Merge remote-tracking branch 'upstream/sycl' into cc1-options

Author: AaronBallman

.github/CODEOWNERS

@@ -13,7 +13,7 @@ llvm-spirv/ @AlexeySotkin @AlexeySachkov
 opencl-aot/ @dm-vodopyanov @AlexeySachkov @romanovvlad
 
 # Device library
-libdevice/ @vzakhari
+libdevice/ @vzakhari @jinge90
 
 # DPC++ runtime library
 sycl/ @intel/llvm-reviewers-runtime

buildbot/dependency.conf

@@ -7,8 +7,8 @@ ocl_cpu_rt_ver_win=2021.11.3.0.02
 # https://github.com/intel/compute-runtime/releases/tag/21.08.19096
 ocl_gpu_rt_ver=21.08.19096
 # Same GPU driver supports Level Zero and OpenCL
-# https://downloadmirror.intel.com/30148/a08/igfx_win10_100.9168.zip
-ocl_gpu_rt_ver_win=27.20.100.9168
+# https://downloadmirror.intel.com/30266/a08/igfx_win10_100.9316.zip
+ocl_gpu_rt_ver_win=27.20.100.9316
 intel_sycl_ver=build
 
 # TBB binaries can be built from sources following instructions under
@@ -31,7 +31,7 @@ ocloc_ver_win=27.20.100.9168
 cpu_driver_lin=2021.11.3.0.02
 cpu_driver_win=2021.11.3.0.02
 gpu_driver_lin=21.08.19096
-gpu_driver_win=27.20.100.9168
+gpu_driver_win=27.20.100.9316
 fpga_driver_lin=2021.11.3.0.02
 fpga_driver_win=2021.11.3.0.02
 # NVidia CUDA driver

clang/include/clang/Basic/AttrDocs.td

@@ -2421,8 +2421,8 @@ device kernel, the attribute is not ignored and it is propagated to the kernel.
 If the`` intel::reqd_work_group_size`` or ``cl::reqd_work_group_size``
 attribute is specified on a declaration along with a
 intel::num_simd_work_items attribute, the work group size attribute
-arguments must all be evenly divisible by the argument specified in
-the ``intel::num_simd_work_items`` attribute.
+argument (the first argument) must be evenly divisible by the argument specified
+in the ``intel::num_simd_work_items`` attribute.
 
 .. code-block:: c++
 

clang/include/clang/Driver/Options.td

@@ -4295,7 +4295,10 @@ def sycl_std_EQ : Joined<["-"], "sycl-std=">, Group<sycl_Group>, Flags<[CC1Optio
   HelpText<"SYCL language standard to compile for.">, Values<"2020,2017,121,1.2.1,sycl-1.2.1">,
   NormalizedValues<["SYCL_2020", "SYCL_2017", "SYCL_2017", "SYCL_2017", "SYCL_2017"]>, NormalizedValuesScope<"LangOptions">,
   MarshallingInfoString<LangOpts<"SYCLVersion">, "SYCL_None">, AutoNormalizeEnum;
-defm sycl_esimd: OptInFFlag<"sycl-explicit-simd", "Enable", "Disable", " SYCL explicit SIMD extension.", [CC1Option,CoreOption], LangOpts<"SYCLExplicitSIMD">>;
+defm sycl_esimd: BoolFOption<"sycl-explicit-simd",
+  LangOpts<"SYCLExplicitSIMD">, DefaultFalse,
+  PosFlag<SetTrue, [CC1Option], "Enable">, NegFlag<SetFalse, [], "Disable">,
+  BothFlags<[NoArgumentUnused, CoreOption], "SYCL explicit SIMD extension.">>;
 defm sycl_early_optimizations : OptOutFFlag<"sycl-early-optimizations", "Enable", "Disable", " standard optimization pipeline for SYCL device compiler", [CoreOption]>;
 def fsycl_dead_args_optimization : Flag<["-"], "fsycl-dead-args-optimization">,
   Group<sycl_Group>, Flags<[NoArgumentUnused, CoreOption]>, HelpText<"Enables "

clang/include/clang/Sema/Sema.h

@@ -3471,8 +3471,6 @@ class Sema final {
   EnforceTCBLeafAttr *mergeEnforceTCBLeafAttr(Decl *D,
                                               const EnforceTCBLeafAttr &AL);
 
-  SYCLIntelLoopFuseAttr *
-  mergeSYCLIntelLoopFuseAttr(Decl *D, const AttributeCommonInfo &CI, Expr *E);
   void mergeDeclAttributes(NamedDecl *New, Decl *Old,
                            AvailabilityMergeKind AMK = AMK_Redeclaration);
   void MergeTypedefNameDecl(Scope *S, TypedefNameDecl *New,
@@ -10232,6 +10230,10 @@ class Sema final {
                                           Expr *E);
   SYCLIntelNoGlobalWorkOffsetAttr *MergeSYCLIntelNoGlobalWorkOffsetAttr(
       Decl *D, const SYCLIntelNoGlobalWorkOffsetAttr &A);
+  void AddSYCLIntelLoopFuseAttr(Decl *D, const AttributeCommonInfo &CI,
+                                Expr *E);
+  SYCLIntelLoopFuseAttr *
+  MergeSYCLIntelLoopFuseAttr(Decl *D, const SYCLIntelLoopFuseAttr &A);
 
   /// AddAlignedAttr - Adds an aligned attribute to a particular declaration.
   void AddAlignedAttr(Decl *D, const AttributeCommonInfo &CI, Expr *E,
@@ -10286,8 +10288,6 @@ class Sema final {
   /// addSYCLIntelPipeIOAttr - Adds a pipe I/O attribute to a particular
   /// declaration.
   void addSYCLIntelPipeIOAttr(Decl *D, const AttributeCommonInfo &CI, Expr *ID);
-  void addSYCLIntelLoopFuseAttr(Decl *D, const AttributeCommonInfo &CI,
-                                Expr *E);
 
   bool checkNSReturnsRetainedReturnType(SourceLocation loc, QualType type);
   bool checkAllowedSYCLInitializer(VarDecl *VD,

clang/lib/CodeGen/BackendUtil.cpp

@@ -943,12 +943,9 @@ void EmitAssemblyHelper::EmitAssembly(BackendAction Action,
   std::unique_ptr<llvm::ToolOutputFile> ThinLinkOS, DwoOS;
 
   // Eliminate dead arguments from SPIR kernels in SYCL environment.
-  // 1. Run DAE when LLVM optimizations are applied as well.
-  // 2. We cannot run DAE for ESIMD since the pointers to SPIR kernel
-  //    functions are saved in !genx.kernels metadata.
-  // 3. DAE pass temporary guarded under option.
+  // Run DAE when LLVM optimizations are applied as well.
   if (LangOpts.SYCLIsDevice && !CodeGenOpts.DisableLLVMPasses &&
-      !LangOpts.SYCLExplicitSIMD && LangOpts.EnableDAEInSpirKernels)
+      LangOpts.EnableDAEInSpirKernels)
     PerModulePasses.add(createDeadArgEliminationSYCLPass());
 
   switch (Action) {

clang/lib/CodeGen/CMakeLists.txt

@@ -98,5 +98,6 @@ add_clang_library(clangCodeGen
   clangBasic
   clangFrontend
   clangLex
+  clangSema
   clangSerialization
   )

clang/lib/CodeGen/CodeGenFunction.cpp

@@ -983,10 +983,11 @@ void CodeGenFunction::StartFunction(GlobalDecl GD, QualType RetTy,
 
   if (getLangOpts().SYCLIsDevice && D) {
     if (const auto *A = D->getAttr<SYCLIntelLoopFuseAttr>()) {
-      Expr *E = A->getValue();
+      const auto *CE = cast<ConstantExpr>(A->getValue());
+      Optional<llvm::APSInt> ArgVal = CE->getResultAsAPSInt();
       llvm::Metadata *AttrMDArgs[] = {
-          llvm::ConstantAsMetadata::get(Builder.getInt32(
-              E->getIntegerConstantExpr(D->getASTContext())->getZExtValue())),
+          llvm::ConstantAsMetadata::get(
+              Builder.getInt32(ArgVal->getZExtValue())),
           llvm::ConstantAsMetadata::get(
               A->isIndependent() ? Builder.getInt32(1) : Builder.getInt32(0))};
       Fn->setMetadata("loop_fuse",

clang/lib/Frontend/Rewrite/CMakeLists.txt

@@ -19,6 +19,7 @@ add_clang_library(clangRewriteFrontend
   clangFrontend
   clangLex
   clangRewrite
+  clangSema
   clangSerialization
 
   DEPENDS

clang/lib/Index/CMakeLists.txt

@@ -24,6 +24,7 @@ add_clang_library(clangIndex
   clangFrontend
   clangLex
   clangRewrite
+  clangSema
   clangSerialization
   clangToolingCore