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