Merge from 'sycl' to 'sycl-web' (intel#30)

.github/workflows/gh_pages.yml

@@ -12,10 +12,6 @@ jobs:
       with:
         ref: sycl
         path: repo
-    - uses: actions/checkout@v2
-      with:
-        repository: intel/llvm-docs
-        path: docs
     - name: Install deps
       run: sudo apt-get install -y doxygen graphviz ssh ninja-build
     - name: Build Docs
@@ -34,6 +30,7 @@ jobs:
         chmod 600 ~/.ssh/id_rsa
         eval "$(ssh-agent -s)"
         ssh-add -k ~/.ssh/id_rsa
+        git clone [email protected]:intel/llvm-docs.git docs
         cd $GITHUB_WORKSPACE/docs
         yes | \cp -rf $GITHUB_WORKSPACE/build/tools/sycl/doc/doxygen/html/* .
         git config --global user.name "iclsrc"

clang/include/clang/Basic/DiagnosticDriverKinds.td

@@ -324,6 +324,8 @@ def warn_drv_treating_input_as_cxx : Warning<
   InGroup<Deprecated>;
 def warn_drv_pch_not_first_include : Warning<
   "precompiled header '%0' was ignored because '%1' is not first '-include'">;
+def warn_drv_existing_archive_append: Warning<
+  "appending to an existing archive '%0'">, InGroup<DiagGroup<"archive-append">>;
 def warn_missing_sysroot : Warning<"no such sysroot directory: '%0'">,
   InGroup<DiagGroup<"missing-sysroot">>;
 def warn_incompatible_sysroot : Warning<"using sysroot for '%0' but targeting '%1'">,

clang/include/clang/Driver/Types.def

@@ -110,4 +110,5 @@ TYPE("wholearchive",             WholeArchive, INVALID,         "a",      phases
 TYPE("fpga_aocx",                FPGA_AOCX,    INVALID,         "aocx",   phases::Compile, phases::Backend, phases::Assemble, phases::Link)
 TYPE("fpga_aocr",                FPGA_AOCR,    INVALID,         "aocr",   phases::Compile, phases::Backend, phases::Assemble, phases::Link)
 TYPE("fpga_aoco",                FPGA_AOCO,    INVALID,         "aoco",   phases::Compile, phases::Backend, phases::Assemble, phases::Link)
+TYPE("fpga_dependencies",        FPGA_Dependencies, INVALID,    "d",      phases::Compile, phases::Backend, phases::Assemble, phases::Link)
 TYPE("none",                     Nothing,      INVALID,         nullptr,  phases::Compile, phases::Backend, phases::Assemble, phases::Link)

clang/lib/CodeGen/CGSYCLRuntime.cpp

@@ -104,7 +104,7 @@ bool Util::matchQualifiedTypeName(const CXXRecordDecl *RecTy,
   // (namespace) and name.
   if (!RecTy)
     return false; // only classes/structs supported
-  const auto *Ctx = dyn_cast<DeclContext>(RecTy);
+  const auto *Ctx = cast<DeclContext>(RecTy);
   StringRef Name = "";
 
   for (const auto &Scope : llvm::reverse(Scopes)) {

clang/lib/CodeGen/SYCLLowerIR/LowerWGScope.cpp

@@ -662,14 +662,14 @@ static void fixupPrivateMemoryPFWILambdaCaptures(CallInst *PFWICall) {
   // now rewrite the captured addresss of a private_memory variables within the
   // PFWI lambda object:
   for (auto &C : PrivMemCaptures) {
-    GetElementPtrInst *NewGEP = dyn_cast<GetElementPtrInst>(C.second->clone());
+    GetElementPtrInst *NewGEP = cast<GetElementPtrInst>(C.second->clone());
     NewGEP->insertBefore(PFWICall);
     IRBuilder<> Bld(PFWICall->getContext());
     Bld.SetInsertPoint(PFWICall);
     Value *Val = C.first;
-    auto ValAS = dyn_cast<PointerType>(Val->getType())->getAddressSpace();
-    auto PtrAS = dyn_cast<PointerType>(NewGEP->getResultElementType())
-                     ->getAddressSpace();
+    auto ValAS = cast<PointerType>(Val->getType())->getAddressSpace();
+    auto PtrAS =
+        cast<PointerType>(NewGEP->getResultElementType())->getAddressSpace();
 
     if (ValAS != PtrAS)
       Val = Bld.CreateAddrSpaceCast(Val, NewGEP->getResultElementType());

clang/lib/Driver/Driver.cpp

@@ -3253,6 +3253,9 @@ class OffloadingActionBuilder final {
     /// Type of output file for FPGA device compilation.
     types::ID FPGAOutType = types::TY_FPGA_AOCX;
 
+    /// List of objects to extract FPGA dependency info from
+    ActionList FPGAObjectInputs;
+
     /// List of CUDA architectures to use in this compilation with NVPTX targets.
     SmallVector<CudaArch, 8> GpuArchList;
 
@@ -3417,8 +3420,12 @@ class OffloadingActionBuilder final {
           // Check if the type of the file is the same as the action. Do not
           // unbundle it if it is not. Do not unbundle .so files, for example,
           // which are not object files.
-          if (IA->getType() == types::TY_Object && !isObjectFile(FileName))
-            return ABRT_Inactive;
+          if (IA->getType() == types::TY_Object) {
+            if (!isObjectFile(FileName))
+              return ABRT_Inactive;
+            if (Args.hasArg(options::OPT_fintelfpga))
+              FPGAObjectInputs.push_back(IA);
+          }
           // When creating FPGA device fat objects, all host objects are
           // partially linked.  Gather that list here.
           if (IA->getType() == types::TY_Object ||
@@ -3598,6 +3605,15 @@ class OffloadingActionBuilder final {
           Action *DeviceBECompileAction;
           ActionList BEActionList;
           BEActionList.push_back(DeviceLinkAction);
+          for (Action *A : FPGAObjectInputs) {
+            // Send any known objects through the unbundler to grab the
+            // dependency file associated.
+            ActionList AL;
+            AL.push_back(A);
+            Action *UnbundleAction = C.MakeAction<OffloadUnbundlingJobAction>(
+                AL, types::TY_FPGA_Dependencies);
+            BEActionList.push_back(UnbundleAction);
+          }
           for (const auto &A : DeviceLibObjects)
             BEActionList.push_back(A);
           DeviceBECompileAction =
@@ -4462,8 +4478,8 @@ void Driver::BuildActions(Compilation &C, DerivedArgList &Args,
             IA->getInputArg().getOption().hasFlag(options::LinkerInput))
           // Pass the Input along to linker only.
           continue;
-        UnbundlerInputs.push_back(LI);
       }
+      UnbundlerInputs.push_back(LI);
     }
     const Arg *LastArg;
     auto addUnbundlerInput = [&](types::ID T, const Arg *A) {
@@ -5484,14 +5500,25 @@ InputInfo Driver::BuildJobsForActionNoCache(
                                                 DependentOffloadKind)}] =
           CurI;
     }
-
-    // Now that we have all the results generated, select the one that should be
-    // returned for the current depending action.
-    std::pair<const Action *, std::string> ActionTC = {
-        A, GetTriplePlusArchString(TC, BoundArch, TargetDeviceOffloadKind)};
-    assert(CachedResults.find(ActionTC) != CachedResults.end() &&
-           "Result does not exist??");
-    Result = CachedResults[ActionTC];
+    // Do a check for a dependency file unbundle for FPGA.  This is out of line
+    // from a regular unbundle, so just create and return the name of the
+    // unbundled file.
+    if (JA->getType() == types::TY_FPGA_Dependencies) {
+      std::string TmpFileName =
+          C.getDriver().GetTemporaryPath(llvm::sys::path::stem(BaseInput), "d");
+      const char *TmpFile =
+          C.addTempFile(C.getArgs().MakeArgString(TmpFileName));
+      Result = InputInfo(types::TY_FPGA_Dependencies, TmpFile, TmpFile);
+      UnbundlingResults.push_back(Result);
+    } else {
+      // Now that we have all the results generated, select the one that should
+      // be returned for the current depending action.
+      std::pair<const Action *, std::string> ActionTC = {
+          A, GetTriplePlusArchString(TC, BoundArch, TargetDeviceOffloadKind)};
+      assert(CachedResults.find(ActionTC) != CachedResults.end() &&
+             "Result does not exist??");
+      Result = CachedResults[ActionTC];
+    }
   } else if (JA->getType() == types::TY_Nothing)
     Result = InputInfo(A, BaseInput);
   else {

clang/lib/Driver/ToolChains/Clang.cpp

@@ -1139,7 +1139,8 @@ void Clang::AddPreprocessingOptions(Compilation &C, const JobAction &JA,
     } else if (!ArgMD) {
       DepFile = "-";
     } else if (ArgMD->getOption().matches(options::OPT_MMD) &&
-               Args.hasArg(options::OPT_fintelfpga)) {
+               Args.hasArg(options::OPT_fintelfpga) &&
+               !Args.hasArg(options::OPT_c)) {
       // When generating dependency files for FPGA AOT, the output files will
       // always be named after the source file.
       DepFile = Args.MakeArgString(Twine(getBaseInputStem(Args, Inputs)) + ".d");
@@ -3944,10 +3945,13 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   // Check number of inputs for sanity. We need at least one input.
   assert(Inputs.size() >= 1 && "Must have at least one input.");
   // CUDA/HIP compilation may have multiple inputs (source file + results of
-  // device-side compilations). OpenMP and SYCL device jobs also take the host
-  // IR as a second input. Module precompilation accepts a list of header files
-  // to include as part of the module. All other jobs are expected to have
-  // exactly one input.
+  // device-side compilations).
+  // OpenMP device jobs take the host IR as a second input.
+  // SYCL host jobs accept the integration header from the device-side
+  // compilation as a second input.
+  // Module precompilation accepts a list of header files to include as part
+  // of the module.
+  // All other jobs are expected to have exactly one input.
   bool IsCuda = JA.isOffloading(Action::OFK_Cuda);
   bool IsHIP = JA.isOffloading(Action::OFK_HIP);
   bool IsOpenMPDevice = JA.isDeviceOffloading(Action::OFK_OpenMP);
@@ -4742,6 +4746,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   // `--gpu-use-aux-triple-only` is specified.
   if (!Args.getLastArg(options::OPT_gpu_use_aux_triple_only) &&
       ((IsCuda && JA.isDeviceOffloading(Action::OFK_Cuda)) ||
+       (IsSYCL && IsSYCLOffloadDevice) ||
        (IsHIP && JA.isDeviceOffloading(Action::OFK_HIP)))) {
     const ArgList &HostArgs =
         C.getArgsForToolChain(nullptr, StringRef(), Action::OFK_None);
@@ -6782,7 +6787,8 @@ const char *Clang::getDependencyFileName(const ArgList &Args,
                                          const InputInfoList &Inputs) {
   // FIXME: Think about this more.
 
-  if (Arg *OutputOpt = Args.getLastArg(options::OPT_o)) {
+  if (Arg *OutputOpt =
+          Args.getLastArg(options::OPT_o, options::OPT__SLASH_Fo)) {
     SmallString<128> OutputFilename(OutputOpt->getValue());
     llvm::sys::path::replace_extension(OutputFilename, llvm::Twine('d'));
     return Args.MakeArgString(OutputFilename);
@@ -7094,6 +7100,20 @@ void OffloadBundler::ConstructJob(Compilation &C, const JobAction &JA,
       Triples += CurDep->getOffloadingArch();
     }
   }
+  bool IsFPGADepBundle = (TCArgs.hasArg(options::OPT_fintelfpga) &&
+                          Output.getType() == types::TY_Object);
+  // For -fintelfpga, when bundling objects we also want to bundle up the
+  // named dependency file.
+  // TODO - We are currently using the target triple inputs to slot a location
+  // of the dependency information into the bundle.  It would be good to
+  // separate this out to an explicit option in the bundler for the dependency
+  // file as it does not match the type being bundled.
+  if (IsFPGADepBundle) {
+    Triples += ',';
+    Triples += Action::GetOffloadKindName(Action::OFK_SYCL);
+    Triples += '-';
+    Triples += llvm::Triple::getArchTypeName(llvm::Triple::fpga_dep);
+  }
   CmdArgs.push_back(TCArgs.MakeArgString(Triples));
 
   // Get bundled file command.
@@ -7118,6 +7138,14 @@ void OffloadBundler::ConstructJob(Compilation &C, const JobAction &JA,
     }
     UB += CurTC->getInputFilename(Inputs[I]);
   }
+  // For -fintelfpga, when bundling objects we also want to bundle up the
+  // named dependency file.
+  if (IsFPGADepBundle) {
+    SmallString<128> CurOutput(Output.getFilename());
+    llvm::sys::path::replace_extension(CurOutput, "d");
+    UB += ',';
+    UB += CurOutput;
+  }
   CmdArgs.push_back(TCArgs.MakeArgString(UB));
 
   // All the inputs are encoded as commands.
@@ -7148,6 +7176,7 @@ void OffloadBundler::ConstructJobMultipleOutputs(
   bool IsMSVCEnv =
       C.getDefaultToolChain().getTriple().isWindowsMSVCEnvironment();
   types::ID InputType(Input.getType());
+  bool IsFPGADepUnbundle = (JA.getType() == types::TY_FPGA_Dependencies);
 
   // For Linux, we have initial support for fat archives (archives which
   // contain bundled objects). We will perform partial linking against the
@@ -7207,6 +7236,8 @@ void OffloadBundler::ConstructJobMultipleOutputs(
   if (InputType == types::TY_FPGA_AOCO ||
       (IsMSVCEnv && types::isArchive(InputType)))
     TypeArg = "aoo";
+  if (IsFPGADepUnbundle)
+    TypeArg = "o";
 
   // Get the type.
   CmdArgs.push_back(TCArgs.MakeArgString(Twine("-type=") + TypeArg));
@@ -7262,7 +7293,15 @@ void OffloadBundler::ConstructJobMultipleOutputs(
       Triples += Dep.DependentBoundArch;
     }
   }
-
+  if (IsFPGADepUnbundle) {
+    // TODO - We are currently using the target triple inputs to slot a location
+    // of the dependency information into the bundle.  It would be good to
+    // separate this out to an explicit option in the bundler for the dependency
+    // file as it does not match the type being bundled.
+    Triples += Action::GetOffloadKindName(Action::OFK_SYCL);
+    Triples += '-';
+    Triples += llvm::Triple::getArchTypeName(llvm::Triple::fpga_dep);
+  }
   CmdArgs.push_back(TCArgs.MakeArgString(Triples));
 
   // Get bundled file command.
@@ -7272,10 +7311,17 @@ void OffloadBundler::ConstructJobMultipleOutputs(
   // Get unbundled files command.
   SmallString<128> UB;
   UB += "-outputs=";
-  for (unsigned I = 0; I < Outputs.size(); ++I) {
-    if (I)
-      UB += ',';
-    UB += DepInfo[I].DependentToolChain->getInputFilename(Outputs[I]);
+  // When dealing with -fintelfpga, there is an additional unbundle step
+  // that occurs for the dependency file.  In that case, do not use the
+  // dependent information, but just the output file.
+  if (IsFPGADepUnbundle)
+    UB += Outputs[0].getFilename();
+  else {
+    for (unsigned I = 0; I < Outputs.size(); ++I) {
+      if (I)
+        UB += ',';
+      UB += DepInfo[I].DependentToolChain->getInputFilename(Outputs[I]);
+    }
   }
   CmdArgs.push_back(TCArgs.MakeArgString(UB));
   CmdArgs.push_back("-unbundle");

clang/lib/Driver/ToolChains/Gnu.cpp

@@ -348,7 +348,12 @@ void tools::gnutools::Linker::constructLLVMARCommand(
     const InputInfoList &Input, const ArgList &Args) const {
   ArgStringList CmdArgs;
   CmdArgs.push_back("cr");
-  CmdArgs.push_back(Output.getFilename());
+  const char *OutputFilename = Output.getFilename();
+  if (llvm::sys::fs::exists(OutputFilename)) {
+    C.getDriver().Diag(clang::diag::warn_drv_existing_archive_append)
+        << OutputFilename;
+  }
+  CmdArgs.push_back(OutputFilename);
   for (const auto &II : Input) {
     if (II.getType() == types::TY_Tempfilelist) {
       // Take the list file and pass it in with '@'.

clang/lib/Driver/ToolChains/SYCL.cpp

@@ -208,6 +208,7 @@ void SYCL::fpga::BackendCompiler::ConstructJob(Compilation &C,
          "Unsupported target");
 
   InputInfoList ForeachInputs;
+  InputInfoList FPGADepFiles;
   ArgStringList CmdArgs{"-o",  Output.getFilename()};
   for (const auto &II : Inputs) {
     std::string Filename(II.getFilename());
@@ -217,6 +218,8 @@ void SYCL::fpga::BackendCompiler::ConstructJob(Compilation &C,
       // Add any FPGA library lists.  These come in as special tempfile lists.
       CmdArgs.push_back(Args.MakeArgString(Twine("-library-list=") +
           Filename));
+    else if (II.getType() == types::TY_FPGA_Dependencies)
+      FPGADepFiles.push_back(II);
     else
       CmdArgs.push_back(C.getArgs().MakeArgString(Filename));
   }
@@ -229,8 +232,6 @@ void SYCL::fpga::BackendCompiler::ConstructJob(Compilation &C,
       ForeachExt = "aocr";
     }
 
-
-  InputInfoList FPGADepFiles;
   for (auto *A : Args) {
     // Any input file is assumed to have a dependency file associated
     if (A->getOption().getKind() == Option::InputClass) {
@@ -240,7 +241,7 @@ void SYCL::fpga::BackendCompiler::ConstructJob(Compilation &C,
         types::ID Ty = getToolChain().LookupTypeForExtension(Ext.drop_front());
         if (Ty == types::TY_INVALID)
           continue;
-        if (types::isSrcFile(Ty) || Ty == types::TY_Object) {
+        if (types::isSrcFile(Ty)) {
           llvm::sys::path::replace_extension(FN, "d");
           FPGADepFiles.push_back(InputInfo(types::TY_Dependencies,
               Args.MakeArgString(FN), Args.MakeArgString(FN)));

clang/lib/Sema/SemaSYCL.cpp

@@ -138,12 +138,8 @@ static bool IsSyclMathFunc(unsigned BuiltinID) {
   case Builtin::BI__builtin_truncl:
   case Builtin::BIlroundl:
   case Builtin::BI__builtin_lroundl:
-  case Builtin::BIceil:
-  case Builtin::BI__builtin_ceil:
   case Builtin::BIcopysign:
   case Builtin::BI__builtin_copysign:
-  case Builtin::BIfabs:
-  case Builtin::BI__builtin_fabs:
   case Builtin::BIfloor:
   case Builtin::BI__builtin_floor:
   case Builtin::BIfmax:
@@ -158,12 +154,8 @@ static bool IsSyclMathFunc(unsigned BuiltinID) {
   case Builtin::BI__builtin_round:
   case Builtin::BItrunc:
   case Builtin::BI__builtin_trunc:
-  case Builtin::BIceilf:
-  case Builtin::BI__builtin_ceilf:
   case Builtin::BIcopysignf:
   case Builtin::BI__builtin_copysignf:
-  case Builtin::BIfabsf:
-  case Builtin::BI__builtin_fabsf:
   case Builtin::BIfloorf:
   case Builtin::BI__builtin_floorf:
   case Builtin::BIfmaxf:
@@ -215,6 +207,8 @@ class MarkDeviceFunction : public RecursiveASTVisitor<MarkDeviceFunction> {
 
     if (FunctionDecl *Callee = e->getDirectCallee()) {
       Callee = Callee->getCanonicalDecl();
+      assert(Callee && "Device function canonical decl must be available");
+
       // Remember that all SYCL kernel functions have deferred
       // instantiation as template functions. It means that
       // all functions used by kernel have already been parsed and have
@@ -254,7 +248,7 @@ class MarkDeviceFunction : public RecursiveASTVisitor<MarkDeviceFunction> {
       }
       // Specifically check if the math library function corresponding to this
       // builtin is supported for SYCL
-      unsigned BuiltinID = (Callee ? Callee->getBuiltinID() : 0);
+      unsigned BuiltinID = Callee->getBuiltinID();
       if (BuiltinID && !IsSyclMathFunc(BuiltinID)) {
         StringRef Name = SemaRef.Context.BuiltinInfo.getName(BuiltinID);
         SemaRef.Diag(e->getExprLoc(), diag::err_builtin_target_unsupported)

clang/lib/Sema/SemaType.cpp

@@ -2344,6 +2344,12 @@ QualType Sema::BuildArrayType(QualType T, ArrayType::ArraySizeModifier ASM,
             << ArraySize->getSourceRange();
         ASM = ArrayType::Normal;
       }
+
+      // Zero length arrays are disallowed in SYCL device code.
+      if (getLangOpts().SYCLIsDevice)
+        SYCLDiagIfDeviceCode(ArraySize->getBeginLoc(),
+                             diag::err_typecheck_zero_array_size)
+            << ArraySize->getSourceRange();
     } else if (!T->isDependentType() && !T->isVariablyModifiedType() &&
                !T->isIncompleteType() && !T->isUndeducedType()) {
       // Is the array too large?