248 lines · plain
1; RUN: opt -mtriple=thumbv8.1m.main-arm-eabihf -mattr=+mve.fp -passes=loop-vectorize -tail-predication=enabled -S < %s | \2; RUN: FileCheck %s3 4target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"5 6; Test that ARMTTIImpl::preferPredicateOverEpilogue triggers tail-folding.7 8define dso_local void @f1(ptr noalias nocapture %A, ptr noalias nocapture readonly %B, ptr noalias nocapture readonly %C, i32 %N) {9; CHECK-LABEL: f1(10; CHECK: entry:11; CHECK: @llvm.get.active.lane.mask12; CHECK: }13entry:14 %cmp8 = icmp sgt i32 %N, 015 br i1 %cmp8, label %for.body.preheader, label %for.cond.cleanup16 17for.body.preheader: ; preds = %entry18 br label %for.body19 20for.cond.cleanup.loopexit: ; preds = %for.body21 br label %for.cond.cleanup22 23for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry24 ret void25 26for.body: ; preds = %for.body.preheader, %for.body27 %i.09 = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]28 %arrayidx = getelementptr inbounds i32, ptr %B, i32 %i.0929 %0 = load i32, ptr %arrayidx, align 430 %arrayidx1 = getelementptr inbounds i32, ptr %C, i32 %i.0931 %1 = load i32, ptr %arrayidx1, align 432 %add = add nsw i32 %1, %033 %arrayidx2 = getelementptr inbounds i32, ptr %A, i32 %i.0934 store i32 %add, ptr %arrayidx2, align 435 %inc = add nuw nsw i32 %i.09, 136 %exitcond.not = icmp eq i32 %inc, %N37 br i1 %exitcond.not, label %for.cond.cleanup.loopexit, label %for.body38}39 40define dso_local void @f32_reduction(ptr nocapture readonly %Input, i32 %N, ptr nocapture %Output) {41; CHECK-LABEL: f32_reduction(42; CHECK: vector.body:43; CHECK: @llvm.masked.load44; CHECK: br i1 %{{.*}}, label {{.*}}, label %vector.body45entry:46 %cmp6 = icmp eq i32 %N, 047 br i1 %cmp6, label %while.end, label %while.body.preheader48 49while.body.preheader: ; preds = %entry50 br label %while.body51 52while.body: ; preds = %while.body.preheader, %while.body53 %blkCnt.09 = phi i32 [ %dec, %while.body ], [ %N, %while.body.preheader ]54 %sum.08 = phi float [ %add, %while.body ], [ 0.000000e+00, %while.body.preheader ]55 %Input.addr.07 = phi ptr [ %incdec.ptr, %while.body ], [ %Input, %while.body.preheader ]56 %incdec.ptr = getelementptr inbounds float, ptr %Input.addr.07, i32 157 %0 = load float, ptr %Input.addr.07, align 458 %add = fadd fast float %0, %sum.0859 %dec = add i32 %blkCnt.09, -160 %cmp = icmp eq i32 %dec, 061 br i1 %cmp, label %while.end.loopexit, label %while.body62 63while.end.loopexit: ; preds = %while.body64 %add.lcssa = phi float [ %add, %while.body ]65 br label %while.end66 67while.end: ; preds = %while.end.loopexit, %entry68 %sum.0.lcssa = phi float [ 0.000000e+00, %entry ], [ %add.lcssa, %while.end.loopexit ]69 %conv = uitofp i32 %N to float70 %div = fdiv fast float %sum.0.lcssa, %conv71 store float %div, ptr %Output, align 472 ret void73}74 75define dso_local void @f16_reduction(ptr nocapture readonly %Input, i32 %N, ptr nocapture %Output) {76; CHECK-LABEL: f16_reduction(77; CHECK: vector.body:78; CHECK: @llvm.masked.load79; CHECK: br i1 %{{.*}}, label {{.*}}, label %vector.body80entry:81 %cmp6 = icmp eq i32 %N, 082 br i1 %cmp6, label %while.end, label %while.body.preheader83 84while.body.preheader: ; preds = %entry85 br label %while.body86 87while.body: ; preds = %while.body.preheader, %while.body88 %blkCnt.09 = phi i32 [ %dec, %while.body ], [ %N, %while.body.preheader ]89 %sum.08 = phi half [ %add, %while.body ], [ 0.000000e+00, %while.body.preheader ]90 %Input.addr.07 = phi ptr [ %incdec.ptr, %while.body ], [ %Input, %while.body.preheader ]91 %incdec.ptr = getelementptr inbounds half, ptr %Input.addr.07, i32 192 %0 = load half, ptr %Input.addr.07, align 293 %add = fadd fast half %0, %sum.0894 %dec = add i32 %blkCnt.09, -195 %cmp = icmp eq i32 %dec, 096 br i1 %cmp, label %while.end.loopexit, label %while.body97 98while.end.loopexit: ; preds = %while.body99 %add.lcssa = phi half [ %add, %while.body ]100 br label %while.end101 102while.end: ; preds = %while.end.loopexit, %entry103 %sum.0.lcssa = phi half [ 0.000000e+00, %entry ], [ %add.lcssa, %while.end.loopexit ]104 %conv = uitofp i32 %N to half105 %div = fdiv fast half %sum.0.lcssa, %conv106 store half %div, ptr %Output, align 2107 ret void108}109 110define dso_local void @mixed_f32_i32_reduction(ptr nocapture readonly %fInput, ptr nocapture readonly %iInput, i32 %N, ptr nocapture %fOutput, ptr nocapture %iOutput) {111; CHECK-LABEL: mixed_f32_i32_reduction(112; CHECK: vector.body:113; CHECK: @llvm.masked.load114; CHECK: br i1 %{{.*}}, label {{.*}}, label %vector.body115entry:116 %cmp15 = icmp eq i32 %N, 0117 br i1 %cmp15, label %while.end, label %while.body.preheader118 119while.body.preheader:120 br label %while.body121 122while.body:123 %blkCnt.020 = phi i32 [ %dec, %while.body ], [ %N, %while.body.preheader ]124 %isum.019 = phi i32 [ %add2, %while.body ], [ 0, %while.body.preheader ]125 %fsum.018 = phi float [ %add, %while.body ], [ 0.000000e+00, %while.body.preheader ]126 %fInput.addr.017 = phi ptr [ %incdec.ptr, %while.body ], [ %fInput, %while.body.preheader ]127 %iInput.addr.016 = phi ptr [ %incdec.ptr1, %while.body ], [ %iInput, %while.body.preheader ]128 %incdec.ptr = getelementptr inbounds float, ptr %fInput.addr.017, i32 1129 %incdec.ptr1 = getelementptr inbounds i32, ptr %iInput.addr.016, i32 1130 %0 = load i32, ptr %iInput.addr.016, align 4131 %add2 = add nsw i32 %0, %isum.019132 %1 = load float, ptr %fInput.addr.017, align 4133 %add = fadd fast float %1, %fsum.018134 %dec = add i32 %blkCnt.020, -1135 %cmp = icmp eq i32 %dec, 0136 br i1 %cmp, label %while.end.loopexit, label %while.body137 138while.end.loopexit:139 %add.lcssa = phi float [ %add, %while.body ]140 %add2.lcssa = phi i32 [ %add2, %while.body ]141 %phitmp = sitofp i32 %add2.lcssa to float142 br label %while.end143 144while.end:145 %fsum.0.lcssa = phi float [ 0.000000e+00, %entry ], [ %add.lcssa, %while.end.loopexit ]146 %isum.0.lcssa = phi float [ 0.000000e+00, %entry ], [ %phitmp, %while.end.loopexit ]147 %conv = uitofp i32 %N to float148 %div = fdiv fast float %fsum.0.lcssa, %conv149 store float %div, ptr %fOutput, align 4150 %div5 = fdiv fast float %isum.0.lcssa, %conv151 %conv6 = fptosi float %div5 to i32152 store i32 %conv6, ptr %iOutput, align 4153 ret void154}155 156define dso_local i32 @i32_mul_reduction(ptr noalias nocapture readonly %B, i32 %N) {157; CHECK-LABEL: i32_mul_reduction(158; CHECK: vector.body:159; CHECK: @llvm.masked.load160; CHECK: br i1 %{{.*}}, label {{.*}}, label %vector.body161entry:162 %cmp6 = icmp sgt i32 %N, 0163 br i1 %cmp6, label %for.body.preheader, label %for.cond.cleanup164 165for.body.preheader:166 br label %for.body167 168for.cond.cleanup.loopexit:169 %mul.lcssa = phi i32 [ %mul, %for.body ]170 br label %for.cond.cleanup171 172for.cond.cleanup:173 %S.0.lcssa = phi i32 [ 1, %entry ], [ %mul.lcssa, %for.cond.cleanup.loopexit ]174 ret i32 %S.0.lcssa175 176for.body:177 %i.08 = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]178 %S.07 = phi i32 [ %mul, %for.body ], [ 1, %for.body.preheader ]179 %arrayidx = getelementptr inbounds i32, ptr %B, i32 %i.08180 %0 = load i32, ptr %arrayidx, align 4181 %mul = mul nsw i32 %0, %S.07182 %inc = add nuw nsw i32 %i.08, 1183 %exitcond = icmp eq i32 %inc, %N184 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body185}186 187define dso_local i32 @i32_or_reduction(ptr noalias nocapture readonly %B, i32 %N) {188; CHECK-LABEL: i32_or_reduction(189; CHECK: vector.body:190; CHECK: @llvm.masked.load191; CHECK: br i1 %{{.*}}, label {{.*}}, label %vector.body192entry:193 %cmp6 = icmp sgt i32 %N, 0194 br i1 %cmp6, label %for.body.preheader, label %for.cond.cleanup195 196for.body.preheader: ; preds = %entry197 br label %for.body198 199for.cond.cleanup.loopexit: ; preds = %for.body200 %or.lcssa = phi i32 [ %or, %for.body ]201 br label %for.cond.cleanup202 203for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry204 %S.0.lcssa = phi i32 [ 1, %entry ], [ %or.lcssa, %for.cond.cleanup.loopexit ]205 ret i32 %S.0.lcssa206 207for.body: ; preds = %for.body.preheader, %for.body208 %i.08 = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]209 %S.07 = phi i32 [ %or, %for.body ], [ 1, %for.body.preheader ]210 %arrayidx = getelementptr inbounds i32, ptr %B, i32 %i.08211 %0 = load i32, ptr %arrayidx, align 4212 %or = or i32 %0, %S.07213 %inc = add nuw nsw i32 %i.08, 1214 %exitcond = icmp eq i32 %inc, %N215 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body216}217 218define dso_local i32 @i32_and_reduction(ptr noalias nocapture readonly %A, i32 %N, i32 %S) {219; CHECK-LABEL: i32_and_reduction(220; CHECK: vector.body:221; CHECK: @llvm.masked.load222; CHECK: br i1 %{{.*}}, label {{.*}}, label %vector.body223entry:224 %cmp5 = icmp sgt i32 %N, 0225 br i1 %cmp5, label %for.body.preheader, label %for.cond.cleanup226 227for.body.preheader: ; preds = %entry228 br label %for.body229 230for.cond.cleanup.loopexit: ; preds = %for.body231 %and.lcssa = phi i32 [ %and, %for.body ]232 br label %for.cond.cleanup233 234for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry235 %S.addr.0.lcssa = phi i32 [ %S, %entry ], [ %and.lcssa, %for.cond.cleanup.loopexit ]236 ret i32 %S.addr.0.lcssa237 238for.body: ; preds = %for.body.preheader, %for.body239 %i.07 = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]240 %S.addr.06 = phi i32 [ %and, %for.body ], [ %S, %for.body.preheader ]241 %arrayidx = getelementptr inbounds i32, ptr %A, i32 %i.07242 %0 = load i32, ptr %arrayidx, align 4243 %and = and i32 %0, %S.addr.06244 %inc = add nuw nsw i32 %i.07, 1245 %exitcond = icmp eq i32 %inc, %N246 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body247}248