1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -passes=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, ptr nocapture %A, ptr 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 [[ITER_CHECK:%.*]], label [[DOT_CRIT_EDGE:%.*]] 11; CHECK: iter.check: 12; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[N]], -3 13; CHECK-NEXT: [[TMP3:%.*]] = zext i32 [[TMP2]] to i64 14; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 4 15; 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]], 32 18; 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]], 32 21; 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: [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 0 29; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[A:%.*]], i64 [[TMP4]] 30; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr [[TMP5]], i32 0 31; CHECK-NEXT: store <32 x i8> [[VEC_IND]], ptr [[TMP6]], align 1 32; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32 33; CHECK-NEXT: [[VEC_IND_NEXT]] = add <32 x i8> [[VEC_IND]], splat (i8 32) 34; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 35; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 36; CHECK: middle.block: 37; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]] 38; CHECK-NEXT: br i1 [[CMP_N]], label [[DOT_CRIT_EDGE_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] 39; CHECK: vec.epilog.iter.check: 40; CHECK-NEXT: [[IND_END5:%.*]] = add i64 3, [[N_VEC]] 41; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[TMP3]], [[N_VEC]] 42; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 4 43; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] 44; CHECK: vec.epilog.ph: 45; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] 46; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ 3, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] 47; CHECK-NEXT: [[N_MOD_VF2:%.*]] = urem i64 [[TMP3]], 4 48; CHECK-NEXT: [[N_VEC3:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF2]] 49; CHECK-NEXT: [[IND_END4:%.*]] = add i64 3, [[N_VEC3]] 50; CHECK-NEXT: [[TMP8:%.*]] = trunc i64 [[BC_RESUME_VAL]] to i8 51; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x i8> poison, i8 [[TMP8]], i64 0 52; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x i8> [[DOTSPLATINSERT]], <4 x i8> poison, <4 x i32> zeroinitializer 53; CHECK-NEXT: [[INDUCTION:%.*]] = add <4 x i8> [[DOTSPLAT]], <i8 0, i8 1, i8 2, i8 3> 54; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] 55; CHECK: vec.epilog.vector.body: 56; CHECK-NEXT: [[INDEX7:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT11:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 57; CHECK-NEXT: [[VEC_IND8:%.*]] = phi <4 x i8> [ [[INDUCTION]], [[VEC_EPILOG_PH]] ], [ [[VEC_IND_NEXT9:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 58; CHECK-NEXT: [[OFFSET_IDX10:%.*]] = add i64 3, [[INDEX7]] 59; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[OFFSET_IDX10]], 0 60; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP9]] 61; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[TMP10]], i32 0 62; CHECK-NEXT: store <4 x i8> [[VEC_IND8]], ptr [[TMP11]], align 1 63; CHECK-NEXT: [[INDEX_NEXT11]] = add nuw i64 [[INDEX7]], 4 64; CHECK-NEXT: [[VEC_IND_NEXT9]] = add <4 x i8> [[VEC_IND8]], splat (i8 4) 65; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT11]], [[N_VEC3]] 66; CHECK-NEXT: br i1 [[TMP12]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 67; CHECK: vec.epilog.middle.block: 68; CHECK-NEXT: [[CMP_N12:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC3]] 69; CHECK-NEXT: br i1 [[CMP_N12]], label [[DOT_CRIT_EDGE_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH]] 70; CHECK: vec.epilog.scalar.ph: 71; CHECK-NEXT: [[BC_RESUME_VAL6:%.*]] = phi i64 [ [[IND_END4]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 3, [[ITER_CHECK]] ], [ [[IND_END5]], [[VEC_EPILOG_ITER_CHECK]] ] 72; CHECK-NEXT: br label [[DOTLR_PH:%.*]] 73; CHECK: .lr.ph: 74; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[DOTLR_PH]] ], [ [[BC_RESUME_VAL6]], [[VEC_EPILOG_SCALAR_PH]] ] 75; CHECK-NEXT: [[TMP13:%.*]] = trunc i64 [[INDVARS_IV]] to i8 76; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDVARS_IV]] 77; CHECK-NEXT: store i8 [[TMP13]], ptr [[TMP14]], align 1 78; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1 79; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32 80; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]] 81; CHECK-NEXT: br i1 [[EXITCOND]], label [[DOT_CRIT_EDGE_LOOPEXIT]], label [[DOTLR_PH]], !llvm.loop [[LOOP4:![0-9]+]] 82; CHECK: ._crit_edge.loopexit: 83; CHECK-NEXT: br label [[DOT_CRIT_EDGE]] 84; CHECK: ._crit_edge: 85; CHECK-NEXT: ret i32 undef 86; 87 %1 = icmp sgt i32 %n, 3 88 br i1 %1, label %.lr.ph, label %._crit_edge 89 90.lr.ph: ; preds = %0, %.lr.ph 91 %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 3, %0 ] 92 %2 = trunc i64 %indvars.iv to i8 93 %3 = getelementptr inbounds i8, ptr %A, i64 %indvars.iv 94 store i8 %2, ptr %3, align 1 95 %indvars.iv.next = add i64 %indvars.iv, 1 96 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 97 %exitcond = icmp eq i32 %lftr.wideiv, %n 98 br i1 %exitcond, label %._crit_edge, label %.lr.ph 99 100._crit_edge: ; preds = %.lr.ph, %0 101 ret i32 undef 102} 103 104define i32 @conversion_cost2(i32 %n, ptr nocapture %A, ptr nocapture %B) nounwind uwtable ssp { 105; CHECK-LABEL: @conversion_cost2( 106; CHECK-NEXT: [[TMP1:%.*]] = icmp sgt i32 [[N:%.*]], 9 107; CHECK-NEXT: br i1 [[TMP1]], label [[DOTLR_PH_PREHEADER:%.*]], label [[DOT_CRIT_EDGE:%.*]] 108; CHECK: .lr.ph.preheader: 109; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[N]], -9 110; CHECK-NEXT: [[TMP3:%.*]] = zext i32 [[TMP2]] to i64 111; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 8 112; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 113; CHECK: vector.ph: 114; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 8 115; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]] 116; CHECK-NEXT: [[IND_END:%.*]] = add i64 9, [[N_VEC]] 117; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 118; CHECK: vector.body: 119; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 120; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 9, i64 10>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] 121; CHECK-NEXT: [[STEP_ADD:%.*]] = add <2 x i64> [[VEC_IND]], splat (i64 2) 122; CHECK-NEXT: [[STEP_ADD_2:%.*]] = add <2 x i64> [[STEP_ADD]], splat (i64 2) 123; CHECK-NEXT: [[STEP_ADD_3:%.*]] = add <2 x i64> [[STEP_ADD_2]], splat (i64 2) 124; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i64 9, [[INDEX]] 125; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 0 126; CHECK-NEXT: [[TMP8:%.*]] = add nsw <2 x i64> [[VEC_IND]], splat (i64 3) 127; CHECK-NEXT: [[TMP9:%.*]] = add nsw <2 x i64> [[STEP_ADD]], splat (i64 3) 128; CHECK-NEXT: [[TMP10:%.*]] = add nsw <2 x i64> [[STEP_ADD_2]], splat (i64 3) 129; CHECK-NEXT: [[TMP11:%.*]] = add nsw <2 x i64> [[STEP_ADD_3]], splat (i64 3) 130; CHECK-NEXT: [[TMP12:%.*]] = sitofp <2 x i64> [[TMP8]] to <2 x float> 131; CHECK-NEXT: [[TMP18:%.*]] = sitofp <2 x i64> [[TMP9]] to <2 x float> 132; CHECK-NEXT: [[TMP19:%.*]] = sitofp <2 x i64> [[TMP10]] to <2 x float> 133; CHECK-NEXT: [[TMP20:%.*]] = sitofp <2 x i64> [[TMP11]] to <2 x float> 134; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 [[TMP4]] 135; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds float, ptr [[TMP13]], i32 0 136; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds float, ptr [[TMP13]], i32 2 137; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds float, ptr [[TMP13]], i32 4 138; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, ptr [[TMP13]], i32 6 139; CHECK-NEXT: store <2 x float> [[TMP12]], ptr [[TMP14]], align 4 140; CHECK-NEXT: store <2 x float> [[TMP18]], ptr [[TMP15]], align 4 141; CHECK-NEXT: store <2 x float> [[TMP19]], ptr [[TMP16]], align 4 142; CHECK-NEXT: store <2 x float> [[TMP20]], ptr [[TMP17]], align 4 143; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 144; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[STEP_ADD_3]], splat (i64 2) 145; CHECK-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 146; CHECK-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] 147; CHECK: middle.block: 148; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]] 149; CHECK-NEXT: br i1 [[CMP_N]], label [[DOT_CRIT_EDGE_LOOPEXIT:%.*]], label [[SCALAR_PH]] 150; CHECK: scalar.ph: 151; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ 9, [[DOTLR_PH_PREHEADER]] ] 152; CHECK-NEXT: br label [[DOTLR_PH:%.*]] 153; CHECK: .lr.ph: 154; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[DOTLR_PH]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 155; CHECK-NEXT: [[ADD:%.*]] = add nsw i64 [[INDVARS_IV]], 3 156; CHECK-NEXT: [[TOFP:%.*]] = sitofp i64 [[ADD]] to float 157; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[INDVARS_IV]] 158; CHECK-NEXT: store float [[TOFP]], ptr [[GEP]], align 4 159; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1 160; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32 161; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]] 162; CHECK-NEXT: br i1 [[EXITCOND]], label [[DOT_CRIT_EDGE_LOOPEXIT]], label [[DOTLR_PH]], !llvm.loop [[LOOP6:![0-9]+]] 163; CHECK: ._crit_edge.loopexit: 164; CHECK-NEXT: br label [[DOT_CRIT_EDGE]] 165; CHECK: ._crit_edge: 166; CHECK-NEXT: ret i32 undef 167; 168 %1 = icmp sgt i32 %n, 9 169 br i1 %1, label %.lr.ph, label %._crit_edge 170 171.lr.ph: ; preds = %0, %.lr.ph 172 %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 9, %0 ] 173 %add = add nsw i64 %indvars.iv, 3 174 %tofp = sitofp i64 %add to float 175 %gep = getelementptr inbounds float, ptr %B, i64 %indvars.iv 176 store float %tofp, ptr %gep, align 4 177 %indvars.iv.next = add i64 %indvars.iv, 1 178 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 179 %exitcond = icmp eq i32 %lftr.wideiv, %n 180 br i1 %exitcond, label %._crit_edge, label %.lr.ph 181 182._crit_edge: ; preds = %.lr.ph, %0 183 ret i32 undef 184} 185