ %inc, %inner_loop_latch ], [ %dst.0.ph.i, %inner_loop_preheader ]
45  %var = load i8, i8* %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, i8* %dst.0.i, i64 1
54  %scevgep679.i = getelementptr i8, i8* %scevgep.i, i64 0
55  br label %outer_loop_top
56
57inner_loop_latch:                                ; preds = %inner_loop_test
58  %cmp75.i = icmp ult i8* %dst.0.i, %c
59  %inc = getelementptr i8, i8* %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(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 {
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 %r13
95; CHECK-NEXT:    pushq %r12
96; CHECK-NEXT:    pushq %rbx
97; CHECK-NEXT:    pushq %rax
98; CHECK-NEXT:    movl $1, %ebx
99; CHECK-NEXT:    xorl %eax, %eax
100; CHECK-NEXT:    testb %al, %al
101; CHECK-NEXT:    jne .LBB1_27
102; CHECK-NEXT:  # %bb.1: # %if.end19
103; CHECK-NEXT:    movl %esi, %r13d
104; CHECK-NEXT:    movq %rdi, %r12
105; CHECK-NEXT:    movl (%rax), %ebp
106; CHECK-NEXT:    leal (,%rbp,4), %r14d
107; CHECK-NEXT:    movl %r14d, %r15d
108; CHECK-NEXT:    movl $1, %esi
109; CHECK-NEXT:    movq %r15, %rdi
110; CHECK-NEXT:    callq cli_calloc@PLT
111; CHECK-NEXT:    testl %r13d, %r13d
112; CHECK-NEXT:    je .LBB1_26
113; CHECK-NEXT:  # %bb.2: # %if.end19
114; CHECK-NEXT:    testl %ebp, %ebp
115; CHECK-NEXT:    je .LBB1_26
116; CHECK-NEXT:  # %bb.3: # %if.end19
117; CHECK-NEXT:    movq %rax, %rbx
118; CHECK-NEXT:    xorl %eax, %eax
119; CHECK-NEXT:    testb %al, %al
120; CHECK-NEXT:    jne .LBB1_26
121; CHECK-NEXT:  # %bb.4: # %if.end19
122; CHECK-NEXT:    cmpq %r12, %rbx
123; CHECK-NEXT:    jb .LBB1_26
124; CHECK-NEXT:  # %bb.5: # %if.end50
125; CHECK-NEXT:    movq %rbx, %rdi
126; CHECK-NEXT:    movq %r15, %rdx
127; CHECK-NEXT:    callq memcpy@PLT
128; CHECK-NEXT:    cmpl $4, %r14d
129; CHECK-NEXT:    jb .LBB1_29
130; CHECK-NEXT:  # %bb.6: # %shared_preheader
131; CHECK-NEXT:    movb $32, %dl
132; CHECK-NEXT:    xorl %eax, %eax
133; CHECK-NEXT:    # implicit-def: $rcx
134; CHECK-NEXT:    jmp .LBB1_9
135; CHECK-NEXT:    .p2align 4, 0x90
136; CHECK-NEXT:  .LBB1_7: # %merge_predecessor_split
137; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
138; CHECK-NEXT:    movb $32, %dl
139; CHECK-NEXT:    xorl %esi, %esi
140; CHECK-NEXT:  .LBB1_8: # %outer_loop_latch
141; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
142; CHECK-NEXT:    movzwl %si, %esi
143; CHECK-NEXT:    decl %esi
144; CHECK-NEXT:    movzwl %si, %esi
145; CHECK-NEXT:    leaq 1(%rcx,%rsi), %rcx
146; CHECK-NEXT:  .LBB1_9: # %outer_loop_header
147; CHECK-NEXT:    # =>This Loop Header: Depth=1
148; CHECK-NEXT:    # Child Loop BB1_10 Depth 2
149; CHECK-NEXT:    testl %ebp, %ebp
150; CHECK-NEXT:    je .LBB1_19
151; CHECK-NEXT:    .p2align 4, 0x90
152; CHECK-NEXT:  .LBB1_10: # %shared_loop_header
153; CHECK-NEXT:    # Parent Loop BB1_9 Depth=1
154; CHECK-NEXT:    # => This Inner Loop Header: Depth=2
155; CHECK-NEXT:    testq %rbx, %rbx
156; CHECK-NEXT:    jne .LBB1_28
157; CHECK-NEXT:  # %bb.11: # %inner_loop_body
158; CHECK-NEXT:    # in Loop: Header=BB1_10 Depth=2
159; CHECK-NEXT:    testl %eax, %eax
160; CHECK-NEXT:    jns .LBB1_10
161; CHECK-NEXT:  # %bb.12: # %if.end96.i
162; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
163; CHECK-NEXT:    cmpl $3, %ebp
164; CHECK-NEXT:    jae .LBB1_23
165; CHECK-NEXT:  # %bb.13: # %if.end287.i
166; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
167; CHECK-NEXT:    xorl %esi, %esi
168; CHECK-NEXT:    cmpl $1, %ebp
169; CHECK-NEXT:    setne %dl
170; CHECK-NEXT:    testb %al, %al
171; CHECK-NEXT:    jne .LBB1_17
172; CHECK-NEXT:  # %bb.14: # %if.end308.i
173; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
174; CHECK-NEXT:    testb %al, %al
175; CHECK-NEXT:    je .LBB1_7
176; CHECK-NEXT:  # %bb.15: # %if.end335.i
177; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
178; CHECK-NEXT:    xorl %edx, %edx
179; CHECK-NEXT:    testb %dl, %dl
180; CHECK-NEXT:    movl $0, %esi
181; CHECK-NEXT:    jne .LBB1_8
182; CHECK-NEXT:  # %bb.16: # %merge_other
183; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
184; CHECK-NEXT:    xorl %esi, %esi
185; CHECK-NEXT:    jmp .LBB1_18
186; CHECK-NEXT:    .p2align 4, 0x90
187; CHECK-NEXT:  .LBB1_17: # in Loop: Header=BB1_9 Depth=1
188; CHECK-NEXT:    movb %dl, %sil
189; CHECK-NEXT:    addl $3, %esi
190; CHECK-NEXT:  .LBB1_18: # %outer_loop_latch
191; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
192; CHECK-NEXT:    # implicit-def: $dl
193; CHECK-NEXT:    jmp .LBB1_8
194; CHECK-NEXT:  .LBB1_26:
195; CHECK-NEXT:    movl $1, %ebx
196; CHECK-NEXT:    jmp .LBB1_27
197; CHECK-NEXT:  .LBB1_19: # %while.cond.us1412.i
198; CHECK-NEXT:    xorl %eax, %eax
199; CHECK-NEXT:    testb %al, %al
200; CHECK-NEXT:    movl $1, %ebx
201; CHECK-NEXT:    jne .LBB1_21
202; CHECK-NEXT:  # %bb.20: # %while.cond.us1412.i
203; CHECK-NEXT:    decb %dl
204; CHECK-NEXT:    jne .LBB1_27
205; CHECK-NEXT:  .LBB1_21: # %if.end41.us1436.i
206; CHECK-NEXT:  .LBB1_23: # %if.then99.i
207; CHECK-NEXT:    movq .str.6@GOTPCREL(%rip), %rdi
208; CHECK-NEXT:    xorl %ebx, %ebx
209; CHECK-NEXT:    xorl %eax, %eax
210; CHECK-NEXT:    callq cli_dbgmsg@PLT
211; CHECK-NEXT:  .LBB1_27: # %cleanup
212; CHECK-NEXT:    movl %ebx, %eax
213; CHECK-NEXT:    addq $8, %rsp
214; CHECK-NEXT:    popq %rbx
215; CHECK-NEXT:    popq %r12
216; CHECK-NEXT:    popq %r13
217; CHECK-NEXT:    popq %r14
218; CHECK-NEXT:    popq %r15
219; CHECK-NEXT:    popq %rbp
220; CHECK-NEXT:    retq
221; CHECK-NEXT:  .LBB1_28: # %wunpsect.exit.thread.loopexit389
222; CHECK-NEXT:  .LBB1_29: # %wunpsect.exit.thread.loopexit391
223entry:
224  %0 = load i32, i32* undef, align 4
225  %mul = shl nsw i32 %0, 2
226  br i1 undef, label %if.end19, label %cleanup
227
228if.end19:                                         ; preds = %entry
229  %conv = zext i32 %mul to i64
230  %call = tail call i8* @cli_calloc(i64 %conv, i64 1)
231  %1 = icmp eq i32 %exesz, 0
232  %notrhs = icmp eq i32 %0, 0
233  %or.cond117.not = or i1 %1, %notrhs
234  %or.cond202 = or i1 %or.cond117.not, undef
235  %cmp35 = icmp ult i8* %call, %exe
236  %or.cond203 = or i1 %or.cond202, %cmp35
237  br i1 %or.cond203, label %cleanup, label %if.end50
238
239if.end50:                                         ; preds = %if.end19
240  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull %call, i8* undef, i64 %conv, i1 false)
241  %cmp1.i.i = icmp ugt i32 %mul, 3
242  br i1 %cmp1.i.i, label %shared_preheader, label %wunpsect.exit.thread.loopexit391
243
244shared_preheader:                                 ; preds = %if.end50
245  br label %outer_loop_header
246
247outer_loop_header:                                ; preds = %outer_loop_latch, %shared_preheader
248  %bits.1.i = phi i8 [ 32, %shared_preheader ], [ %bits.43.i, %outer_loop_latch ]
249  %dst.0.ph.i = phi i8* [ undef, %shared_preheader ], [ %scevgep679.i, %outer_loop_latch ]
250  %2 = icmp eq i32 %0, 0
251  br i1 %2, label %while.cond.us1412.i, label %shared_loop_header
252
253while.cond.us1412.i:                              ; preds = %outer_loop_header
254  %.pre.i = add i8 %bits.1.i, -1
255  %tobool2.us1420.i = icmp eq i8 %.pre.i, 0
256  %or.cond.us1421.i = or i1 undef, %tobool2.us1420.i
257  br i1 %or.cond.us1421.i, label %if.end41.us1436.i, label %cleanup
258
259if.end41.us1436.i:                                ; preds = %while.cond.us1412.i
260  unreachable
261
262shared_loop_header:                               ; preds = %dup_early2, %dup_early1
263  %dst.0.i = phi i8* [ undef, %inner_loop_body ], [ %dst.0.ph.i, %outer_loop_header ], [ undef, %dead_block ]
264  %cmp3.i1172.i = icmp ult i8* null, %call
265  br i1 %cmp3.i1172.i, label %wunpsect.exit.thread.loopexit389, label %inner_loop_body
266
267inner_loop_body:                                  ; preds = %shared_loop_header
268  %3 = icmp slt i32 undef, 0
269  br i1 %3, label %if.end96.i, label %shared_loop_header
270
271dead_block:                                       ; preds = %inner_loop_body
272  %cmp75.i = icmp ult i8* %dst.0.i, null
273  br label %shared_loop_header
274
275if.end96.i:                                       ; preds = %inner_loop_body
276  %cmp97.i = icmp ugt i32 %0, 2
277  br i1 %cmp97.i, label %if.then99.i, label %if.end287.i
278
279if.then99.i:                                      ; preds = %if.end96.i
280  tail call void (i8*, ...) @cli_dbgmsg(i8* getelementptr inbounds ([23 x i8], [23 x i8]* @.str.6, i64 0, i64 0), i32 undef)
281  br label %cleanup
282
283if.end287.i:                                      ; preds = %if.end96.i
284  %cmp291.i = icmp ne i32 %0, 1
285  %conv294.i = select i1 %cmp291.i, i16 4, i16 3
286  br i1 undef, label %if.end308.i, label %outer_loop_latch
287
288if.end308.i:                                      ; preds = %if.end287.i
289  br i1 undef, label %if.end335.i, label %merge_predecessor_split
290
291merge_predecessor_split:                          ; preds = %if.end308.i
292  %4 = bitcast i8* undef to i32*
293  br label %outer_loop_latch
294
295if.end335.i:                                      ; preds = %if.end308.i
296  br i1 undef, label %outer_loop_latch, label %merge_other
297
298merge_other:                                      ; preds = %if.end335.i
299  br label %outer_loop_latch
300
301outer_loop_latch:                                 ; preds = %merge_other, %if.end335.i, %merge_predecessor_split, %if.end287.i
302  %bits.43.i = phi i8 [ undef, %if.end287.i ], [ undef, %merge_other ], [ 32, %merge_predecessor_split ], [ 0, %if.end335.i ]
303  %backsize.0.i = phi i16 [ %conv294.i, %if.end287.i ], [ 0, %merge_other ], [ 0, %merge_predecessor_split ], [ 0, %if.end335.i ]
304  %5 = add i16 %backsize.0.i, -1
305  %6 = zext i16 %5 to i64
306  %scevgep.i = getelementptr i8, i8* %dst.0.ph.i, i64 1
307  %scevgep679.i = getelementptr i8, i8* %scevgep.i, i64 %6
308  br label %outer_loop_header
309
310wunpsect.exit.thread.loopexit389:                 ; preds = %shared_loop_header
311  unreachable
312
313wunpsect.exit.thread.loopexit391:                 ; preds = %if.end50
314  unreachable
315
316cleanup:                                          ; preds = %if.then99.i, %while.cond.us1412.i, %if.end19, %entry
317  %retval.0 = phi i32 [ 0, %if.then99.i ], [ 1, %entry ], [ 1, %if.end19 ], [ 1, %while.cond.us1412.i ]
318  ret i32 %retval.0
319}
320
321; Function Attrs: nounwind
322declare void @cli_dbgmsg(i8*, ...) local_unnamed_addr #0
323
324; Function Attrs: nounwind
325declare i8* @cli_calloc(i64, i64) local_unnamed_addr #0
326
327; Function Attrs: argmemonly nounwind
328declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i1) #1
329attributes #0 = { nounwind }
330attributes #1 = { argmemonly nounwind }
331