|
252
|
1 ; RUN: opt < %s -S -passes=loop-unroll -unroll-runtime=true -unroll-runtime-epilog=true | FileCheck %s -check-prefixes=EPILOG,COMMON
|
|
|
2 ; RUN: opt < %s -S -passes=loop-unroll -unroll-runtime=true -unroll-runtime-epilog=false | FileCheck %s -check-prefixes=PROLOG,COMMON
|
|
150
|
3 ;
|
|
221
|
4 ; RUN: opt < %s -S -passes='require<opt-remark-emit>,loop-unroll' -unroll-runtime=true -unroll-runtime-epilog=true | FileCheck %s -check-prefixes=EPILOG,COMMON
|
|
|
5 ; RUN: opt < %s -S -passes='require<opt-remark-emit>,loop-unroll' -unroll-runtime=true -unroll-runtime-epilog=false | FileCheck %s -check-prefixes=PROLOG,COMMON
|
|
150
|
6 ;
|
|
|
7 ; Restricted versions of unroll (unroll<peeling;noruntime>, unroll-full) should not be doing runtime unrolling
|
|
|
8 ; even if it is globally enabled through -unroll-runtime option
|
|
|
9 ;
|
|
221
|
10 ; RUN: opt < %s -S -passes='require<opt-remark-emit>,loop-unroll<peeling;no-runtime>' -unroll-runtime=true -unroll-runtime-epilog=true | FileCheck %s -check-prefixes=NOEPILOG,COMMON
|
|
|
11 ; RUN: opt < %s -S -passes='require<opt-remark-emit>,loop-unroll<peeling;no-runtime>' -unroll-runtime=true -unroll-runtime-epilog=false | FileCheck %s -check-prefixes=NOPROLOG,COMMON
|
|
|
12 ; RUN: opt < %s -S -passes='require<opt-remark-emit>,loop(loop-unroll-full)' -unroll-runtime=true -unroll-runtime-epilog=true | FileCheck %s -check-prefixes=NOEPILOG,COMMON
|
|
|
13 ; RUN: opt < %s -S -passes='require<opt-remark-emit>,loop(loop-unroll-full)' -unroll-runtime=true -unroll-runtime-epilog=false | FileCheck %s -check-prefixes=NOPROLOG,COMMON
|
|
150
|
14
|
|
|
15 target 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"
|
|
|
16
|
|
|
17 ; Tests for unrolling loops with run-time trip counts
|
|
|
18
|
|
|
19 ; COMMON-LABEL: @test(
|
|
|
20
|
|
|
21 ; EPILOG: %xtraiter = and i32 %n
|
|
|
22 ; EPILOG: %lcmp.mod = icmp ne i32 %xtraiter, 0
|
|
|
23 ; EPILOG: br i1 %lcmp.mod, label %for.body.epil.preheader, label %for.end.loopexit
|
|
|
24
|
|
|
25 ; NOEPILOG-NOT: %xtraiter = and i32 %n
|
|
|
26
|
|
|
27 ; PROLOG: %xtraiter = and i32 %n
|
|
|
28 ; PROLOG: %lcmp.mod = icmp ne i32 %xtraiter, 0
|
|
|
29 ; PROLOG: br i1 %lcmp.mod, label %for.body.prol.preheader, label %for.body.prol.loopexit
|
|
|
30
|
|
|
31 ; NOPROLOG-NOT: %xtraiter = and i32 %n
|
|
|
32
|
|
|
33 ; EPILOG: for.body.epil:
|
|
|
34 ; EPILOG: %indvars.iv.epil = phi i64 [ %indvars.iv.next.epil, %for.body.epil ], [ %indvars.iv.unr, %for.body.epil.preheader ]
|
|
236
|
35 ; EPILOG: %epil.iter.next = add i32 %epil.iter, 1
|
|
|
36 ; EPILOG: %epil.iter.cmp = icmp ne i32 %epil.iter.next, %xtraiter
|
|
150
|
37 ; EPILOG: br i1 %epil.iter.cmp, label %for.body.epil, label %for.end.loopexit.epilog-lcssa, !llvm.loop !0
|
|
|
38
|
|
|
39 ; NOEPILOG: for.body:
|
|
|
40 ; NOEPILOG-NOT: for.body.epil:
|
|
|
41
|
|
|
42 ; PROLOG: for.body.prol:
|
|
|
43 ; PROLOG: %indvars.iv.prol = phi i64 [ %indvars.iv.next.prol, %for.body.prol ], [ 0, %for.body.prol.preheader ]
|
|
236
|
44 ; PROLOG: %prol.iter.next = add i32 %prol.iter, 1
|
|
|
45 ; PROLOG: %prol.iter.cmp = icmp ne i32 %prol.iter.next, %xtraiter
|
|
150
|
46 ; PROLOG: br i1 %prol.iter.cmp, label %for.body.prol, label %for.body.prol.loopexit.unr-lcssa, !llvm.loop !0
|
|
|
47
|
|
|
48 ; NOPROLOG: for.body:
|
|
|
49 ; NOPROLOG-NOT: for.body.prol:
|
|
|
50
|
|
|
51
|
|
252
|
52 define i32 @test(ptr nocapture %a, i32 %n) nounwind uwtable readonly {
|
|
150
|
53 entry:
|
|
|
54 %cmp1 = icmp eq i32 %n, 0
|
|
|
55 br i1 %cmp1, label %for.end, label %for.body
|
|
|
56
|
|
|
57 for.body: ; preds = %for.body, %entry
|
|
|
58 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
|
|
|
59 %sum.02 = phi i32 [ %add, %for.body ], [ 0, %entry ]
|
|
252
|
60 %arrayidx = getelementptr inbounds i32, ptr %a, i64 %indvars.iv
|
|
|
61 %0 = load i32, ptr %arrayidx, align 4
|
|
150
|
62 %add = add nsw i32 %0, %sum.02
|
|
|
63 %indvars.iv.next = add i64 %indvars.iv, 1
|
|
|
64 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
|
|
|
65 %exitcond = icmp eq i32 %lftr.wideiv, %n
|
|
|
66 br i1 %exitcond, label %for.end, label %for.body
|
|
|
67
|
|
|
68 for.end: ; preds = %for.body, %entry
|
|
|
69 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add, %for.body ]
|
|
|
70 ret i32 %sum.0.lcssa
|
|
|
71 }
|
|
|
72
|
|
|
73
|
|
|
74 ; Still try to completely unroll loops with compile-time trip counts
|
|
|
75 ; even if the -unroll-runtime is specified
|
|
|
76
|
|
|
77 ; COMMON-LABEL: @test1(
|
|
|
78 ; COMMON: for.body:
|
|
|
79 ; COMMON-NOT: for.body.epil:
|
|
|
80 ; COMMON-NOT: for.body.prol:
|
|
|
81
|
|
252
|
82 define i32 @test1(ptr nocapture %a) nounwind uwtable readonly {
|
|
150
|
83 entry:
|
|
|
84 br label %for.body
|
|
|
85
|
|
|
86 for.body: ; preds = %for.body, %entry
|
|
|
87 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
|
|
|
88 %sum.01 = phi i32 [ 0, %entry ], [ %add, %for.body ]
|
|
252
|
89 %arrayidx = getelementptr inbounds i32, ptr %a, i64 %indvars.iv
|
|
|
90 %0 = load i32, ptr %arrayidx, align 4
|
|
150
|
91 %add = add nsw i32 %0, %sum.01
|
|
|
92 %indvars.iv.next = add i64 %indvars.iv, 1
|
|
|
93 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
|
|
|
94 %exitcond = icmp eq i32 %lftr.wideiv, 5
|
|
|
95 br i1 %exitcond, label %for.end, label %for.body
|
|
|
96
|
|
|
97 for.end: ; preds = %for.body
|
|
|
98 ret i32 %add
|
|
|
99 }
|
|
|
100
|
|
|
101 ; This is test 2007-05-09-UnknownTripCount.ll which can be unrolled now
|
|
|
102 ; if the -unroll-runtime option is turned on
|
|
|
103
|
|
|
104 ; COMMON-LABEL: @foo(
|
|
|
105 ; EPILOG: bb72.2:
|
|
|
106 ; PROLOG: bb72.2:
|
|
|
107 ; NOEPILOG-NOT: bb72.2:
|
|
|
108 ; NOPROLOG-NOT: bb72.2:
|
|
|
109
|
|
|
110 define void @foo(i32 %trips) {
|
|
|
111 entry:
|
|
|
112 br label %cond_true.outer
|
|
|
113
|
|
|
114 cond_true.outer:
|
|
|
115 %indvar1.ph = phi i32 [ 0, %entry ], [ %indvar.next2, %bb72 ]
|
|
|
116 br label %bb72
|
|
|
117
|
|
|
118 bb72:
|
|
|
119 %indvar.next2 = add i32 %indvar1.ph, 1
|
|
|
120 %exitcond3 = icmp eq i32 %indvar.next2, %trips
|
|
|
121 br i1 %exitcond3, label %cond_true138, label %cond_true.outer
|
|
|
122
|
|
|
123 cond_true138:
|
|
|
124 ret void
|
|
|
125 }
|
|
|
126
|
|
|
127
|
|
|
128 ; Test run-time unrolling for a loop that counts down by -2.
|
|
|
129
|
|
|
130 ; COMMON-LABEL: @down(
|
|
|
131 ; EPILOG: for.body.epil:
|
|
|
132 ; EPILOG: br i1 %epil.iter.cmp, label %for.body.epil, label %for.cond.for.end_crit_edge.epilog-lcssa
|
|
|
133
|
|
|
134 ; NOEPILOG: for.body:
|
|
|
135 ; NOEPILOG-NOT: for.body.epil:
|
|
|
136
|
|
|
137 ; PROLOG: for.body.prol:
|
|
|
138 ; PROLOG: br i1 %prol.iter.cmp, label %for.body.prol, label %for.body.prol.loopexit
|
|
|
139
|
|
|
140 ; NOPROLOG: for.body:
|
|
|
141 ; NOPROLOG-NOT: for.body.prol:
|
|
|
142
|
|
252
|
143 define zeroext i16 @down(ptr nocapture %p, i32 %len) nounwind uwtable readonly {
|
|
150
|
144 entry:
|
|
|
145 %cmp2 = icmp eq i32 %len, 0
|
|
|
146 br i1 %cmp2, label %for.end, label %for.body
|
|
|
147
|
|
|
148 for.body: ; preds = %for.body, %entry
|
|
252
|
149 %p.addr.05 = phi ptr [ %incdec.ptr, %for.body ], [ %p, %entry ]
|
|
150
|
150 %len.addr.04 = phi i32 [ %sub, %for.body ], [ %len, %entry ]
|
|
|
151 %res.03 = phi i32 [ %add, %for.body ], [ 0, %entry ]
|
|
252
|
152 %incdec.ptr = getelementptr inbounds i16, ptr %p.addr.05, i64 1
|
|
|
153 %0 = load i16, ptr %p.addr.05, align 2
|
|
150
|
154 %conv = zext i16 %0 to i32
|
|
|
155 %add = add i32 %conv, %res.03
|
|
|
156 %sub = add nsw i32 %len.addr.04, -2
|
|
|
157 %cmp = icmp eq i32 %sub, 0
|
|
|
158 br i1 %cmp, label %for.cond.for.end_crit_edge, label %for.body
|
|
|
159
|
|
|
160 for.cond.for.end_crit_edge: ; preds = %for.body
|
|
|
161 %phitmp = trunc i32 %add to i16
|
|
|
162 br label %for.end
|
|
|
163
|
|
|
164 for.end: ; preds = %for.cond.for.end_crit_edge, %entry
|
|
|
165 %res.0.lcssa = phi i16 [ %phitmp, %for.cond.for.end_crit_edge ], [ 0, %entry ]
|
|
|
166 ret i16 %res.0.lcssa
|
|
|
167 }
|
|
|
168
|
|
|
169 ; Test run-time unrolling disable metadata.
|
|
|
170 ; COMMON-LABEL: @test2(
|
|
|
171
|
|
|
172 ; EPILOG: for.body:
|
|
|
173 ; EPILOG-NOT: for.body.epil:
|
|
|
174
|
|
|
175 ; NOEPILOG: for.body:
|
|
|
176 ; NOEPILOG-NOT: for.body.epil:
|
|
|
177
|
|
|
178 ; PROLOG: for.body:
|
|
|
179 ; PROLOG-NOT: for.body.prol:
|
|
|
180
|
|
|
181 ; NOPROLOG: for.body:
|
|
|
182 ; NOPROLOG-NOT: for.body.prol:
|
|
|
183
|
|
252
|
184 define zeroext i16 @test2(ptr nocapture %p, i32 %len) nounwind uwtable readonly {
|
|
150
|
185 entry:
|
|
|
186 %cmp2 = icmp eq i32 %len, 0
|
|
|
187 br i1 %cmp2, label %for.end, label %for.body
|
|
|
188
|
|
|
189 for.body: ; preds = %for.body, %entry
|
|
252
|
190 %p.addr.05 = phi ptr [ %incdec.ptr, %for.body ], [ %p, %entry ]
|
|
150
|
191 %len.addr.04 = phi i32 [ %sub, %for.body ], [ %len, %entry ]
|
|
|
192 %res.03 = phi i32 [ %add, %for.body ], [ 0, %entry ]
|
|
252
|
193 %incdec.ptr = getelementptr inbounds i16, ptr %p.addr.05, i64 1
|
|
|
194 %0 = load i16, ptr %p.addr.05, align 2
|
|
150
|
195 %conv = zext i16 %0 to i32
|
|
|
196 %add = add i32 %conv, %res.03
|
|
|
197 %sub = add nsw i32 %len.addr.04, -2
|
|
|
198 %cmp = icmp eq i32 %sub, 0
|
|
|
199 br i1 %cmp, label %for.cond.for.end_crit_edge, label %for.body, !llvm.loop !0
|
|
|
200
|
|
|
201 for.cond.for.end_crit_edge: ; preds = %for.body
|
|
|
202 %phitmp = trunc i32 %add to i16
|
|
|
203 br label %for.end
|
|
|
204
|
|
|
205 for.end: ; preds = %for.cond.for.end_crit_edge, %entry
|
|
|
206 %res.0.lcssa = phi i16 [ %phitmp, %for.cond.for.end_crit_edge ], [ 0, %entry ]
|
|
|
207 ret i16 %res.0.lcssa
|
|
|
208 }
|
|
|
209
|
|
|
210 ; dont unroll loop with multiple exit/exiting blocks, unless
|
|
|
211 ; -runtime-unroll-multi-exit=true
|
|
|
212 ; single exit, multiple exiting blocks.
|
|
|
213 define void @unique_exit(i32 %arg) {
|
|
|
214 ; COMMON-LABEL: @unique_exit(
|
|
|
215 ; COMMON-NOT: .unr
|
|
|
216
|
|
|
217 entry:
|
|
|
218 %tmp = icmp sgt i32 undef, %arg
|
|
|
219 br i1 %tmp, label %preheader, label %returnblock
|
|
|
220
|
|
|
221 preheader: ; preds = %entry
|
|
|
222 br label %header
|
|
|
223
|
|
|
224 LoopExit: ; preds = %header, %latch
|
|
|
225 %tmp2.ph = phi i32 [ %tmp4, %header ], [ -1, %latch ]
|
|
|
226 br label %returnblock
|
|
|
227
|
|
|
228 returnblock: ; preds = %LoopExit, %entry
|
|
|
229 %tmp2 = phi i32 [ -1, %entry ], [ %tmp2.ph, %LoopExit ]
|
|
|
230 ret void
|
|
|
231
|
|
|
232 header: ; preds = %preheader, %latch
|
|
|
233 %tmp4 = phi i32 [ %inc, %latch ], [ %arg, %preheader ]
|
|
|
234 %inc = add nsw i32 %tmp4, 1
|
|
|
235 br i1 true, label %LoopExit, label %latch
|
|
|
236
|
|
|
237 latch: ; preds = %header
|
|
|
238 %cmp = icmp slt i32 %inc, undef
|
|
|
239 br i1 %cmp, label %header, label %LoopExit
|
|
|
240 }
|
|
|
241
|
|
|
242 ; multiple exit blocks. don't unroll
|
|
|
243 define void @multi_exit(i64 %trip, i1 %cond) {
|
|
|
244 ; COMMON-LABEL: @multi_exit(
|
|
|
245 ; COMMON-NOT: .unr
|
|
|
246
|
|
|
247 entry:
|
|
|
248 br label %loop_header
|
|
|
249
|
|
|
250 loop_header:
|
|
|
251 %iv = phi i64 [ 0, %entry ], [ %iv_next, %loop_latch ]
|
|
|
252 br i1 %cond, label %loop_latch, label %loop_exiting_bb1
|
|
|
253
|
|
|
254 loop_exiting_bb1:
|
|
|
255 br i1 false, label %loop_exiting_bb2, label %exit1
|
|
|
256
|
|
|
257 loop_exiting_bb2:
|
|
|
258 br i1 false, label %loop_latch, label %exit3
|
|
|
259
|
|
|
260 exit3:
|
|
|
261 ret void
|
|
|
262
|
|
|
263 loop_latch:
|
|
|
264 %iv_next = add i64 %iv, 1
|
|
|
265 %cmp = icmp ne i64 %iv_next, %trip
|
|
|
266 br i1 %cmp, label %loop_header, label %exit2.loopexit
|
|
|
267
|
|
|
268 exit1:
|
|
|
269 ret void
|
|
|
270
|
|
|
271 exit2.loopexit:
|
|
|
272 ret void
|
|
|
273 }
|
|
|
274
|
|
|
275 !0 = distinct !{!0, !1}
|
|
|
276 !1 = !{!"llvm.loop.unroll.runtime.disable"}
|
|
|
277
|
|
|
278 ; need to use LABEL here to separate function IR matching from metadata matching
|
|
|
279 ; COMMON-LABEL: {{^}}!0 =
|
|
|
280
|
|
|
281 ; EPILOG-SAME: distinct !{!0, !1}
|
|
|
282 ; EPILOG: !1 = !{!"llvm.loop.unroll.disable"}
|
|
|
283
|
|
|
284 ; PROLOG-SAME: distinct !{!0, !1}
|
|
|
285 ; PROLOG: !1 = !{!"llvm.loop.unroll.disable"}
|
