[NFC][SYCL] Prepare misc source/ files for getSyclObjImpl to return raw ref (#19253)

I'm planning to change `getSyclObjImpl` to return a raw reference in a
later patch, uploading a bunch of PRs in preparation to that to make the
subsequent review easier.

Author: aelovikov-intel

sycl/source/detail/device_global_map_entry.cpp

@@ -104,19 +104,18 @@ DeviceGlobalMapEntry::getOrAllocateDeviceGlobalUSM(const context &Context) {
          "USM allocations should not be acquired for device_global with "
          "device_image_scope property.");
   context_impl &CtxImpl = *getSyclObjImpl(Context);
-  const std::shared_ptr<device_impl> &DevImpl =
-      getSyclObjImpl(CtxImpl.getDevices().front());
+  device_impl &DevImpl = *getSyclObjImpl(CtxImpl.getDevices().front());
   std::lock_guard<std::mutex> Lock(MDeviceToUSMPtrMapMutex);
 
-  auto DGUSMPtr = MDeviceToUSMPtrMap.find({DevImpl.get(), &CtxImpl});
+  auto DGUSMPtr = MDeviceToUSMPtrMap.find({&DevImpl, &CtxImpl});
   if (DGUSMPtr != MDeviceToUSMPtrMap.end())
     return DGUSMPtr->second;
 
   void *NewDGUSMPtr = detail::usm::alignedAllocInternal(
-      0, MDeviceGlobalTSize, &CtxImpl, DevImpl.get(), sycl::usm::alloc::device);
+      0, MDeviceGlobalTSize, &CtxImpl, &DevImpl, sycl::usm::alloc::device);
 
   auto NewAllocIt = MDeviceToUSMPtrMap.emplace(
-      std::piecewise_construct, std::forward_as_tuple(DevImpl.get(), &CtxImpl),
+      std::piecewise_construct, std::forward_as_tuple(&DevImpl, &CtxImpl),
       std::forward_as_tuple(NewDGUSMPtr));
   assert(NewAllocIt.second &&
          "USM allocation for device and context already happened.");

sycl/source/detail/queue_impl.cpp

@@ -52,8 +52,8 @@ static std::vector<ur_event_handle_t>
 getUrEvents(const std::vector<sycl::event> &DepEvents) {
   std::vector<ur_event_handle_t> RetUrEvents;
   for (const sycl::event &Event : DepEvents) {
-    const EventImplPtr &EventImpl = detail::getSyclObjImpl(Event);
-    auto Handle = EventImpl->getHandle();
+    event_impl &EventImpl = *detail::getSyclObjImpl(Event);
+    auto Handle = EventImpl.getHandle();
     if (Handle != nullptr)
       RetUrEvents.push_back(Handle);
   }
@@ -313,7 +313,7 @@ queue_impl::submit_impl(const detail::type_erased_cgfo_ty &CGF,
 #else
   handler Handler(shared_from_this(), SecondaryQueue, CallerNeedsEvent);
 #endif
-  auto &HandlerImpl = detail::getSyclObjImpl(Handler);
+  detail::handler_impl &HandlerImpl = *detail::getSyclObjImpl(Handler);
 
 #ifdef XPTI_ENABLE_INSTRUMENTATION
   if (xptiTraceEnabled()) {
@@ -329,16 +329,16 @@ queue_impl::submit_impl(const detail::type_erased_cgfo_ty &CGF,
   // Scheduler will later omit events, that are not required to execute tasks.
   // Host and interop tasks, however, are not submitted to low-level runtimes
   // and require separate dependency management.
-  const CGType Type = HandlerImpl->MCGType;
+  const CGType Type = HandlerImpl.MCGType;
   std::vector<StreamImplPtr> Streams;
   if (Type == CGType::Kernel)
     Streams = std::move(Handler.MStreamStorage);
 
-  HandlerImpl->MEventMode = SubmitInfo.EventMode();
+  HandlerImpl.MEventMode = SubmitInfo.EventMode();
 
   auto isHostTask = Type == CGType::CodeplayHostTask ||
                     (Type == CGType::ExecCommandBuffer &&
-                     HandlerImpl->MExecGraph->containsHostTask());
+                     HandlerImpl.MExecGraph->containsHostTask());
 
   auto requiresPostProcess = SubmitInfo.PostProcessorFunc() || Streams.size();
   auto noLastEventPath = !isHostTask &&

sycl/source/detail/ur.cpp

@@ -49,11 +49,11 @@ namespace pi {
 void contextSetExtendedDeleter(const sycl::context &context,
                                pi_context_extended_deleter func,
                                void *user_data) {
-  const auto &impl = getSyclObjImpl(context);
-  const auto &Adapter = impl->getAdapter();
-  Adapter->call<UrApiKind::urContextSetExtendedDeleter>(
-      impl->getHandleRef(),
-      reinterpret_cast<ur_context_extended_deleter_t>(func), user_data);
+  context_impl &Ctx = *getSyclObjImpl(context);
+  adapter_impl &Adapter = *Ctx.getAdapter();
+  Adapter.call<UrApiKind::urContextSetExtendedDeleter>(
+      Ctx.getHandleRef(), reinterpret_cast<ur_context_extended_deleter_t>(func),
+      user_data);
 }
 } // namespace pi
 

sycl/source/handler.cpp

@@ -1077,24 +1077,21 @@ void handler::processArg(void *Ptr, const detail::kernel_param_kind_t &Kind,
 
     detail::AccessorBaseHost *GBufBase =
         static_cast<detail::AccessorBaseHost *>(&S->GlobalBuf);
-    detail::AccessorImplPtr GBufImpl = detail::getSyclObjImpl(*GBufBase);
-    detail::Requirement *GBufReq = GBufImpl.get();
+    detail::Requirement *GBufReq = &*detail::getSyclObjImpl(*GBufBase);
     addArgsForGlobalAccessor(GBufReq, Index, IndexShift, Size,
                              IsKernelCreatedFromSource, GlobalSize, impl->MArgs,
                              IsESIMD);
     ++IndexShift;
     detail::AccessorBaseHost *GOffsetBase =
         static_cast<detail::AccessorBaseHost *>(&S->GlobalOffset);
-    detail::AccessorImplPtr GOfssetImpl = detail::getSyclObjImpl(*GOffsetBase);
-    detail::Requirement *GOffsetReq = GOfssetImpl.get();
+    detail::Requirement *GOffsetReq = &*detail::getSyclObjImpl(*GOffsetBase);
     addArgsForGlobalAccessor(GOffsetReq, Index, IndexShift, Size,
                              IsKernelCreatedFromSource, GlobalSize, impl->MArgs,
                              IsESIMD);
     ++IndexShift;
     detail::AccessorBaseHost *GFlushBase =
         static_cast<detail::AccessorBaseHost *>(&S->GlobalFlushBuf);
-    detail::AccessorImplPtr GFlushImpl = detail::getSyclObjImpl(*GFlushBase);
-    detail::Requirement *GFlushReq = GFlushImpl.get();
+    detail::Requirement *GFlushReq = &*detail::getSyclObjImpl(*GFlushBase);
 
     // If work group size wasn't set explicitly then it must be recieved
     // from kernel attribute or set to default values.

sycl/source/queue.cpp

@@ -369,9 +369,9 @@ event queue::ext_oneapi_submit_barrier(const std::vector<event> &WaitList,
                                        const detail::code_location &CodeLoc) {
   bool AllEventsEmptyOrNop = std::all_of(
       begin(WaitList), end(WaitList), [&](const event &Event) -> bool {
-        auto EventImpl = detail::getSyclObjImpl(Event);
-        return (EventImpl->isDefaultConstructed() || EventImpl->isNOP()) &&
-               !EventImpl->hasCommandGraph();
+        detail::event_impl &EventImpl = *detail::getSyclObjImpl(Event);
+        return (EventImpl.isDefaultConstructed() || EventImpl.isNOP()) &&
+               !EventImpl.hasCommandGraph();
       });
   if (is_in_order() && !impl->hasCommandGraph() && !impl->MIsProfilingEnabled &&
       AllEventsEmptyOrNop) {