188 lines · plain
1; RUN: opt < %s -passes=sample-profile -sample-profile-file=%S/Inputs/profile-correlation-irreducible-loops.prof -sample-profile-use-profi=0 | opt -passes='print<block-freq>' -disable-output -use-iterative-bfi-inference 2>&1 | FileCheck %s2; RUN: opt < %s -passes=sample-profile -sample-profile-file=%S/Inputs/profile-correlation-irreducible-loops.prof -sample-profile-use-profi=0 -S | FileCheck %s --check-prefix=CHECK23; RUN: opt < %s -passes='print<block-freq>' -use-iterative-bfi-inference -disable-output 2>&1 | FileCheck %s --check-prefix=CHECK34 5; The C++ code for this test case is from c-parse.c in 403.gcc (SPEC2006)6; The problem with BFI for the test is solved by applying iterative inference.7; The corresponding CFG graph is shown below, with intended counts for every8; basic block. The hot loop, b3->b4->b2, is not getting proper (large) counts9; unless the -use-iterative-bfi-inference option is specified.10;11; +-------------------------------------------+12; | |13; | +----------+ |14; | | b1 [1] | |15; | +----------+ |16; | | |17; | | |18; | v |19; | +----------+ |20; | +------------> | b2 [625] | -+ |21; | | +----------+ | |22; | | | | |23; | | | | |24; | | v | |25; | +----------+ +----------+ | |26; | | b4 [624] | <-- | b3 [625] | <+---------+27; | +----------+ +----------+ |28; | | |29; +----+ | |30; | v v31; +----------+ +--------------------+32; | b8 [1] | <-- | b7 [2] |33; +----------+ +--------------------+34; | ^35; | |36; v |37; +----------+ +----------+ |38; | b9 [1] | <-- | b5 [2] | |39; +----------+ +----------+ |40; | |41; | |42; v |43; +----------+ |44; | b6 [1] | -+45; +----------+46 47@yydebug = dso_local global i32 0, align 448 49; Function Attrs: noinline nounwind uwtable50define dso_local i32 @yyparse_1() #0 {51b1:52 call void @llvm.pseudoprobe(i64 -7702751003264189226, i64 1, i32 0, i64 -1)53 %0 = load i32, ptr @yydebug, align 454 %cmp = icmp ne i32 %0, 055 br label %b256; CHECK: - b1: float = {{.*}}, int = {{.*}}, count = 157 58b2:59 call void @llvm.pseudoprobe(i64 -7702751003264189226, i64 2, i32 0, i64 -1)60 br i1 %cmp, label %b7, label %b361; CHECK: - b2: float = {{.*}}, int = {{.*}}, count = 58662 63b3:64 call void @llvm.pseudoprobe(i64 -7702751003264189226, i64 3, i32 0, i64 -1)65 br i1 %cmp, label %b7, label %b466; CHECK: - b3: float = {{.*}}, int = {{.*}}, count = 58667; CHECK2: br i1 %cmp, label %b7, label %b4,68; CHECK2-SAME: !prof ![[END172_PROF:[0-9]+]]69 70b4:71 call void @llvm.pseudoprobe(i64 -7702751003264189226, i64 4, i32 0, i64 -1)72 br label %b273; CHECK: - b4: float = {{.*}}, int = {{.*}}, count = 58574 75b5:76 call void @llvm.pseudoprobe(i64 -7702751003264189226, i64 5, i32 0, i64 -1)77 br i1 %cmp, label %b9, label %b678; CHECK: - b5: float = {{.*}}, int = {{.*}}, count = 279 80b6:81 call void @llvm.pseudoprobe(i64 -7702751003264189226, i64 6, i32 0, i64 -1)82 br label %b783; CHECK: - b6: float = {{.*}}, int = {{.*}}, count = 184 85b7:86 call void @llvm.pseudoprobe(i64 -7702751003264189226, i64 7, i32 0, i64 -1)87 br i1 %cmp, label %b5, label %b888; CHECK: - b7: float = {{.*}}, int = {{.*}}, count = 289; CHECK2: br i1 %cmp, label %b5, label %b8,90; CHECK2-SAME: !prof ![[FALSE4858_PROF:[0-9]+]]91 92b8:93 call void @llvm.pseudoprobe(i64 -7702751003264189226, i64 8, i32 0, i64 -1)94 br label %b395; CHECK: - b8: float = {{.*}}, int = {{.*}}, count = 196 97b9:98 call void @llvm.pseudoprobe(i64 -7702751003264189226, i64 9, i32 0, i64 -1)99 %1 = load i32, ptr @yydebug, align 4100 ret i32 %1101; CHECK: - b9: float = {{.*}}, int = {{.*}}, count = 1102 103}104 105; Another difficult (for BFI) instance with irreducible loops,106; containing 'indirectbr'. The corresponding CFG graph is shown below, with107; intended counts for every basic block.108;109; +-----------+110; | b1 [1] |111; +-----------+112; |113; |114; v115; +------------------------+116; +- | b2 [86] | <+117; | +------------------------+ |118; | | | |119; | | | |120; | v | |121; | +-----------+ | |122; | | b3 [8212] | <+-------+ |123; | +-----------+ | | |124; | | | | |125; | | | | |126; | v v | |127; | +------------------------+ |128; | | indirectgoto [17747] | -+129; | +------------------------+130; | | ^ |131; | | +--+132; | v133; | +-----------+134; +> | b4 [1] |135; +-----------+136 137; Function Attrs: nounwind uwtable138define dso_local i32 @foo1() #0 !prof !132 {139b1:140 call void @llvm.pseudoprobe(i64 7682762345278052905, i64 1, i32 0, i64 -1)141 %0 = load i32, ptr @yydebug, align 4142 %cmp = icmp ne i32 %0, 0143 br label %b2144; CHECK3: - b1: float = {{.*}}, int = {{.*}}, count = 1145 146b2:147 call void @llvm.pseudoprobe(i64 7682762345278052905, i64 2, i32 0, i64 -1)148 %1 = load i32, ptr @yydebug, align 4149 switch i32 %1, label %b4 [150 i32 1, label %indirectgoto151 i32 2, label %b3152 ], !prof !133153; CHECK3: - b2: float = {{.*}}, int = {{.*}}, count = 86154 155b3:156 call void @llvm.pseudoprobe(i64 7682762345278052905, i64 3, i32 0, i64 -1)157 br label %indirectgoto158; CHECK3: - b3: float = {{.*}}, int = {{.*}}, count = 8212159 160b4:161 call void @llvm.pseudoprobe(i64 7682762345278052905, i64 4, i32 0, i64 -1)162 %2 = load i32, ptr @yydebug, align 4163 ret i32 %2164; CHECK3: - b4: float = {{.*}}, int = {{.*}}, count = 1165 166indirectgoto:167 %indirect.goto.dest = alloca i8, align 4168 call void @llvm.pseudoprobe(i64 7682762345278052905, i64 5, i32 0, i64 -1)169 indirectbr ptr %indirect.goto.dest, [label %b2, label %indirectgoto, label %b4, label %b3], !prof !134170; CHECK3: - indirectgoto: float = {{.*}}, int = {{.*}}, count = 17747171 172}173 174declare void @llvm.pseudoprobe(i64, i64, i32, i64) #1175 176attributes #0 = { noinline nounwind uwtable "use-sample-profile"}177attributes #1 = { nounwind }178 179!llvm.pseudo_probe_desc = !{!1079, !4496}180!1079 = !{i64 -7702751003264189226, i64 158496288380146391, !"yyparse_1", null}181!4496 = !{i64 7682762345278052905, i64 404850113186107133, !"foo1", null}182!132 = !{!"function_entry_count", i64 1}183!133 = !{!"branch_weights", i32 0, i32 86, i32 0}184!134 = !{!"branch_weights", i32 85, i32 9449, i32 1, i32 8212}185 186; CHECK2: ![[END172_PROF]] = !{!"branch_weights", i32 1, i32 1003}187; CHECK2: ![[FALSE4858_PROF]] = !{!"branch_weights", i32 2, i32 1}188