199 lines · plain
1; REQUIRES: aarch64-registered-target2; RUN: opt -passes='require<profile-summary>,function(codegenprepare)' < %s -mtriple=aarch64-none-linux-gnu -S | FileCheck %s3 4target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"5target triple = "aarch64--linux-gnu"6 7; Expect to skip merging two empty blocks (sw.bb and sw.bb2) into sw.epilog8; as both of them are unlikely executed.9define i32 @f_switch(i32 %c) {10; CHECK-LABEL: @f_switch11; CHECK-LABEL: entry:12; CHECK: i32 10, label %sw.bb13; CHECK: i32 20, label %sw.bb214entry:15 switch i32 %c, label %sw.default [16 i32 10, label %sw.bb17 i32 20, label %sw.bb218 i32 30, label %sw.bb319 i32 40, label %sw.bb420 ], !prof !021 22sw.bb: ; preds = %entry23 br label %sw.epilog24 25sw.bb2: ; preds = %entry26 br label %sw.epilog27 28sw.bb3: ; preds = %entry29 call void @callcase3()30 br label %sw.epilog31 32sw.bb4: ; preds = %entry33 call void @callcase4()34 br label %sw.epilog35 36sw.default: ; preds = %entry37 call void @calldefault()38 br label %sw.epilog39 40; CHECK-LABEL: sw.epilog:41; CHECK: %fp.0 = phi ptr [ @FD, %sw.default ], [ @F4, %sw.bb4 ], [ @F3, %sw.bb3 ], [ @F2, %sw.bb2 ], [ @F1, %sw.bb ]42sw.epilog: ; preds = %sw.default, %sw.bb3, %sw.bb2, %sw.bb43 %fp.0 = phi ptr [ @FD, %sw.default ], [ @F4, %sw.bb4 ], [ @F3, %sw.bb3 ], [ @F2, %sw.bb2 ], [ @F1, %sw.bb ]44 call void %fp.0()45 ret i32 046}47 48; Expect not to merge sw.bb2 because of the conflict in the incoming value from49; sw.bb which is already merged.50define i32 @f_switch2(i32 %c) {51; CHECK-LABEL: @f_switch252; CHECK-LABEL: entry:53; CHECK: i32 10, label %sw.epilog54; CHECK: i32 20, label %sw.bb255entry:56 switch i32 %c, label %sw.default [57 i32 10, label %sw.bb58 i32 20, label %sw.bb259 i32 30, label %sw.bb360 i32 40, label %sw.bb461 ], !prof !162 63sw.bb: ; preds = %entry64 br label %sw.epilog65 66sw.bb2: ; preds = %entry67 br label %sw.epilog68 69sw.bb3: ; preds = %entry70 call void @callcase3()71 br label %sw.epilog72 73sw.bb4: ; preds = %entry74 call void @callcase4()75 br label %sw.epilog76 77sw.default: ; preds = %entry78 call void @calldefault()79 br label %sw.epilog80 81; CHECK-LABEL: sw.epilog:82; CHECK: %fp.0 = phi ptr [ @FD, %sw.default ], [ @F4, %sw.bb4 ], [ @F3, %sw.bb3 ], [ @F2, %sw.bb2 ], [ @F1, %entry ]83sw.epilog: ; preds = %sw.default, %sw.bb3, %sw.bb2, %sw.bb84 %fp.0 = phi ptr [ @FD, %sw.default ], [ @F4, %sw.bb4 ], [ @F3, %sw.bb3 ], [ @F2, %sw.bb2 ], [ @F1, %sw.bb ]85 call void %fp.0()86 ret i32 087}88 89; Multiple empty blocks should be considered together if all incoming values90; from them are same. We expect to merge both empty blocks (sw.bb and sw.bb2)91; because the sum of frequencies are higer than the threshold.92define i32 @f_switch3(i32 %c) {93; CHECK-LABEL: @f_switch394; CHECK-LABEL: entry:95; CHECK: i32 10, label %sw.epilog96; CHECK: i32 20, label %sw.epilog97entry:98 switch i32 %c, label %sw.default [99 i32 10, label %sw.bb100 i32 20, label %sw.bb2101 i32 30, label %sw.bb3102 i32 40, label %sw.bb4103 ], !prof !2104 105sw.bb: ; preds = %entry106 br label %sw.epilog107 108sw.bb2: ; preds = %entry109 br label %sw.epilog110 111sw.bb3: ; preds = %entry112 call void @callcase3()113 br label %sw.epilog114 115sw.bb4: ; preds = %entry116 call void @callcase4()117 br label %sw.epilog118 119sw.default: ; preds = %entry120 call void @calldefault()121 br label %sw.epilog122 123; CHECK-LABEL: sw.epilog:124; CHECK: %fp.0 = phi ptr [ @FD, %sw.default ], [ @F4, %sw.bb4 ], [ @F3, %sw.bb3 ], [ @F1, %entry ], [ @F1, %entry ]125sw.epilog: ; preds = %sw.default, %sw.bb3, %sw.bb2, %sw.bb126 %fp.0 = phi ptr [ @FD, %sw.default ], [ @F4, %sw.bb4 ], [ @F3, %sw.bb3 ], [ @F1, %sw.bb2 ], [ @F1, %sw.bb ]127 call void %fp.0()128 ret i32 0129}130 131declare void @F1(...) local_unnamed_addr132declare void @F2(...) local_unnamed_addr133declare void @F3(...) local_unnamed_addr134declare void @F4(...) local_unnamed_addr135declare void @FD(...) local_unnamed_addr136declare void @callcase3(...) local_unnamed_addr137declare void @callcase4(...) local_unnamed_addr138declare void @calldefault(...) local_unnamed_addr139 140!0 = !{!"branch_weights", i32 5, i32 1, i32 1,i32 5, i32 5}141!1 = !{!"branch_weights", i32 1 , i32 5, i32 1,i32 1, i32 1}142!2 = !{!"branch_weights", i32 1 , i32 4, i32 1,i32 1, i32 1}143 144 145; This test that BFI/BPI is created without any assertion in isMergingEmptyBlockProfitable()146; in the case where empty blocks are removed before creating BFI/BPI.147@b = common global i32 0, align 4148@a = common global ptr null, align 8149define i32 @should_not_assert(i32 %i) local_unnamed_addr {150entry:151 %0 = load i32, ptr @b, align 4152 %cond = icmp eq i32 %0, 6153 br i1 %cond, label %while.cond.preheader, label %sw.epilog154 155while.cond.preheader: ; preds = %entry156 %1 = load ptr, ptr @a, align 8157 %magicptr = ptrtoint ptr %1 to i64158 %arrayidx = getelementptr inbounds i32, ptr %1, i64 1159 br label %while.cond160 161while.cond: ; preds = %while.cond.preheader, %land.rhs162 switch i64 %magicptr, label %land.rhs [163 i64 32, label %while.cond2.loopexit164 i64 0, label %while.cond2.loopexit165 ]166 167land.rhs: ; preds = %while.cond168 %2 = load i32, ptr %arrayidx, align 4169 %tobool1 = icmp eq i32 %2, 0170 br i1 %tobool1, label %while.cond2thread-pre-split.loopexit, label %while.cond171 172while.cond2thread-pre-split.loopexit: ; preds = %land.rhs173 br label %while.cond2thread-pre-split174 175while.cond2thread-pre-split: ; preds = %while.cond2thread-pre-split.loopexit, %while.body4176 %.pr = phi ptr [ %.pr.pre, %while.body4 ], [ %1, %while.cond2thread-pre-split.loopexit ]177 br label %while.cond2178 179while.cond2.loopexit: ; preds = %while.cond, %while.cond180 br label %while.cond2181 182while.cond2: ; preds = %while.cond2.loopexit, %while.cond2thread-pre-split183 %3 = phi ptr [ %.pr, %while.cond2thread-pre-split ], [ %1, %while.cond2.loopexit ]184 %tobool3 = icmp eq ptr %3, null185 br i1 %tobool3, label %sw.epilog, label %while.body4186 187while.body4: ; preds = %while.cond2188 tail call void @fn2()189 %.pr.pre = load ptr, ptr @a, align 8190 br label %while.cond2thread-pre-split191 192sw.epilog: ; preds = %while.cond2, %entry193 ret i32 undef194}195 196 197declare void @fn2(...) local_unnamed_addr198 199