153 lines · plain
1;; Test to make sure that the memprof ThinLTO backend finds the correct summary2;; for an imported promoted local, so that we can perform the correct cloning.3;; In particular, we should be able to use the thinlto_src_file metadata to4;; recreate its original GUID. In particular, this test contains promoted5;; internal functions with the same original name as those that were imported,6;; and we want to ensure we don't use those by mistake.7 8;; The original code looks something like:9;;10;; src1.cc:11;; extern void external1();12;; extern void external2();13;; static void internal1() {14;; external2();15;; }16;; static void internal2() {17;; external2();18;; }19;; int main() {20;; internal1();21;; internal2();22;; external1();23;; return 0;24;; }25;;26;; src2.cc:27;; extern void external2();28;; static void internal1() {29;; external2();30;; }31;; static void internal2() {32;; external2();33;; }34;; void external1() {35;; internal1();36;; internal2();37;; }38;;39;; The assembly for src1 shown below was dumped after function importing, with40;; some hand modification to ensure we import the definitions of src2.cc's41;; external1 and internal1 functions, and the declaration only for its42;; internal2 function. I also hand modified it to add !callsite metadata43;; to a few calls, and the distributed ThinLTO summary in src1.o.thinlto.ll to44;; contain callsite metadata records with cloning results.45 46; RUN: rm -rf %t && split-file %s %t && cd %t47; RUN: llvm-as src1.ll -o src1.o48; RUN: llvm-as src1.o.thinlto.ll -o src1.o.thinlto.bc49 50; RUN: opt -passes=memprof-context-disambiguation src1.o -S -memprof-import-summary=src1.o.thinlto.bc | FileCheck %s51 52;; Per the cloning results in the summary, none of the original functions should53;; call any memprof clones.54; CHECK-NOT: memprof55;; We should have one clone of src1.cc's internal1 that calls a clone of56;; external2.57; CHECK-LABEL: define void @_ZL9internal1v.llvm.5985484347676238233.memprof.1()58; CHECK: tail call void @_Z9external2v.memprof.1()59; CHECK-LABEL: declare void @_Z9external2v.memprof.1()60;; We should have one clone of external1 that calls a clone of internal2 from61;; a synthesized callsite record (for a tail call with a missing frame).62; CHECK-LABEL: define available_externally {{.*}} void @_Z9external1v.memprof.1()63; CHECK: tail call void @_ZL9internal1v.llvm.3267420853450984672()64; CHECK: tail call void @_ZL9internal2v.llvm.3267420853450984672.memprof.1()65; CHECK-LABEL: declare void @_ZL9internal2v.llvm.3267420853450984672.memprof.1()66;; We should have one clone of src2.cc's internal1 function, calling a single67;; clone of external2, and a second clone that was detected to be a duplicate68;; of the first that becomes a declaration (since this is available_externally -69;; in the module with the prevailing copy it would be an alias to clone 1).70; CHECK-LABEL: define available_externally void @_ZL9internal1v.llvm.3267420853450984672.memprof.1()71; CHECK: tail call void @_Z9external2v.memprof.1()72; CHECK: tail call void @_Z9external2v.memprof.1()73; CHECK: declare void @_ZL9internal1v.llvm.3267420853450984672.memprof.2()74; CHECK-NOT: memprof75 76;--- src1.ll77; ModuleID = 'src1.o'78source_filename = "src1.cc"79target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"80target triple = "x86_64-unknown-linux-gnu"81 82define dso_local noundef i32 @main() {83entry:84 tail call void @_ZL9internal1v.llvm.5985484347676238233()85 tail call void @_ZL9internal2v.llvm.5985484347676238233()86 tail call void @_Z9external1v()87 ret i32 088}89 90define void @_ZL9internal1v.llvm.5985484347676238233() {91entry:92 tail call void @_Z9external2v(), !callsite !893 ret void94}95 96define void @_ZL9internal2v.llvm.5985484347676238233() {97entry:98 tail call void @_Z9external2v()99 ret void100}101 102declare void @_Z9external2v()103 104define available_externally dso_local void @_Z9external1v() !thinlto_src_module !6 !thinlto_src_file !7 {105entry:106 tail call void @_ZL9internal1v.llvm.3267420853450984672()107 tail call void @_ZL9internal2v.llvm.3267420853450984672()108 ret void109}110 111define available_externally void @_ZL9internal1v.llvm.3267420853450984672() !thinlto_src_module !6 !thinlto_src_file !7 {112entry:113 ;; This one has more callsite records than the other version of internal1,114 ;; which would cause the code to iterate past the end of the callsite115 ;; records if we incorrectly got the other internal1's summary.116 tail call void @_Z9external2v(), !callsite !9117 tail call void @_Z9external2v(), !callsite !10118 ret void119}120 121declare void @_ZL9internal2v.llvm.3267420853450984672()122 123!6 = !{!"src2.o"}124!7 = !{!"src2.cc"}125!8 = !{i64 12345}126!9 = !{i64 23456}127!10 = !{i64 34567}128 129;--- src1.o.thinlto.ll130; ModuleID = 'src1.o.thinlto.bc'131source_filename = "src1.o.thinlto.bc"132 133^0 = module: (path: "src1.o", hash: (1393604173, 1072112025, 2857473630, 2016801496, 3238735916))134^1 = module: (path: "src2.o", hash: (760755700, 1705397472, 4198605753, 677969311, 2408738824))135;; src2.o:internal1. It specifies that we should have 3 clones total (including136;; original).137^3 = gv: (guid: 1143217136900127394, summaries: (function: (module: ^1, flags: (linkage: available_externally, visibility: default, notEligibleToImport: 0, live: 1, dsoLocal: 1, canAutoHide: 0, importType: definition), insts: 2, funcFlags: (readNone: 0, readOnly: 0, noRecurse: 0, returnDoesNotAlias: 0, noInline: 1, alwaysInline: 0, noUnwind: 0, mayThrow: 0, hasUnknownCall: 0, mustBeUnreachable: 0), calls: ((callee: ^6, tail: 1), (callee: ^6, tail: 1)), callsites: ((callee: ^6, clones: (0, 1, 1), stackIds: (23456)), (callee: ^6, clones: (0, 1, 1), stackIds: (34567))))))138;; src2.o:internal2. It was manually modified to have importType = declaration.139^4 = gv: (guid: 3599593882704738259, summaries: (function: (module: ^1, flags: (linkage: available_externally, visibility: default, notEligibleToImport: 0, live: 1, dsoLocal: 1, canAutoHide: 0, importType: declaration), insts: 2, funcFlags: (readNone: 0, readOnly: 0, noRecurse: 0, returnDoesNotAlias: 0, noInline: 1, alwaysInline: 0, noUnwind: 0, mayThrow: 0, hasUnknownCall: 0, mustBeUnreachable: 0), calls: ((callee: ^6, tail: 1)))))140;; src1.o:internal1.141^5 = gv: (guid: 6084810090198994915, summaries: (function: (module: ^0, flags: (linkage: internal, visibility: default, notEligibleToImport: 0, live: 1, dsoLocal: 1, canAutoHide: 0, importType: definition), insts: 2, funcFlags: (readNone: 0, readOnly: 0, noRecurse: 0, returnDoesNotAlias: 0, noInline: 1, alwaysInline: 0, noUnwind: 0, mayThrow: 0, hasUnknownCall: 0, mustBeUnreachable: 0), calls: ((callee: ^6, tail: 1)), callsites: ((callee: ^6, clones: (0, 1), stackIds: (12345))))))142^6 = gv: (guid: 8596367375252297795)143;; src1.o:internal2.144^7 = gv: (guid: 11092151021205906565, summaries: (function: (module: ^0, flags: (linkage: internal, visibility: default, notEligibleToImport: 0, live: 1, dsoLocal: 1, canAutoHide: 0, importType: definition), insts: 2, funcFlags: (readNone: 0, readOnly: 0, noRecurse: 0, returnDoesNotAlias: 0, noInline: 1, alwaysInline: 0, noUnwind: 0, mayThrow: 0, hasUnknownCall: 0, mustBeUnreachable: 0), calls: ((callee: ^6, tail: 1)))))145;; src2.o:external1. It contains a synthesized callsite record for the tail call146;; to internal2 (the empty stackId list indicates it is synthesized for a147;; discovered missing tail call frame.148^8 = gv: (guid: 12313225385227428720, summaries: (function: (module: ^1, flags: (linkage: external, visibility: default, notEligibleToImport: 0, live: 1, dsoLocal: 1, canAutoHide: 0, importType: definition), insts: 3, calls: ((callee: ^3, tail: 1), (callee: ^4, tail: 1)), callsites: ((callee: ^4, clones: (0, 1), stackIds: ())))))149;; src1.o:main.150^9 = gv: (guid: 15822663052811949562, summaries: (function: (module: ^0, flags: (linkage: external, visibility: default, notEligibleToImport: 0, live: 1, dsoLocal: 1, canAutoHide: 0, importType: definition), insts: 4, funcFlags: (readNone: 0, readOnly: 0, noRecurse: 1, returnDoesNotAlias: 0, noInline: 0, alwaysInline: 0, noUnwind: 0, mayThrow: 0, hasUnknownCall: 0, mustBeUnreachable: 0), calls: ((callee: ^5, tail: 1), (callee: ^7, tail: 1), (callee: ^8, tail: 1)))))151^10 = flags: 97152^11 = blockcount: 0153