92 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --filter-out-after "^for.body:"2; RUN: opt < %s -passes=loop-vectorize -mtriple aarch64-unknown-linux-gnu -mattr=+sve -epilogue-vectorization-force-VF=2 -S | FileCheck %s3 4;5; Integer reduction with interleaving & a start value of 56;7define i64 @int_reduction_add(ptr %a, i64 %N) {8; CHECK-LABEL: @int_reduction_add(9; CHECK-NEXT: iter.check:10; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 211; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]12; CHECK: vector.main.loop.iter.check:13; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()14; CHECK-NEXT: [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 215; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[N]], [[TMP1]]16; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]17; CHECK: vector.ph:18; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()19; CHECK-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 420; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]]21; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]22; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]23; CHECK: vector.body:24; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]25; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i64> [ insertelement (<vscale x 2 x i64> zeroinitializer, i64 5, i32 0), [[VECTOR_PH]] ], [ [[TMP16:%.*]], [[VECTOR_BODY]] ]26; CHECK-NEXT: [[VEC_PHI2:%.*]] = phi <vscale x 2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP17:%.*]], [[VECTOR_BODY]] ]27; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[INDEX]]28; CHECK-NEXT: [[TMP13:%.*]] = call i64 @llvm.vscale.i64()29; CHECK-NEXT: [[TMP14:%.*]] = shl nuw i64 [[TMP13]], 130; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i64, ptr [[TMP10]], i64 [[TMP14]]31; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 2 x i64>, ptr [[TMP10]], align 832; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 2 x i64>, ptr [[TMP15]], align 833; CHECK-NEXT: [[TMP16]] = add <vscale x 2 x i64> [[WIDE_LOAD]], [[VEC_PHI]]34; CHECK-NEXT: [[TMP17]] = add <vscale x 2 x i64> [[WIDE_LOAD3]], [[VEC_PHI2]]35; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]]36; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]37; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]38; CHECK: middle.block:39; CHECK-NEXT: [[BIN_RDX:%.*]] = add <vscale x 2 x i64> [[TMP17]], [[TMP16]]40; CHECK-NEXT: [[TMP21:%.*]] = call i64 @llvm.vector.reduce.add.nxv2i64(<vscale x 2 x i64> [[BIN_RDX]])41; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]42; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]43; CHECK: vec.epilog.iter.check:44; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 245; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]46; CHECK: vec.epilog.ph:47; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]48; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP21]], [[VEC_EPILOG_ITER_CHECK]] ], [ 5, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]49; CHECK-NEXT: [[N_MOD_VF4:%.*]] = urem i64 [[N]], 250; CHECK-NEXT: [[N_VEC5:%.*]] = sub i64 [[N]], [[N_MOD_VF4]]51; CHECK-NEXT: [[TMP22:%.*]] = insertelement <2 x i64> zeroinitializer, i64 [[BC_MERGE_RDX]], i32 052; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]53; CHECK: vec.epilog.vector.body:54; CHECK-NEXT: [[INDEX7:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT10:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]55; CHECK-NEXT: [[VEC_PHI8:%.*]] = phi <2 x i64> [ [[TMP22]], [[VEC_EPILOG_PH]] ], [ [[TMP26:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]56; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDEX7]]57; CHECK-NEXT: [[WIDE_LOAD9:%.*]] = load <2 x i64>, ptr [[TMP24]], align 858; CHECK-NEXT: [[TMP26]] = add <2 x i64> [[WIDE_LOAD9]], [[VEC_PHI8]]59; CHECK-NEXT: [[INDEX_NEXT10]] = add nuw i64 [[INDEX7]], 260; CHECK-NEXT: [[TMP27:%.*]] = icmp eq i64 [[INDEX_NEXT10]], [[N_VEC5]]61; CHECK-NEXT: br i1 [[TMP27]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]62; CHECK: vec.epilog.middle.block:63; CHECK-NEXT: [[TMP28:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[TMP26]])64; CHECK-NEXT: [[CMP_N6:%.*]] = icmp eq i64 [[N]], [[N_VEC5]]65; CHECK-NEXT: br i1 [[CMP_N6]], label [[FOR_END]], label [[VEC_EPILOG_SCALAR_PH]]66; CHECK: vec.epilog.scalar.ph:67; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC5]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]68; CHECK-NEXT: [[BC_MERGE_RDX11:%.*]] = phi i64 [ [[TMP28]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[TMP21]], [[VEC_EPILOG_ITER_CHECK]] ], [ 5, [[ITER_CHECK]] ]69; CHECK-NEXT: br label [[FOR_BODY:%.*]]70; CHECK: for.body:71;72entry:73 br label %for.body74 75for.body:76 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]77 %sum = phi i64 [ 5, %entry ], [ %add, %for.body ]78 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv79 %0 = load i64, ptr %arrayidx80 %add = add i64 %0, %sum81 %iv.next = add nuw nsw i64 %iv, 182 %exitcond.not = icmp eq i64 %iv.next, %N83 br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !084 85for.end:86 ret i64 %add87}88 89!0 = distinct !{!0, !1, !2}90!1 = !{!"llvm.loop.interleave.count", i32 2}91!2 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}92