Merge from 'sycl' to 'sycl-web' (intel#33)

buildbot/dependency.conf

@@ -4,11 +4,11 @@ ocl_cpu_rt_ver=2020.11.8.0.27
 # https://github.com/intel/llvm/releases/download/2020-WW36/win-oclcpuexp-2020.11.8.0.27_rel.zip
 ocl_cpu_rt_ver_win=2020.11.8.0.27
 # Same GPU driver supports Level Zero and OpenCL:
-# https://github.com/intel/compute-runtime/releases/tag/20.29.17408
-ocl_gpu_rt_ver=20.29.17408
+# https://github.com/intel/compute-runtime/releases/tag/20.35.17767
+ocl_gpu_rt_ver=20.35.17767
 # Same GPU driver supports Level Zero and OpenCL:
-# https://downloadmirror.intel.com/29674/a08/igfx_win10_100.8336.zip
-ocl_gpu_rt_ver_win=27.20.100.8336
+# https://downloadmirror.intel.com/29817/a08/igfx_win10_100.8673.zip
+ocl_gpu_rt_ver_win=27.20.100.8673
 intel_sycl_ver=build
 # https://github.com/oneapi-src/oneTBB/releases/download/v2021.1-beta08/oneapi-tbb-2021.1-beta08-lin.tgz
 tbb_ver=2021.1.9.636
@@ -24,8 +24,8 @@ fpga_ver_win=20200811_000006
 [DRIVER VERSIONS]
 cpu_driver_lin=2020.11.8.0.27
 cpu_driver_win=2020.11.8.0.27
-gpu_driver_lin=20.29.17408
-gpu_driver_win=27.20.100.8336
+gpu_driver_lin=20.35.17767
+gpu_driver_win=27.20.100.8673
 fpga_driver_lin=2020.11.8.0.27
 fpga_driver_win=2020.11.8.0.27
 # NVidia CUDA driver

clang/include/clang/Basic/Attr.td

@@ -1299,7 +1299,6 @@ def LoopUnrollHint : InheritableAttr {
 
 def IntelReqdSubGroupSize: InheritableAttr {
   let Spellings = [GNU<"intel_reqd_sub_group_size">,
-                   CXX11<"cl", "intel_reqd_sub_group_size">,
                    CXX11<"intel", "reqd_sub_group_size">];
   let Args = [ExprArgument<"SubGroupSize">];
   let Subjects = SubjectList<[Function, CXXMethod], ErrorDiag>;

clang/include/clang/Basic/AttrDocs.td

@@ -3476,30 +3476,16 @@ code. See `cl_intel_required_subgroup_size
 for details.
 
 SYCL documentation:
-The [[cl::intel_reqd_sub_group_size(n)]] and [[intel::reqd_sub_group_size(n)]]
-attribute indicates that the kernel must be compiled and executed with a
-sub-group of size n. The value of n must be set to a sub-group size supported
-by the device, or device compilation will fail.
+The [[intel::reqd_sub_group_size(n)]] attribute indicates that the kernel must
+be compiled and executed with a sub-group of size n. The value of n must be set
+to a sub-group size supported by the device, or device compilation will fail.
 
 In addition to device functions, the required sub-group size attribute may also
 be specified in the definition of a named functor object and lambda functions,
 as in the examples below:
 
 .. code-block:: c++
 
-  class Functor
-  {
-      void operator()(item<1> item) [[cl::intel_reqd_sub_group_size(16)]]
-      {
-          /* kernel code */
-      }
-  }
-
-  kernel<class kernel_name>(
-  []() [[cl::intel_reqd_sub_group_size(n)]] {
-       /* kernel code */
-  });
-
   class Functor
   {
       [[intel::reqd_sub_group_size(16)]] void operator()(item<1> item)

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -11002,11 +11002,9 @@ def err_sycl_restrict : Error<
   "|use a const static or global variable that is neither zero-initialized "
   "nor constant-initialized"
   "}0">;
-def warn_sycl_kernel_too_many_args : Warning<
-  "kernel argument count (%0) exceeds supported maximum of %1 on GPU">,
-  InGroup<SyclStrict>;
-def note_sycl_kernel_args_count : Note<"array elements and fields of a "
-  "class/struct may be counted separately">;
+def warn_sycl_kernel_too_big_args : Warning<
+  "size of kernel arguments (%0 bytes) exceeds supported maximum of %1 bytes "
+  "on GPU">, InGroup<SyclStrict>;
 def err_sycl_virtual_types : Error<
   "No class with a vtable can be used in a SYCL kernel or any code included in the kernel">;
 def note_sycl_recursive_function_declared_here: Note<"function implemented using recursion declared here">;
@@ -11073,11 +11071,6 @@ def err_ivdep_declrefexpr_arg : Error<
 def warn_ivdep_redundant : Warning <"ignoring redundant Intel FPGA loop "
   "attribute 'ivdep': safelen %select{INF|%1}0 >= safelen %select{INF|%3}2">,
   InGroup<IgnoredAttributes>;
-def warn_attribute_spelling_deprecated : Warning<
-  "attribute %0 is deprecated">,
-  InGroup<DeprecatedAttributes>;
-def note_spelling_suggestion : Note<
-  "did you mean to use %0 instead?">;
 
 // errors of expect.with.probability
 def err_probability_not_constant_float : Error<

clang/lib/Driver/Driver.cpp

@@ -6119,8 +6119,9 @@ const char *Driver::GetNamedOutputPath(Compilation &C, const JobAction &JA,
   }
 
   // For /P, preprocess to file named after BaseInput.
-  if (C.getArgs().hasArg(options::OPT__SLASH_P)) {
-    assert(AtTopLevel && isa<PreprocessJobAction>(JA));
+  if (C.getArgs().hasArg(options::OPT__SLASH_P) &&
+      ((AtTopLevel && isa<PreprocessJobAction>(JA)) ||
+       isa<OffloadBundlingJobAction>(JA))) {
     StringRef BaseName = llvm::sys::path::filename(BaseInput);
     StringRef NameArg;
     if (Arg *A = C.getArgs().getLastArg(options::OPT__SLASH_Fi))

clang/lib/Driver/ToolChains/Clang.cpp

@@ -4201,8 +4201,8 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
       SYCLStdArg->render(Args, CmdArgs);
       CmdArgs.push_back("-fsycl-std-layout-kernel-params");
     } else {
-      // Ensure the default version in SYCL mode is 1.2.1 (aka 2017)
-      CmdArgs.push_back("-sycl-std=2017");
+      // Ensure the default version in SYCL mode is 2020
+      CmdArgs.push_back("-sycl-std=2020");
     }
   }
 
@@ -5996,10 +5996,17 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
                     options::OPT_fno_gnu_inline_asm, true))
     CmdArgs.push_back("-fno-gnu-inline-asm");
 
+  bool EnableSYCLEarlyOptimizations =
+      Args.hasFlag(options::OPT_fsycl_early_optimizations,
+                   options::OPT_fno_sycl_early_optimizations,
+                   Triple.getSubArch() != llvm::Triple::SPIRSubArch_fpga);
+
   // Enable vectorization per default according to the optimization level
   // selected. For optimization levels that want vectorization we use the alias
   // option to simplify the hasFlag logic.
   bool EnableVec = shouldEnableVectorizerAtOLevel(Args, false);
+  if (UseSYCLTriple && EnableSYCLEarlyOptimizations)
+    EnableVec = false; // But disable vectorization for SYCL device code
   OptSpecifier VectorizeAliasOption =
       EnableVec ? options::OPT_O_Group : options::OPT_fvectorize;
   if (Args.hasFlag(options::OPT_fvectorize, VectorizeAliasOption,
@@ -6008,6 +6015,8 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
 
   // -fslp-vectorize is enabled based on the optimization level selected.
   bool EnableSLPVec = shouldEnableVectorizerAtOLevel(Args, true);
+  if (UseSYCLTriple && EnableSYCLEarlyOptimizations)
+    EnableSLPVec = false; // But disable vectorization for SYCL device code
   OptSpecifier SLPVectAliasOption =
       EnableSLPVec ? options::OPT_O_Group : options::OPT_fslp_vectorize;
   if (Args.hasFlag(options::OPT_fslp_vectorize, SLPVectAliasOption,
@@ -7710,6 +7719,7 @@ void SPIRVTranslator::ConstructJob(Compilation &C, const JobAction &JA,
   TranslatorArgs.push_back(Output.getFilename());
   if (getToolChain().getTriple().isSYCLDeviceEnvironment()) {
     TranslatorArgs.push_back("-spirv-max-version=1.1");
+    TranslatorArgs.push_back("-spirv-debug-info-version=legacy");
     if (C.getArgs().hasArg(options::OPT_fsycl_esimd))
       TranslatorArgs.push_back("-spirv-allow-unknown-intrinsics");
 

clang/lib/Driver/ToolChains/SYCL.cpp

@@ -42,6 +42,7 @@ const char *SYCL::Linker::constructLLVMSpirvCommand(Compilation &C,
   } else {
     CmdArgs.push_back("-spirv-max-version=1.1");
     CmdArgs.push_back("-spirv-ext=+all");
+    CmdArgs.push_back("-spirv-debug-info-version=legacy");
     if (C.getArgs().hasArg(options::OPT_fsycl_esimd))
       CmdArgs.push_back("-spirv-allow-unknown-intrinsics");
     CmdArgs.push_back("-o");
@@ -204,6 +205,13 @@ void SYCL::Linker::ConstructJob(Compilation &C, const JobAction &JA,
                            SpirvInputs);
 }
 
+static const char *makeExeName(Compilation &C, StringRef Name) {
+  llvm::SmallString<8> ExeName(Name);
+  if (C.getDriver().IsCLMode())
+    ExeName.append(".exe");
+  return C.getArgs().MakeArgString(ExeName);
+}
+
 void SYCL::fpga::BackendCompiler::ConstructJob(Compilation &C,
                                          const JobAction &JA,
                                          const InputInfo &Output,
@@ -312,7 +320,8 @@ void SYCL::fpga::BackendCompiler::ConstructJob(Compilation &C,
     CmdArgs.push_back(Args.MakeArgString(A->getAsString(Args)));
   }
 
-  SmallString<128> ExecPath(getToolChain().GetProgramPath("aoc"));
+  SmallString<128> ExecPath(
+      getToolChain().GetProgramPath(makeExeName(C, "aoc")));
   const char *Exec = C.getArgs().MakeArgString(ExecPath);
   auto Cmd = std::make_unique<Command>(
       JA, *this, ResponseFileSupport::None(), Exec, CmdArgs, None);
@@ -349,7 +358,8 @@ void SYCL::gen::BackendCompiler::ConstructJob(Compilation &C,
       static_cast<const toolchains::SYCLToolChain &>(getToolChain());
   TC.TranslateBackendTargetArgs(Args, CmdArgs);
   TC.TranslateLinkerTargetArgs(Args, CmdArgs);
-  SmallString<128> ExecPath(getToolChain().GetProgramPath("ocloc"));
+  SmallString<128> ExecPath(
+      getToolChain().GetProgramPath(makeExeName(C, "ocloc")));
   const char *Exec = C.getArgs().MakeArgString(ExecPath);
   auto Cmd = std::make_unique<Command>(
       JA, *this, ResponseFileSupport::None(), Exec, CmdArgs, None);
@@ -382,7 +392,8 @@ void SYCL::x86_64::BackendCompiler::ConstructJob(Compilation &C,
 
   TC.TranslateBackendTargetArgs(Args, CmdArgs);
   TC.TranslateLinkerTargetArgs(Args, CmdArgs);
-  SmallString<128> ExecPath(getToolChain().GetProgramPath("opencl-aot"));
+  SmallString<128> ExecPath(
+      getToolChain().GetProgramPath(makeExeName(C, "opencl-aot")));
   const char *Exec = C.getArgs().MakeArgString(ExecPath);
   auto Cmd = std::make_unique<Command>(
       JA, *this, ResponseFileSupport::None(), Exec, CmdArgs, None);

clang/lib/Sema/SemaDecl.cpp

@@ -3212,6 +3212,22 @@ static void adjustDeclContextForDeclaratorDecl(DeclaratorDecl *NewD,
     FixSemaDC(VD->getDescribedVarTemplate());
 }
 
+template <typename AttributeType>
+static void checkDimensionsAndSetDiagnostics(Sema &S, FunctionDecl *New,
+                                             FunctionDecl *Old) {
+  AttributeType *NewDeclAttr = New->getAttr<AttributeType>();
+  AttributeType *OldDeclAttr = Old->getAttr<AttributeType>();
+  if ((NewDeclAttr->getXDim() != OldDeclAttr->getXDim()) ||
+      (NewDeclAttr->getYDim() != OldDeclAttr->getYDim()) ||
+      (NewDeclAttr->getZDim() != OldDeclAttr->getZDim())) {
+    S.Diag(New->getLocation(), diag::err_conflicting_sycl_function_attributes)
+        << OldDeclAttr << NewDeclAttr;
+    S.Diag(New->getLocation(), diag::warn_duplicate_attribute) << OldDeclAttr;
+    S.Diag(OldDeclAttr->getLocation(), diag::note_conflicting_attribute);
+    S.Diag(NewDeclAttr->getLocation(), diag::note_conflicting_attribute);
+  }
+}
+
 /// MergeFunctionDecl - We just parsed a function 'New' from
 /// declarator D which has the same name and scope as a previous
 /// declaration 'Old'.  Figure out how to resolve this situation,
@@ -3286,6 +3302,15 @@ bool Sema::MergeFunctionDecl(FunctionDecl *New, NamedDecl *&OldD,
     }
   }
 
+  if (New->hasAttr<ReqdWorkGroupSizeAttr>() &&
+      Old->hasAttr<ReqdWorkGroupSizeAttr>())
+    checkDimensionsAndSetDiagnostics<ReqdWorkGroupSizeAttr>(*this, New, Old);
+
+  if (New->hasAttr<SYCLIntelMaxWorkGroupSizeAttr>() &&
+      Old->hasAttr<SYCLIntelMaxWorkGroupSizeAttr>())
+    checkDimensionsAndSetDiagnostics<SYCLIntelMaxWorkGroupSizeAttr>(*this, New,
+                                                                    Old);
+
   if (New->hasAttr<InternalLinkageAttr>() &&
       !Old->hasAttr<InternalLinkageAttr>()) {
     Diag(New->getLocation(), diag::err_internal_linkage_redeclaration)

clang/lib/Sema/SemaDeclAttr.cpp

@@ -3013,13 +3013,6 @@ static void handleSubGroupSize(Sema &S, Decl *D, const ParsedAttr &AL) {
   if (D->getAttr<IntelReqdSubGroupSizeAttr>())
     S.Diag(AL.getLoc(), diag::warn_duplicate_attribute) << AL;
 
-  if (AL.getAttributeSpellingListIndex() ==
-      IntelReqdSubGroupSizeAttr::CXX11_cl_intel_reqd_sub_group_size) {
-    S.Diag(AL.getLoc(), diag::warn_attribute_spelling_deprecated) << AL;
-    S.Diag(AL.getLoc(), diag::note_spelling_suggestion)
-        << "'intel::reqd_sub_group_size'";
-  }
-
   S.addIntelReqdSubGroupSizeAttr(D, AL, E);
 }