152 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=loop-vectorize -mtriple=aarch64--linux-gnu -mattr=+neon -force-vector-width=2 -force-vector-interleave=1 \3; RUN: -prefer-predicate-over-epilogue=scalar-epilogue -S -o - | FileCheck %s4; RUN: opt < %s -passes=loop-vectorize -mtriple=aarch64--linux-gnu -mattr=+sve -force-vector-width=2 -force-vector-interleave=1 \5; RUN: -prefer-predicate-over-epilogue=scalar-epilogue -scalable-vectorization=on -S -o - | FileCheck --check-prefix=SVE %s6 7define dso_local double @test(ptr nocapture noundef readonly %data, ptr nocapture noundef readonly %offset, i32 noundef %size) local_unnamed_addr {8; CHECK-LABEL: @test(9; CHECK-NEXT: entry:10; CHECK-NEXT: [[CMP6:%.*]] = icmp sgt i32 [[SIZE:%.*]], 011; CHECK-NEXT: br i1 [[CMP6]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]12; CHECK: for.body.preheader:13; CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[SIZE]] to i6414; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 215; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]16; CHECK: vector.ph:17; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 218; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]]19; 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: [[VEC_PHI:%.*]] = phi <2 x double> [ <double 0.000000e+00, double -0.000000e+00>, [[VECTOR_PH]] ], [ [[TMP14:%.*]], [[VECTOR_BODY]] ]23; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 024; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 125; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[OFFSET:%.*]], i64 [[TMP0]]26; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[OFFSET]], i64 [[TMP1]]27; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr [[TMP2]], align 428; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[TMP3]], align 429; CHECK-NEXT: [[TMP6:%.*]] = sext i32 [[TMP4]] to i6430; CHECK-NEXT: [[TMP7:%.*]] = sext i32 [[TMP5]] to i6431; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds double, ptr [[DATA:%.*]], i64 [[TMP6]]32; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds double, ptr [[DATA]], i64 [[TMP7]]33; CHECK-NEXT: [[TMP10:%.*]] = load double, ptr [[TMP8]], align 834; CHECK-NEXT: [[TMP11:%.*]] = load double, ptr [[TMP9]], align 835; CHECK-NEXT: [[TMP12:%.*]] = insertelement <2 x double> poison, double [[TMP10]], i32 036; CHECK-NEXT: [[TMP13:%.*]] = insertelement <2 x double> [[TMP12]], double [[TMP11]], i32 137; CHECK-NEXT: [[TMP14]] = fadd <2 x double> [[VEC_PHI]], [[TMP13]]38; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 239; CHECK-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]40; CHECK-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]41; CHECK: middle.block:42; CHECK-NEXT: [[TMP16:%.*]] = call double @llvm.vector.reduce.fadd.v2f64(double -0.000000e+00, <2 x double> [[TMP14]])43; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]44; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]45; CHECK: scalar.ph:46; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ]47; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi double [ [[TMP16]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[FOR_BODY_PREHEADER]] ]48; CHECK-NEXT: br label [[FOR_BODY:%.*]]49; CHECK: for.cond.cleanup.loopexit:50; CHECK-NEXT: [[ADD_LCSSA:%.*]] = phi double [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[TMP16]], [[MIDDLE_BLOCK]] ]51; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]52; CHECK: for.cond.cleanup:53; CHECK-NEXT: [[RES_0_LCSSA:%.*]] = phi double [ 0.000000e+00, [[ENTRY:%.*]] ], [ [[ADD_LCSSA]], [[FOR_COND_CLEANUP_LOOPEXIT]] ]54; CHECK-NEXT: ret double [[RES_0_LCSSA]]55; CHECK: for.body:56; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]57; CHECK-NEXT: [[RES_07:%.*]] = phi double [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[ADD]], [[FOR_BODY]] ]58; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[OFFSET]], i64 [[INDVARS_IV]]59; CHECK-NEXT: [[TMP17:%.*]] = load i32, ptr [[ARRAYIDX]], align 460; CHECK-NEXT: [[IDXPROM1:%.*]] = sext i32 [[TMP17]] to i6461; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds double, ptr [[DATA]], i64 [[IDXPROM1]]62; CHECK-NEXT: [[TMP18:%.*]] = load double, ptr [[ARRAYIDX2]], align 863; CHECK-NEXT: [[ADD]] = fadd double [[RES_07]], [[TMP18]]64; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 165; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]66; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]67;68; SVE-LABEL: @test(69; SVE-NEXT: entry:70; SVE-NEXT: [[CMP6:%.*]] = icmp sgt i32 [[SIZE:%.*]], 071; SVE-NEXT: br i1 [[CMP6]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]72; SVE: for.body.preheader:73; SVE-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[SIZE]] to i6474; SVE-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()75; SVE-NEXT: [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 176; SVE-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], [[TMP1]]77; SVE-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]78; SVE: vector.ph:79; SVE-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()80; SVE-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 281; SVE-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], [[TMP3]]82; SVE-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]]83; SVE-NEXT: br label [[VECTOR_BODY:%.*]]84; SVE: vector.body:85; SVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]86; SVE-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x double> [ insertelement (<vscale x 2 x double> splat (double -0.000000e+00), double 0.000000e+00, i32 0), [[VECTOR_PH]] ], [ [[TMP9:%.*]], [[VECTOR_BODY]] ]87; SVE-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[OFFSET:%.*]], i64 [[INDEX]]88; SVE-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 2 x i32>, ptr [[TMP5]], align 489; SVE-NEXT: [[TMP7:%.*]] = sext <vscale x 2 x i32> [[WIDE_LOAD]] to <vscale x 2 x i64>90; SVE-NEXT: [[TMP8:%.*]] = getelementptr inbounds double, ptr [[DATA:%.*]], <vscale x 2 x i64> [[TMP7]]91; SVE-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 2 x double> @llvm.masked.gather.nxv2f64.nxv2p0(<vscale x 2 x ptr> align 8 [[TMP8]], <vscale x 2 x i1> splat (i1 true), <vscale x 2 x double> poison)92; SVE-NEXT: [[TMP9]] = fadd <vscale x 2 x double> [[VEC_PHI]], [[WIDE_MASKED_GATHER]]93; SVE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]94; SVE-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]95; SVE-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]96; SVE: middle.block:97; SVE-NEXT: [[TMP15:%.*]] = call double @llvm.vector.reduce.fadd.nxv2f64(double -0.000000e+00, <vscale x 2 x double> [[TMP9]])98; SVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]99; SVE-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]100; SVE: scalar.ph:101; SVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ]102; SVE-NEXT: [[BC_MERGE_RDX:%.*]] = phi double [ [[TMP15]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[FOR_BODY_PREHEADER]] ]103; SVE-NEXT: br label [[FOR_BODY:%.*]]104; SVE: for.cond.cleanup.loopexit:105; SVE-NEXT: [[ADD_LCSSA:%.*]] = phi double [ [[ADD:%.*]], [[FOR_BODY]] ], [ [[TMP15]], [[MIDDLE_BLOCK]] ]106; SVE-NEXT: br label [[FOR_COND_CLEANUP]]107; SVE: for.cond.cleanup:108; SVE-NEXT: [[RES_0_LCSSA:%.*]] = phi double [ 0.000000e+00, [[ENTRY:%.*]] ], [ [[ADD_LCSSA]], [[FOR_COND_CLEANUP_LOOPEXIT]] ]109; SVE-NEXT: ret double [[RES_0_LCSSA]]110; SVE: for.body:111; SVE-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]112; SVE-NEXT: [[RES_07:%.*]] = phi double [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[ADD]], [[FOR_BODY]] ]113; SVE-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[OFFSET]], i64 [[INDVARS_IV]]114; SVE-NEXT: [[TMP13:%.*]] = load i32, ptr [[ARRAYIDX]], align 4115; SVE-NEXT: [[IDXPROM1:%.*]] = sext i32 [[TMP13]] to i64116; SVE-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds double, ptr [[DATA]], i64 [[IDXPROM1]]117; SVE-NEXT: [[TMP14:%.*]] = load double, ptr [[ARRAYIDX2]], align 8118; SVE-NEXT: [[ADD]] = fadd double [[RES_07]], [[TMP14]]119; SVE-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1120; SVE-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]121; SVE-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]122;123entry:124 %cmp6 = icmp sgt i32 %size, 0125 br i1 %cmp6, label %for.body.preheader, label %for.cond.cleanup126 127for.body.preheader: ; preds = %entry128 %wide.trip.count = zext i32 %size to i64129 br label %for.body130 131for.cond.cleanup.loopexit: ; preds = %for.body132 %add.lcssa = phi double [ %add, %for.body ]133 br label %for.cond.cleanup134 135for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry136 %res.0.lcssa = phi double [ 0.000000e+00, %entry ], [ %add.lcssa, %for.cond.cleanup.loopexit ]137 ret double %res.0.lcssa138 139for.body: ; preds = %for.body.preheader, %for.body140 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]141 %res.07 = phi double [ 0.000000e+00, %for.body.preheader ], [ %add, %for.body ]142 %arrayidx = getelementptr inbounds i32, ptr %offset, i64 %indvars.iv143 %0 = load i32, ptr %arrayidx, align 4144 %idxprom1 = sext i32 %0 to i64145 %arrayidx2 = getelementptr inbounds double, ptr %data, i64 %idxprom1146 %1 = load double, ptr %arrayidx2, align 8147 %add = fadd double %res.07, %1148 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1149 %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count150 br i1 %exitcond.not, label %for.cond.cleanup.loopexit, label %for.body151}152