[GCOV] Avoid emitting profile arcs for module and skeleton CU's

[vsk]

Summary

Do not emit profile arc files and note files for module and skeleton
CU's.

Our users report seeing unexpected *.gcda and *.gcno files in their
projects when using gcov-style profiling with modules or frameworks.
The unwanted files come from these modules. This is not very helpful
for end-users. Further, we've seen reports of instrumented programs
crashing while writing these files out (due to I/O failures).

rdar://problem/22838296

[aprantl]

lib/Transforms/Instrumentation/GCOVProfiling.cpp

498 ↗

(On Diff #44334)

// Skip module and module skeleton CUs.

would be more accurate. Regular split-dwarf skeleton CUs won't (and shouldn't) be affected because they are generated later and don't exist in the IR.

[vsk]

Updated comments. Would it be appropriate to add a isModuleOrModuleSkeleton predicate to DICompileUnit?

[vsk]

Ping.

[aprantl]

I'm not sure — this seems to be a fairly special use-case.

If modules did contain code (which they currently don't) we probably wouldn't want to skip them. But at the moment we cannot distinguish a skeleton CU from a module CU at the IR level.

Let's just change the comment to
// Skip module skeleton (and module) CUs.
and be done with it.

llvm/trunk/lib/Transforms/Instrumentation/GCOVProfiling.cpp

void GCOVProfiler::emitProfileNotes() {
   if (!CU_Nodes) return;
 
   for (unsigned i = 0, e = CU_Nodes->getNumOperands(); i != e; ++i) {
     // Each compile unit gets its own .gcno file. This means that whether we run
     // this pass over the original .o's as they're produced, or run it after
     // LTO, we'll generate the same .gcno files.
 
     auto *CU = cast<DICompileUnit>(CU_Nodes->getOperand(i));
+
+    // Skip module skeleton (and module) CUs.
+    if (CU->getDWOId())
+      continue;
+
     std::error_code EC;
     raw_fd_ostream out(mangleName(CU, "gcno"), EC, sys::fs::F_None);
     std::string EdgeDestinations;
 
     unsigned FunctionIdent = 0;
     for (auto *SP : CU->getSubprograms()) {
       Function *F = FnMap[SP];
       if (!F) continue;
Function *GCOVProfiler::insertCounterWriteout(
   Constant *EmitArcs = getEmitArcsFunc();
   Constant *SummaryInfo = getSummaryInfoFunc();
   Constant *EndFile = getEndFileFunc();
 
   NamedMDNode *CU_Nodes = M->getNamedMetadata("llvm.dbg.cu");
   if (CU_Nodes) {
     for (unsigned i = 0, e = CU_Nodes->getNumOperands(); i != e; ++i) {
       auto *CU = cast<DICompileUnit>(CU_Nodes->getOperand(i));
+
+      // Skip module skeleton (and module) CUs.
+      if (CU->getDWOId())
+        continue;
+
       std::string FilenameGcda = mangleName(CU, "gcda");
       uint32_t CfgChecksum = FileChecksums.empty() ? 0 : FileChecksums[i];
       Builder.CreateCall(StartFile,
                          {Builder.CreateGlobalStringPtr(FilenameGcda),
                           Builder.CreateGlobalStringPtr(ReversedVersion),
                           Builder.getInt32(CfgChecksum)});
       for (unsigned j = 0, e = CountersBySP.size(); j != e; ++j) {
         auto *SP = cast_or_null<DISubprogram>(CountersBySP[j].second);

llvm/trunk/test/Transforms/GCOVProfiling/modules.ll

+; RUN: opt -insert-gcov-profiling -o - < %s | llvm-dis | FileCheck -check-prefix=EMIT-ARCS %s
+
+; EMIT-ARCS-NOT: call void @llvm_gcda_start_file
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!3, !4}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "LLVM", isOptimized: false, runtimeVersion: 2, splitDebugFilename: "my.dwo", emissionKind: 1, enums: !2, retainedTypes: !2, subprograms: !2, globals: !2, imports: !2, dwoId: 43981)
+!1 = !DIFile(filename: "<stdin>", directory: "/")
+!2 = !{}
+!3 = !{i32 2, !"Dwarf Version", i32 4}
+!4 = !{i32 2, !"Debug Info Version", i32 3}