Merge branch 'fix/111854' of https://github.com/a-tarasyuk/llvm-project into fix/111854

Author: a-tarasyuk

.gitattributes

@@ -1,3 +1,10 @@
+# Checkout as native, commit as LF except in specific circumstances
+* text=auto
+*.bat text eol=crlf
+*.rc text eol=crlf
+*.sln text eol=crlf
+*.natvis text eol=crlf
+
 libcxx/src/**/*.cpp     merge=libcxx-reformat
 libcxx/include/**/*.h   merge=libcxx-reformat
 

bolt/include/bolt/Core/DIEBuilder.h

@@ -314,7 +314,7 @@ class DIEBuilder {
 
     BC.errs()
         << "BOLT-ERROR: unable to find TypeUnit for Type Unit at offset 0x"
-        << DU.getOffset() << "\n";
+        << Twine::utohexstr(DU.getOffset()) << "\n";
     return nullptr;
   }
 

bolt/lib/Core/BinaryContext.cpp

@@ -1294,8 +1294,8 @@ bool BinaryContext::handleAArch64Veneer(uint64_t Address, bool MatchOnly) {
     Veneer->getOrCreateLocalLabel(Address);
     Veneer->setMaxSize(TotalSize);
     Veneer->updateState(BinaryFunction::State::Disassembled);
-    LLVM_DEBUG(dbgs() << "BOLT-DEBUG: handling veneer function at 0x" << Address
-                      << "\n");
+    LLVM_DEBUG(dbgs() << "BOLT-DEBUG: handling veneer function at 0x"
+                      << Twine::utohexstr(Address) << "\n");
     return true;
   };
 

bolt/lib/Core/BinaryFunction.cpp

@@ -3684,9 +3684,8 @@ BinaryFunction::BasicBlockListType BinaryFunction::dfs() const {
     BinaryBasicBlock *BB = Stack.top();
     Stack.pop();
 
-    if (Visited.find(BB) != Visited.end())
+    if (!Visited.insert(BB).second)
       continue;
-    Visited.insert(BB);
     DFS.push_back(BB);
 
     for (BinaryBasicBlock *SuccBB : BB->landing_pads()) {
@@ -3879,11 +3878,8 @@ void BinaryFunction::disambiguateJumpTables(
       JumpTable *JT = getJumpTable(Inst);
       if (!JT)
         continue;
-      auto Iter = JumpTables.find(JT);
-      if (Iter == JumpTables.end()) {
-        JumpTables.insert(JT);
+      if (JumpTables.insert(JT).second)
         continue;
-      }
       // This instruction is an indirect jump using a jump table, but it is
       // using the same jump table of another jump. Try all our tricks to
       // extract the jump table symbol and make it point to a new, duplicated JT

bolt/lib/Core/DIEBuilder.cpp

@@ -57,11 +57,9 @@ getDWOName(llvm::DWARFUnit &CU,
          "DW_AT_dwo_name/DW_AT_GNU_dwo_name does not exist.");
   if (DwarfOutputPath) {
     DWOName = std::string(sys::path::filename(DWOName));
-    auto Iter = NameToIndexMap.find(DWOName);
-    if (Iter == NameToIndexMap.end())
-      Iter = NameToIndexMap.insert({DWOName, 0}).first;
-    DWOName.append(std::to_string(Iter->second));
-    ++Iter->second;
+    uint32_t &Index = NameToIndexMap[DWOName];
+    DWOName.append(std::to_string(Index));
+    ++Index;
   }
   DWOName.append(".dwo");
   return DWOName;

bolt/lib/Passes/Instrumentation.cpp

@@ -109,9 +109,8 @@ static bool hasAArch64ExclusiveMemop(
     BinaryBasicBlock *BB = BBQueue.front().first;
     bool IsLoad = BBQueue.front().second;
     BBQueue.pop();
-    if (Visited.find(BB) != Visited.end())
+    if (!Visited.insert(BB).second)
       continue;
-    Visited.insert(BB);
 
     for (const MCInst &Inst : *BB) {
       // Two loads one after another - skip whole function
@@ -126,8 +125,7 @@ static bool hasAArch64ExclusiveMemop(
       if (BC.MIB->isAArch64ExclusiveLoad(Inst))
         IsLoad = true;
 
-      if (IsLoad && BBToSkip.find(BB) == BBToSkip.end()) {
-        BBToSkip.insert(BB);
+      if (IsLoad && BBToSkip.insert(BB).second) {
         if (opts::Verbosity >= 2) {
           outs() << "BOLT-INSTRUMENTER: skip BB " << BB->getName()
                  << " due to exclusive instruction in function "

bolt/lib/Passes/LongJmp.cpp

@@ -324,9 +324,8 @@ uint64_t LongJmpPass::tentativeLayoutRelocColdPart(
 uint64_t LongJmpPass::tentativeLayoutRelocMode(
     const BinaryContext &BC, std::vector<BinaryFunction *> &SortedFunctions,
     uint64_t DotAddress) {
-
   // Compute hot cold frontier
-  uint32_t LastHotIndex = -1u;
+  int64_t LastHotIndex = -1u;
   uint32_t CurrentIndex = 0;
   if (opts::HotFunctionsAtEnd) {
     for (BinaryFunction *BF : SortedFunctions) {
@@ -351,19 +350,20 @@ uint64_t LongJmpPass::tentativeLayoutRelocMode(
   // Hot
   CurrentIndex = 0;
   bool ColdLayoutDone = false;
+  auto runColdLayout = [&]() {
+    DotAddress = tentativeLayoutRelocColdPart(BC, SortedFunctions, DotAddress);
+    ColdLayoutDone = true;
+    if (opts::HotFunctionsAtEnd)
+      DotAddress = alignTo(DotAddress, opts::AlignText);
+  };
   for (BinaryFunction *Func : SortedFunctions) {
     if (!BC.shouldEmit(*Func)) {
       HotAddresses[Func] = Func->getAddress();
       continue;
     }
 
-    if (!ColdLayoutDone && CurrentIndex >= LastHotIndex) {
-      DotAddress =
-          tentativeLayoutRelocColdPart(BC, SortedFunctions, DotAddress);
-      ColdLayoutDone = true;
-      if (opts::HotFunctionsAtEnd)
-        DotAddress = alignTo(DotAddress, opts::AlignText);
-    }
+    if (!ColdLayoutDone && CurrentIndex >= LastHotIndex)
+      runColdLayout();
 
     DotAddress = alignTo(DotAddress, Func->getMinAlignment());
     uint64_t Pad =
@@ -382,6 +382,11 @@ uint64_t LongJmpPass::tentativeLayoutRelocMode(
     DotAddress += Func->estimateConstantIslandSize();
     ++CurrentIndex;
   }
+
+  // Ensure that tentative code layout always runs for cold blocks.
+  if (!ColdLayoutDone)
+    runColdLayout();
+
   // BBs
   for (BinaryFunction *Func : SortedFunctions)
     tentativeBBLayout(*Func);

bolt/lib/Rewrite/DWARFRewriter.cpp

@@ -1362,7 +1362,7 @@ void DWARFRewriter::updateDWARFObjectAddressRanges(
                Die.getTag() == dwarf::DW_TAG_compile_unit)) {
     if (opts::Verbosity >= 1)
       errs() << "BOLT-WARNING: cannot update ranges for DIE in Unit offset 0x"
-             << Unit.getOffset() << '\n';
+             << Twine::utohexstr(Unit.getOffset()) << '\n';
   }
 }
 

bolt/test/AArch64/split-funcs-lite.s

@@ -0,0 +1,27 @@
+# This test checks that tentative code layout for cold blocks always runs.
+# It commonly happens when using lite mode with split functions.
+
+# REQUIRES: system-linux, asserts
+
+# RUN: %clang %cflags -o %t %s
+# RUN: %clang %s %cflags -Wl,-q -o %t
+# RUN: link_fdata --no-lbr %s %t %t.fdata
+# RUN: llvm-bolt %t -o %t.bolt --data %t.fdata -split-functions \
+# RUN:   -debug 2>&1 | FileCheck %s
+
+  .text
+  .globl  foo
+  .type foo, %function
+foo:
+.entry_bb:
+# FDATA: 1 foo #.entry_bb# 10
+    cmp x0, #0
+    b.eq .Lcold_bb1
+    ret
+.Lcold_bb1:
+    ret
+
+## Force relocation mode.
+.reloc 0, R_AARCH64_NONE
+
+# CHECK: foo{{.*}} cold tentative: {{.*}}

bolt/tools/driver/llvm-bolt.cpp

@@ -202,9 +202,9 @@ int main(int argc, char **argv) {
 
   ToolName = argv[0];
 
-  if (llvm::sys::path::filename(ToolName) == "perf2bolt")
+  if (llvm::sys::path::filename(ToolName).starts_with("perf2bolt"))
     perf2boltMode(argc, argv);
-  else if (llvm::sys::path::filename(ToolName) == "llvm-boltdiff")
+  else if (llvm::sys::path::filename(ToolName).starts_with("llvm-boltdiff"))
     boltDiffMode(argc, argv);
   else
     boltMode(argc, argv);