merge internal development externally

amd/comgr/cmake/opencl_pch.cmake

@@ -1,26 +1,8 @@
 if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR)
   find_package(Clang REQUIRED CONFIG)
-
-  # FIXME: CLANG_CMAKE_DIR seems like the most stable way to find this, but
-  # really there is no way to reliably discover this header.
-  #
-  # We effectively back up to the Clang output directory (for the case of a build
-  # tree) or install prefix (for the case of an installed copy), and then search
-  # for a file named opencl-c.h anywhere below that. We take the first result in
-  # the case where there are multiple (e.g. if there is an installed copy nested
-  # in a build directory). This is a bit imprecise, but it covers cases like MSVC
-  # adding some additional configuration-specific subdirectories to the build
-  # tree but not to an installed copy.
-  file(GLOB_RECURSE OPENCL_C_H_LIST "${CLANG_CMAKE_DIR}/../../../*/opencl-c.h")
-
-  list(GET OPENCL_C_H_LIST 0 OPENCL_C_H)
-
-  if (NOT EXISTS "${OPENCL_C_H}" OR IS_DIRECTORY "${OPENCL_C_H}")
-    message(FATAL_ERROR "Unable to locate opencl-c.h from the supplied Clang. The path '${CLANG_CMAKE_DIR}/../../../*' was searched.")
-  endif()
-else()
-  get_target_property(clang_build_header_dir clang-resource-headers RUNTIME_OUTPUT_DIRECTORY)
-  set(OPENCL_C_H "${clang_build_header_dir}/opencl-c.h")
+  execute_process(COMMAND "${CLANG_CMAKE_DIR}/../../../bin/clang" -print-resource-dir OUTPUT_VARIABLE CLANG_RESOURCE_DIR)
+  string(STRIP ${CLANG_RESOURCE_DIR} CLANG_RESOURCE_DIR)
+  set(OPENCL_C_H "${CLANG_RESOURCE_DIR}/include/opencl-c.h")
 endif()
 
 # Macro to create and install a custom target for generating PCH for given

amd/comgr/src/comgr-compiler.cpp

@@ -948,6 +948,8 @@ AMDGPUCompiler::processFiles(amd_comgr_data_kind_t OutputKind,
     if (auto Status = amd_comgr_create_data(OutputKind, &OutputT)) {
       return Status;
     }
+
+    // OutputT can be released after addition to the data_set
     ScopedDataObjectReleaser SDOR(OutputT);
 
     DataObject *Output = DataObject::convert(OutputT);
@@ -1365,6 +1367,9 @@ amd_comgr_status_t AMDGPUCompiler::unbundle() {
                                               &ResultT))
         return Status;
 
+      // ResultT can be released after addition to the data_set
+      ScopedDataObjectReleaser SDOR(ResultT);
+
       DataObject *Result = DataObject::convert(ResultT);
       if (auto Status = inputFromFile(Result, StringRef(output_file_path)))
         return Status;
@@ -1520,6 +1525,9 @@ amd_comgr_status_t AMDGPUCompiler::linkBitcodeToBitcode() {
       if (auto Status = amd_comgr_create_data(AMD_COMGR_DATA_KIND_BC, &ResultT))
         return Status;
 
+      // ResultT can be released after addition to the data_set
+      ScopedDataObjectReleaser SDOR(ResultT);
+
       DataObject *Result = DataObject::convert(ResultT);
       if (auto Status = inputFromFile(Result, StringRef(output_file_path)))
         return Status;
@@ -1539,8 +1547,6 @@ amd_comgr_status_t AMDGPUCompiler::linkBitcodeToBitcode() {
         return AMD_COMGR_STATUS_ERROR;
       if (L.linkInModule(std::move(Mod), ApplicableFlags))
         return AMD_COMGR_STATUS_ERROR;
-
-      Result->release();
     }
     // Unbundle bitcode archive
     else if (Input->DataKind == AMD_COMGR_DATA_KIND_AR_BUNDLE) {
@@ -1620,6 +1626,9 @@ amd_comgr_status_t AMDGPUCompiler::linkBitcodeToBitcode() {
       if (auto Status = amd_comgr_create_data(AMD_COMGR_DATA_KIND_AR, &ResultT))
         return Status;
 
+      // ResultT can be released after addition to the data_set
+      ScopedDataObjectReleaser SDOR(ResultT);
+
       DataObject *Result = DataObject::convert(ResultT);
       if (auto Status = inputFromFile(Result, StringRef(output_file_path)))
         return Status;
@@ -1699,6 +1708,8 @@ amd_comgr_status_t AMDGPUCompiler::linkBitcodeToBitcode() {
   if (auto Status = amd_comgr_create_data(AMD_COMGR_DATA_KIND_BC, &OutputT)) {
     return Status;
   }
+
+  // OutputT can be released after addition to the data_set
   ScopedDataObjectReleaser SDOR(OutputT);
 
   DataObject *Output = DataObject::convert(OutputT);
@@ -1795,6 +1806,8 @@ amd_comgr_status_t AMDGPUCompiler::linkToRelocatable() {
           amd_comgr_create_data(AMD_COMGR_DATA_KIND_RELOCATABLE, &OutputT)) {
     return Status;
   }
+
+  // OutputT can be released after addition to the data_set
   ScopedDataObjectReleaser SDOR(OutputT);
 
   DataObject *Output = DataObject::convert(OutputT);
@@ -1849,6 +1862,7 @@ amd_comgr_status_t AMDGPUCompiler::linkToExecutable() {
           amd_comgr_create_data(AMD_COMGR_DATA_KIND_EXECUTABLE, &OutputT)) {
     return Status;
   }
+  // OutputT can be released after addition to the data_set
   ScopedDataObjectReleaser SDOR(OutputT);
 
   DataObject *Output = DataObject::convert(OutputT);

clang/docs/ClangLinkerWrapper.rst

@@ -46,6 +46,8 @@ only for the linker wrapper will be forwarded to the wrapped linker job.
     -l <libname>           Search for library <libname>
     --opt-level=<O0, O1, O2, or O3>
                            Optimization level for LTO
+    --override-image=<kind=file>
+                            Uses the provided file as if it were the output of the device link step
     -o <path>              Path to file to write output
     --pass-remarks-analysis=<value>
                            Pass remarks for LTO
@@ -87,6 +89,42 @@ other. Generally, this requires that the target triple and architecture match.
 An exception is made when the architecture is listed as ``generic``, which will
 cause it be linked with any other device code with the same target triple.
 
+Debugging
+=========
+
+The linker wrapper performs a lot of steps internally, such as input matching,
+symbol resolution, and image registration. This makes it difficult to debug in
+some scenarios. The behavior of the linker-wrapper is controlled mostly through
+metadata, described in `clang documentation
+<https://clang.llvm.org/docs/OffloadingDesign.html>`_. Intermediate output can
+be obtained from the linker-wrapper using the ``--save-temps`` flag. These files
+can then be modified.
+
+.. code-block:: sh
+
+  $> clang openmp.c -fopenmp --offload-arch=gfx90a -c
+  $> clang openmp.o -fopenmp --offload-arch=gfx90a -Wl,--save-temps
+  $> ; Modify temp files.
+  $> llvm-objcopy --update-section=.llvm.offloading=out.bc openmp.o
+
+Doing this will allow you to override one of the input files by replacing its
+embedded offloading metadata with a user-modified version. However, this will be
+more difficult when there are multiple input files. For a very large hammer, the
+``--override-image=<kind>=<file>`` flag can be used.
+
+In the following example, we use the ``--save-temps`` to obtain the LLVM-IR just
+before running the backend. We then modify it to test altered behavior, and then
+compile it to a binary. This can then be passed to the linker-wrapper which will
+then ignore all embedded metadata and use the provided image as if it were the
+result of the device linking phase.
+
+.. code-block:: sh
+
+  $> clang openmp.c -fopenmp --offload-arch=gfx90a -Wl,--save-temps
+  $> ; Modify temp files.
+  $> clang --target=amdgcn-amd-amdhsa -mcpu=gfx90a -nogpulib out.bc -o a.out
+  $> clang openmp.c -fopenmp --offload-arch=gfx90a -Wl,--override-image=openmp=a.out
+
 Example
 =======
 

clang/lib/AST/Interp/ByteCodeExprGen.cpp

@@ -2488,6 +2488,16 @@ bool ByteCodeExprGen<Emitter>::VisitRecoveryExpr(const RecoveryExpr *E) {
   return this->emitError(E);
 }
 
+template <class Emitter>
+bool ByteCodeExprGen<Emitter>::VisitAddrLabelExpr(const AddrLabelExpr *E) {
+  assert(E->getType()->isVoidPointerType());
+
+  unsigned Offset = allocateLocalPrimitive(
+      E->getLabel(), PT_Ptr, /*IsConst=*/true, /*IsExtended=*/false);
+
+  return this->emitGetLocal(PT_Ptr, Offset, E);
+}
+
 template <class Emitter> bool ByteCodeExprGen<Emitter>::discard(const Expr *E) {
   OptionScope<Emitter> Scope(this, /*NewDiscardResult=*/true,
                              /*NewInitializing=*/false);

clang/lib/AST/Interp/ByteCodeExprGen.h

@@ -122,6 +122,7 @@ class ByteCodeExprGen : public ConstStmtVisitor<ByteCodeExprGen<Emitter>, bool>,
   bool VisitPseudoObjectExpr(const PseudoObjectExpr *E);
   bool VisitPackIndexingExpr(const PackIndexingExpr *E);
   bool VisitRecoveryExpr(const RecoveryExpr *E);
+  bool VisitAddrLabelExpr(const AddrLabelExpr *E);
 
 protected:
   bool visitExpr(const Expr *E) override;

clang/lib/AST/Interp/Interp.cpp

@@ -302,7 +302,9 @@ bool CheckConstant(InterpState &S, CodePtr OpPC, const Descriptor *Desc) {
 
     QualType T = VD->getType();
     if (S.getLangOpts().CPlusPlus && !S.getLangOpts().CPlusPlus11)
-      return T->isSignedIntegerOrEnumerationType() || T->isUnsignedIntegerOrEnumerationType();
+      return (T->isSignedIntegerOrEnumerationType() ||
+              T->isUnsignedIntegerOrEnumerationType()) &&
+             T.isConstQualified();
 
     if (T.isConstQualified())
       return true;
@@ -316,12 +318,10 @@ bool CheckConstant(InterpState &S, CodePtr OpPC, const Descriptor *Desc) {
     return false;
   };
 
-  if (const auto *D = Desc->asValueDecl()) {
-    if (const auto *VD = dyn_cast<VarDecl>(D);
-        VD && VD->hasGlobalStorage() && !IsConstType(VD)) {
-      diagnoseNonConstVariable(S, OpPC, VD);
-      return S.inConstantContext();
-    }
+  if (const auto *D = Desc->asVarDecl();
+      D && D->hasGlobalStorage() && !IsConstType(D)) {
+    diagnoseNonConstVariable(S, OpPC, D);
+    return S.inConstantContext();
   }
 
   return true;

clang/lib/AST/Interp/Pointer.cpp

@@ -191,7 +191,7 @@ void Pointer::print(llvm::raw_ostream &OS) const {
     else
       OS << Offset << ", ";
 
-    if (isBlockPointer() && PointeeStorage.BS.Pointee)
+    if (PointeeStorage.BS.Pointee)
       OS << PointeeStorage.BS.Pointee->getSize();
     else
       OS << "nullptr";

clang/lib/AST/Interp/Program.cpp

@@ -152,8 +152,7 @@ std::optional<unsigned> Program::getOrCreateDummy(const ValueDecl *VD) {
   if (std::optional<PrimType> T = Ctx.classify(QT))
     Desc = createDescriptor(VD, *T, std::nullopt, true, false);
   else
-    Desc = createDescriptor(VD, VD->getType().getTypePtr(), std::nullopt, true,
-                            false);
+    Desc = createDescriptor(VD, QT.getTypePtr(), std::nullopt, true, false);
   if (!Desc)
     Desc = allocateDescriptor(VD);
 

clang/lib/CodeGen/Targets/AArch64.cpp

@@ -317,7 +317,7 @@ AArch64ABIInfo::classifyArgumentType(QualType Ty, bool IsVariadic,
 
     if (const auto *EIT = Ty->getAs<BitIntType>())
       if (EIT->getNumBits() > 128)
-        return getNaturalAlignIndirect(Ty);
+        return getNaturalAlignIndirect(Ty, false);
 
     return (isPromotableIntegerTypeForABI(Ty) && isDarwinPCS()
                 ? ABIArgInfo::getExtend(Ty)

clang/lib/Format/TokenAnnotator.cpp

@@ -5405,6 +5405,9 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
     return true;
   }
 
+  const auto *BeforeLeft = Left.Previous;
+  const auto *AfterRight = Right.Next;
+
   if (Style.isCSharp()) {
     if (Left.is(TT_FatArrow) && Right.is(tok::l_brace) &&
         Style.BraceWrapping.AfterFunction) {
@@ -5416,7 +5419,7 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
     }
     if (Right.is(TT_CSharpGenericTypeConstraint))
       return true;
-    if (Right.Next && Right.Next->is(TT_FatArrow) &&
+    if (AfterRight && AfterRight->is(TT_FatArrow) &&
         (Right.is(tok::numeric_constant) ||
          (Right.is(tok::identifier) && Right.TokenText == "_"))) {
       return true;
@@ -5433,15 +5436,14 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
         Left.is(tok::r_square) && Right.is(tok::l_square)) {
       return true;
     }
-
   } else if (Style.isJavaScript()) {
     // FIXME: This might apply to other languages and token kinds.
-    if (Right.is(tok::string_literal) && Left.is(tok::plus) && Left.Previous &&
-        Left.Previous->is(tok::string_literal)) {
+    if (Right.is(tok::string_literal) && Left.is(tok::plus) && BeforeLeft &&
+        BeforeLeft->is(tok::string_literal)) {
       return true;
     }
     if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) && Line.Level == 0 &&
-        Left.Previous && Left.Previous->is(tok::equal) &&
+        BeforeLeft && BeforeLeft->is(tok::equal) &&
         Line.First->isOneOf(tok::identifier, Keywords.kw_import, tok::kw_export,
                             tok::kw_const) &&
         // kw_var/kw_let are pseudo-tokens that are tok::identifier, so match
@@ -5460,8 +5462,8 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
       // instead of bin-packing.
       return true;
     }
-    if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && Left.Previous &&
-        Left.Previous->is(TT_FatArrow)) {
+    if (Right.is(tok::r_brace) && Left.is(tok::l_brace) && BeforeLeft &&
+        BeforeLeft->is(TT_FatArrow)) {
       // JS arrow function (=> {...}).
       switch (Style.AllowShortLambdasOnASingleLine) {
       case FormatStyle::SLS_All:
@@ -5489,8 +5491,8 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
                   FormatStyle::SFS_InlineOnly);
     }
   } else if (Style.Language == FormatStyle::LK_Java) {
-    if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next &&
-        Right.Next->is(tok::string_literal)) {
+    if (Right.is(tok::plus) && Left.is(tok::string_literal) && AfterRight &&
+        AfterRight->is(tok::string_literal)) {
       return true;
     }
   } else if (Style.isVerilog()) {
@@ -5543,8 +5545,7 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
       }
       return Style.BreakArrays;
     }
-  }
-  if (Style.isTableGen()) {
+  } else if (Style.isTableGen()) {
     // Break the comma in side cond operators.
     // !cond(case1:1,
     //       case2:0);
@@ -5600,8 +5601,8 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
     return true;
   if (Left.IsUnterminatedLiteral)
     return true;
-  if (Right.is(tok::lessless) && Right.Next && Left.is(tok::string_literal) &&
-      Right.Next->is(tok::string_literal)) {
+  if (Right.is(tok::lessless) && AfterRight && Left.is(tok::string_literal) &&
+      AfterRight->is(tok::string_literal)) {
     return true;
   }
   if (Right.is(TT_RequiresClause)) {
@@ -5678,8 +5679,8 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
     // string literal accordingly. Thus, we try keep existing line breaks.
     return Right.IsMultiline && Right.NewlinesBefore > 0;
   }
-  if ((Left.is(tok::l_brace) || (Left.is(tok::less) && Left.Previous &&
-                                 Left.Previous->is(tok::equal))) &&
+  if ((Left.is(tok::l_brace) ||
+       (Left.is(tok::less) && BeforeLeft && BeforeLeft->is(tok::equal))) &&
       Right.NestingLevel == 1 && Style.Language == FormatStyle::LK_Proto) {
     // Don't put enums or option definitions onto single lines in protocol
     // buffers.
@@ -5793,7 +5794,7 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
   //
   // We ensure elsewhere that extensions are always on their own line.
   if (Style.isProto() && Right.is(TT_SelectorName) &&
-      Right.isNot(tok::r_square) && Right.Next) {
+      Right.isNot(tok::r_square) && AfterRight) {
     // Keep `@submessage` together in:
     // @submessage { key: value }
     if (Left.is(tok::at))
@@ -5802,7 +5803,7 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
     // selector { ...
     // selector: { ...
     // selector: @base { ...
-    FormatToken *LBrace = Right.Next;
+    const auto *LBrace = AfterRight;
     if (LBrace && LBrace->is(tok::colon)) {
       LBrace = LBrace->Next;
       if (LBrace && LBrace->is(tok::at)) {

clang/lib/StaticAnalyzer/Checkers/ErrnoChecker.cpp

@@ -205,7 +205,7 @@ void ErrnoChecker::checkPreCall(const CallEvent &Call,
   // Probably 'strerror'?
   if (CallF->isExternC() && CallF->isGlobal() &&
       C.getSourceManager().isInSystemHeader(CallF->getLocation()) &&
-      !isErrno(CallF)) {
+      !isErrnoLocationCall(Call)) {
     if (getErrnoState(C.getState()) == MustBeChecked) {
       std::optional<ento::Loc> ErrnoLoc = getErrnoLoc(C.getState());
       assert(ErrnoLoc && "ErrnoLoc should exist if an errno state is set.");