186 lines · plain
1; RUN: opt < %s -passes='print<loop-cache-cost>' -disable-output 2>&1 | FileCheck %s2 3target datalayout = "e-m:e-i64:64-n32:64"4target triple = "powerpc64le-unknown-linux-gnu"5 6; void matvecmul(const double *__restrict y, const double * __restrict x, const double * __restrict b,7; const int * __restrict nb, const int * __restrict nx, const int * __restrict ny, const int * __restrict nz) {8;9; for (int k=1;k<nz,++k) 10; for (int j=1;j<ny,++j)11; for (int i=1;i<nx,++i)12; for (int l=1;l<nb,++l)13; for (int m=1;m<nb,++m)14; y[k+1][j][i][l] = y[k+1][j][i][l] + b[k][j][i][m][l]*x[k][j][i][m]15; }16 17; CHECK: Loop 'k_loop' has cost = 1020000000000000018; CHECK-NEXT: Loop 'j_loop' has cost = 10200000000000019; CHECK-NEXT: Loop 'i_loop' has cost = 102000000000020; CHECK-NEXT: Loop 'm_loop' has cost = 1080000000021; CHECK-NEXT: Loop 'l_loop' has cost = 150000000022 23%_elem_type_of_double = type <{ double }>24 25; Function Attrs: norecurse nounwind26define void @mat_vec_mpy(ptr noalias %y, ptr noalias readonly %x,27 ptr noalias readonly %b, ptr noalias readonly %nb, ptr noalias readonly %nx, 28 ptr noalias readonly %ny, ptr noalias readonly %nz) {29mat_times_vec_entry:30 %_ind_val = load i32, ptr %nb, align 431 %_conv = sext i32 %_ind_val to i6432 %_grt_tmp.i = icmp sgt i64 %_conv, 033 %a_b.i = select i1 %_grt_tmp.i, i64 %_conv, i64 034 %_ind_val1 = load i32, ptr %nx, align 435 %_conv2 = sext i32 %_ind_val1 to i6436 %_grt_tmp.i266 = icmp sgt i64 %_conv2, 037 %a_b.i267 = select i1 %_grt_tmp.i266, i64 %_conv2, i64 038 %_ind_val3 = load i32, ptr %ny, align 439 %_conv4 = sext i32 %_ind_val3 to i6440 %_grt_tmp.i264 = icmp sgt i64 %_conv4, 041 %a_b.i265 = select i1 %_grt_tmp.i264, i64 %_conv4, i64 042 %_ind_val5 = load i32, ptr %nz, align 443 %_mult_tmp = shl nsw i64 %a_b.i, 344 %_mult_tmp7 = mul i64 %_mult_tmp, %a_b.i26745 %_mult_tmp8 = mul i64 %_mult_tmp7, %a_b.i26546 %_sub_tmp = sub nuw nsw i64 -8, %_mult_tmp47 %_sub_tmp21 = sub i64 %_sub_tmp, %_mult_tmp748 %_sub_tmp23 = sub i64 %_sub_tmp21, %_mult_tmp849 %_mult_tmp73 = mul i64 %_mult_tmp, %a_b.i50 %_mult_tmp74 = mul i64 %_mult_tmp73, %a_b.i26751 %_mult_tmp75 = mul i64 %_mult_tmp74, %a_b.i26552 %_sub_tmp93 = sub i64 %_sub_tmp, %_mult_tmp7353 %_sub_tmp95 = sub i64 %_sub_tmp93, %_mult_tmp7454 %_sub_tmp97 = sub i64 %_sub_tmp95, %_mult_tmp7555 %_grt_tmp853288 = icmp slt i32 %_ind_val5, 156 br i1 %_grt_tmp853288, label %_return_bb, label %k_loop.lr.ph57 58k_loop.lr.ph: ; preds = %mat_times_vec_entry59 %_grt_tmp851279 = icmp slt i32 %_ind_val3, 160 %_grt_tmp847270 = icmp slt i32 %_ind_val, 161 %_aa_conv = bitcast ptr %y to ptr62 %_adda_ = getelementptr inbounds i8, ptr %_aa_conv, i64 %_sub_tmp2363 %_aa_conv434 = bitcast ptr %x to ptr64 %_adda_435 = getelementptr inbounds i8, ptr %_aa_conv434, i64 %_sub_tmp2365 %_aa_conv785 = bitcast ptr %b to ptr66 %_adda_786 = getelementptr inbounds i8, ptr %_aa_conv785, i64 %_sub_tmp9767 br i1 %_grt_tmp851279, label %k_loop.us.preheader, label %k_loop.lr.ph.split68 69k_loop.us.preheader: ; preds = %k_loop.lr.ph70 br label %_return_bb.loopexit71 72k_loop.lr.ph.split: ; preds = %k_loop.lr.ph73 %_grt_tmp849273 = icmp slt i32 %_ind_val1, 174 br i1 %_grt_tmp849273, label %k_loop.us291.preheader, label %k_loop.lr.ph.split.split75 76k_loop.us291.preheader: ; preds = %k_loop.lr.ph.split77 br label %_return_bb.loopexit30078 79k_loop.lr.ph.split.split: ; preds = %k_loop.lr.ph.split80 br i1 %_grt_tmp847270, label %k_loop.us294.preheader, label %k_loop.preheader81 82k_loop.preheader: ; preds = %k_loop.lr.ph.split.split83 %0 = add i32 %_ind_val, 184 %1 = add i32 %_ind_val1, 185 %2 = add i32 %_ind_val3, 186 %3 = add i32 %_ind_val5, 187 br label %k_loop88 89k_loop.us294.preheader: ; preds = %k_loop.lr.ph.split.split90 br label %_return_bb.loopexit30191 92k_loop: ; preds = %k_loop._label_18_crit_edge.split.split.split, %k_loop.preheader93 %indvars.iv316 = phi i64 [ 1, %k_loop.preheader ], [ %indvars.iv.next317, %k_loop._label_18_crit_edge.split.split.split ]94 %indvars.iv.next317 = add nuw nsw i64 %indvars.iv316, 195 %_ix_x_len = mul i64 %_mult_tmp8, %indvars.iv.next31796 %_ix_x_len410 = mul i64 %_mult_tmp75, %indvars.iv31697 %_ix_x_len822 = mul i64 %_mult_tmp8, %indvars.iv31698 br label %j_loop99 100j_loop: ; preds = %j_loop._label_15_crit_edge.split.split, %k_loop101 %indvars.iv312 = phi i64 [ %indvars.iv.next313, %j_loop._label_15_crit_edge.split.split ], [ 1, %k_loop ]102 %_ix_x_len371 = mul i64 %_mult_tmp7, %indvars.iv312103 %_ix_x_len415 = mul i64 %_mult_tmp74, %indvars.iv312104 br label %i_loop105 106i_loop: ; preds = %i_loop._label_12_crit_edge.split, %j_loop107 %indvars.iv307 = phi i64 [ %indvars.iv.next308, %i_loop._label_12_crit_edge.split ], [ 1, %j_loop ]108 %_ix_x_len375 = mul i64 %_mult_tmp, %indvars.iv307109 %_ix_x_len420 = mul i64 %_mult_tmp73, %indvars.iv307110 br label %l_loop111 112l_loop: ; preds = %l_loop._label_9_crit_edge, %i_loop113 %indvars.iv303 = phi i64 [ %indvars.iv.next304, %l_loop._label_9_crit_edge ], [ 1, %i_loop ]114 %_ix_x_len378 = shl nuw nsw i64 %indvars.iv303, 3115 br label %m_loop116 117m_loop: ; preds = %m_loop, %l_loop118 %indvars.iv = phi i64 [ %indvars.iv.next, %m_loop ], [ 1, %l_loop ]119 %_ix_x_len424 = mul i64 %_mult_tmp, %indvars.iv120 %_ix_x_len454 = shl nuw nsw i64 %indvars.iv, 3121 %_ixa_gep = getelementptr inbounds i8, ptr %_adda_, i64 %_ix_x_len122 %_ixa_gep791 = getelementptr inbounds i8, ptr %_adda_786, i64 %_ix_x_len410123 %_ixa_gep823 = getelementptr inbounds i8, ptr %_adda_435, i64 %_ix_x_len822124 %_ixa_gep372 = getelementptr inbounds i8, ptr %_ixa_gep, i64 %_ix_x_len371125 %_ixa_gep376 = getelementptr inbounds i8, ptr %_ixa_gep372, i64 %_ix_x_len375126 %_ixa_gep796 = getelementptr inbounds i8, ptr %_ixa_gep791, i64 %_ix_x_len415127 %_ixa_gep828 = getelementptr inbounds i8, ptr %_ixa_gep823, i64 %_ix_x_len371128 %_ixa_gep379 = getelementptr inbounds i8, ptr %_ixa_gep376, i64 %_ix_x_len378129 %_ixa_gep801 = getelementptr inbounds i8, ptr %_ixa_gep796, i64 %_ix_x_len420130 %_ixa_gep833 = getelementptr inbounds i8, ptr %_ixa_gep828, i64 %_ix_x_len375131 %_ixa_gep806 = getelementptr inbounds i8, ptr %_ixa_gep801, i64 %_ix_x_len378132 %_ixa_gep810 = getelementptr inbounds i8, ptr %_ixa_gep806, i64 %_ix_x_len424133 %_gepp = bitcast ptr %_ixa_gep379 to ptr134 %_gepp813 = bitcast ptr %_ixa_gep810 to ptr135 %_ind_val814 = load double, ptr %_gepp813, align 8136 %_ixa_gep837 = getelementptr inbounds i8, ptr %_ixa_gep833, i64 %_ix_x_len454137 %_gepp840 = bitcast ptr %_ixa_gep837 to ptr138 %_ind_val841 = load double, ptr %_gepp840, align 8139 %_mult_tmp842 = fmul double %_ind_val814, %_ind_val841140 store double %_mult_tmp842, ptr %_gepp, align 8141 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1142 %wide.trip.count = zext i32 %0 to i64143 %wide.trip.count305 = zext i32 %0 to i64144 %wide.trip.count309 = zext i32 %1 to i64145 %wide.trip.count314 = zext i32 %2 to i64146 %wide.trip.count319 = zext i32 %3 to i64147 %exitcond = icmp ne i64 %indvars.iv.next, %wide.trip.count148 br i1 %exitcond, label %m_loop, label %l_loop._label_9_crit_edge149 150l_loop._label_9_crit_edge: ; preds = %m_loop151 %indvars.iv.next304 = add nuw nsw i64 %indvars.iv303, 1152 %exitcond306 = icmp ne i64 %indvars.iv.next304, %wide.trip.count305153 br i1 %exitcond306, label %l_loop, label %i_loop._label_12_crit_edge.split154 155i_loop._label_12_crit_edge.split: ; preds = %l_loop._label_9_crit_edge156 %indvars.iv.next308 = add nuw nsw i64 %indvars.iv307, 1157 %exitcond310 = icmp ne i64 %indvars.iv.next308, %wide.trip.count309158 br i1 %exitcond310, label %i_loop, label %j_loop._label_15_crit_edge.split.split159 160j_loop._label_15_crit_edge.split.split: ; preds = %i_loop._label_12_crit_edge.split161 %indvars.iv.next313 = add nuw nsw i64 %indvars.iv312, 1162 %exitcond315 = icmp ne i64 %indvars.iv.next313, %wide.trip.count314163 br i1 %exitcond315, label %j_loop, label %k_loop._label_18_crit_edge.split.split.split164 165k_loop._label_18_crit_edge.split.split.split: ; preds = %j_loop._label_15_crit_edge.split.split166 %exitcond320 = icmp ne i64 %indvars.iv.next317, %wide.trip.count319167 br i1 %exitcond320, label %k_loop, label %_return_bb.loopexit302168 169_return_bb.loopexit: ; preds = %k_loop.us.preheader170 br label %_return_bb171 172_return_bb.loopexit300: ; preds = %k_loop.us291.preheader173 br label %_return_bb174 175_return_bb.loopexit301: ; preds = %k_loop.us294.preheader176 br label %_return_bb177 178_return_bb.loopexit302: ; preds = %k_loop._label_18_crit_edge.split.split.split179 br label %_return_bb180 181_return_bb: ; preds = %_return_bb.loopexit302, %_return_bb.loopexit301, %_return_bb.loopexit300, %_return_bb.loopexit, %mat_times_vec_entry182 ret void183}184 185 186