190 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --filter-out-after "^scalar.ph:"2; RUN: opt -passes=loop-vectorize,dce,instcombine -mtriple aarch64-linux-gnu -mattr=+sve \3; RUN: -prefer-predicate-over-epilogue=scalar-epilogue -S %s -o - | FileCheck %s4 5define void @cond_inv_load_i32i32i16(ptr noalias nocapture %a, ptr noalias nocapture readonly %cond, ptr noalias nocapture readonly %inv, i64 %n) #0 {6; CHECK-LABEL: @cond_inv_load_i32i32i16(7; CHECK-NEXT: entry:8; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()9; CHECK-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 210; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], [[TMP1]]11; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]12; CHECK: vector.ph:13; CHECK-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64()14; CHECK-NEXT: [[TMP4:%.*]] = shl nuw nsw i64 [[TMP3]], 215; CHECK-NEXT: [[DOTNOT:%.*]] = sub nsw i64 0, [[TMP4]]16; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[N]], [[DOTNOT]]17; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x ptr> poison, ptr [[INV:%.*]], i64 018; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x ptr> [[BROADCAST_SPLATINSERT]], <vscale x 4 x ptr> poison, <vscale x 4 x i32> zeroinitializer19; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]20; CHECK: vector.body:21; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]22; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[COND:%.*]], i64 [[INDEX]]23; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP5]], align 424; CHECK-NEXT: [[TMP6:%.*]] = icmp ne <vscale x 4 x i32> [[WIDE_LOAD]], zeroinitializer25; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x i16> @llvm.masked.gather.nxv4i16.nxv4p0(<vscale x 4 x ptr> align 2 [[BROADCAST_SPLAT]], <vscale x 4 x i1> [[TMP6]], <vscale x 4 x i16> poison)26; CHECK-NEXT: [[TMP7:%.*]] = sext <vscale x 4 x i16> [[WIDE_MASKED_GATHER]] to <vscale x 4 x i32>27; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i32, ptr [[A:%.*]], i64 [[INDEX]]28; CHECK-NEXT: call void @llvm.masked.store.nxv4i32.p0(<vscale x 4 x i32> [[TMP7]], ptr align 4 [[TMP8]], <vscale x 4 x i1> [[TMP6]])29; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP4]]30; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]31; CHECK-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]32; CHECK: middle.block:33; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]34; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]35; CHECK: scalar.ph:36;37entry:38 br label %for.body39 40for.body: ; preds = %entry, %for.inc41 %i.07 = phi i64 [ %inc, %for.inc ], [ 0, %entry ]42 %arrayidx = getelementptr inbounds i32, ptr %cond, i64 %i.0743 %0 = load i32, ptr %arrayidx, align 444 %tobool.not = icmp eq i32 %0, 045 br i1 %tobool.not, label %for.inc, label %if.then46 47if.then: ; preds = %for.body48 %1 = load i16, ptr %inv, align 249 %conv = sext i16 %1 to i3250 %arrayidx1 = getelementptr inbounds i32, ptr %a, i64 %i.0751 store i32 %conv, ptr %arrayidx1, align 452 br label %for.inc53 54for.inc: ; preds = %for.body, %if.then55 %inc = add nuw nsw i64 %i.07, 156 %exitcond.not = icmp eq i64 %inc, %n57 br i1 %exitcond.not, label %exit, label %for.body, !llvm.loop !058 59exit: ; preds = %for.inc60 ret void61}62 63define void @cond_inv_load_f64f64f64(ptr noalias nocapture %a, ptr noalias nocapture readonly %cond, ptr noalias nocapture readonly %inv, i64 %n) #0 {64; CHECK-LABEL: @cond_inv_load_f64f64f64(65; CHECK-NEXT: entry:66; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()67; CHECK-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 268; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], [[TMP1]]69; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]70; CHECK: vector.ph:71; CHECK-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64()72; CHECK-NEXT: [[TMP4:%.*]] = shl nuw nsw i64 [[TMP3]], 273; CHECK-NEXT: [[DOTNOT:%.*]] = sub nsw i64 0, [[TMP4]]74; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[N]], [[DOTNOT]]75; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x ptr> poison, ptr [[INV:%.*]], i64 076; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x ptr> [[BROADCAST_SPLATINSERT]], <vscale x 4 x ptr> poison, <vscale x 4 x i32> zeroinitializer77; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]78; CHECK: vector.body:79; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]80; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds double, ptr [[COND:%.*]], i64 [[INDEX]]81; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x double>, ptr [[TMP5]], align 882; CHECK-NEXT: [[TMP6:%.*]] = fcmp ogt <vscale x 4 x double> [[WIDE_LOAD]], splat (double 4.000000e-01)83; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x double> @llvm.masked.gather.nxv4f64.nxv4p0(<vscale x 4 x ptr> align 8 [[BROADCAST_SPLAT]], <vscale x 4 x i1> [[TMP6]], <vscale x 4 x double> poison)84; CHECK-NEXT: [[TMP7:%.*]] = getelementptr double, ptr [[A:%.*]], i64 [[INDEX]]85; CHECK-NEXT: call void @llvm.masked.store.nxv4f64.p0(<vscale x 4 x double> [[WIDE_MASKED_GATHER]], ptr align 8 [[TMP7]], <vscale x 4 x i1> [[TMP6]])86; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP4]]87; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]88; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]89; CHECK: middle.block:90; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]91; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]92; CHECK: scalar.ph:93;94entry:95 br label %for.body96 97for.body: ; preds = %entry, %for.inc98 %i.08 = phi i64 [ %inc, %for.inc ], [ 0, %entry ]99 %arrayidx = getelementptr inbounds double, ptr %cond, i64 %i.08100 %0 = load double, ptr %arrayidx, align 8101 %cmp1 = fcmp ogt double %0, 4.000000e-01102 br i1 %cmp1, label %if.then, label %for.inc103 104if.then: ; preds = %for.body105 %1 = load double, ptr %inv, align 8106 %arrayidx2 = getelementptr inbounds double, ptr %a, i64 %i.08107 store double %1, ptr %arrayidx2, align 8108 br label %for.inc109 110for.inc: ; preds = %for.body, %if.then111 %inc = add nuw nsw i64 %i.08, 1112 %exitcond.not = icmp eq i64 %inc, %n113 br i1 %exitcond.not, label %exit, label %for.body, !llvm.loop !0114 115exit: ; preds = %for.inc116 ret void117}118 119define void @invariant_load_cond(ptr noalias nocapture %a, ptr nocapture readonly %b, ptr nocapture readonly %cond, i64 %n) #0 {120; CHECK-LABEL: @invariant_load_cond(121; CHECK-NEXT: entry:122; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()123; CHECK-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 2124; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], [[TMP1]]125; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]126; CHECK: vector.ph:127; CHECK-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64()128; CHECK-NEXT: [[TMP4:%.*]] = shl nuw nsw i64 [[TMP3]], 2129; CHECK-NEXT: [[DOTNOT:%.*]] = sub nsw i64 0, [[TMP4]]130; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[N]], [[DOTNOT]]131; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds nuw i8, ptr [[B:%.*]], i64 168132; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x ptr> poison, ptr [[TMP5]], i64 0133; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x ptr> [[BROADCAST_SPLATINSERT]], <vscale x 4 x ptr> poison, <vscale x 4 x i32> zeroinitializer134; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]135; CHECK: vector.body:136; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]137; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[COND:%.*]], i64 [[INDEX]]138; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP6]], align 4139; CHECK-NEXT: [[TMP7:%.*]] = icmp ne <vscale x 4 x i32> [[WIDE_LOAD]], zeroinitializer140; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i32, ptr [[B]], i64 [[INDEX]]141; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0(ptr align 4 [[TMP8]], <vscale x 4 x i1> [[TMP7]], <vscale x 4 x i32> poison)142; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x i32> @llvm.masked.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> align 4 [[BROADCAST_SPLAT]], <vscale x 4 x i1> [[TMP7]], <vscale x 4 x i32> poison)143; CHECK-NEXT: [[TMP9:%.*]] = add nsw <vscale x 4 x i32> [[WIDE_MASKED_GATHER]], [[WIDE_MASKED_LOAD]]144; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i32, ptr [[A:%.*]], i64 [[INDEX]]145; CHECK-NEXT: call void @llvm.masked.store.nxv4i32.p0(<vscale x 4 x i32> [[TMP9]], ptr align 4 [[TMP10]], <vscale x 4 x i1> [[TMP7]])146; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP4]]147; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]148; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]]149; CHECK: middle.block:150; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]151; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]]152; CHECK: scalar.ph:153;154entry:155 br label %for.body156 157for.body:158 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.inc ]159 %arrayidx1 = getelementptr inbounds i32, ptr %b, i64 42160 %arrayidx2 = getelementptr inbounds i32, ptr %cond, i64 %iv161 %0 = load i32, ptr %arrayidx2, align 4162 %tobool.not = icmp eq i32 %0, 0163 br i1 %tobool.not, label %for.inc, label %if.then164 165if.then:166 %arrayidx3 = getelementptr inbounds i32, ptr %b, i64 %iv167 %1 = load i32, ptr %arrayidx3, align 4168 %2 = load i32, ptr %arrayidx1, align 4169 %add = add nsw i32 %2, %1170 %arrayidx4 = getelementptr inbounds i32, ptr %a, i64 %iv171 store i32 %add, ptr %arrayidx4, align 4172 br label %for.inc173 174for.inc:175 %iv.next = add nuw nsw i64 %iv, 1176 %exitcond.not = icmp eq i64 %iv.next, %n177 br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !0178 179for.end:180 ret void181}182 183attributes #0 = { vscale_range(1, 16) }184!0 = distinct !{!0, !1, !2, !3, !4, !5}185!1 = !{!"llvm.loop.mustprogress"}186!2 = !{!"llvm.loop.vectorize.width", i32 4}187!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}188!4 = !{!"llvm.loop.interleave.count", i32 1}189!5 = !{!"llvm.loop.vectorize.enable", i1 true}190