merge main into amd-staging (llvm#1644)

Author: ronlieb

.github/workflows/containers/github-action-ci/Dockerfile

@@ -32,7 +32,7 @@ RUN cmake -B ./build -G Ninja ./llvm \
   -DLLVM_ENABLE_RUNTIMES="compiler-rt" \
   -DCMAKE_INSTALL_PREFIX="$LLVM_SYSROOT" \
   -DLLVM_ENABLE_PROJECTS="bolt;clang;lld;clang-tools-extra" \
-  -DLLVM_DISTRIBUTION_COMPONENTS="lld;compiler-rt;clang-format;scan-build" \
+  -DLLVM_DISTRIBUTION_COMPONENTS="lld;compiler-rt;clang-format;scan-build;llvm-symbolizer" \
   -DCLANG_DEFAULT_LINKER="lld"
 
 RUN ninja -C ./build stage2-clang-bolt stage2-install-distribution && ninja -C ./build install-distribution

bolt/lib/Profile/DataAggregator.cpp

@@ -871,18 +871,27 @@ DataAggregator::getFallthroughsInTrace(BinaryFunction &BF,
 
   BinaryContext &BC = BF.getBinaryContext();
 
-  if (!BF.isSimple())
-    return std::nullopt;
-
-  assert(BF.hasCFG() && "can only record traces in CFG state");
-
   // Offsets of the trace within this function.
   const uint64_t From = FirstLBR.To - BF.getAddress();
   const uint64_t To = SecondLBR.From - BF.getAddress();
 
   if (From > To)
     return std::nullopt;
 
+  // Accept fall-throughs inside pseudo functions (PLT/thunks).
+  // This check has to be above BF.empty as pseudo functions would pass it:
+  // pseudo => ignored => CFG not built => empty.
+  // If we return nullopt, trace would be reported as mismatching disassembled
+  // function contents which it is not. To avoid this, return an empty
+  // fall-through list instead.
+  if (BF.isPseudo())
+    return Branches;
+
+  if (!BF.isSimple())
+    return std::nullopt;
+
+  assert(BF.hasCFG() && "can only record traces in CFG state");
+
   const BinaryBasicBlock *FromBB = BF.getBasicBlockContainingOffset(From);
   const BinaryBasicBlock *ToBB = BF.getBasicBlockContainingOffset(To);
 

bolt/test/X86/callcont-fallthru.s

@@ -9,6 +9,7 @@
 # RUN: link_fdata %s %t %t.pa3 PREAGG3
 # RUN: link_fdata %s %t %t.pat PREAGGT1
 # RUN: link_fdata %s %t %t.pat2 PREAGGT2
+# RUN: link_fdata %s %t %t.patplt PREAGGPLT
 
 ## Check normal case: fallthrough is not LP or secondary entry.
 # RUN: llvm-strip --strip-unneeded %t -o %t.strip
@@ -42,6 +43,12 @@
 # RUN: llvm-bolt %t.strip --pa -p %t.pat2 -o %t.out \
 # RUN:   --print-cfg --print-only=main | FileCheck %s --check-prefix=CHECK3
 
+## Check pre-aggregated traces don't report zero-sized PLT fall-through as
+## invalid trace
+# RUN: llvm-bolt %t.strip --pa -p %t.patplt -o %t.out | FileCheck %s \
+# RUN:   --check-prefix=CHECK-PLT
+# CHECK-PLT: traces mismatching disassembled function contents: 0
+
   .globl foo
   .type foo, %function
 foo:
@@ -65,7 +72,10 @@ main:
 	movl	$0x0, -0x4(%rbp)
 	movl	%edi, -0x8(%rbp)
 	movq	%rsi, -0x10(%rbp)
+Ltmp0_br:
 	callq	puts@PLT
+## Check PLT traces are accepted
+# PREAGGPLT: T #Ltmp0_br# #puts@plt# #puts@plt# 3
 ## Target is an external-origin call continuation
 # PREAGG1: B X:0 #Ltmp1# 2 0
 # PREAGGT1: T X:0 #Ltmp1# #Ltmp4_br# 2

bolt/test/link_fdata.py

@@ -8,6 +8,7 @@
 """
 
 import argparse
+import os
 import subprocess
 import sys
 import re
@@ -84,8 +85,16 @@
             exit("ERROR: unexpected input:\n%s" % line)
 
 # Read nm output: <symbol value> <symbol type> <symbol name>
+is_llvm_nm = os.path.basename(args.nmtool) == "llvm-nm"
 nm_output = subprocess.run(
-    [args.nmtool, "--defined-only", args.objfile], text=True, capture_output=True
+    [
+        args.nmtool,
+        "--defined-only",
+        "--special-syms" if is_llvm_nm else "--synthetic",
+        args.objfile,
+    ],
+    text=True,
+    capture_output=True,
 ).stdout
 # Populate symbol map
 symbols = {}

clang/include/clang/Frontend/CompilerInstance.h

@@ -871,6 +871,14 @@ class CompilerInstance : public ModuleLoader {
                                                  bool IsInclusionDirective);
 
 public:
+  /// Creates a new \c CompilerInstance for compiling a module.
+  ///
+  /// This takes care of creating appropriate \c FrontendInputFile for
+  /// public/private frameworks, inferred modules and such.
+  std::unique_ptr<CompilerInstance>
+  cloneForModuleCompile(SourceLocation ImportLoc, Module *Module,
+                        StringRef ModuleFileName);
+
   ModuleLoadResult loadModule(SourceLocation ImportLoc, ModuleIdPath Path,
                               Module::NameVisibilityKind Visibility,
                               bool IsInclusionDirective) override;

clang/lib/CodeGen/CGExpr.cpp

@@ -1970,7 +1970,7 @@ bool CodeGenFunction::EmitScalarRangeCheck(llvm::Value *Value, QualType Ty,
   SanitizerKind::SanitizerOrdinal Kind =
       NeedsEnumCheck ? SanitizerKind::SO_Enum : SanitizerKind::SO_Bool;
   EmitCheck(std::make_pair(Check, Kind), SanitizerHandler::LoadInvalidValue,
-            StaticArgs, EmitCheckValue(Value));
+            StaticArgs, Value);
   return true;
 }
 

clang/lib/CodeGen/CodeGenFunction.cpp

@@ -3197,9 +3197,7 @@ void CodeGenFunction::emitAlignmentAssumptionCheck(
     llvm::Constant *StaticData[] = {EmitCheckSourceLocation(Loc),
                                     EmitCheckSourceLocation(SecondaryLoc),
                                     EmitCheckTypeDescriptor(Ty)};
-    llvm::Value *DynamicData[] = {EmitCheckValue(Ptr),
-                                  EmitCheckValue(Alignment),
-                                  EmitCheckValue(OffsetValue)};
+    llvm::Value *DynamicData[] = {Ptr, Alignment, OffsetValue};
     EmitCheck({std::make_pair(TheCheck, SanitizerKind::SO_Alignment)},
               SanitizerHandler::AlignmentAssumption, StaticData, DynamicData);
   }

clang/lib/Frontend/CompilerInstance.cpp

@@ -1335,23 +1335,15 @@ static OptionalFileEntryRef getPublicModuleMap(FileEntryRef File,
   return FileMgr.getOptionalFileRef(PublicFilename);
 }
 
-/// Creates a \c CompilerInstance for compiling a module.
-///
-/// This takes care of creating appropriate \c FrontendInputFile for
-/// public/private frameworks, inferred modules and such.
-static std::unique_ptr<CompilerInstance>
-createCompilerInstanceForModuleCompile(CompilerInstance &ImportingInstance,
-                                       SourceLocation ImportLoc, Module *Module,
-                                       StringRef ModuleFileName) {
+std::unique_ptr<CompilerInstance> CompilerInstance::cloneForModuleCompile(
+    SourceLocation ImportLoc, Module *Module, StringRef ModuleFileName) {
   StringRef ModuleName = Module->getTopLevelModuleName();
 
-  InputKind IK(getLanguageFromOptions(ImportingInstance.getLangOpts()),
-               InputKind::ModuleMap);
+  InputKind IK(getLanguageFromOptions(getLangOpts()), InputKind::ModuleMap);
 
   // Get or create the module map that we'll use to build this module.
-  ModuleMap &ModMap =
-      ImportingInstance.getPreprocessor().getHeaderSearchInfo().getModuleMap();
-  SourceManager &SourceMgr = ImportingInstance.getSourceManager();
+  ModuleMap &ModMap = getPreprocessor().getHeaderSearchInfo().getModuleMap();
+  SourceManager &SourceMgr = getSourceManager();
 
   if (FileID ModuleMapFID = ModMap.getContainingModuleMapFileID(Module);
       ModuleMapFID.isValid()) {
@@ -1372,8 +1364,8 @@ createCompilerInstanceForModuleCompile(CompilerInstance &ImportingInstance,
     // Canonicalize compilation to start with the public module map. This is
     // vital for submodules declarations in the private module maps to be
     // correctly parsed when depending on a top level module in the public one.
-    if (OptionalFileEntryRef PublicMMFile = getPublicModuleMap(
-            *ModuleMapFile, ImportingInstance.getFileManager()))
+    if (OptionalFileEntryRef PublicMMFile =
+            getPublicModuleMap(*ModuleMapFile, getFileManager()))
       ModuleMapFile = PublicMMFile;
 
     StringRef ModuleMapFilePath = ModuleMapFile->getNameAsRequested();
@@ -1387,7 +1379,7 @@ createCompilerInstanceForModuleCompile(CompilerInstance &ImportingInstance,
 
     // Use the module map where this module resides.
     return createCompilerInstanceForModuleCompileImpl(
-        ImportingInstance, ImportLoc, ModuleName,
+        *this, ImportLoc, ModuleName,
         FrontendInputFile(ModuleMapFilePath, IK, IsSystem),
         ModMap.getModuleMapFileForUniquing(Module)->getName(), ModuleFileName);
   }
@@ -1404,7 +1396,7 @@ createCompilerInstanceForModuleCompile(CompilerInstance &ImportingInstance,
   Module->print(OS);
 
   auto Instance = createCompilerInstanceForModuleCompileImpl(
-      ImportingInstance, ImportLoc, ModuleName,
+      *this, ImportLoc, ModuleName,
       FrontendInputFile(FakeModuleMapFile, IK, +Module->IsSystem),
       ModMap.getModuleMapFileForUniquing(Module)->getName(), ModuleFileName);
 
@@ -1465,8 +1457,8 @@ static bool compileModuleAndReadASTImpl(CompilerInstance &ImportingInstance,
                                         SourceLocation ModuleNameLoc,
                                         Module *Module,
                                         StringRef ModuleFileName) {
-  auto Instance = createCompilerInstanceForModuleCompile(
-      ImportingInstance, ModuleNameLoc, Module, ModuleFileName);
+  auto Instance = ImportingInstance.cloneForModuleCompile(ModuleNameLoc, Module,
+                                                          ModuleFileName);
 
   if (!compileModule(ImportingInstance, ModuleNameLoc,
                      Module->getTopLevelModuleName(), ModuleFileName,

clang/test/CodeGen/catch-alignment-assumption-array.c

@@ -16,9 +16,9 @@ void *caller(void) {
   // CHECK-SANITIZE-NEXT:               %[[PTRINT:.*]] = ptrtoint ptr %[[ARRAYDECAY]] to i64
   // CHECK-SANITIZE-NEXT:               %[[MASKEDPTR:.*]] = and i64 %[[PTRINT]], 0
   // CHECK-SANITIZE-NEXT:               %[[MASKCOND:.*]] = icmp eq i64 %[[MASKEDPTR]], 0
-  // CHECK-SANITIZE-NEXT:               %[[PTRINT_DUP:.*]] = ptrtoint ptr %[[ARRAYDECAY]] to i64, !nosanitize
   // CHECK-SANITIZE-NEXT:               br i1 %[[MASKCOND]], label %[[CONT:.*]], label %[[HANDLER_ALIGNMENT_ASSUMPTION:[^,]+]],{{.*}} !nosanitize
   // CHECK-SANITIZE:                  [[HANDLER_ALIGNMENT_ASSUMPTION]]:
+  // CHECK-SANITIZE-ANYRECOVER-NEXT:    %[[PTRINT_DUP:.*]] = ptrtoint ptr %[[ARRAYDECAY]] to i64, !nosanitize
   // CHECK-SANITIZE-NORECOVER-NEXT:     call void @__ubsan_handle_alignment_assumption_abort(ptr @[[ALIGNMENT_ASSUMPTION]], i64 %[[PTRINT_DUP]], i64 1, i64 0){{.*}}, !nosanitize
   // CHECK-SANITIZE-RECOVER-NEXT:       call void @__ubsan_handle_alignment_assumption(ptr @[[ALIGNMENT_ASSUMPTION]], i64 %[[PTRINT_DUP]], i64 1, i64 0){{.*}}, !nosanitize
   // CHECK-SANITIZE-TRAP-NEXT:          call void @llvm.ubsantrap(i8 23){{.*}}, !nosanitize

clang/test/CodeGen/catch-alignment-assumption-attribute-align_value-on-lvalue.cpp

@@ -24,9 +24,9 @@ char **load_from_ac_struct(struct ac_struct *x) {
   // CHECK-SANITIZE-NEXT:               %[[PTRINT:.*]] = ptrtoint ptr %[[A]] to i64
   // CHECK-SANITIZE-NEXT:               %[[MASKEDPTR:.*]] = and i64 %[[PTRINT]], 4294967295
   // CHECK-SANITIZE-NEXT:               %[[MASKCOND:.*]] = icmp eq i64 %[[MASKEDPTR]], 0
-  // CHECK-SANITIZE-NEXT:               %[[PTRINT_DUP:.*]] = ptrtoint ptr %[[A]] to i64, !nosanitize
   // CHECK-SANITIZE-NEXT:               br i1 %[[MASKCOND]], label %[[CONT:.*]], label %[[HANDLER_ALIGNMENT_ASSUMPTION:[^,]+]],{{.*}} !nosanitize
   // CHECK-SANITIZE:                  [[HANDLER_ALIGNMENT_ASSUMPTION]]:
+  // CHECK-SANITIZE-ANYRECOVER-NEXT:    %[[PTRINT_DUP:.*]] = ptrtoint ptr %[[A]] to i64, !nosanitize
   // CHECK-SANITIZE-NORECOVER-NEXT:     call void @__ubsan_handle_alignment_assumption_abort(ptr @[[LINE_100_ALIGNMENT_ASSUMPTION]], i64 %[[PTRINT_DUP]], i64 4294967296, i64 0){{.*}}, !nosanitize
   // CHECK-SANITIZE-RECOVER-NEXT:       call void @__ubsan_handle_alignment_assumption(ptr @[[LINE_100_ALIGNMENT_ASSUMPTION]], i64 %[[PTRINT_DUP]], i64 4294967296, i64 0){{.*}}, !nosanitize
   // CHECK-SANITIZE-TRAP-NEXT:          call void @llvm.ubsantrap(i8 23){{.*}}, !nosanitize