1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -loop-vectorize -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx -S | FileCheck %s 3 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 i32 @conversion_cost1(i32 %n, i8* nocapture %A, float* nocapture %B) nounwind uwtable ssp { 8; CHECK-LABEL: @conversion_cost1( 9; CHECK-NEXT: [[TMP1:%.*]] = icmp sgt i32 [[N:%.*]], 3 10; CHECK-NEXT: br i1 [[TMP1]], label [[DOTLR_PH_PREHEADER:%.*]], label [[DOT_CRIT_EDGE:%.*]] 11; CHECK: .lr.ph.preheader: 12; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[N]], -4 13; CHECK-NEXT: [[TMP3:%.*]] = zext i32 [[TMP2]] to i64 14; CHECK-NEXT: [[TMP4:%.*]] = add nuw nsw i64 [[TMP3]], 1 15; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP4]], 32 16; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 17; CHECK: vector.ph: 18; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP4]], 32 19; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP4]], [[N_MOD_VF]] 20; CHECK-NEXT: [[IND_END:%.*]] = add i64 3, [[N_VEC]] 21; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 22; CHECK: vector.body: 23; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 24; CHECK-NEXT: [[VEC_IND1:%.*]] = 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_NEXT2:%.*]], [[VECTOR_BODY]] ] 25; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i64 3, [[INDEX]] 26; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 0 27; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, i8* [[A:%.*]], i64 [[TMP5]] 28; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i8, i8* [[TMP6]], i32 0 29; CHECK-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to <32 x i8>* 30; CHECK-NEXT: store <32 x i8> [[VEC_IND1]], <32 x i8>* [[TMP8]], align 1 31; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32 32; CHECK-NEXT: [[VEC_IND_NEXT2]] = add <32 x i8> [[VEC_IND1]], <i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32, i8 32> 33; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 34; CHECK-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 35; CHECK: middle.block: 36; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP4]], [[N_VEC]] 37; CHECK-NEXT: br i1 [[CMP_N]], label [[DOT_CRIT_EDGE_LOOPEXIT:%.*]], label [[SCALAR_PH]] 38; CHECK: scalar.ph: 39; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 3, [[DOTLR_PH_PREHEADER]] ] 40; CHECK-NEXT: br label [[DOTLR_PH:%.*]] 41; CHECK: .lr.ph: 42; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[DOTLR_PH]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 43; CHECK-NEXT: [[TMP10:%.*]] = trunc i64 [[INDVARS_IV]] to i8 44; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, i8* [[A]], i64 [[INDVARS_IV]] 45; CHECK-NEXT: store i8 [[TMP10]], i8* [[TMP11]], align 1 46; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1 47; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32 48; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]] 49; CHECK-NEXT: br i1 [[EXITCOND]], label [[DOT_CRIT_EDGE_LOOPEXIT]], label [[DOTLR_PH]], !llvm.loop [[LOOP2:![0-9]+]] 50; CHECK: ._crit_edge.loopexit: 51; CHECK-NEXT: br label [[DOT_CRIT_EDGE]] 52; CHECK: ._crit_edge: 53; CHECK-NEXT: ret i32 undef 54; 55 %1 = icmp sgt i32 %n, 3 56 br i1 %1, label %.lr.ph, label %._crit_edge 57 58.lr.ph: ; preds = %0, %.lr.ph 59 %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 3, %0 ] 60 %2 = trunc i64 %indvars.iv to i8 61 %3 = getelementptr inbounds i8, i8* %A, i64 %indvars.iv 62 store i8 %2, i8* %3, align 1 63 %indvars.iv.next = add i64 %indvars.iv, 1 64 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 65 %exitcond = icmp eq i32 %lftr.wideiv, %n 66 br i1 %exitcond, label %._crit_edge, label %.lr.ph 67 68._crit_edge: ; preds = %.lr.ph, %0 69 ret i32 undef 70} 71 72define i32 @conversion_cost2(i32 %n, i8* nocapture %A, float* nocapture %B) nounwind uwtable ssp { 73; CHECK-LABEL: @conversion_cost2( 74; CHECK-NEXT: [[TMP1:%.*]] = icmp sgt i32 [[N:%.*]], 9 75; CHECK-NEXT: br i1 [[TMP1]], label [[DOTLR_PH_PREHEADER:%.*]], label [[DOT_CRIT_EDGE:%.*]] 76; CHECK: .lr.ph.preheader: 77; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[N]], -10 78; CHECK-NEXT: [[TMP3:%.*]] = zext i32 [[TMP2]] to i64 79; CHECK-NEXT: [[TMP4:%.*]] = add nuw nsw i64 [[TMP3]], 1 80; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP4]], 8 81; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 82; CHECK: vector.ph: 83; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP4]], 8 84; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP4]], [[N_MOD_VF]] 85; CHECK-NEXT: [[IND_END:%.*]] = add i64 9, [[N_VEC]] 86; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 87; CHECK: vector.body: 88; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 89; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 9, i64 10>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] 90; CHECK-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2> 91; CHECK-NEXT: [[STEP_ADD1:%.*]] = add <2 x i64> [[STEP_ADD]], <i64 2, i64 2> 92; CHECK-NEXT: [[STEP_ADD2:%.*]] = add <2 x i64> [[STEP_ADD1]], <i64 2, i64 2> 93; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i64 9, [[INDEX]] 94; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 0 95; CHECK-NEXT: [[TMP7:%.*]] = add i64 [[OFFSET_IDX]], 2 96; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[OFFSET_IDX]], 4 97; CHECK-NEXT: [[TMP11:%.*]] = add i64 [[OFFSET_IDX]], 6 98; CHECK-NEXT: [[TMP13:%.*]] = add nsw <2 x i64> [[VEC_IND]], <i64 3, i64 3> 99; CHECK-NEXT: [[TMP14:%.*]] = add nsw <2 x i64> [[STEP_ADD]], <i64 3, i64 3> 100; CHECK-NEXT: [[TMP15:%.*]] = add nsw <2 x i64> [[STEP_ADD1]], <i64 3, i64 3> 101; CHECK-NEXT: [[TMP16:%.*]] = add nsw <2 x i64> [[STEP_ADD2]], <i64 3, i64 3> 102; CHECK-NEXT: [[TMP17:%.*]] = sitofp <2 x i64> [[TMP13]] to <2 x float> 103; CHECK-NEXT: [[TMP18:%.*]] = sitofp <2 x i64> [[TMP14]] to <2 x float> 104; CHECK-NEXT: [[TMP19:%.*]] = sitofp <2 x i64> [[TMP15]] to <2 x float> 105; CHECK-NEXT: [[TMP20:%.*]] = sitofp <2 x i64> [[TMP16]] to <2 x float> 106; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, float* [[B:%.*]], i64 [[TMP5]] 107; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP7]] 108; CHECK-NEXT: [[TMP23:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP9]] 109; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[TMP11]] 110; CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds float, float* [[TMP21]], i32 0 111; CHECK-NEXT: [[TMP26:%.*]] = bitcast float* [[TMP25]] to <2 x float>* 112; CHECK-NEXT: store <2 x float> [[TMP17]], <2 x float>* [[TMP26]], align 4 113; CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds float, float* [[TMP21]], i32 2 114; CHECK-NEXT: [[TMP28:%.*]] = bitcast float* [[TMP27]] to <2 x float>* 115; CHECK-NEXT: store <2 x float> [[TMP18]], <2 x float>* [[TMP28]], align 4 116; CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, float* [[TMP21]], i32 4 117; CHECK-NEXT: [[TMP30:%.*]] = bitcast float* [[TMP29]] to <2 x float>* 118; CHECK-NEXT: store <2 x float> [[TMP19]], <2 x float>* [[TMP30]], align 4 119; CHECK-NEXT: [[TMP31:%.*]] = getelementptr inbounds float, float* [[TMP21]], i32 6 120; CHECK-NEXT: [[TMP32:%.*]] = bitcast float* [[TMP31]] to <2 x float>* 121; CHECK-NEXT: store <2 x float> [[TMP20]], <2 x float>* [[TMP32]], align 4 122; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 123; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[STEP_ADD2]], <i64 2, i64 2> 124; CHECK-NEXT: [[TMP33:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 125; CHECK-NEXT: br i1 [[TMP33]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] 126; CHECK: middle.block: 127; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP4]], [[N_VEC]] 128; CHECK-NEXT: br i1 [[CMP_N]], label [[DOT_CRIT_EDGE_LOOPEXIT:%.*]], label [[SCALAR_PH]] 129; CHECK: scalar.ph: 130; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 9, [[DOTLR_PH_PREHEADER]] ] 131; CHECK-NEXT: br label [[DOTLR_PH:%.*]] 132; CHECK: .lr.ph: 133; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[DOTLR_PH]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 134; CHECK-NEXT: [[ADD:%.*]] = add nsw i64 [[INDVARS_IV]], 3 135; CHECK-NEXT: [[TOFP:%.*]] = sitofp i64 [[ADD]] to float 136; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[INDVARS_IV]] 137; CHECK-NEXT: store float [[TOFP]], float* [[GEP]], align 4 138; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1 139; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32 140; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]] 141; CHECK-NEXT: br i1 [[EXITCOND]], label [[DOT_CRIT_EDGE_LOOPEXIT]], label [[DOTLR_PH]], !llvm.loop [[LOOP5:![0-9]+]] 142; CHECK: ._crit_edge.loopexit: 143; CHECK-NEXT: br label [[DOT_CRIT_EDGE]] 144; CHECK: ._crit_edge: 145; CHECK-NEXT: ret i32 undef 146; 147 %1 = icmp sgt i32 %n, 9 148 br i1 %1, label %.lr.ph, label %._crit_edge 149 150.lr.ph: ; preds = %0, %.lr.ph 151 %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 9, %0 ] 152 %add = add nsw i64 %indvars.iv, 3 153 %tofp = sitofp i64 %add to float 154 %gep = getelementptr inbounds float, float* %B, i64 %indvars.iv 155 store float %tofp, float* %gep, align 4 156 %indvars.iv.next = add i64 %indvars.iv, 1 157 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 158 %exitcond = icmp eq i32 %lftr.wideiv, %n 159 br i1 %exitcond, label %._crit_edge, label %.lr.ph 160 161._crit_edge: ; preds = %.lr.ph, %0 162 ret i32 undef 163} 164