203 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 32; PR41445: This test checks the case when LSR split critical edge3; and phi node has other pending fixup operands4 5; RUN: opt -S -loop-reduce < %s | FileCheck %s6 7target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"8target triple = "x86_64-unknown-linux-gnu"9 10; We have %indvars.iv.lcssa phi node where 4 input operands11; need to be rewritten: %tmp1, %tmp2, %tmp3, %tmp4.12; When we try to rewrite %tmp1, we first split the critical edge.13; All the other PHI inputs besides %tmp1 go to a new phi node.14; This test checks that LSR is still able to rewrite %tmp2, %tmp3, %tmp4.15define i32 @foo(ptr %A, i32 %t) {16; CHECK-LABEL: define i32 @foo(17; CHECK-SAME: ptr [[A:%.*]], i32 [[T:%.*]]) {18; CHECK-NEXT: entry:19; CHECK-NEXT: br label [[LOOP_32:%.*]]20; CHECK: loop.exit.loopexitsplitsplitsplit:21; CHECK-NEXT: [[LSR_IV:%.*]] = phi i64 [ [[LSR_IV1:%.*]], [[IFMERGE_34:%.*]] ]22; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[LSR_IV]], -123; CHECK-NEXT: br label [[LOOP_EXIT_LOOPEXITSPLITSPLIT:%.*]]24; CHECK: ifmerge.38.loop.exit.loopexitsplitsplit_crit_edge:25; CHECK-NEXT: [[LSR_IV_LCSSA10:%.*]] = phi i64 [ [[LSR_IV1]], [[IFMERGE_38:%.*]] ]26; CHECK-NEXT: br label [[LOOP_EXIT_LOOPEXITSPLITSPLIT]]27; CHECK: loop.exit.loopexitsplitsplit:28; CHECK-NEXT: [[INDVARS_IV_LCSSA_PH_PH_PH:%.*]] = phi i64 [ [[LSR_IV_LCSSA10]], [[IFMERGE_38_LOOP_EXIT_LOOPEXITSPLITSPLIT_CRIT_EDGE:%.*]] ], [ [[TMP0]], [[LOOP_EXIT_LOOPEXITSPLITSPLITSPLIT:%.*]] ]29; CHECK-NEXT: br label [[LOOP_EXIT_LOOPEXITSPLIT:%.*]]30; CHECK: ifmerge.42.loop.exit.loopexitsplit_crit_edge:31; CHECK-NEXT: [[LSR_IV_LCSSA11:%.*]] = phi i64 [ [[LSR_IV1]], [[IFMERGE_42:%.*]] ]32; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[LSR_IV_LCSSA11]], 133; CHECK-NEXT: br label [[LOOP_EXIT_LOOPEXITSPLIT]]34; CHECK: loop.exit.loopexitsplit:35; CHECK-NEXT: [[INDVARS_IV_LCSSA_PH_PH:%.*]] = phi i64 [ [[TMP1]], [[IFMERGE_42_LOOP_EXIT_LOOPEXITSPLIT_CRIT_EDGE:%.*]] ], [ [[INDVARS_IV_LCSSA_PH_PH_PH]], [[LOOP_EXIT_LOOPEXITSPLITSPLIT]] ]36; CHECK-NEXT: br label [[LOOP_EXIT_LOOPEXIT:%.*]]37; CHECK: then.34.loop.exit.loopexit_crit_edge:38; CHECK-NEXT: [[LSR_IV_LCSSA:%.*]] = phi i64 [ [[LSR_IV1]], [[THEN_34:%.*]] ]39; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[LSR_IV_LCSSA]], -240; CHECK-NEXT: br label [[LOOP_EXIT_LOOPEXIT]]41; CHECK: loop.exit.loopexit:42; CHECK-NEXT: [[INDVARS_IV_LCSSA_PH:%.*]] = phi i64 [ [[TMP2]], [[THEN_34_LOOP_EXIT_LOOPEXIT_CRIT_EDGE:%.*]] ], [ [[INDVARS_IV_LCSSA_PH_PH]], [[LOOP_EXIT_LOOPEXITSPLIT]] ]43; CHECK-NEXT: br label [[LOOP_EXIT:%.*]]44; CHECK: loop.exit:45; CHECK-NEXT: [[INDVARS_IV_LCSSA:%.*]] = phi i64 [ 48, [[THEN_8:%.*]] ], [ 49, [[THEN_8_1:%.*]] ], [ [[INDVARS_IV_LCSSA_PH]], [[LOOP_EXIT_LOOPEXIT]] ]46; CHECK-NEXT: [[TMP:%.*]] = trunc i64 [[INDVARS_IV_LCSSA]] to i3247; CHECK-NEXT: br label [[FOR_END:%.*]]48; CHECK: for.end:49; CHECK-NEXT: [[I_0_LCSSA:%.*]] = phi i32 [ [[TMP]], [[LOOP_EXIT]] ], [ 50, [[THEN_8_1]] ], [ 50, [[IFMERGE_8:%.*]] ]50; CHECK-NEXT: ret i32 [[I_0_LCSSA]]51; CHECK: loop.32:52; CHECK-NEXT: [[LSR_IV1]] = phi i64 [ [[LSR_IV_NEXT:%.*]], [[IFMERGE_46:%.*]] ], [ 2, [[ENTRY:%.*]] ]53; CHECK-NEXT: [[I1_I64_0:%.*]] = phi i64 [ 0, [[ENTRY]] ], [ [[NEXTIVLOOP_32:%.*]], [[IFMERGE_46]] ]54; CHECK-NEXT: [[TMP3:%.*]] = shl nuw nsw i64 [[LSR_IV1]], 255; CHECK-NEXT: [[SCEVGEP7:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP3]]56; CHECK-NEXT: [[SCEVGEP8:%.*]] = getelementptr i8, ptr [[SCEVGEP7]], i64 -457; CHECK-NEXT: [[GEPLOAD:%.*]] = load i32, ptr [[SCEVGEP8]], align 458; CHECK-NEXT: [[CMP_34:%.*]] = icmp sgt i32 [[GEPLOAD]], [[T]]59; CHECK-NEXT: br i1 [[CMP_34]], label [[THEN_34]], label [[IFMERGE_34]]60; CHECK: then.34:61; CHECK-NEXT: [[TMP4:%.*]] = shl nuw nsw i64 [[LSR_IV1]], 262; CHECK-NEXT: [[SCEVGEP5:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP4]]63; CHECK-NEXT: [[SCEVGEP6:%.*]] = getelementptr i8, ptr [[SCEVGEP5]], i64 -864; CHECK-NEXT: [[GEPLOAD18:%.*]] = load i32, ptr [[SCEVGEP6]], align 465; CHECK-NEXT: [[CMP_35:%.*]] = icmp slt i32 [[GEPLOAD18]], [[T]]66; CHECK-NEXT: br i1 [[CMP_35]], label [[THEN_34_LOOP_EXIT_LOOPEXIT_CRIT_EDGE]], label [[IFMERGE_34]]67; CHECK: ifmerge.34:68; CHECK-NEXT: [[TMP5:%.*]] = shl nuw nsw i64 [[LSR_IV1]], 269; CHECK-NEXT: [[SCEVGEP4:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP5]]70; CHECK-NEXT: [[GEPLOAD20:%.*]] = load i32, ptr [[SCEVGEP4]], align 471; CHECK-NEXT: [[CMP_38:%.*]] = icmp sgt i32 [[GEPLOAD20]], [[T]]72; CHECK-NEXT: [[CMP_39:%.*]] = icmp slt i32 [[GEPLOAD]], [[T]]73; CHECK-NEXT: [[OR_COND:%.*]] = and i1 [[CMP_38]], [[CMP_39]]74; CHECK-NEXT: br i1 [[OR_COND]], label [[LOOP_EXIT_LOOPEXITSPLITSPLITSPLIT]], label [[IFMERGE_38]]75; CHECK: ifmerge.38:76; CHECK-NEXT: [[TMP6:%.*]] = shl nuw nsw i64 [[LSR_IV1]], 277; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP6]]78; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[SCEVGEP2]], i64 479; CHECK-NEXT: [[GEPLOAD24:%.*]] = load i32, ptr [[SCEVGEP3]], align 480; CHECK-NEXT: [[CMP_42:%.*]] = icmp sgt i32 [[GEPLOAD24]], [[T]]81; CHECK-NEXT: [[CMP_43:%.*]] = icmp slt i32 [[GEPLOAD20]], [[T]]82; CHECK-NEXT: [[OR_COND55:%.*]] = and i1 [[CMP_42]], [[CMP_43]]83; CHECK-NEXT: br i1 [[OR_COND55]], label [[IFMERGE_38_LOOP_EXIT_LOOPEXITSPLITSPLIT_CRIT_EDGE]], label [[IFMERGE_42]]84; CHECK: ifmerge.42:85; CHECK-NEXT: [[TMP7:%.*]] = shl nuw nsw i64 [[LSR_IV1]], 286; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP7]]87; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[SCEVGEP]], i64 888; CHECK-NEXT: [[GEPLOAD28:%.*]] = load i32, ptr [[SCEVGEP1]], align 489; CHECK-NEXT: [[CMP_46:%.*]] = icmp sgt i32 [[GEPLOAD28]], [[T]]90; CHECK-NEXT: [[CMP_47:%.*]] = icmp slt i32 [[GEPLOAD24]], [[T]]91; CHECK-NEXT: [[OR_COND56:%.*]] = and i1 [[CMP_46]], [[CMP_47]]92; CHECK-NEXT: br i1 [[OR_COND56]], label [[IFMERGE_42_LOOP_EXIT_LOOPEXITSPLIT_CRIT_EDGE]], label [[IFMERGE_46]]93; CHECK: ifmerge.46:94; CHECK-NEXT: [[NEXTIVLOOP_32]] = add nuw nsw i64 [[I1_I64_0]], 195; CHECK-NEXT: [[LSR_IV_NEXT]] = add nuw nsw i64 [[LSR_IV1]], 496; CHECK-NEXT: [[CONDLOOP_32:%.*]] = icmp ult i64 [[NEXTIVLOOP_32]], 1297; CHECK-NEXT: br i1 [[CONDLOOP_32]], label [[LOOP_32]], label [[LOOP_25:%.*]]98; CHECK: loop.25:99; CHECK-NEXT: [[ARRAYIDX31:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 49100; CHECK-NEXT: [[GEPLOAD32:%.*]] = load i32, ptr [[ARRAYIDX31]], align 4101; CHECK-NEXT: [[CMP_8:%.*]] = icmp sgt i32 [[GEPLOAD32]], [[T]]102; CHECK-NEXT: br i1 [[CMP_8]], label [[THEN_8]], label [[IFMERGE_8]]103; CHECK: then.8:104; CHECK-NEXT: [[ARRAYIDX33:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 48105; CHECK-NEXT: [[GEPLOAD34:%.*]] = load i32, ptr [[ARRAYIDX33]], align 4106; CHECK-NEXT: [[CMP_15:%.*]] = icmp slt i32 [[GEPLOAD34]], [[T]]107; CHECK-NEXT: br i1 [[CMP_15]], label [[LOOP_EXIT]], label [[IFMERGE_8]]108; CHECK: ifmerge.8:109; CHECK-NEXT: [[ARRAYIDX31_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 50110; CHECK-NEXT: [[GEPLOAD32_1:%.*]] = load i32, ptr [[ARRAYIDX31_1]], align 4111; CHECK-NEXT: [[CMP_8_1:%.*]] = icmp sgt i32 [[GEPLOAD32_1]], [[T]]112; CHECK-NEXT: br i1 [[CMP_8_1]], label [[THEN_8_1]], label [[FOR_END]]113; CHECK: then.8.1:114; CHECK-NEXT: [[ARRAYIDX33_1:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 49115; CHECK-NEXT: [[GEPLOAD34_1:%.*]] = load i32, ptr [[ARRAYIDX33_1]], align 4116; CHECK-NEXT: [[CMP_15_1:%.*]] = icmp slt i32 [[GEPLOAD34_1]], [[T]]117; CHECK-NEXT: br i1 [[CMP_15_1]], label [[LOOP_EXIT]], label [[FOR_END]]118;119entry:120 br label %loop.32121 122loop.exit: ; preds = %then.8.1, %then.8, %ifmerge.42, %ifmerge.38, %ifmerge.34, %then.34123 %indvars.iv.lcssa = phi i64 [ 48, %then.8 ], [ 49, %then.8.1 ], [ %tmp4, %ifmerge.42 ], [ %tmp3, %ifmerge.38 ], [ %tmp2, %ifmerge.34 ], [ %tmp1, %then.34 ]124 %tmp = trunc i64 %indvars.iv.lcssa to i32125 br label %for.end126 127for.end: ; preds = %then.8.1, %ifmerge.8, %loop.exit128 %i.0.lcssa = phi i32 [ %tmp, %loop.exit ], [ 50, %then.8.1 ], [ 50, %ifmerge.8 ]129 ret i32 %i.0.lcssa130 131; shl instruction will be dead eliminated when all it's uses will be rewritten.132loop.32: ; preds = %ifmerge.46, %entry133 %i1.i64.0 = phi i64 [ 0, %entry ], [ %nextivloop.32, %ifmerge.46 ]134 %tmp1 = shl i64 %i1.i64.0, 2135 %tmp2 = or disjoint i64 %tmp1, 1136 %arrayIdx = getelementptr inbounds i32, ptr %A, i64 %tmp2137 %gepload = load i32, ptr %arrayIdx, align 4138 %cmp.34 = icmp sgt i32 %gepload, %t139 br i1 %cmp.34, label %then.34, label %ifmerge.34140 141then.34: ; preds = %loop.32142 %arrayIdx17 = getelementptr inbounds i32, ptr %A, i64 %tmp1143 %gepload18 = load i32, ptr %arrayIdx17, align 4144 %cmp.35 = icmp slt i32 %gepload18, %t145 br i1 %cmp.35, label %loop.exit, label %ifmerge.34146 147ifmerge.34: ; preds = %then.34, %loop.32148 %tmp3 = or disjoint i64 %tmp1, 2149 %arrayIdx19 = getelementptr inbounds i32, ptr %A, i64 %tmp3150 %gepload20 = load i32, ptr %arrayIdx19, align 4151 %cmp.38 = icmp sgt i32 %gepload20, %t152 %cmp.39 = icmp slt i32 %gepload, %t153 %or.cond = and i1 %cmp.38, %cmp.39154 br i1 %or.cond, label %loop.exit, label %ifmerge.38155 156ifmerge.38: ; preds = %ifmerge.34157 %tmp4 = or disjoint i64 %tmp1, 3158 %arrayIdx23 = getelementptr inbounds i32, ptr %A, i64 %tmp4159 %gepload24 = load i32, ptr %arrayIdx23, align 4160 %cmp.42 = icmp sgt i32 %gepload24, %t161 %cmp.43 = icmp slt i32 %gepload20, %t162 %or.cond55 = and i1 %cmp.42, %cmp.43163 br i1 %or.cond55, label %loop.exit, label %ifmerge.42164 165ifmerge.42: ; preds = %ifmerge.38166 %tmp5 = add i64 %tmp1, 4167 %arrayIdx27 = getelementptr inbounds i32, ptr %A, i64 %tmp5168 %gepload28 = load i32, ptr %arrayIdx27, align 4169 %cmp.46 = icmp sgt i32 %gepload28, %t170 %cmp.47 = icmp slt i32 %gepload24, %t171 %or.cond56 = and i1 %cmp.46, %cmp.47172 br i1 %or.cond56, label %loop.exit, label %ifmerge.46173 174ifmerge.46: ; preds = %ifmerge.42175 %nextivloop.32 = add nuw nsw i64 %i1.i64.0, 1176 %condloop.32 = icmp ult i64 %nextivloop.32, 12177 br i1 %condloop.32, label %loop.32, label %loop.25178 179loop.25: ; preds = %ifmerge.46180 %arrayIdx31 = getelementptr inbounds i32, ptr %A, i64 49181 %gepload32 = load i32, ptr %arrayIdx31, align 4182 %cmp.8 = icmp sgt i32 %gepload32, %t183 br i1 %cmp.8, label %then.8, label %ifmerge.8184 185then.8: ; preds = %loop.25186 %arrayIdx33 = getelementptr inbounds i32, ptr %A, i64 48187 %gepload34 = load i32, ptr %arrayIdx33, align 4188 %cmp.15 = icmp slt i32 %gepload34, %t189 br i1 %cmp.15, label %loop.exit, label %ifmerge.8190 191ifmerge.8: ; preds = %then.8, %loop.25192 %arrayIdx31.1 = getelementptr inbounds i32, ptr %A, i64 50193 %gepload32.1 = load i32, ptr %arrayIdx31.1, align 4194 %cmp.8.1 = icmp sgt i32 %gepload32.1, %t195 br i1 %cmp.8.1, label %then.8.1, label %for.end196 197then.8.1: ; preds = %ifmerge.8198 %arrayIdx33.1 = getelementptr inbounds i32, ptr %A, i64 49199 %gepload34.1 = load i32, ptr %arrayIdx33.1, align 4200 %cmp.15.1 = icmp slt i32 %gepload34.1, %t201 br i1 %cmp.15.1, label %loop.exit, label %for.end202}203