[MemProf] Fix combined index handling for locals

[tejohnson]

Summary

Since the linker does not resolve local symbols, we cannot look up
whether they are prevailing. The prior check was blocking all locals
from getting memprof summaries in the combined index.

Modified the existing test case to contain a local. This necessitated
some other fixes as the order of summary entries changed.

[snehasish]

lgtm

(Unrelated to this patch)
I guess we should update https://llvm.org/docs/LangRef.html#thinlto-summary at some point with the memProf extensions.

[tejohnson]

In D140786#4024114, @snehasish wrote:

lgtm

(Unrelated to this patch)
I guess we should update https://llvm.org/docs/LangRef.html#thinlto-summary at some point with the memProf extensions.

Good point, will send a patch for that.

llvm/lib/Bitcode/Reader/BitcodeReader.cpp

case bitc::FS_PERMODULE_PROFILE: {
       bool HasProfile = (BitCode == bitc::FS_PERMODULE_PROFILE);
       bool HasRelBF = (BitCode == bitc::FS_PERMODULE_RELBF);
       std::vector<FunctionSummary::EdgeTy> Calls = makeCallList(
           ArrayRef<uint64_t>(Record).slice(CallGraphEdgeStartIndex),
           IsOldProfileFormat, HasProfile, HasRelBF);
       setSpecialRefs(Refs, NumRORefs, NumWORefs);
       auto VIAndOriginalGUID = getValueInfoFromValueId(ValueID);
       // In order to save memory, only record the memprof summaries if this is
-      // the prevailing copy of a symbol.
-      if (IsPrevailing && !IsPrevailing(std::get<2>(VIAndOriginalGUID))) {
+      // the prevailing copy of a symbol. The linker doesn't resolve local
+      // linkage values so don't check whether those are prevailing.
+      auto LT = (GlobalValue::LinkageTypes)Flags.Linkage;
+      if (IsPrevailing &&
+          !GlobalValue::isLocalLinkage(LT) &&
+          !IsPrevailing(std::get<2>(VIAndOriginalGUID))) {
         PendingCallsites.clear();
         PendingAllocs.clear();
       }
       auto FS = std::make_unique<FunctionSummary>(
           Flags, InstCount, getDecodedFFlags(RawFunFlags), /*EntryCount=*/0,
           std::move(Refs), std::move(Calls), std::move(PendingTypeTests),
           std::move(PendingTypeTestAssumeVCalls),
           std::move(PendingTypeCheckedLoadVCalls),

llvm/test/ThinLTO/X86/memprof-summary.ll

 ;; Check memprof summaries (per module, combined index, and distributed indexes)
 
 ; RUN: split-file %s %t
 ; RUN: opt -module-summary %t/a.ll -o %ta.bc
 ; RUN: opt -module-summary %t/b.ll -o %tb.bc
 
 ; RUN: llvm-dis -o - %ta.bc | FileCheck %s --check-prefix=PRELINKDISA
 ; PRELINKDISA: gv: (name: "main", {{.*}} callsites: ((callee: ^2, clones: (0), stackIds: (8632435727821051414)), (callee: ^2, clones: (0), stackIds: (15025054523792398438)))))) ; guid = 15822663052811949562
 
 ; RUN: llvm-dis -o - %tb.bc | FileCheck %s --check-prefix=PRELINKDISB
-; PRELINKDISB: gv: (name: "_Z3foov", {{.*}} callsites: ((callee: ^2, clones: (0), stackIds: (2732490490862098848)))))) ; guid = 9191153033785521275
-; PRELINKDISB: gv: (name: "_Z3bazv", {{.*}} callsites: ((callee: ^3, clones: (0), stackIds: (12481870273128938184)))))) ; guid = 15176620447596392000
-; PRELINKDISB: gv: (name: "_Z3barv", {{.*}} allocs: ((versions: (none), memProf: ((type: notcold, stackIds: (12481870273128938184, 2732490490862098848, 8632435727821051414)), (type: cold, stackIds: (12481870273128938184, 2732490490862098848, 15025054523792398438)))))))) ; guid = 17377440600225628772
+; PRELINKDISB: ^[[PLBAR:[0-9]+]] = gv: (name: "_Z3barv", {{.*}} allocs: ((versions: (none), memProf: ((type: notcold, stackIds: (12481870273128938184, 2732490490862098848, 8632435727821051414)), (type: cold, stackIds: (12481870273128938184, 2732490490862098848, 15025054523792398438)))))))) ; guid = 4555904644815367798
+; PRELINKDISB: ^[[PLFOO:[0-9]+]] = gv: (name: "_Z3foov", {{.*}} callsites: ((callee: ^[[PLBAZ:[0-9]+]], clones: (0), stackIds: (2732490490862098848)))))) ; guid = 9191153033785521275
+; PRELINKDISB: ^[[PLBAZ]] = gv: (name: "_Z3bazv", {{.*}} callsites: ((callee: ^[[PLBAR]], clones: (0), stackIds: (12481870273128938184)))))) ; guid = 15176620447596392000
 
 ; RUN: llvm-bcanalyzer -dump %ta.bc | FileCheck %s --check-prefix=PRELINKBCANA
 ; PRELINKBCANA: <STACK_IDS abbrevid=4 op0=8632435727821051414 op1=-3421689549917153178/>
 
 ; RUN: llvm-bcanalyzer -dump %tb.bc | FileCheck %s --check-prefix=PRELINKBCANB
 ; PRELINKBCANB: <STACK_IDS abbrevid=4 op0=-5964873800580613432 op1=2732490490862098848 op2=8632435727821051414 op3=-3421689549917153178/>
 
 ; RUN: llvm-lto2 run %ta.bc %tb.bc -o %t -save-temps \
 ; RUN:     -thinlto-distributed-indexes \
 ; RUN:     -r=%ta.bc,main,plx \
 ; RUN:     -r=%ta.bc,_Z3foov, \
 ; RUN:     -r=%ta.bc,free, \
 ; RUN:     -r=%ta.bc,sleep, \
 ; RUN:     -r=%tb.bc,_Z3foov,pl \
 ; RUN:     -r=%tb.bc,_Znam, \
-; RUN:     -r=%tb.bc,_Z3barv,pl \
 ; RUN:     -r=%tb.bc,_Z3bazv,pl
 
 ; RUN: llvm-dis -o - %t.index.bc | FileCheck %s --check-prefix=COMBINEDDIS
-; COMBINEDDIS: gv: (guid: 9191153033785521275, {{.*}} callsites: ((callee: ^3, clones: (0), stackIds: (2732490490862098848))))))
-; COMBINEDDIS: gv: (guid: 15176620447596392000, {{.*}} callsites: ((callee: ^5, clones: (0), stackIds: (12481870273128938184))))))
-; COMBINEDDIS: gv: (guid: 15822663052811949562, {{.*}} callsites: ((callee: ^2, clones: (0), stackIds: (8632435727821051414)), (callee: ^2, clones: (0), stackIds: (15025054523792398438))))))
-; COMBINEDDIS: gv: (guid: 17377440600225628772, {{.*}} allocs: ((versions: (none), memProf: ((type: notcold, stackIds: (12481870273128938184, 2732490490862098848, 8632435727821051414)), (type: cold, stackIds: (12481870273128938184, 2732490490862098848, 15025054523792398438))))))))
+; COMBINEDDIS: ^[[COMBBAR:[0-9]+]] = gv: (guid: 4555904644815367798, {{.*}} allocs: ((versions: (none), memProf: ((type: notcold, stackIds: (12481870273128938184, 2732490490862098848, 8632435727821051414)), (type: cold, stackIds: (12481870273128938184, 2732490490862098848, 15025054523792398438))))))))
+; COMBINEDDIS: ^[[COMBFOO:[0-9]+]] = gv: (guid: 9191153033785521275, {{.*}} callsites: ((callee: ^[[COMBBAZ:[0-9]+]], clones: (0), stackIds: (2732490490862098848))))))
+; COMBINEDDIS: ^[[COMBBAZ]] = gv: (guid: 15176620447596392000, {{.*}} callsites: ((callee: ^[[COMBBAR]], clones: (0), stackIds: (12481870273128938184))))))
+; COMBINEDDIS: ^[[COMBMAIN:[0-9]+]] = gv: (guid: 15822663052811949562, {{.*}} callsites: ((callee: ^[[COMBFOO]], clones: (0), stackIds: (8632435727821051414)), (callee: ^[[COMBFOO]], clones: (0), stackIds: (15025054523792398438))))))
 
 ; RUN: llvm-bcanalyzer -dump %t.index.bc | FileCheck %s --check-prefix=COMBINEDBCAN
 ; COMBINEDBCAN: <STACK_IDS abbrevid=4 op0=8632435727821051414 op1=-3421689549917153178 op2=-5964873800580613432 op3=2732490490862098848/>
 
 ; RUN: llvm-dis -o - %ta.bc.thinlto.bc | FileCheck %s --check-prefix=DISTRIBUTEDDISA
 ; DISTRIBUTEDDISA: gv: (guid: 9191153033785521275, {{.*}} callsites: ((callee: null, clones: (0), stackIds: (2732490490862098848))))))
 ; DISTRIBUTEDDISA: gv: (guid: 15822663052811949562, {{.*}} callsites: ((callee: ^2, clones: (0), stackIds: (8632435727821051414)), (callee: ^2, clones: (0), stackIds: (15025054523792398438))))))
 
 ; RUN: llvm-dis -o - %tb.bc.thinlto.bc | FileCheck %s --check-prefix=DISTRIBUTEDDISB
-; DISTRIBUTEDDISB: gv: (guid: 9191153033785521275, {{.*}} callsites: ((callee: ^2, clones: (0), stackIds: (2732490490862098848))))))
-; DISTRIBUTEDDISB: gv: (guid: 15176620447596392000, {{.*}} callsites: ((callee: ^3, clones: (0), stackIds: (12481870273128938184))))))
-; DISTRIBUTEDDISB: gv: (guid: 17377440600225628772, {{.*}} allocs: ((versions: (none), memProf: ((type: notcold, stackIds: (12481870273128938184, 2732490490862098848, 8632435727821051414)), (type: cold, stackIds: (12481870273128938184, 2732490490862098848, 15025054523792398438))))))))
+; DISTRIBUTEDDISB: ^[[DISTRBAR:[0-9]+]] = gv: (guid: 4555904644815367798, {{.*}} allocs: ((versions: (none), memProf: ((type: notcold, stackIds: (12481870273128938184, 2732490490862098848, 8632435727821051414)), (type: cold, stackIds: (12481870273128938184, 2732490490862098848, 15025054523792398438))))))))
+; DISTRIBUTEDDISB: ^[[DISTRFOO:[0-9]+]] = gv: (guid: 9191153033785521275, {{.*}} callsites: ((callee: ^[[DISTRBAZ:[0-9]+]], clones: (0), stackIds: (2732490490862098848))))))
+; DISTRIBUTEDDISB: ^[[DISTRBAZ]] = gv: (guid: 15176620447596392000, {{.*}} callsites: ((callee: ^[[DISTRBAR]], clones: (0), stackIds: (12481870273128938184))))))
 
 ; RUN: llvm-bcanalyzer -dump %ta.bc.thinlto.bc | FileCheck %s --check-prefix=DISTRIBUTEDBCANA
 ; DISTRIBUTEDBCANA: <STACK_IDS abbrevid=4 op0=8632435727821051414 op1=-3421689549917153178 op2=2732490490862098848/>
 
 ; RUN: llvm-bcanalyzer -dump %tb.bc.thinlto.bc | FileCheck %s --check-prefix=DISTRIBUTEDBCANB
 ; DISTRIBUTEDBCANB: <STACK_IDS abbrevid=4 op0=8632435727821051414 op1=-3421689549917153178 op2=-5964873800580613432 op3=2732490490862098848/>
 
 ;--- a.ll

 
 ;--- b.ll
 ; ModuleID = 'b.cc'
 source_filename = "b.cc"
 target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
 ; Function Attrs: mustprogress noinline uwtable
-define dso_local noalias noundef nonnull ptr @_Z3barv() local_unnamed_addr #0 !dbg !39 {
+define internal noalias noundef nonnull ptr @_Z3barv() local_unnamed_addr #0 !dbg !39 {
 entry:
   %call = call noalias noundef nonnull dereferenceable(10) ptr @_Znam(i64 noundef 10) #2, !dbg !42, !memprof !43, !callsite !48
   ret ptr %call, !dbg !49
 }
 
 ; Function Attrs: nobuiltin allocsize(0)
 declare noundef nonnull ptr @_Znam(i64 noundef) local_unnamed_addr #1