1 ; RUN: opt -codegenprepare < %s -mtriple=aarch64-none-linux-gnu -S | FileCheck %s 2 3 target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128" 4 target triple = "aarch64--linux-gnu" 5 6 ; Expect to skip merging two empty blocks (sw.bb and sw.bb2) into sw.epilog 7 ; as both of them are unlikely executed. 8 define i32 @f_switch(i32 %c) { 9 ; CHECK-LABEL: @f_switch 10 ; CHECK-LABEL: entry: 11 ; CHECK: i32 10, label %sw.bb 12 ; CHECK: i32 20, label %sw.bb2 13 entry: 14 switch i32 %c, label %sw.default [ 15 i32 10, label %sw.bb 16 i32 20, label %sw.bb2 17 i32 30, label %sw.bb3 18 i32 40, label %sw.bb4 19 ], !prof !0 20 21 sw.bb: ; preds = %entry 22 br label %sw.epilog 23 24 sw.bb2: ; preds = %entry 25 br label %sw.epilog 26 27 sw.bb3: ; preds = %entry 28 call void bitcast (void (...)* @callcase3 to void ()*)() 29 br label %sw.epilog 30 31 sw.bb4: ; preds = %entry 32 call void bitcast (void (...)* @callcase4 to void ()*)() 33 br label %sw.epilog 34 35 sw.default: ; preds = %entry 36 call void bitcast (void (...)* @calldefault to void ()*)() 37 br label %sw.epilog 38 39 ; CHECK-LABEL: sw.epilog: 40 ; CHECK: %fp.0 = phi void (...)* [ @FD, %sw.default ], [ @F4, %sw.bb4 ], [ @F3, %sw.bb3 ], [ @F2, %sw.bb2 ], [ @F1, %sw.bb ] 41 sw.epilog: ; preds = %sw.default, %sw.bb3, %sw.bb2, %sw.bb 42 %fp.0 = phi void (...)* [ @FD, %sw.default ], [ @F4, %sw.bb4 ], [ @F3, %sw.bb3 ], [ @F2, %sw.bb2 ], [ @F1, %sw.bb ] 43 %callee.knr.cast = bitcast void (...)* %fp.0 to void ()* 44 call void %callee.knr.cast() 45 ret i32 0 46 } 47 48 ; Expect not to merge sw.bb2 because of the conflict in the incoming value from 49 ; sw.bb which is already merged. 50 define i32 @f_switch2(i32 %c) { 51 ; CHECK-LABEL: @f_switch2 52 ; CHECK-LABEL: entry: 53 ; CHECK: i32 10, label %sw.epilog 54 ; CHECK: i32 20, label %sw.bb2 55 entry: 56 switch i32 %c, label %sw.default [ 57 i32 10, label %sw.bb 58 i32 20, label %sw.bb2 59 i32 30, label %sw.bb3 60 i32 40, label %sw.bb4 61 ], !prof !1 62 63 sw.bb: ; preds = %entry 64 br label %sw.epilog 65 66 sw.bb2: ; preds = %entry 67 br label %sw.epilog 68 69 sw.bb3: ; preds = %entry 70 call void bitcast (void (...)* @callcase3 to void ()*)() 71 br label %sw.epilog 72 73 sw.bb4: ; preds = %entry 74 call void bitcast (void (...)* @callcase4 to void ()*)() 75 br label %sw.epilog 76 77 sw.default: ; preds = %entry 78 call void bitcast (void (...)* @calldefault to void ()*)() 79 br label %sw.epilog 80 81 ; CHECK-LABEL: sw.epilog: 82 ; CHECK: %fp.0 = phi void (...)* [ @FD, %sw.default ], [ @F4, %sw.bb4 ], [ @F3, %sw.bb3 ], [ @F2, %sw.bb2 ], [ @F1, %entry ] 83 sw.epilog: ; preds = %sw.default, %sw.bb3, %sw.bb2, %sw.bb 84 %fp.0 = phi void (...)* [ @FD, %sw.default ], [ @F4, %sw.bb4 ], [ @F3, %sw.bb3 ], [ @F2, %sw.bb2 ], [ @F1, %sw.bb ] 85 %callee.knr.cast = bitcast void (...)* %fp.0 to void ()* 86 call void %callee.knr.cast() 87 ret i32 0 88 } 89 90 ; Multiple empty blocks should be considered together if all incoming values 91 ; from them are same. We expect to merge both empty blocks (sw.bb and sw.bb2) 92 ; because the sum of frequencies are higer than the threshold. 93 define i32 @f_switch3(i32 %c) { 94 ; CHECK-LABEL: @f_switch3 95 ; CHECK-LABEL: entry: 96 ; CHECK: i32 10, label %sw.epilog 97 ; CHECK: i32 20, label %sw.epilog 98 entry: 99 switch i32 %c, label %sw.default [ 100 i32 10, label %sw.bb 101 i32 20, label %sw.bb2 102 i32 30, label %sw.bb3 103 i32 40, label %sw.bb4 104 ], !prof !2 105 106 sw.bb: ; preds = %entry 107 br label %sw.epilog 108 109 sw.bb2: ; preds = %entry 110 br label %sw.epilog 111 112 sw.bb3: ; preds = %entry 113 call void bitcast (void (...)* @callcase3 to void ()*)() 114 br label %sw.epilog 115 116 sw.bb4: ; preds = %entry 117 call void bitcast (void (...)* @callcase4 to void ()*)() 118 br label %sw.epilog 119 120 sw.default: ; preds = %entry 121 call void bitcast (void (...)* @calldefault to void ()*)() 122 br label %sw.epilog 123 124 ; CHECK-LABEL: sw.epilog: 125 ; CHECK: %fp.0 = phi void (...)* [ @FD, %sw.default ], [ @F4, %sw.bb4 ], [ @F3, %sw.bb3 ], [ @F1, %entry ], [ @F1, %entry ] 126 sw.epilog: ; preds = %sw.default, %sw.bb3, %sw.bb2, %sw.bb 127 %fp.0 = phi void (...)* [ @FD, %sw.default ], [ @F4, %sw.bb4 ], [ @F3, %sw.bb3 ], [ @F1, %sw.bb2 ], [ @F1, %sw.bb ] 128 %callee.knr.cast = bitcast void (...)* %fp.0 to void ()* 129 call void %callee.knr.cast() 130 ret i32 0 131 } 132 133 declare void @F1(...) local_unnamed_addr 134 declare void @F2(...) local_unnamed_addr 135 declare void @F3(...) local_unnamed_addr 136 declare void @F4(...) local_unnamed_addr 137 declare void @FD(...) local_unnamed_addr 138 declare void @callcase3(...) local_unnamed_addr 139 declare void @callcase4(...) local_unnamed_addr 140 declare void @calldefault(...) local_unnamed_addr 141 142 !0 = !{!"branch_weights", i32 5, i32 1, i32 1,i32 5, i32 5} 143 !1 = !{!"branch_weights", i32 1 , i32 5, i32 1,i32 1, i32 1} 144 !2 = !{!"branch_weights", i32 1 , i32 4, i32 1,i32 1, i32 1} 145 146 147 ; This test that BFI/BPI is created without any assertion in isMergingEmptyBlockProfitable() 148 ; in the case where empty blocks are removed before creating BFI/BPI. 149 @b = common global i32 0, align 4 150 @a = common global i32* null, align 8 151 define i32 @should_not_assert(i32 %i) local_unnamed_addr { 152 entry: 153 %0 = load i32, i32* @b, align 4 154 %cond = icmp eq i32 %0, 6 155 br i1 %cond, label %while.cond.preheader, label %sw.epilog 156 157 while.cond.preheader: ; preds = %entry 158 %1 = load i32*, i32** @a, align 8 159 %magicptr = ptrtoint i32* %1 to i64 160 %arrayidx = getelementptr inbounds i32, i32* %1, i64 1 161 br label %while.cond 162 163 while.cond: ; preds = %while.cond.preheader, %land.rhs 164 switch i64 %magicptr, label %land.rhs [ 165 i64 32, label %while.cond2.loopexit 166 i64 0, label %while.cond2.loopexit 167 ] 168 169 land.rhs: ; preds = %while.cond 170 %2 = load i32, i32* %arrayidx, align 4 171 %tobool1 = icmp eq i32 %2, 0 172 br i1 %tobool1, label %while.cond2thread-pre-split.loopexit, label %while.cond 173 174 while.cond2thread-pre-split.loopexit: ; preds = %land.rhs 175 br label %while.cond2thread-pre-split 176 177 while.cond2thread-pre-split: ; preds = %while.cond2thread-pre-split.loopexit, %while.body4 178 %.pr = phi i32* [ %.pr.pre, %while.body4 ], [ %1, %while.cond2thread-pre-split.loopexit ] 179 br label %while.cond2 180 181 while.cond2.loopexit: ; preds = %while.cond, %while.cond 182 br label %while.cond2 183 184 while.cond2: ; preds = %while.cond2.loopexit, %while.cond2thread-pre-split 185 %3 = phi i32* [ %.pr, %while.cond2thread-pre-split ], [ %1, %while.cond2.loopexit ] 186 %tobool3 = icmp eq i32* %3, null 187 br i1 %tobool3, label %sw.epilog, label %while.body4 188 189 while.body4: ; preds = %while.cond2 190 tail call void bitcast (void (...)* @fn2 to void ()*)() 191 %.pr.pre = load i32*, i32** @a, align 8 192 br label %while.cond2thread-pre-split 193 194 sw.epilog: ; preds = %while.cond2, %entry 195 ret i32 undef 196 } 197 198 199 declare void @fn2(...) local_unnamed_addr 200 201
