xref: /llvm-project/llvm/test/CodeGen/X86/tail-dup-merge-loop-headers.ll (revision a45f301f7a5d0f62910d0ed93c96d221555697c9)
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:    testl %edi, %edi
9; CHECK-NEXT:    jne .LBB0_2
10; CHECK-NEXT:    jmp .LBB0_5
11; CHECK-NEXT:  .LBB0_3: # %inner_loop_exit
12; CHECK-NEXT:    # in Loop: Header=BB0_2 Depth=1
13; CHECK-NEXT:    incq %rsi
14; CHECK-NEXT:    testl %edi, %edi
15; CHECK-NEXT:    je .LBB0_5
16; CHECK-NEXT:    .p2align 4, 0x90
17; CHECK-NEXT:  .LBB0_2: # %inner_loop_top
18; CHECK-NEXT:    # =>This Loop Header: Depth=1
19; CHECK-NEXT:    # Child Loop BB0_4 Depth 2
20; CHECK-NEXT:    cmpb $0, (%rsi)
21; CHECK-NEXT:    js .LBB0_3
22; CHECK-NEXT:  .LBB0_4: # %inner_loop_latch
23; CHECK-NEXT:    # Parent Loop BB0_2 Depth=1
24; CHECK-NEXT:    # => This Inner Loop Header: Depth=2
25; CHECK-NEXT:    addq $2, %rsi
26; CHECK-NEXT:    cmpb $0, (%rsi)
27; CHECK-NEXT:    jns .LBB0_4
28; CHECK-NEXT:    jmp .LBB0_3
29; CHECK-NEXT:  .LBB0_5: # %exit
30; CHECK-NEXT:    retq
31entry:
32  %notlhs674.i = icmp eq i32 %a, 0
33  br label %outer_loop_top
34
35outer_loop_top:                         ; preds = %inner_loop_exit, %entry
36  %dst.0.ph.i = phi i8* [ %b, %entry ], [ %scevgep679.i, %inner_loop_exit ]
37  br i1 %notlhs674.i, label %exit, label %inner_loop_preheader
38
39inner_loop_preheader:                           ; preds = %outer_loop_top
40  br label %inner_loop_top
41
42inner_loop_top:                                     ; preds = %inner_loop_latch, %inner_loop_preheader
43  %dst.0.i = phi i8* [ %inc, %inner_loop_latch ], [ %dst.0.ph.i, %inner_loop_preheader ]
44  %var = load i8, i8* %dst.0.i
45  %tobool1.i = icmp slt i8 %var, 0
46  br label %inner_loop_test
47
48inner_loop_test:                                       ; preds = %inner_loop_top
49  br i1 %tobool1.i, label %inner_loop_exit, label %inner_loop_latch
50
51inner_loop_exit:                       ; preds = %inner_loop_test
52  %scevgep.i = getelementptr i8, i8* %dst.0.i, i64 1
53  %scevgep679.i = getelementptr i8, i8* %scevgep.i, i64 0
54  br label %outer_loop_top
55
56inner_loop_latch:                                ; preds = %inner_loop_test
57  %cmp75.i = icmp ult i8* %dst.0.i, %c
58  %inc = getelementptr i8, i8* %dst.0.i, i64 2
59  br label %inner_loop_top
60
61exit:                              ; preds = %outer_loop_top
62  ret void
63}
64
65@.str.6 = external unnamed_addr constant [23 x i8], align 1
66
67; There is an erroneus check in LoopBase::addBasicBlockToLoop(), where it
68; assumes that the header block for a loop is unique.
69; For most of compilation this assumption is true, but during layout we allow
70; this assumption to be violated. The following code will trigger the bug:
71
72; The loops in question is eventually headed by the block shared_loop_header
73;
74; During layout The block labeled outer_loop_header gets tail-duplicated into
75; outer_loop_latch, and into shared_preheader, and then removed. This leaves
76; shared_loop_header as the header of both loops. The end result
77; is that there are 2 valid loops, and that they share a header. If we re-ran
78; the loop analysis, it would classify this as a single loop.
79; So far this is fine as far as layout is concerned.
80; After layout we tail merge blocks merge_other and merge_predecessor_split.
81; We do this even though they share only a single instruction, because
82; merge_predecessor_split falls through to their shared successor:
83; outer_loop_latch.
84; The rest of the blocks in the function are noise unfortunately. Bugpoint
85; couldn't shrink the test any further.
86
87define 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 {
88; CHECK-LABEL: loop_shared_header:
89; CHECK:       # %bb.0: # %entry
90; CHECK-NEXT:    pushq %rbp
91; CHECK-NEXT:    pushq %r15
92; CHECK-NEXT:    pushq %r14
93; CHECK-NEXT:    pushq %r13
94; CHECK-NEXT:    pushq %r12
95; CHECK-NEXT:    pushq %rbx
96; CHECK-NEXT:    pushq %rax
97; CHECK-NEXT:    movl $1, %ebx
98; CHECK-NEXT:    xorl %eax, %eax
99; CHECK-NEXT:    testb %al, %al
100; CHECK-NEXT:    jne .LBB1_26
101; CHECK-NEXT:  # %bb.1: # %if.end19
102; CHECK-NEXT:    movl %esi, %r13d
103; CHECK-NEXT:    movq %rdi, %r12
104; CHECK-NEXT:    movl (%rax), %ebp
105; CHECK-NEXT:    leal (,%rbp,4), %r14d
106; CHECK-NEXT:    movl %r14d, %r15d
107; CHECK-NEXT:    movl $1, %esi
108; CHECK-NEXT:    movq %r15, %rdi
109; CHECK-NEXT:    callq cli_calloc
110; CHECK-NEXT:    testl %r13d, %r13d
111; CHECK-NEXT:    je .LBB1_25
112; CHECK-NEXT:  # %bb.2: # %if.end19
113; CHECK-NEXT:    testl %ebp, %ebp
114; CHECK-NEXT:    je .LBB1_25
115; CHECK-NEXT:  # %bb.3: # %if.end19
116; CHECK-NEXT:    movq %rax, %rbx
117; CHECK-NEXT:    xorl %eax, %eax
118; CHECK-NEXT:    testb %al, %al
119; CHECK-NEXT:    jne .LBB1_25
120; CHECK-NEXT:  # %bb.4: # %if.end19
121; CHECK-NEXT:    cmpq %r12, %rbx
122; CHECK-NEXT:    jb .LBB1_25
123; CHECK-NEXT:  # %bb.5: # %if.end50
124; CHECK-NEXT:    movq %rbx, %rdi
125; CHECK-NEXT:    movq %r15, %rdx
126; CHECK-NEXT:    callq memcpy
127; CHECK-NEXT:    cmpl $4, %r14d
128; CHECK-NEXT:    jb .LBB1_28
129; CHECK-NEXT:  # %bb.6: # %shared_preheader
130; CHECK-NEXT:    movb $32, %dl
131; CHECK-NEXT:    xorl %eax, %eax
132; CHECK-NEXT:    # implicit-def: $rcx
133; CHECK-NEXT:    testl %ebp, %ebp
134; CHECK-NEXT:    je .LBB1_18
135; CHECK-NEXT:    .p2align 4, 0x90
136; CHECK-NEXT:  .LBB1_9: # %shared_loop_header
137; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
138; CHECK-NEXT:    testq %rbx, %rbx
139; CHECK-NEXT:    jne .LBB1_27
140; CHECK-NEXT:  # %bb.10: # %inner_loop_body
141; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
142; CHECK-NEXT:    testl %eax, %eax
143; CHECK-NEXT:    jns .LBB1_9
144; CHECK-NEXT:  # %bb.11: # %if.end96.i
145; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
146; CHECK-NEXT:    cmpl $3, %ebp
147; CHECK-NEXT:    jae .LBB1_22
148; CHECK-NEXT:  # %bb.12: # %if.end287.i
149; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
150; CHECK-NEXT:    xorl %esi, %esi
151; CHECK-NEXT:    cmpl $1, %ebp
152; CHECK-NEXT:    setne %dl
153; CHECK-NEXT:    testb %al, %al
154; CHECK-NEXT:    jne .LBB1_16
155; CHECK-NEXT:  # %bb.13: # %if.end308.i
156; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
157; CHECK-NEXT:    testb %al, %al
158; CHECK-NEXT:    je .LBB1_7
159; CHECK-NEXT:  # %bb.14: # %if.end335.i
160; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
161; CHECK-NEXT:    xorl %edx, %edx
162; CHECK-NEXT:    testb %dl, %dl
163; CHECK-NEXT:    movl $0, %esi
164; CHECK-NEXT:    jne .LBB1_8
165; CHECK-NEXT:  # %bb.15: # %merge_other
166; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
167; CHECK-NEXT:    xorl %esi, %esi
168; CHECK-NEXT:    jmp .LBB1_17
169; CHECK-NEXT:  .LBB1_16: # in Loop: Header=BB1_9 Depth=1
170; CHECK-NEXT:    movb %dl, %sil
171; CHECK-NEXT:    addl $3, %esi
172; CHECK-NEXT:  .LBB1_17: # %outer_loop_latch
173; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
174; CHECK-NEXT:    # implicit-def: $dl
175; CHECK-NEXT:    jmp .LBB1_8
176; CHECK-NEXT:  .LBB1_7: # %merge_predecessor_split
177; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
178; CHECK-NEXT:    movb $32, %dl
179; CHECK-NEXT:    xorl %esi, %esi
180; CHECK-NEXT:  .LBB1_8: # %outer_loop_latch
181; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=1
182; CHECK-NEXT:    movzwl %si, %esi
183; CHECK-NEXT:    decl %esi
184; CHECK-NEXT:    movzwl %si, %esi
185; CHECK-NEXT:    leaq 1(%rcx,%rsi), %rcx
186; CHECK-NEXT:    testl %ebp, %ebp
187; CHECK-NEXT:    jne .LBB1_9
188; CHECK-NEXT:  .LBB1_18: # %while.cond.us1412.i
189; CHECK-NEXT:    xorl %eax, %eax
190; CHECK-NEXT:    testb %al, %al
191; CHECK-NEXT:    movl $1, %ebx
192; CHECK-NEXT:    jne .LBB1_20
193; CHECK-NEXT:  # %bb.19: # %while.cond.us1412.i
194; CHECK-NEXT:    decb %dl
195; CHECK-NEXT:    jne .LBB1_26
196; CHECK-NEXT:  .LBB1_20: # %if.end41.us1436.i
197; CHECK-NEXT:  .LBB1_25:
198; CHECK-NEXT:    movl $1, %ebx
199; CHECK-NEXT:    jmp .LBB1_26
200; CHECK-NEXT:  .LBB1_22: # %if.then99.i
201; CHECK-NEXT:    xorl %ebx, %ebx
202; CHECK-NEXT:    movl $.str.6, %edi
203; CHECK-NEXT:    xorl %eax, %eax
204; CHECK-NEXT:    callq cli_dbgmsg
205; CHECK-NEXT:  .LBB1_26: # %cleanup
206; CHECK-NEXT:    movl %ebx, %eax
207; CHECK-NEXT:    addq $8, %rsp
208; CHECK-NEXT:    popq %rbx
209; CHECK-NEXT:    popq %r12
210; CHECK-NEXT:    popq %r13
211; CHECK-NEXT:    popq %r14
212; CHECK-NEXT:    popq %r15
213; CHECK-NEXT:    popq %rbp
214; CHECK-NEXT:    retq
215; CHECK-NEXT:  .LBB1_27: # %wunpsect.exit.thread.loopexit389
216; CHECK-NEXT:  .LBB1_28: # %wunpsect.exit.thread.loopexit391
217entry:
218  %0 = load i32, i32* undef, align 4
219  %mul = shl nsw i32 %0, 2
220  br i1 undef, label %if.end19, label %cleanup
221
222if.end19:                                         ; preds = %entry
223  %conv = zext i32 %mul to i64
224  %call = tail call i8* @cli_calloc(i64 %conv, i64 1)
225  %1 = icmp eq i32 %exesz, 0
226  %notrhs = icmp eq i32 %0, 0
227  %or.cond117.not = or i1 %1, %notrhs
228  %or.cond202 = or i1 %or.cond117.not, undef
229  %cmp35 = icmp ult i8* %call, %exe
230  %or.cond203 = or i1 %or.cond202, %cmp35
231  br i1 %or.cond203, label %cleanup, label %if.end50
232
233if.end50:                                         ; preds = %if.end19
234  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull %call, i8* undef, i64 %conv, i1 false)
235  %cmp1.i.i = icmp ugt i32 %mul, 3
236  br i1 %cmp1.i.i, label %shared_preheader, label %wunpsect.exit.thread.loopexit391
237
238shared_preheader:                                 ; preds = %if.end50
239  br label %outer_loop_header
240
241outer_loop_header:                                ; preds = %outer_loop_latch, %shared_preheader
242  %bits.1.i = phi i8 [ 32, %shared_preheader ], [ %bits.43.i, %outer_loop_latch ]
243  %dst.0.ph.i = phi i8* [ undef, %shared_preheader ], [ %scevgep679.i, %outer_loop_latch ]
244  %2 = icmp eq i32 %0, 0
245  br i1 %2, label %while.cond.us1412.i, label %shared_loop_header
246
247while.cond.us1412.i:                              ; preds = %outer_loop_header
248  %.pre.i = add i8 %bits.1.i, -1
249  %tobool2.us1420.i = icmp eq i8 %.pre.i, 0
250  %or.cond.us1421.i = or i1 undef, %tobool2.us1420.i
251  br i1 %or.cond.us1421.i, label %if.end41.us1436.i, label %cleanup
252
253if.end41.us1436.i:                                ; preds = %while.cond.us1412.i
254  unreachable
255
256shared_loop_header:                               ; preds = %dup_early2, %dup_early1
257  %dst.0.i = phi i8* [ undef, %inner_loop_body ], [ %dst.0.ph.i, %outer_loop_header ], [ undef, %dead_block ]
258  %cmp3.i1172.i = icmp ult i8* null, %call
259  br i1 %cmp3.i1172.i, label %wunpsect.exit.thread.loopexit389, label %inner_loop_body
260
261inner_loop_body:                                  ; preds = %shared_loop_header
262  %3 = icmp slt i32 undef, 0
263  br i1 %3, label %if.end96.i, label %shared_loop_header
264
265dead_block:                                       ; preds = %inner_loop_body
266  %cmp75.i = icmp ult i8* %dst.0.i, null
267  br label %shared_loop_header
268
269if.end96.i:                                       ; preds = %inner_loop_body
270  %cmp97.i = icmp ugt i32 %0, 2
271  br i1 %cmp97.i, label %if.then99.i, label %if.end287.i
272
273if.then99.i:                                      ; preds = %if.end96.i
274  tail call void (i8*, ...) @cli_dbgmsg(i8* getelementptr inbounds ([23 x i8], [23 x i8]* @.str.6, i64 0, i64 0), i32 undef)
275  br label %cleanup
276
277if.end287.i:                                      ; preds = %if.end96.i
278  %cmp291.i = icmp ne i32 %0, 1
279  %conv294.i = select i1 %cmp291.i, i16 4, i16 3
280  br i1 undef, label %if.end308.i, label %outer_loop_latch
281
282if.end308.i:                                      ; preds = %if.end287.i
283  br i1 undef, label %if.end335.i, label %merge_predecessor_split
284
285merge_predecessor_split:                          ; preds = %if.end308.i
286  %4 = bitcast i8* undef to i32*
287  br label %outer_loop_latch
288
289if.end335.i:                                      ; preds = %if.end308.i
290  br i1 undef, label %outer_loop_latch, label %merge_other
291
292merge_other:                                      ; preds = %if.end335.i
293  br label %outer_loop_latch
294
295outer_loop_latch:                                 ; preds = %merge_other, %if.end335.i, %merge_predecessor_split, %if.end287.i
296  %bits.43.i = phi i8 [ undef, %if.end287.i ], [ undef, %merge_other ], [ 32, %merge_predecessor_split ], [ 0, %if.end335.i ]
297  %backsize.0.i = phi i16 [ %conv294.i, %if.end287.i ], [ 0, %merge_other ], [ 0, %merge_predecessor_split ], [ 0, %if.end335.i ]
298  %5 = add i16 %backsize.0.i, -1
299  %6 = zext i16 %5 to i64
300  %scevgep.i = getelementptr i8, i8* %dst.0.ph.i, i64 1
301  %scevgep679.i = getelementptr i8, i8* %scevgep.i, i64 %6
302  br label %outer_loop_header
303
304wunpsect.exit.thread.loopexit389:                 ; preds = %shared_loop_header
305  unreachable
306
307wunpsect.exit.thread.loopexit391:                 ; preds = %if.end50
308  unreachable
309
310cleanup:                                          ; preds = %if.then99.i, %while.cond.us1412.i, %if.end19, %entry
311  %retval.0 = phi i32 [ 0, %if.then99.i ], [ 1, %entry ], [ 1, %if.end19 ], [ 1, %while.cond.us1412.i ]
312  ret i32 %retval.0
313}
314
315; Function Attrs: nounwind
316declare void @cli_dbgmsg(i8*, ...) local_unnamed_addr #0
317
318; Function Attrs: nounwind
319declare i8* @cli_calloc(i64, i64) local_unnamed_addr #0
320
321; Function Attrs: argmemonly nounwind
322declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i1) #1
323attributes #0 = { nounwind }
324attributes #1 = { argmemonly nounwind }
325