1; RUN: opt < %s -S -loop-unroll -unroll-runtime=true -unroll-runtime-epilog=true  | FileCheck %s -check-prefixes=EPILOG,COMMON
2; RUN: opt < %s -S -loop-unroll -unroll-runtime=true -unroll-runtime-epilog=false | FileCheck %s -check-prefixes=PROLOG,COMMON
3
4; RUN: opt < %s -S -passes='require<opt-remark-emit>,unroll' -unroll-runtime=true -unroll-runtime-epilog=true  | FileCheck %s -check-prefixes=EPILOG,COMMON
5; RUN: opt < %s -S -passes='require<opt-remark-emit>,unroll' -unroll-runtime=true -unroll-runtime-epilog=false | FileCheck %s -check-prefixes=PROLOG,COMMON
6
7target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
8
9; Tests for unrolling loops with run-time trip counts
10
11; COMMON-LABEL: @test(
12
13; EPILOG: %xtraiter = and i32 %n
14; EPILOG:  %lcmp.mod = icmp ne i32 %xtraiter, 0
15; EPILOG:  br i1 %lcmp.mod, label %for.body.epil.preheader, label %for.end.loopexit
16
17; PROLOG: %xtraiter = and i32 %n
18; PROLOG:  %lcmp.mod = icmp ne i32 %xtraiter, 0
19; PROLOG:  br i1 %lcmp.mod, label %for.body.prol.preheader, label %for.body.prol.loopexit
20
21; EPILOG: for.body.epil:
22; EPILOG: %indvars.iv.epil = phi i64 [ %indvars.iv.next.epil, %for.body.epil ],  [ %indvars.iv.unr, %for.body.epil.preheader ]
23; EPILOG:  %epil.iter.sub = sub i32 %epil.iter, 1
24; EPILOG:  %epil.iter.cmp = icmp ne i32 %epil.iter.sub, 0
25; EPILOG:  br i1 %epil.iter.cmp, label %for.body.epil, label %for.end.loopexit.epilog-lcssa, !llvm.loop !0
26
27; PROLOG: for.body.prol:
28; PROLOG: %indvars.iv.prol = phi i64 [ %indvars.iv.next.prol, %for.body.prol ], [ 0, %for.body.prol.preheader ]
29; PROLOG:  %prol.iter.sub = sub i32 %prol.iter, 1
30; PROLOG:  %prol.iter.cmp = icmp ne i32 %prol.iter.sub, 0
31; PROLOG:  br i1 %prol.iter.cmp, label %for.body.prol, label %for.body.prol.loopexit.unr-lcssa, !llvm.loop !0
32
33
34define i32 @test(i32* nocapture %a, i32 %n) nounwind uwtable readonly {
35entry:
36  %cmp1 = icmp eq i32 %n, 0
37  br i1 %cmp1, label %for.end, label %for.body
38
39for.body:                                         ; preds = %for.body, %entry
40  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
41  %sum.02 = phi i32 [ %add, %for.body ], [ 0, %entry ]
42  %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
43  %0 = load i32, i32* %arrayidx, align 4
44  %add = add nsw i32 %0, %sum.02
45  %indvars.iv.next = add i64 %indvars.iv, 1
46  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
47  %exitcond = icmp eq i32 %lftr.wideiv, %n
48  br i1 %exitcond, label %for.end, label %for.body
49
50for.end:                                          ; preds = %for.body, %entry
51  %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add, %for.body ]
52  ret i32 %sum.0.lcssa
53}
54
55
56; Still try to completely unroll loops with compile-time trip counts
57; even if the -unroll-runtime is specified
58
59; COMMON-LABEL: @test1(
60; COMMON: for.body:
61; COMMON-NOT: for.body.epil:
62; COMMON-NOT: for.body.prol:
63
64define i32 @test1(i32* nocapture %a) nounwind uwtable readonly {
65entry:
66  br label %for.body
67
68for.body:                                         ; preds = %for.body, %entry
69  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
70  %sum.01 = phi i32 [ 0, %entry ], [ %add, %for.body ]
71  %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
72  %0 = load i32, i32* %arrayidx, align 4
73  %add = add nsw i32 %0, %sum.01
74  %indvars.iv.next = add i64 %indvars.iv, 1
75  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
76  %exitcond = icmp eq i32 %lftr.wideiv, 5
77  br i1 %exitcond, label %for.end, label %for.body
78
79for.end:                                          ; preds = %for.body
80  ret i32 %add
81}
82
83; This is test 2007-05-09-UnknownTripCount.ll which can be unrolled now
84; if the -unroll-runtime option is turned on
85
86; COMMON-LABEL: @foo(
87; EPILOG: bb72.2:
88; PROLOG: bb72.2:
89
90define void @foo(i32 %trips) {
91entry:
92        br label %cond_true.outer
93
94cond_true.outer:
95        %indvar1.ph = phi i32 [ 0, %entry ], [ %indvar.next2, %bb72 ]
96        br label %bb72
97
98bb72:
99        %indvar.next2 = add i32 %indvar1.ph, 1
100        %exitcond3 = icmp eq i32 %indvar.next2, %trips
101        br i1 %exitcond3, label %cond_true138, label %cond_true.outer
102
103cond_true138:
104        ret void
105}
106
107
108; Test run-time unrolling for a loop that counts down by -2.
109
110; COMMON-LABEL: @down(
111; EPILOG: for.body.epil:
112; EPILOG: br i1 %epil.iter.cmp, label %for.body.epil, label %for.cond.for.end_crit_edge.epilog-lcssa
113
114; PROLOG: for.body.prol:
115; PROLOG: br i1 %prol.iter.cmp, label %for.body.prol, label %for.body.prol.loopexit
116
117define zeroext i16 @down(i16* nocapture %p, i32 %len) nounwind uwtable readonly {
118entry:
119  %cmp2 = icmp eq i32 %len, 0
120  br i1 %cmp2, label %for.end, label %for.body
121
122for.body:                                         ; preds = %for.body, %entry
123  %p.addr.05 = phi i16* [ %incdec.ptr, %for.body ], [ %p, %entry ]
124  %len.addr.04 = phi i32 [ %sub, %for.body ], [ %len, %entry ]
125  %res.03 = phi i32 [ %add, %for.body ], [ 0, %entry ]
126  %incdec.ptr = getelementptr inbounds i16, i16* %p.addr.05, i64 1
127  %0 = load i16, i16* %p.addr.05, align 2
128  %conv = zext i16 %0 to i32
129  %add = add i32 %conv, %res.03
130  %sub = add nsw i32 %len.addr.04, -2
131  %cmp = icmp eq i32 %sub, 0
132  br i1 %cmp, label %for.cond.for.end_crit_edge, label %for.body
133
134for.cond.for.end_crit_edge:                       ; preds = %for.body
135  %phitmp = trunc i32 %add to i16
136  br label %for.end
137
138for.end:                                          ; preds = %for.cond.for.end_crit_edge, %entry
139  %res.0.lcssa = phi i16 [ %phitmp, %for.cond.for.end_crit_edge ], [ 0, %entry ]
140  ret i16 %res.0.lcssa
141}
142
143; Test run-time unrolling disable metadata.
144; COMMON-LABEL: @test2(
145
146; EPILOG: for.body:
147; EPILOG-NOT: for.body.epil:
148
149; PROLOG: for.body:
150; PROLOG-NOT: for.body.prol:
151
152define zeroext i16 @test2(i16* nocapture %p, i32 %len) nounwind uwtable readonly {
153entry:
154  %cmp2 = icmp eq i32 %len, 0
155  br i1 %cmp2, label %for.end, label %for.body
156
157for.body:                                         ; preds = %for.body, %entry
158  %p.addr.05 = phi i16* [ %incdec.ptr, %for.body ], [ %p, %entry ]
159  %len.addr.04 = phi i32 [ %sub, %for.body ], [ %len, %entry ]
160  %res.03 = phi i32 [ %add, %for.body ], [ 0, %entry ]
161  %incdec.ptr = getelementptr inbounds i16, i16* %p.addr.05, i64 1
162  %0 = load i16, i16* %p.addr.05, align 2
163  %conv = zext i16 %0 to i32
164  %add = add i32 %conv, %res.03
165  %sub = add nsw i32 %len.addr.04, -2
166  %cmp = icmp eq i32 %sub, 0
167  br i1 %cmp, label %for.cond.for.end_crit_edge, label %for.body, !llvm.loop !0
168
169for.cond.for.end_crit_edge:                       ; preds = %for.body
170  %phitmp = trunc i32 %add to i16
171  br label %for.end
172
173for.end:                                          ; preds = %for.cond.for.end_crit_edge, %entry
174  %res.0.lcssa = phi i16 [ %phitmp, %for.cond.for.end_crit_edge ], [ 0, %entry ]
175  ret i16 %res.0.lcssa
176}
177
178; dont unroll loop with multiple exit/exiting blocks, unless
179; -runtime-unroll-multi-exit=true
180; single exit, multiple exiting blocks.
181define void @unique_exit(i32 %arg) {
182; COMMON-LABEL: @unique_exit(
183; COMMON-NOT: .unr
184
185entry:
186  %tmp = icmp sgt i32 undef, %arg
187  br i1 %tmp, label %preheader, label %returnblock
188
189preheader:                                 ; preds = %entry
190  br label %header
191
192LoopExit:                                ; preds = %header, %latch
193  %tmp2.ph = phi i32 [ %tmp4, %header ], [ -1, %latch ]
194  br label %returnblock
195
196returnblock:                                         ; preds = %LoopExit, %entry
197  %tmp2 = phi i32 [ -1, %entry ], [ %tmp2.ph, %LoopExit ]
198  ret void
199
200header:                                           ; preds = %preheader, %latch
201  %tmp4 = phi i32 [ %inc, %latch ], [ %arg, %preheader ]
202  %inc = add nsw i32 %tmp4, 1
203  br i1 true, label %LoopExit, label %latch
204
205latch:                                            ; preds = %header
206  %cmp = icmp slt i32 %inc, undef
207  br i1 %cmp, label %header, label %LoopExit
208}
209
210; multiple exit blocks. don't unroll
211define void @multi_exit(i64 %trip, i1 %cond) {
212; COMMON-LABEL: @multi_exit(
213; COMMON-NOT: .unr
214
215entry:
216  br label %loop_header
217
218loop_header:
219  %iv = phi i64 [ 0, %entry ], [ %iv_next, %loop_latch ]
220  br i1 %cond, label %loop_latch, label %loop_exiting_bb1
221
222loop_exiting_bb1:
223  br i1 false, label %loop_exiting_bb2, label %exit1
224
225loop_exiting_bb2:
226  br i1 false, label %loop_latch, label %exit3
227
228exit3:
229  ret void
230
231loop_latch:
232  %iv_next = add i64 %iv, 1
233  %cmp = icmp ne i64 %iv_next, %trip
234  br i1 %cmp, label %loop_header, label %exit2.loopexit
235
236exit1:
237 ret void
238
239exit2.loopexit:
240  ret void
241}
242
243!0 = distinct !{!0, !1}
244!1 = !{!"llvm.loop.unroll.runtime.disable"}
245
246; need to use LABEL here to separate function IR matching from metadata matching
247; COMMON-LABEL: {{^}}!0 =
248
249; EPILOG-SAME: distinct !{!0, !1}
250; EPILOG: !1 = !{!"llvm.loop.unroll.disable"}
251
252; PROLOG-SAME: distinct !{!0, !1}
253; PROLOG: !1 = !{!"llvm.loop.unroll.disable"}
254