xref: /llvm-project/llvm/test/CodeGen/X86/tail-dup-merge-loop-headers.ll (revision a701b7e368b70688bb4b84dafcaa43fa7c9a3649)
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_24
102; CHECK-NEXT:  # %bb.1: # %if.end19
103; CHECK-NEXT:    movl %esi, %ebp
104; CHECK-NEXT:    movq %rdi, %r15
105; CHECK-NEXT:    movl (%rax), %r13d
106; CHECK-NEXT:    leal (,%r13,4), %ebx
107; CHECK-NEXT:    movl %ebx, %r12d
108; CHECK-NEXT:    movl $1, %esi
109; CHECK-NEXT:    movq %r12, %rdi
110; CHECK-NEXT:    callq cli_calloc@PLT
111; CHECK-NEXT:    testl %ebp, %ebp
112; CHECK-NEXT:    je .LBB1_23
113; CHECK-NEXT:  # %bb.2: # %if.end19
114; CHECK-NEXT:    testl %r13d, %r13d
115; CHECK-NEXT:    je .LBB1_23
116; CHECK-NEXT:  # %bb.3: # %if.end19
117; CHECK-NEXT:    movq %rax, %r14
118; CHECK-NEXT:    xorl %eax, %eax
119; CHECK-NEXT:    testb %al, %al
120; CHECK-NEXT:    jne .LBB1_23
121; CHECK-NEXT:  # %bb.4: # %if.end19
122; CHECK-NEXT:    cmpq %r15, %r14
123; CHECK-NEXT:    jb .LBB1_23
124; CHECK-NEXT:  # %bb.5: # %if.end50
125; CHECK-NEXT:    movq %r14, %rdi
126; CHECK-NEXT:    movq %r12, %rdx
127; CHECK-NEXT:    callq memcpy@PLT
128; CHECK-NEXT:    cmpl $4, %ebx
129; CHECK-NEXT:    jb .LBB1_26
130; CHECK-NEXT:  # %bb.6: # %shared_preheader
131; CHECK-NEXT:    movb $32, %cl
132; CHECK-NEXT:    xorl %eax, %eax
133; CHECK-NEXT:    jmp .LBB1_8
134; CHECK-NEXT:    .p2align 4, 0x90
135; CHECK-NEXT:  .LBB1_7: # %merge_predecessor_split
136; CHECK-NEXT:    # in Loop: Header=BB1_8 Depth=1
137; CHECK-NEXT:    movb $32, %cl
138; CHECK-NEXT:  .LBB1_8: # %outer_loop_header
139; CHECK-NEXT:    # =>This Loop Header: Depth=1
140; CHECK-NEXT:    # Child Loop BB1_9 Depth 2
141; CHECK-NEXT:    testl %r13d, %r13d
142; CHECK-NEXT:    je .LBB1_16
143; CHECK-NEXT:    .p2align 4, 0x90
144; CHECK-NEXT:  .LBB1_9: # %shared_loop_header
145; CHECK-NEXT:    # Parent Loop BB1_8 Depth=1
146; CHECK-NEXT:    # => This Inner Loop Header: Depth=2
147; CHECK-NEXT:    testq %r14, %r14
148; CHECK-NEXT:    jne .LBB1_25
149; CHECK-NEXT:  # %bb.10: # %inner_loop_body
150; CHECK-NEXT:    # in Loop: Header=BB1_9 Depth=2
151; CHECK-NEXT:    testl %eax, %eax
152; CHECK-NEXT:    jns .LBB1_9
153; CHECK-NEXT:  # %bb.11: # %if.end96.i
154; CHECK-NEXT:    # in Loop: Header=BB1_8 Depth=1
155; CHECK-NEXT:    cmpl $3, %r13d
156; CHECK-NEXT:    jae .LBB1_20
157; CHECK-NEXT:  # %bb.12: # %if.end287.i
158; CHECK-NEXT:    # in Loop: Header=BB1_8 Depth=1
159; CHECK-NEXT:    testb %al, %al
160; CHECK-NEXT:    # implicit-def: $cl
161; CHECK-NEXT:    jne .LBB1_8
162; CHECK-NEXT:  # %bb.13: # %if.end308.i
163; CHECK-NEXT:    # in Loop: Header=BB1_8 Depth=1
164; CHECK-NEXT:    testb %al, %al
165; CHECK-NEXT:    je .LBB1_7
166; CHECK-NEXT:  # %bb.14: # %if.end335.i
167; CHECK-NEXT:    # in Loop: Header=BB1_8 Depth=1
168; CHECK-NEXT:    xorl %ecx, %ecx
169; CHECK-NEXT:    testb %cl, %cl
170; CHECK-NEXT:    jne .LBB1_8
171; CHECK-NEXT:  # %bb.15: # %merge_other
172; CHECK-NEXT:    # in Loop: Header=BB1_8 Depth=1
173; CHECK-NEXT:    # implicit-def: $cl
174; CHECK-NEXT:    jmp .LBB1_8
175; CHECK-NEXT:  .LBB1_23:
176; CHECK-NEXT:    movl $1, %ebx
177; CHECK-NEXT:    jmp .LBB1_24
178; CHECK-NEXT:  .LBB1_16: # %while.cond.us1412.i
179; CHECK-NEXT:    xorl %eax, %eax
180; CHECK-NEXT:    testb %al, %al
181; CHECK-NEXT:    movl $1, %ebx
182; CHECK-NEXT:    jne .LBB1_18
183; CHECK-NEXT:  # %bb.17: # %while.cond.us1412.i
184; CHECK-NEXT:    decb %cl
185; CHECK-NEXT:    jne .LBB1_24
186; CHECK-NEXT:  .LBB1_18: # %if.end41.us1436.i
187; CHECK-NEXT:  .LBB1_20: # %if.then99.i
188; CHECK-NEXT:    movq .str.6@GOTPCREL(%rip), %rdi
189; CHECK-NEXT:    xorl %ebx, %ebx
190; CHECK-NEXT:    xorl %eax, %eax
191; CHECK-NEXT:    callq cli_dbgmsg@PLT
192; CHECK-NEXT:  .LBB1_24: # %cleanup
193; CHECK-NEXT:    movl %ebx, %eax
194; CHECK-NEXT:    addq $8, %rsp
195; CHECK-NEXT:    popq %rbx
196; CHECK-NEXT:    popq %r12
197; CHECK-NEXT:    popq %r13
198; CHECK-NEXT:    popq %r14
199; CHECK-NEXT:    popq %r15
200; CHECK-NEXT:    popq %rbp
201; CHECK-NEXT:    retq
202; CHECK-NEXT:  .LBB1_25: # %wunpsect.exit.thread.loopexit389
203; CHECK-NEXT:  .LBB1_26: # %wunpsect.exit.thread.loopexit391
204entry:
205  %0 = load i32, i32* undef, align 4
206  %mul = shl nsw i32 %0, 2
207  br i1 undef, label %if.end19, label %cleanup
208
209if.end19:                                         ; preds = %entry
210  %conv = zext i32 %mul to i64
211  %call = tail call i8* @cli_calloc(i64 %conv, i64 1)
212  %1 = icmp eq i32 %exesz, 0
213  %notrhs = icmp eq i32 %0, 0
214  %or.cond117.not = or i1 %1, %notrhs
215  %or.cond202 = or i1 %or.cond117.not, undef
216  %cmp35 = icmp ult i8* %call, %exe
217  %or.cond203 = or i1 %or.cond202, %cmp35
218  br i1 %or.cond203, label %cleanup, label %if.end50
219
220if.end50:                                         ; preds = %if.end19
221  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull %call, i8* undef, i64 %conv, i1 false)
222  %cmp1.i.i = icmp ugt i32 %mul, 3
223  br i1 %cmp1.i.i, label %shared_preheader, label %wunpsect.exit.thread.loopexit391
224
225shared_preheader:                                 ; preds = %if.end50
226  br label %outer_loop_header
227
228outer_loop_header:                                ; preds = %outer_loop_latch, %shared_preheader
229  %bits.1.i = phi i8 [ 32, %shared_preheader ], [ %bits.43.i, %outer_loop_latch ]
230  %dst.0.ph.i = phi i8* [ undef, %shared_preheader ], [ %scevgep679.i, %outer_loop_latch ]
231  %2 = icmp eq i32 %0, 0
232  br i1 %2, label %while.cond.us1412.i, label %shared_loop_header
233
234while.cond.us1412.i:                              ; preds = %outer_loop_header
235  %.pre.i = add i8 %bits.1.i, -1
236  %tobool2.us1420.i = icmp eq i8 %.pre.i, 0
237  %or.cond.us1421.i = or i1 undef, %tobool2.us1420.i
238  br i1 %or.cond.us1421.i, label %if.end41.us1436.i, label %cleanup
239
240if.end41.us1436.i:                                ; preds = %while.cond.us1412.i
241  unreachable
242
243shared_loop_header:                               ; preds = %dup_early2, %dup_early1
244  %dst.0.i = phi i8* [ undef, %inner_loop_body ], [ %dst.0.ph.i, %outer_loop_header ], [ undef, %dead_block ]
245  %cmp3.i1172.i = icmp ult i8* null, %call
246  br i1 %cmp3.i1172.i, label %wunpsect.exit.thread.loopexit389, label %inner_loop_body
247
248inner_loop_body:                                  ; preds = %shared_loop_header
249  %3 = icmp slt i32 undef, 0
250  br i1 %3, label %if.end96.i, label %shared_loop_header
251
252dead_block:                                       ; preds = %inner_loop_body
253  %cmp75.i = icmp ult i8* %dst.0.i, null
254  br label %shared_loop_header
255
256if.end96.i:                                       ; preds = %inner_loop_body
257  %cmp97.i = icmp ugt i32 %0, 2
258  br i1 %cmp97.i, label %if.then99.i, label %if.end287.i
259
260if.then99.i:                                      ; preds = %if.end96.i
261  tail call void (i8*, ...) @cli_dbgmsg(i8* getelementptr inbounds ([23 x i8], [23 x i8]* @.str.6, i64 0, i64 0), i32 undef)
262  br label %cleanup
263
264if.end287.i:                                      ; preds = %if.end96.i
265  %cmp291.i = icmp ne i32 %0, 1
266  %conv294.i = select i1 %cmp291.i, i16 4, i16 3
267  br i1 undef, label %if.end308.i, label %outer_loop_latch
268
269if.end308.i:                                      ; preds = %if.end287.i
270  br i1 undef, label %if.end335.i, label %merge_predecessor_split
271
272merge_predecessor_split:                          ; preds = %if.end308.i
273  %4 = bitcast i8* undef to i32*
274  br label %outer_loop_latch
275
276if.end335.i:                                      ; preds = %if.end308.i
277  br i1 undef, label %outer_loop_latch, label %merge_other
278
279merge_other:                                      ; preds = %if.end335.i
280  br label %outer_loop_latch
281
282outer_loop_latch:                                 ; preds = %merge_other, %if.end335.i, %merge_predecessor_split, %if.end287.i
283  %bits.43.i = phi i8 [ undef, %if.end287.i ], [ undef, %merge_other ], [ 32, %merge_predecessor_split ], [ 0, %if.end335.i ]
284  %backsize.0.i = phi i16 [ %conv294.i, %if.end287.i ], [ 0, %merge_other ], [ 0, %merge_predecessor_split ], [ 0, %if.end335.i ]
285  %5 = add i16 %backsize.0.i, -1
286  %6 = zext i16 %5 to i64
287  %scevgep.i = getelementptr i8, i8* %dst.0.ph.i, i64 1
288  %scevgep679.i = getelementptr i8, i8* %scevgep.i, i64 %6
289  br label %outer_loop_header
290
291wunpsect.exit.thread.loopexit389:                 ; preds = %shared_loop_header
292  unreachable
293
294wunpsect.exit.thread.loopexit391:                 ; preds = %if.end50
295  unreachable
296
297cleanup:                                          ; preds = %if.then99.i, %while.cond.us1412.i, %if.end19, %entry
298  %retval.0 = phi i32 [ 0, %if.then99.i ], [ 1, %entry ], [ 1, %if.end19 ], [ 1, %while.cond.us1412.i ]
299  ret i32 %retval.0
300}
301
302; Function Attrs: nounwind
303declare void @cli_dbgmsg(i8*, ...) local_unnamed_addr #0
304
305; Function Attrs: nounwind
306declare i8* @cli_calloc(i64, i64) local_unnamed_addr #0
307
308; Function Attrs: argmemonly nounwind
309declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i1) #1
310attributes #0 = { nounwind }
311attributes #1 = { argmemonly nounwind }
312