178 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=loop-vectorize -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx -S | FileCheck %s3 4target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"5target triple = "x86_64-apple-macosx10.8.0"6 7define void @conversion_cost1(i32 %n, ptr nocapture %A, ptr nocapture %B) nounwind uwtable ssp {8; CHECK-LABEL: @conversion_cost1(9; CHECK-NEXT: [[TMP1:%.*]] = icmp sgt i32 [[N:%.*]], 310; CHECK-NEXT: br i1 [[TMP1]], label [[ITER_CHECK:%.*]], label [[DOT_CRIT_EDGE:%.*]]11; CHECK: iter.check:12; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[N]], -313; CHECK-NEXT: [[TMP3:%.*]] = zext i32 [[TMP2]] to i6414; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 415; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]16; CHECK: vector.main.loop.iter.check:17; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[TMP3]], 3218; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]19; CHECK: vector.ph:20; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 3221; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]22; CHECK-NEXT: [[IND_END:%.*]] = add i64 3, [[N_VEC]]23; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]24; CHECK: vector.body:25; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]26; CHECK-NEXT: [[VEC_IND:%.*]] = phi <32 x i8> [ <i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 16, i8 17, i8 18, i8 19, i8 20, i8 21, i8 22, i8 23, i8 24, i8 25, i8 26, i8 27, i8 28, i8 29, i8 30, i8 31, i8 32, i8 33, i8 34>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]27; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i64 3, [[INDEX]]28; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[A:%.*]], i64 [[OFFSET_IDX]]29; CHECK-NEXT: store <32 x i8> [[VEC_IND]], ptr [[TMP5]], align 130; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 3231; CHECK-NEXT: [[VEC_IND_NEXT]] = add <32 x i8> [[VEC_IND]], splat (i8 32)32; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]33; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]34; CHECK: middle.block:35; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]36; CHECK-NEXT: br i1 [[CMP_N]], label [[DOT_CRIT_EDGE_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]37; CHECK: vec.epilog.iter.check:38; CHECK-NEXT: [[IND_END5:%.*]] = add i64 3, [[N_VEC]]39; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 440; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3:![0-9]+]]41; CHECK: vec.epilog.ph:42; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]43; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ 3, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]44; CHECK-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[TMP3]], 445; CHECK-NEXT: [[N_VEC3:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF2]]46; CHECK-NEXT: [[IND_END4:%.*]] = add i64 3, [[N_VEC3]]47; CHECK-NEXT: [[TMP8:%.*]] = trunc i64 [[BC_RESUME_VAL]] to i848; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x i8> poison, i8 [[TMP8]], i64 049; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x i8> [[DOTSPLATINSERT]], <4 x i8> poison, <4 x i32> zeroinitializer50; CHECK-NEXT: [[INDUCTION:%.*]] = add <4 x i8> [[DOTSPLAT]], <i8 0, i8 1, i8 2, i8 3>51; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]52; CHECK: vec.epilog.vector.body:53; CHECK-NEXT: [[INDEX7:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT11:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]54; CHECK-NEXT: [[VEC_IND8:%.*]] = phi <4 x i8> [ [[INDUCTION]], [[VEC_EPILOG_PH]] ], [ [[VEC_IND_NEXT9:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]55; CHECK-NEXT: [[OFFSET_IDX10:%.*]] = add i64 3, [[INDEX7]]56; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[OFFSET_IDX10]]57; CHECK-NEXT: store <4 x i8> [[VEC_IND8]], ptr [[TMP10]], align 158; CHECK-NEXT: [[INDEX_NEXT11]] = add nuw i64 [[INDEX7]], 459; CHECK-NEXT: [[VEC_IND_NEXT9]] = add <4 x i8> [[VEC_IND8]], splat (i8 4)60; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT11]], [[N_VEC3]]61; CHECK-NEXT: br i1 [[TMP12]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]62; CHECK: vec.epilog.middle.block:63; CHECK-NEXT: [[CMP_N12:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC3]]64; CHECK-NEXT: br i1 [[CMP_N12]], label [[DOT_CRIT_EDGE_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH]]65; CHECK: vec.epilog.scalar.ph:66; CHECK-NEXT: [[BC_RESUME_VAL6:%.*]] = phi i64 [ [[IND_END4]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END5]], [[VEC_EPILOG_ITER_CHECK]] ], [ 3, [[ITER_CHECK]] ]67; CHECK-NEXT: br label [[DOTLR_PH:%.*]]68; CHECK: .lr.ph:69; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[DOTLR_PH]] ], [ [[BC_RESUME_VAL6]], [[VEC_EPILOG_SCALAR_PH]] ]70; CHECK-NEXT: [[TMP13:%.*]] = trunc i64 [[INDVARS_IV]] to i871; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDVARS_IV]]72; CHECK-NEXT: store i8 [[TMP13]], ptr [[TMP14]], align 173; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 174; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i3275; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]]76; CHECK-NEXT: br i1 [[EXITCOND]], label [[DOT_CRIT_EDGE_LOOPEXIT]], label [[DOTLR_PH]], !llvm.loop [[LOOP5:![0-9]+]]77; CHECK: ._crit_edge.loopexit:78; CHECK-NEXT: br label [[DOT_CRIT_EDGE]]79; CHECK: ._crit_edge:80; CHECK-NEXT: ret void81;82 %1 = icmp sgt i32 %n, 383 br i1 %1, label %.lr.ph, label %._crit_edge84 85.lr.ph: ; preds = %0, %.lr.ph86 %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 3, %0 ]87 %2 = trunc i64 %indvars.iv to i888 %3 = getelementptr inbounds i8, ptr %A, i64 %indvars.iv89 store i8 %2, ptr %3, align 190 %indvars.iv.next = add i64 %indvars.iv, 191 %lftr.wideiv = trunc i64 %indvars.iv.next to i3292 %exitcond = icmp eq i32 %lftr.wideiv, %n93 br i1 %exitcond, label %._crit_edge, label %.lr.ph94 95._crit_edge: ; preds = %.lr.ph, %096 ret void97}98 99define void @conversion_cost2(i32 %n, ptr nocapture %A, ptr nocapture %B) nounwind uwtable ssp {100; CHECK-LABEL: @conversion_cost2(101; CHECK-NEXT: [[TMP1:%.*]] = icmp sgt i32 [[N:%.*]], 9102; CHECK-NEXT: br i1 [[TMP1]], label [[DOTLR_PH_PREHEADER:%.*]], label [[DOT_CRIT_EDGE:%.*]]103; CHECK: .lr.ph.preheader:104; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[N]], -9105; CHECK-NEXT: [[TMP3:%.*]] = zext i32 [[TMP2]] to i64106; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 8107; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]108; CHECK: vector.ph:109; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 8110; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]111; CHECK-NEXT: [[IND_END:%.*]] = add i64 9, [[N_VEC]]112; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]113; CHECK: vector.body:114; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]115; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 9, i64 10>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]116; CHECK-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], splat (i64 2)117; CHECK-NEXT: [[STEP_ADD_2:%.*]] = add <2 x i64> [[STEP_ADD]], splat (i64 2)118; CHECK-NEXT: [[STEP_ADD_3:%.*]] = add <2 x i64> [[STEP_ADD_2]], splat (i64 2)119; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i64 9, [[INDEX]]120; CHECK-NEXT: [[TMP8:%.*]] = add nsw <2 x i64> [[VEC_IND]], splat (i64 3)121; CHECK-NEXT: [[TMP9:%.*]] = add nsw <2 x i64> [[STEP_ADD]], splat (i64 3)122; CHECK-NEXT: [[TMP10:%.*]] = add nsw <2 x i64> [[STEP_ADD_2]], splat (i64 3)123; CHECK-NEXT: [[TMP11:%.*]] = add nsw <2 x i64> [[STEP_ADD_3]], splat (i64 3)124; CHECK-NEXT: [[TMP12:%.*]] = sitofp <2 x i64> [[TMP8]] to <2 x float>125; CHECK-NEXT: [[TMP18:%.*]] = sitofp <2 x i64> [[TMP9]] to <2 x float>126; CHECK-NEXT: [[TMP19:%.*]] = sitofp <2 x i64> [[TMP10]] to <2 x float>127; CHECK-NEXT: [[TMP20:%.*]] = sitofp <2 x i64> [[TMP11]] to <2 x float>128; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 [[OFFSET_IDX]]129; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds float, ptr [[TMP13]], i64 2130; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds float, ptr [[TMP13]], i64 4131; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, ptr [[TMP13]], i64 6132; CHECK-NEXT: store <2 x float> [[TMP12]], ptr [[TMP13]], align 4133; CHECK-NEXT: store <2 x float> [[TMP18]], ptr [[TMP15]], align 4134; CHECK-NEXT: store <2 x float> [[TMP19]], ptr [[TMP16]], align 4135; CHECK-NEXT: store <2 x float> [[TMP20]], ptr [[TMP17]], align 4136; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8137; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[STEP_ADD_3]], splat (i64 2)138; CHECK-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]139; CHECK-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]140; CHECK: middle.block:141; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]142; CHECK-NEXT: br i1 [[CMP_N]], label [[DOT_CRIT_EDGE_LOOPEXIT:%.*]], label [[SCALAR_PH]]143; CHECK: scalar.ph:144; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 9, [[DOTLR_PH_PREHEADER]] ]145; CHECK-NEXT: br label [[DOTLR_PH:%.*]]146; CHECK: .lr.ph:147; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[DOTLR_PH]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]148; CHECK-NEXT: [[ADD:%.*]] = add nsw i64 [[INDVARS_IV]], 3149; CHECK-NEXT: [[TOFP:%.*]] = sitofp i64 [[ADD]] to float150; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[INDVARS_IV]]151; CHECK-NEXT: store float [[TOFP]], ptr [[GEP]], align 4152; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1153; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32154; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]]155; CHECK-NEXT: br i1 [[EXITCOND]], label [[DOT_CRIT_EDGE_LOOPEXIT]], label [[DOTLR_PH]], !llvm.loop [[LOOP7:![0-9]+]]156; CHECK: ._crit_edge.loopexit:157; CHECK-NEXT: br label [[DOT_CRIT_EDGE]]158; CHECK: ._crit_edge:159; CHECK-NEXT: ret void160;161 %1 = icmp sgt i32 %n, 9162 br i1 %1, label %.lr.ph, label %._crit_edge163 164.lr.ph: ; preds = %0, %.lr.ph165 %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 9, %0 ]166 %add = add nsw i64 %indvars.iv, 3167 %tofp = sitofp i64 %add to float168 %gep = getelementptr inbounds float, ptr %B, i64 %indvars.iv169 store float %tofp, ptr %gep, align 4170 %indvars.iv.next = add i64 %indvars.iv, 1171 %lftr.wideiv = trunc i64 %indvars.iv.next to i32172 %exitcond = icmp eq i32 %lftr.wideiv, %n173 br i1 %exitcond, label %._crit_edge, label %.lr.ph174 175._crit_edge: ; preds = %.lr.ph, %0176 ret void177}178