Resolving merge conflicts

Signed-off-by: Gail Lyons <[email protected]>

Author: glyons-intel

clang/include/clang/Basic/Attr.td

@@ -1288,7 +1288,9 @@ def LoopUnrollHint : InheritableAttr {
 }
 
 def IntelReqdSubGroupSize: InheritableAttr {
-  let Spellings = [GNU<"intel_reqd_sub_group_size">, CXX11<"cl", "intel_reqd_sub_group_size">];
+  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>;
   let Documentation = [IntelReqdSubGroupSizeDocs];

clang/include/clang/Basic/AttrDocs.td

@@ -3472,9 +3472,10 @@ code. See `cl_intel_required_subgroup_size
 for details.
 
 SYCL documentation:
-The [[cl::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 [[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.
 
 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,
@@ -3495,6 +3496,19 @@ as in the examples below:
        /* kernel code */
   });
 
+  class Functor
+  {
+      [[intel::reqd_sub_group_size(16)]] void operator()(item<1> item)
+      {
+          /* kernel code */
+      }
+  }
+
+  kernel<class kernel_name>(
+  []() [[intel::reqd_sub_group_size(n)]] {
+       /* kernel code */
+  });
+
 See Sub-groups for NDRange Parallelism proposal in sycl/doc/extensions/sub_group_ndrange/sub_group_ndrange.md
   }];
 }

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -10979,6 +10979,9 @@ def warn_sycl_implicit_decl
               "declaration for a kernel type name; your program may not "
               "be portable">,
       InGroup<SyclStrict>, DefaultIgnore;
+def warn_sycl_restrict_recursion
+    : Warning<"SYCL kernel cannot call a recursive function">,
+      InGroup<SyclStrict>, DefaultError;
 def err_ivdep_duplicate_arg : Error<
   "duplicate argument to 'ivdep'. attribute requires one or both of a safelen "
   "and array">;
@@ -10990,6 +10993,11 @@ 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/CodeGen/BackendUtil.cpp

@@ -27,6 +27,7 @@
 #include "llvm/CodeGen/RegAllocRegistry.h"
 #include "llvm/CodeGen/SchedulerRegistry.h"
 #include "llvm/CodeGen/TargetSubtargetInfo.h"
+#include "llvm/GenXIntrinsics/GenXSPIRVWriterAdaptor.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/IRPrintingPasses.h"
 #include "llvm/IR/LegacyPassManager.h"
@@ -890,6 +891,9 @@ void EmitAssemblyHelper::EmitAssembly(BackendAction Action,
   if (LangOpts.SYCLIsDevice && CodeGenOpts.DisableLLVMPasses)
     PerModulePasses.add(createDeadCodeEliminationPass());
 
+  if (LangOpts.SYCLIsDevice && LangOpts.SYCLExplicitSIMD)
+    PerModulePasses.add(createGenXSPIRVWriterAdaptorPass());
+
   switch (Action) {
   case Backend_EmitNothing:
     break;

clang/lib/CodeGen/CMakeLists.txt

@@ -31,6 +31,13 @@ if (MSVC)
   set_source_files_properties(CodeGenModule.cpp PROPERTIES COMPILE_FLAGS /bigobj)
 endif()
 
+get_property(LLVMGenXIntrinsics_SOURCE_DIR GLOBAL PROPERTY LLVMGenXIntrinsics_SOURCE_PROP)
+get_property(LLVMGenXIntrinsics_BINARY_DIR GLOBAL PROPERTY LLVMGenXIntrinsics_BINARY_PROP)
+
+include_directories(
+  ${LLVMGenXIntrinsics_SOURCE_DIR}/GenXIntrinsics/include
+  ${LLVMGenXIntrinsics_BINARY_DIR}/GenXIntrinsics/include)
+
 add_clang_library(clangCodeGen
   BackendUtil.cpp
   CGAtomic.cpp
@@ -91,8 +98,14 @@ add_clang_library(clangCodeGen
   TargetInfo.cpp
   VarBypassDetector.cpp
 
+  ADDITIONAL_HEADER_DIRS
+  ${LLVMGenXIntrinsics_SOURCE_DIR}/GenXIntrinsics/include
+  ${LLVMGenXIntrinsics_BINARY_DIR}/GenXIntrinsics/include
+
   DEPENDS
+  ${codegen_deps}
   intrinsics_gen
+  LLVMGenXIntrinsics
 
   LINK_LIBS
   clangAnalysis
@@ -102,4 +115,5 @@ add_clang_library(clangCodeGen
   clangFrontend
   clangLex
   clangSerialization
+  LLVMGenXIntrinsics
   )

clang/lib/CodeGen/CodeGenModule.cpp

@@ -641,12 +641,12 @@ void CodeGenModule::Release() {
     SPIRVerMD->addOperand(llvm::MDNode::get(Ctx, SPIRVerElts));
     // We are trying to look like OpenCL C++ for SPIR-V translator.
     // 4 - OpenCL_CPP, 100000 - OpenCL C++ version 1.0
-    // 6 - ESIMD, if any kernel or function is an explicit SIMD one
+    // 0 - ESIMD, if any kernel or function is an explicit SIMD one
     int Lang = llvm::any_of(TheModule,
                             [](const auto &F) {
                               return F.getMetadata("sycl_explicit_simd");
                             })
-                   ? 6
+                   ? 0
                    : 4;
 
     llvm::Metadata *SPIRVSourceElts[] = {

clang/lib/Driver/ToolChains/Clang.cpp

@@ -7658,6 +7658,8 @@ void SPIRVTranslator::ConstructJob(Compilation &C, const JobAction &JA,
   if (getToolChain().getTriple().isSYCLDeviceEnvironment()) {
     TranslatorArgs.push_back("-spirv-max-version=1.1");
     std::string ExtArg("-spirv-ext=+all");
+    if (C.getArgs().hasArg(options::OPT_fsycl_esimd))
+      TranslatorArgs.push_back("-spirv-allow-unknown-intrinsics");
     // Disable SPV_INTEL_usm_storage_classes by default since it adds new
     // storage classes that represent global_device and global_host address
     // spaces, which are not supported for all targets. With the extension

clang/lib/Driver/ToolChains/SYCL.cpp

@@ -42,6 +42,8 @@ const char *SYCL::Linker::constructLLVMSpirvCommand(Compilation &C,
   } else {
     CmdArgs.push_back("-spirv-max-version=1.1");
     CmdArgs.push_back("-spirv-ext=+all");
+    if (C.getArgs().hasArg(options::OPT_fsycl_esimd))
+      CmdArgs.push_back("-spirv-allow-unknown-intrinsics");
     CmdArgs.push_back("-o");
     CmdArgs.push_back(Output.getFilename());
   }

clang/lib/Sema/SemaDeclAttr.cpp

@@ -3014,6 +3014,13 @@ 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);
 }
 

clang/lib/Sema/SemaSYCL.cpp

@@ -354,8 +354,7 @@ class MarkDeviceFunction : public RecursiveASTVisitor<MarkDeviceFunction> {
       // all functions used by kernel have already been parsed and have
       // definitions.
       if (RecursiveSet.count(Callee) && !ConstexprDepth) {
-        SemaRef.Diag(e->getExprLoc(), diag::err_sycl_restrict)
-            << Sema::KernelCallRecursiveFunction;
+        SemaRef.Diag(e->getExprLoc(), diag::warn_sycl_restrict_recursion);
         SemaRef.Diag(Callee->getSourceRange().getBegin(),
                      diag::note_sycl_recursive_function_declared_here)
             << Sema::KernelCallRecursiveFunction;