109 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 %s -check-prefix=CHECK3 4; RUN: opt -mtriple=thumbv8.1m.main-arm-eabihf -mattr=+mve.fp -passes=loop-vectorize -tail-predication=enabled \5; RUN: -prefer-predicate-over-epilogue=predicate-dont-vectorize -S < %s | \6; RUN: FileCheck -check-prefix=PREDFLAG %s7 8target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"9 10; This test has a loop hint "predicate.predicate" set to false, so shouldn't11; get tail-folded, except with -prefer-predicate-over-epilog which then12; overrules this.13;14define dso_local void @flag_overrules_hint(ptr noalias nocapture %A, ptr noalias nocapture readonly %B, ptr noalias nocapture readonly %C) local_unnamed_addr #0 {15; CHECK-LABEL: flag_overrules_hint(16; CHECK: vector.body:17; CHECK-NOT: @llvm.masked.load.v8i32.p0(18; CHECK-NOT: @llvm.masked.store.v8i32.p0(19; CHECK: br i1 %{{.*}}, label {{.*}}, label %vector.body20 21; PREDFLAG-LABEL: flag_overrules_hint(22; PREDFLAG: vector.body:23; PREDFLAG: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]24; PREDFLAG: %active.lane.mask = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i64(i64 %index, i64 430)25; PREDFLAG: %wide.masked.load = call <4 x i32> @llvm.masked.load.v4i32.p0({{.*}}, <4 x i1> %active.lane.mask26; PREDFLAG: %wide.masked.load1 = call <4 x i32> @llvm.masked.load.v4i32.p0({{.*}}, <4 x i1> %active.lane.mask27; PREDFLAG: %{{.*}} = add nsw <4 x i32> %wide.masked.load1, %wide.masked.load28; PREDFLAG: call void @llvm.masked.store.v4i32.p0({{.*}}, <4 x i1> %active.lane.mask29; PREDFLAG: %index.next = add nuw i64 %index, 430; PREDFLAG: %[[CMP:.*]] = icmp eq i64 %index.next, 43231; PREDFLAG: br i1 %[[CMP]], label %middle.block, label %vector.body, !llvm.loop !032entry:33 br label %for.body34 35for.cond.cleanup:36 ret void37 38for.body:39 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]40 %arrayidx = getelementptr inbounds i32, ptr %B, i64 %indvars.iv41 %0 = load i32, ptr %arrayidx, align 442 %arrayidx2 = getelementptr inbounds i32, ptr %C, i64 %indvars.iv43 %1 = load i32, ptr %arrayidx2, align 444 %add = add nsw i32 %1, %045 %arrayidx4 = getelementptr inbounds i32, ptr %A, i64 %indvars.iv46 store i32 %add, ptr %arrayidx4, align 447 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 148 %exitcond = icmp eq i64 %indvars.iv.next, 43049 br i1 %exitcond, label %for.cond.cleanup, label %for.body, !llvm.loop !1050}51 52define dso_local void @interleave4(ptr noalias nocapture %A, ptr noalias nocapture readonly %B, ptr noalias nocapture readonly %C, i32 %N) local_unnamed_addr #0 {53; PREDFLAG-LABEL: interleave4(54; PREDFLAG: %[[ADD2:.*]] = add i32 %index, 455; PREDFLAG: %[[ADD3:.*]] = add i32 %index, 856; PREDFLAG: %[[ADD4:.*]] = add i32 %index, 1257; PREDFLAG: %[[ALM1:active.lane.mask.*]] = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %index, i32 %N)58; PREDFLAG: %[[ALM2:active.lane.mask.*]] = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %[[ADD2]], i32 %N)59; PREDFLAG: %[[ALM3:active.lane.mask.*]] = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %[[ADD3]], i32 %N)60; PREDFLAG: %[[ALM4:active.lane.mask.*]] = call <4 x i1> @llvm.get.active.lane.mask.v4i1.i32(i32 %[[ADD4]], i32 %N)61;62; PREDFLAG: call <4 x i32> @llvm.masked.load.v4i32.p0({{.*}}, <4 x i1> %[[ALM1]],{{.*}}63; PREDFLAG: call <4 x i32> @llvm.masked.load.v4i32.p0({{.*}}, <4 x i1> %[[ALM2]],{{.*}}64; PREDFLAG: call <4 x i32> @llvm.masked.load.v4i32.p0({{.*}}, <4 x i1> %[[ALM3]],{{.*}}65; PREDFLAG: call <4 x i32> @llvm.masked.load.v4i32.p0({{.*}}, <4 x i1> %[[ALM4]],{{.*}}66; PREDFLAG: call <4 x i32> @llvm.masked.load.v4i32.p0({{.*}}, <4 x i1> %[[ALM1]],{{.*}}67; PREDFLAG: call <4 x i32> @llvm.masked.load.v4i32.p0({{.*}}, <4 x i1> %[[ALM2]],{{.*}}68; PREDFLAG: call <4 x i32> @llvm.masked.load.v4i32.p0({{.*}}, <4 x i1> %[[ALM3]],{{.*}}69; PREDFLAG: call <4 x i32> @llvm.masked.load.v4i32.p0({{.*}}, <4 x i1> %[[ALM4]],{{.*}}70;71; PREDFLAG: call void @llvm.masked.store.v4i32.p0({{.*}}, <4 x i1> %[[ALM1]])72; PREDFLAG: call void @llvm.masked.store.v4i32.p0({{.*}}, <4 x i1> %[[ALM2]])73; PREDFLAG: call void @llvm.masked.store.v4i32.p0({{.*}}, <4 x i1> %[[ALM3]])74; PREDFLAG: call void @llvm.masked.store.v4i32.p0({{.*}}, <4 x i1> %[[ALM4]])75;76entry:77 %cmp8 = icmp sgt i32 %N, 078 br i1 %cmp8, label %for.body.preheader, label %for.cond.cleanup79 80for.body.preheader: ; preds = %entry81 br label %for.body82 83for.cond.cleanup.loopexit: ; preds = %for.body84 br label %for.cond.cleanup85 86for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry87 ret void88 89for.body: ; preds = %for.body.preheader, %for.body90 %i.09 = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]91 %arrayidx = getelementptr inbounds i32, ptr %B, i32 %i.0992 %0 = load i32, ptr %arrayidx, align 493 %arrayidx1 = getelementptr inbounds i32, ptr %C, i32 %i.0994 %1 = load i32, ptr %arrayidx1, align 495 %add = add nsw i32 %1, %096 %arrayidx2 = getelementptr inbounds i32, ptr %A, i32 %i.0997 store i32 %add, ptr %arrayidx2, align 498 %inc = add nuw nsw i32 %i.09, 199 %exitcond = icmp eq i32 %inc, %N100 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body, !llvm.loop !14101}102 103!10 = distinct !{!10, !11, !12}104!11 = !{!"llvm.loop.vectorize.predicate.enable", i1 false}105!12 = !{!"llvm.loop.vectorize.enable", i1 true}106 107!14 = distinct !{!14, !15}108!15 = !{!"llvm.loop.interleave.count", i32 4}109