Merge pull request #592 from tangjj11/SYCLomatic-20230213

[SYCLomatic] Fix sycl to SYCLomatic pull down failed.

Author: 0x12CC

.github/CODEOWNERS

@@ -1,2 +1,2 @@
 ##SYCLomatic
-clang/  @oneapi-src/syclomatic-reviewers 
+clang/  @oneapi-src/syclomatic-reviewers

.github/workflows/windows_test_comment_trigger.yml

@@ -10,27 +10,25 @@ jobs:
     if: ${{ github.event.issue.pull_request && github.event.comment.body == '/testwin' }}
     steps:
     - name: react_to_comment
-      uses: actions/github-script@v4
+      uses: actions/github-script@v6
       with:
         script: |
-          const {owner, repo} = context.issue;
-          github.reactions.createForIssueComment({
-            owner,
-            repo,
+          github.rest.reactions.createForIssueComment({
+            owner: context.issue.owner,
+            repo: context.issue.repo,
             comment_id: context.payload.comment.id,
-            content: "rocket",
+            content: "rocket"
           });
     - name: get_pr_sha
       id: sha
-      uses: actions/github-script@v4
+      uses: actions/github-script@v6
       with:
         result-encoding: string
         script: |
-          const { owner, repo, number } = context.issue;
-          const pr = await github.pulls.get({
-            owner,
-            repo,
-            pull_number: number,
+          const pr = await github.rest.pulls.get({
+            owner: context.issue.owner,
+            repo: context.issue.repo,
+            pull_number: number
           });
           return pr.data.head.sha
     - name: update_pr_status_pending

clang/include/clang/Basic/Attr.td

@@ -1333,12 +1333,12 @@ def SYCLType: InheritableAttr {
                             "specialization_id", "kernel_handler", "buffer_location",
                             "no_alias", "accessor_property_list", "group",
                             "private_memory", "aspect", "annotated_ptr", "annotated_arg",
-                            "stream", "sampler"],
+                            "stream", "sampler", "host_pipe"],
                            ["accessor", "local_accessor", "spec_constant",
                             "specialization_id", "kernel_handler", "buffer_location",
                             "no_alias", "accessor_property_list", "group",
                             "private_memory", "aspect", "annotated_ptr", "annotated_arg",
-                            "stream", "sampler"]>];
+                            "stream", "sampler", "host_pipe"]>];
   // Only used internally by SYCL implementation
   let Documentation = [InternalOnly];
 }

clang/include/clang/Basic/CodeGenOptions.def

@@ -515,6 +515,10 @@ CODEGENOPT(CtorDtorReturnThis, 1, 0)
 /// Whether to disable the standard optimization pipeline for the SYCL device compiler.
 CODEGENOPT(DisableSYCLEarlyOpts, 1, 0)
 
+/// Optimize SYCL Framework functions. These are functions
+/// which do not contain "user" code.
+CODEGENOPT(OptimizeSYCLFramework, 1, 0)
+
 #undef CODEGENOPT
 #undef ENUM_CODEGENOPT
 #undef VALUE_CODEGENOPT

clang/include/clang/Basic/DiagnosticDriverKinds.td

@@ -364,6 +364,8 @@ def warn_drv_sycl_target_missing : Warning<
   InGroup<SyclTarget>;
 def err_drv_no_rdc_sycl_target_missing : Error<
   "linked binaries do not contain expected '%0' target; found targets: '%1', this is not supported with '-fno-sycl-rdc'">;
+def err_drv_fsycl_wrong_optimization_options : Error<
+  "-fsycl-optimize-non-user-code option can be used only in conjunction with %0">;
 def err_drv_multiple_target_with_forced_target : Error<
   "multiple target usage with '%0' is not supported with '%1'">;
 def err_drv_failed_to_deduce_target_from_arch : Error<

clang/include/clang/Driver/Options.td

@@ -3034,6 +3034,10 @@ def fsycl_max_parallel_jobs_EQ : Joined<["-"], "fsycl-max-parallel-link-jobs=">,
   "or AOT compilation of each device image.">;
 def : Flag<["-"], "fsycl-rdc">, Flags<[CoreOption]>, Alias<fgpu_rdc>;
 def : Flag<["-"], "fno-sycl-rdc">, Flags<[CoreOption]>, Alias<fno_gpu_rdc>;
+def fsycl_optimize_non_user_code : Flag<["-"], "fsycl-optimize-non-user-code">,
+  Flags<[CC1Option, CoreOption]>, MarshallingInfoFlag<CodeGenOpts<"OptimizeSYCLFramework">>,
+  HelpText<"Option used in conjunction with -O0 to "
+  "optimize SYCL framework utility functions and leave user's kernel code unoptimized. (experimental)">;
 def fsyntax_only : Flag<["-"], "fsyntax-only">,
   Flags<[NoXarchOption,CoreOption,CC1Option,FC1Option,FlangOption]>, Group<Action_Group>,
   HelpText<"Run the preprocessor, parser and semantic analysis stages">;

clang/include/clang/Sema/Sema.h

@@ -371,6 +371,13 @@ class SYCLIntegrationHeader {
     NeedToEmitDeviceGlobalRegistration = true;
   }
 
+  /// Signals that emission of __sycl_host_pipe_registration type and
+  /// declaration of variable __sycl_host_pipe_registrar of this type in
+  /// integration header is required.
+  void addHostPipeRegistration() {
+    NeedToEmitHostPipeRegistration = true;
+  }
+
 private:
   // Kernel actual parameter descriptor.
   struct KernelParamDesc {
@@ -454,6 +461,10 @@ class SYCLIntegrationHeader {
   /// Keeps track of whether declaration of __sycl_device_global_registration
   /// type and __sycl_device_global_registrar variable are required to emit.
   bool NeedToEmitDeviceGlobalRegistration = false;
+
+  /// Keeps track of whether declaration of __sycl_host_pipe_registration
+  /// type and __sycl_host_pipe_registrar variable are required to emit.
+  bool NeedToEmitHostPipeRegistration = false;
 };
 
 class SYCLIntegrationFooter {

clang/lib/CodeGen/BackendUtil.cpp

@@ -46,6 +46,7 @@
 #include "llvm/Passes/PassBuilder.h"
 #include "llvm/Passes/PassPlugin.h"
 #include "llvm/Passes/StandardInstrumentations.h"
+#include "llvm/SYCLLowerIR/CompileTimePropertiesPass.h"
 #include "llvm/SYCLLowerIR/ESIMD/ESIMDVerifier.h"
 #include "llvm/SYCLLowerIR/LowerWGLocalMemory.h"
 #include "llvm/SYCLLowerIR/MutatePrintfAddrspace.h"
@@ -850,6 +851,9 @@ void EmitAssemblyHelper::RunOptimizationPipeline(
   // Only enable CGProfilePass when using integrated assembler, since
   // non-integrated assemblers don't recognize .cgprofile section.
   PTO.CallGraphProfile = !CodeGenOpts.DisableIntegratedAS;
+  // Enable a custom optimization pipeline for non-user SYCL code.
+  PTO.OptimizeSYCLFramework =
+      CodeGenOpts.OptimizeSYCLFramework && !CodeGenOpts.DisableLLVMPasses;
 
   LoopAnalysisManager LAM;
   FunctionAnalysisManager FAM;
@@ -1053,6 +1057,9 @@ void EmitAssemblyHelper::RunOptimizationPipeline(
       // Allocate static local memory in SYCL kernel scope for each allocation
       // call.
       MPM.addPass(SYCLLowerWGLocalMemoryPass());
+
+      // Process properties and annotations
+      MPM.addPass(CompileTimePropertiesPass());
     }
   }
 

clang/lib/CodeGen/CGBuiltin.cpp

@@ -12397,8 +12397,8 @@ static Value *getMaskVecValue(CodeGenFunction &CGF, Value *Mask,
 static Value *EmitX86MaskedStore(CodeGenFunction &CGF, ArrayRef<Value *> Ops,
                                  Align Alignment) {
   // Cast the pointer to right type.
-  Value *Ptr = CGF.Builder.CreateBitCast(Ops[0],
-                               llvm::PointerType::getUnqual(Ops[1]->getType()));
+  Value *Ptr = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+      Ops[0], llvm::PointerType::getUnqual(Ops[1]->getType()));
 
   Value *MaskVec = getMaskVecValue(
       CGF, Ops[2],
@@ -12411,8 +12411,8 @@ static Value *EmitX86MaskedLoad(CodeGenFunction &CGF, ArrayRef<Value *> Ops,
                                 Align Alignment) {
   // Cast the pointer to right type.
   llvm::Type *Ty = Ops[1]->getType();
-  Value *Ptr =
-      CGF.Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
+  Value *Ptr = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+      Ops[0], llvm::PointerType::getUnqual(Ty));
 
   Value *MaskVec = getMaskVecValue(
       CGF, Ops[2], cast<llvm::FixedVectorType>(Ty)->getNumElements());

clang/lib/CodeGen/CodeGenFunction.cpp

@@ -2789,19 +2789,21 @@ Address CodeGenFunction::EmitFieldAnnotations(const FieldDecl *D,
   llvm::PointerType *IntrinTy =
       llvm::PointerType::getWithSamePointeeType(CGM.Int8PtrTy, AS);
 
-  llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::ptr_annotation,
-                                       {IntrinTy, CGM.ConstGlobalsPtrTy});
-
   // llvm.ptr.annotation intrinsic accepts a pointer to integer of any width -
   // don't perform bitcasts if value is integer
   if (Addr.getElementType()->isIntegerTy()) {
+    llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::ptr_annotation,
+                                         {VTy, CGM.ConstGlobalsPtrTy});
 
     for (const auto *I : D->specific_attrs<AnnotateAttr>())
       V = EmitAnnotationCall(F, V, I->getAnnotation(), D->getLocation(), I);
 
     return Address(V, Addr.getElementType(), Addr.getAlignment());
   }
 
+  llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::ptr_annotation,
+                                       {IntrinTy, CGM.ConstGlobalsPtrTy});
+
   for (const auto *I : D->specific_attrs<AnnotateAttr>()) {
     V = Builder.CreateBitCast(V, IntrinTy);
     V = EmitAnnotationCall(F, V, I->getAnnotation(), D->getLocation(), I);