1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -simplifycfg -simplifycfg-require-and-preserve-domtree=1 -simplifycfg-max-small-block-size=6 -S < %s | FileCheck %s 3; RUN: opt -passes=simplifycfg -simplifycfg-max-small-block-size=6 -S < %s | FileCheck %s 4 5target datalayout = "e-p:64:64-p5:32:32-A5" 6 7declare void @llvm.assume(i1) 8declare i1 @llvm.type.test(i8*, metadata) nounwind readnone 9 10define void @test_01(i1 %c, i64* align 1 %ptr) local_unnamed_addr #0 { 11; CHECK-LABEL: @test_01( 12; CHECK-NEXT: br i1 [[C:%.*]], label [[TRUE2_CRITEDGE:%.*]], label [[FALSE1:%.*]] 13; CHECK: false1: 14; CHECK-NEXT: store volatile i64 1, i64* [[PTR:%.*]], align 4 15; CHECK-NEXT: [[PTRINT:%.*]] = ptrtoint i64* [[PTR]] to i64 16; CHECK-NEXT: [[MASKEDPTR:%.*]] = and i64 [[PTRINT]], 7 17; CHECK-NEXT: [[MASKCOND:%.*]] = icmp eq i64 [[MASKEDPTR]], 0 18; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) 19; CHECK-NEXT: store volatile i64 0, i64* [[PTR]], align 8 20; CHECK-NEXT: store volatile i64 3, i64* [[PTR]], align 8 21; CHECK-NEXT: br label [[COMMON_RET:%.*]] 22; CHECK: common.ret: 23; CHECK-NEXT: ret void 24; CHECK: true2.critedge: 25; CHECK-NEXT: [[PTRINT_C:%.*]] = ptrtoint i64* [[PTR]] to i64 26; CHECK-NEXT: [[MASKEDPTR_C:%.*]] = and i64 [[PTRINT_C]], 7 27; CHECK-NEXT: [[MASKCOND_C:%.*]] = icmp eq i64 [[MASKEDPTR_C]], 0 28; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND_C]]) 29; CHECK-NEXT: store volatile i64 0, i64* [[PTR]], align 8 30; CHECK-NEXT: store volatile i64 2, i64* [[PTR]], align 8 31; CHECK-NEXT: br label [[COMMON_RET]] 32; 33 br i1 %c, label %true1, label %false1 34 35true1: ; preds = %false1, %0 36 %ptrint = ptrtoint i64* %ptr to i64 37 %maskedptr = and i64 %ptrint, 7 38 %maskcond = icmp eq i64 %maskedptr, 0 39 tail call void @llvm.assume(i1 %maskcond) 40 store volatile i64 0, i64* %ptr, align 8 41 br i1 %c, label %true2, label %false2 42 43false1: ; preds = %0 44 store volatile i64 1, i64* %ptr, align 4 45 br label %true1 46 47true2: ; preds = %true1 48 store volatile i64 2, i64* %ptr, align 8 49 ret void 50 51false2: ; preds = %true1 52 store volatile i64 3, i64* %ptr, align 8 53 ret void 54} 55 56; Corner case: the block has max possible size for which we still do PRE. 57define void @test_02(i1 %c, i64* align 1 %ptr) local_unnamed_addr #0 { 58; CHECK-LABEL: @test_02( 59; CHECK-NEXT: br i1 [[C:%.*]], label [[TRUE2_CRITEDGE:%.*]], label [[FALSE1:%.*]] 60; CHECK: false1: 61; CHECK-NEXT: store volatile i64 1, i64* [[PTR:%.*]], align 4 62; CHECK-NEXT: [[PTRINT:%.*]] = ptrtoint i64* [[PTR]] to i64 63; CHECK-NEXT: [[MASKEDPTR:%.*]] = and i64 [[PTRINT]], 7 64; CHECK-NEXT: [[MASKCOND:%.*]] = icmp eq i64 [[MASKEDPTR]], 0 65; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) 66; CHECK-NEXT: store volatile i64 0, i64* [[PTR]], align 8 67; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 68; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 69; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 70; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 71; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 72; CHECK-NEXT: store volatile i64 3, i64* [[PTR]], align 8 73; CHECK-NEXT: br label [[COMMON_RET:%.*]] 74; CHECK: common.ret: 75; CHECK-NEXT: ret void 76; CHECK: true2.critedge: 77; CHECK-NEXT: [[PTRINT_C:%.*]] = ptrtoint i64* [[PTR]] to i64 78; CHECK-NEXT: [[MASKEDPTR_C:%.*]] = and i64 [[PTRINT_C]], 7 79; CHECK-NEXT: [[MASKCOND_C:%.*]] = icmp eq i64 [[MASKEDPTR_C]], 0 80; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND_C]]) 81; CHECK-NEXT: store volatile i64 0, i64* [[PTR]], align 8 82; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 83; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 84; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 85; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 86; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 87; CHECK-NEXT: store volatile i64 2, i64* [[PTR]], align 8 88; CHECK-NEXT: br label [[COMMON_RET]] 89; 90 br i1 %c, label %true1, label %false1 91 92true1: ; preds = %false1, %0 93 %ptrint = ptrtoint i64* %ptr to i64 94 %maskedptr = and i64 %ptrint, 7 95 %maskcond = icmp eq i64 %maskedptr, 0 96 tail call void @llvm.assume(i1 %maskcond) 97 store volatile i64 0, i64* %ptr, align 8 98 store volatile i64 -1, i64* %ptr, align 8 99 store volatile i64 -1, i64* %ptr, align 8 100 store volatile i64 -1, i64* %ptr, align 8 101 store volatile i64 -1, i64* %ptr, align 8 102 store volatile i64 -1, i64* %ptr, align 8 103 br i1 %c, label %true2, label %false2 104 105false1: ; preds = %0 106 store volatile i64 1, i64* %ptr, align 4 107 br label %true1 108 109true2: ; preds = %true1 110 store volatile i64 2, i64* %ptr, align 8 111 ret void 112 113false2: ; preds = %true1 114 store volatile i64 3, i64* %ptr, align 8 115 ret void 116} 117 118; This block is too huge for PRE. 119define void @test_03(i1 %c, i64* align 1 %ptr) local_unnamed_addr #0 { 120; CHECK-LABEL: @test_03( 121; CHECK-NEXT: br i1 [[C:%.*]], label [[TRUE1:%.*]], label [[FALSE1:%.*]] 122; CHECK: true1: 123; CHECK-NEXT: [[PTRINT:%.*]] = ptrtoint i64* [[PTR:%.*]] to i64 124; CHECK-NEXT: [[MASKEDPTR:%.*]] = and i64 [[PTRINT]], 7 125; CHECK-NEXT: [[MASKCOND:%.*]] = icmp eq i64 [[MASKEDPTR]], 0 126; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) 127; CHECK-NEXT: store volatile i64 0, i64* [[PTR]], align 8 128; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 129; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 130; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 131; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 132; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 133; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 134; CHECK-NEXT: br i1 [[C]], label [[TRUE2:%.*]], label [[FALSE2:%.*]] 135; CHECK: false1: 136; CHECK-NEXT: store volatile i64 1, i64* [[PTR]], align 4 137; CHECK-NEXT: br label [[TRUE1]] 138; CHECK: common.ret: 139; CHECK-NEXT: ret void 140; CHECK: true2: 141; CHECK-NEXT: store volatile i64 2, i64* [[PTR]], align 8 142; CHECK-NEXT: br label [[COMMON_RET:%.*]] 143; CHECK: false2: 144; CHECK-NEXT: store volatile i64 3, i64* [[PTR]], align 8 145; CHECK-NEXT: br label [[COMMON_RET]] 146; 147 br i1 %c, label %true1, label %false1 148 149true1: ; preds = %false1, %0 150 %ptrint = ptrtoint i64* %ptr to i64 151 %maskedptr = and i64 %ptrint, 7 152 %maskcond = icmp eq i64 %maskedptr, 0 153 tail call void @llvm.assume(i1 %maskcond) 154 store volatile i64 0, i64* %ptr, align 8 155 store volatile i64 -1, i64* %ptr, align 8 156 store volatile i64 -1, i64* %ptr, align 8 157 store volatile i64 -1, i64* %ptr, align 8 158 store volatile i64 -1, i64* %ptr, align 8 159 store volatile i64 -1, i64* %ptr, align 8 160 store volatile i64 -1, i64* %ptr, align 8 161 br i1 %c, label %true2, label %false2 162 163false1: ; preds = %0 164 store volatile i64 1, i64* %ptr, align 4 165 br label %true1 166 167true2: ; preds = %true1 168 store volatile i64 2, i64* %ptr, align 8 169 ret void 170 171false2: ; preds = %true1 172 store volatile i64 3, i64* %ptr, align 8 173 ret void 174} 175 176; Try the max block size for PRE again but with the bitcast/type test/assume 177; sequence used for whole program devirt. 178define void @test_04(i1 %c, i64* align 1 %ptr, [3 x i8*]* %vtable) local_unnamed_addr #0 { 179; CHECK-LABEL: @test_04( 180; CHECK-NEXT: br i1 [[C:%.*]], label [[TRUE2_CRITEDGE:%.*]], label [[FALSE1:%.*]] 181; CHECK: false1: 182; CHECK-NEXT: store volatile i64 1, i64* [[PTR:%.*]], align 4 183; CHECK-NEXT: [[VTABLEI8:%.*]] = bitcast [3 x i8*]* [[VTABLE:%.*]] to i8* 184; CHECK-NEXT: [[P:%.*]] = call i1 @llvm.type.test(i8* [[VTABLEI8]], metadata !"foo") 185; CHECK-NEXT: tail call void @llvm.assume(i1 [[P]]) 186; CHECK-NEXT: store volatile i64 0, i64* [[PTR]], align 8 187; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 188; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 189; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 190; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 191; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 192; CHECK-NEXT: store volatile i64 3, i64* [[PTR]], align 8 193; CHECK-NEXT: br label [[COMMON_RET:%.*]] 194; CHECK: common.ret: 195; CHECK-NEXT: ret void 196; CHECK: true2.critedge: 197; CHECK-NEXT: [[VTABLEI8_C:%.*]] = bitcast [3 x i8*]* [[VTABLE]] to i8* 198; CHECK-NEXT: [[P_C:%.*]] = call i1 @llvm.type.test(i8* [[VTABLEI8_C]], metadata !"foo") 199; CHECK-NEXT: tail call void @llvm.assume(i1 [[P_C]]) 200; CHECK-NEXT: store volatile i64 0, i64* [[PTR]], align 8 201; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 202; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 203; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 204; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 205; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 206; CHECK-NEXT: store volatile i64 2, i64* [[PTR]], align 8 207; CHECK-NEXT: br label [[COMMON_RET]] 208; 209 br i1 %c, label %true1, label %false1 210 211true1: ; preds = %false1, %0 212 %vtablei8 = bitcast [3 x i8*]* %vtable to i8* 213 %p = call i1 @llvm.type.test(i8* %vtablei8, metadata !"foo") 214 tail call void @llvm.assume(i1 %p) 215 store volatile i64 0, i64* %ptr, align 8 216 store volatile i64 -1, i64* %ptr, align 8 217 store volatile i64 -1, i64* %ptr, align 8 218 store volatile i64 -1, i64* %ptr, align 8 219 store volatile i64 -1, i64* %ptr, align 8 220 store volatile i64 -1, i64* %ptr, align 8 221 br i1 %c, label %true2, label %false2 222 223false1: ; preds = %0 224 store volatile i64 1, i64* %ptr, align 4 225 br label %true1 226 227true2: ; preds = %true1 228 store volatile i64 2, i64* %ptr, align 8 229 ret void 230 231false2: ; preds = %true1 232 store volatile i64 3, i64* %ptr, align 8 233 ret void 234} 235 236; The load, icmp and assume should not count towards the limit, they are 237; ephemeral. 238define void @test_non_speculatable(i1 %c, i64* align 1 %ptr, i8* %ptr2) local_unnamed_addr #0 { 239; CHECK-LABEL: @test_non_speculatable( 240; CHECK-NEXT: br i1 [[C:%.*]], label [[TRUE2_CRITEDGE:%.*]], label [[FALSE1:%.*]] 241; CHECK: false1: 242; CHECK-NEXT: store volatile i64 1, i64* [[PTR:%.*]], align 4 243; CHECK-NEXT: [[V:%.*]] = load i8, i8* [[PTR2:%.*]], align 1 244; CHECK-NEXT: [[C2:%.*]] = icmp eq i8 [[V]], 42 245; CHECK-NEXT: call void @llvm.assume(i1 [[C2]]) 246; CHECK-NEXT: store volatile i64 0, i64* [[PTR]], align 8 247; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 248; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 249; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 250; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 251; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 252; CHECK-NEXT: store volatile i64 3, i64* [[PTR]], align 8 253; CHECK-NEXT: br label [[COMMON_RET:%.*]] 254; CHECK: common.ret: 255; CHECK-NEXT: ret void 256; CHECK: true2.critedge: 257; CHECK-NEXT: [[V_C:%.*]] = load i8, i8* [[PTR2]], align 1 258; CHECK-NEXT: [[C2_C:%.*]] = icmp eq i8 [[V_C]], 42 259; CHECK-NEXT: call void @llvm.assume(i1 [[C2_C]]) 260; CHECK-NEXT: store volatile i64 0, i64* [[PTR]], align 8 261; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 262; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 263; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 264; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 265; CHECK-NEXT: store volatile i64 -1, i64* [[PTR]], align 8 266; CHECK-NEXT: store volatile i64 2, i64* [[PTR]], align 8 267; CHECK-NEXT: br label [[COMMON_RET]] 268; 269 br i1 %c, label %true1, label %false1 270 271true1: ; preds = %false1, %0 272 %v = load i8, i8* %ptr2 273 %c2 = icmp eq i8 %v, 42 274 call void @llvm.assume(i1 %c2) 275 store volatile i64 0, i64* %ptr, align 8 276 store volatile i64 -1, i64* %ptr, align 8 277 store volatile i64 -1, i64* %ptr, align 8 278 store volatile i64 -1, i64* %ptr, align 8 279 store volatile i64 -1, i64* %ptr, align 8 280 store volatile i64 -1, i64* %ptr, align 8 281 br i1 %c, label %true2, label %false2 282 283false1: ; preds = %0 284 store volatile i64 1, i64* %ptr, align 4 285 br label %true1 286 287true2: ; preds = %true1 288 store volatile i64 2, i64* %ptr, align 8 289 ret void 290 291false2: ; preds = %true1 292 store volatile i64 3, i64* %ptr, align 8 293 ret void 294} 295