[SYCL] Drop kernel name type backward compatibility (#20713)

Author: sergey-semenov

sycl/include/sycl/detail/kernel_name_str_t.hpp

@@ -19,7 +19,6 @@
 #include <sycl/detail/impl_utils.hpp>
 #include <sycl/detail/kernel_desc.hpp>
 #include <sycl/detail/kernel_launch_helper.hpp>
-#include <sycl/detail/kernel_name_str_t.hpp>
 #include <sycl/detail/reduction_forward.hpp>
 #include <sycl/detail/string.hpp>
 #include <sycl/detail/string_view.hpp>
@@ -2795,7 +2794,9 @@ class __SYCL_EXPORT handler {
   detail::handler_impl *impl;
   std::vector<detail::LocalAccessorImplPtr> MLocalAccStorage;
   std::vector<std::shared_ptr<detail::stream_impl>> MStreamStorage;
-  detail::ABINeutralKernelNameStrT MKernelName;
+  // std::string_view ABI differs under `-D_GLIBCXX_USE_CXX11_ABI=0`,
+  // use our implementation instead.
+  detail::string_view MKernelName;
   /// Storage for a sycl::kernel object.
   std::shared_ptr<detail::kernel_impl> MKernel;
   /// Pointer to the source host memory or accessor(depending on command type).

sycl/include/sycl/handler.hpp

@@ -9,13 +9,13 @@
 #pragma once
 
 #include <mutex>
+#include <string_view>
 #include <unordered_map>
 
 #include <detail/compiler.hpp>
 #include <detail/device_binary_image.hpp>
 #include <detail/device_global_map_entry.hpp>
 #include <sycl/detail/defines_elementary.hpp>
-#include <sycl/detail/kernel_name_str_t.hpp>
 #include <sycl/kernel_bundle.hpp>
 
 namespace sycl {
@@ -161,7 +161,7 @@ class DeviceGlobalMap {
 
   size_t size() const { return MDeviceGlobals.size(); }
 
-  size_t count(const KernelNameStrT &UniqueId) const {
+  size_t count(std::string_view UniqueId) const {
     return MDeviceGlobals.count(UniqueId);
   }
 
@@ -173,7 +173,7 @@ class DeviceGlobalMap {
   bool MOwnerControlledCleanup = true;
 
   // Maps between device_global identifiers and associated information.
-  std::unordered_map<KernelNameStrT, std::unique_ptr<DeviceGlobalMapEntry>>
+  std::unordered_map<std::string_view, std::unique_ptr<DeviceGlobalMapEntry>>
       MDeviceGlobals;
   std::unordered_map<const void *, DeviceGlobalMapEntry *> MPtr2DeviceGlobal;
 

sycl/source/detail/device_global_map.hpp

@@ -32,7 +32,7 @@ std::shared_ptr<kernel_impl> device_image_impl::tryGetExtensionKernel(
 
       auto UrProgram = get_ur_program();
       auto [UrKernel, CacheMutex, ArgMask] =
-          PM.getOrCreateKernel(Context, KernelNameStrT(AdjustedName),
+          PM.getOrCreateKernel(Context, AdjustedName,
                                /*PropList=*/{}, UrProgram);
       return std::make_shared<kernel_impl>(
           std::move(UrKernel), *getSyclObjImpl(Context), shared_from_this(),

sycl/source/detail/device_image_impl.cpp

@@ -13,16 +13,11 @@ inline namespace _V1 {
 namespace detail {
 
 DeviceKernelInfo::DeviceKernelInfo(const CompileTimeKernelInfoTy &Info)
-    : CompileTimeKernelInfoTy(Info)
-#ifndef __INTEL_PREVIEW_BREAKING_CHANGES
-      ,
-      Name(Info.Name.data())
-#endif
-{
+    : CompileTimeKernelInfoTy(Info) {
   init(Name.data());
 }
 
-void DeviceKernelInfo::init(KernelNameStrRefT KernelName) {
+void DeviceKernelInfo::init(std::string_view KernelName) {
   auto &PM = detail::ProgramManager::getInstance();
   MImplicitLocalArgPos = PM.kernelImplicitLocalArgPos(KernelName);
 #ifndef __INTEL_PREVIEW_BREAKING_CHANGES

sycl/source/detail/device_kernel_info.cpp

@@ -11,12 +11,12 @@
 #include <detail/kernel_arg_mask.hpp>
 #include <hash_table8.hpp>
 #include <sycl/detail/compile_time_kernel_info.hpp>
-#include <sycl/detail/kernel_name_str_t.hpp>
 #include <sycl/detail/spinlock.hpp>
 #include <sycl/detail/ur.hpp>
 
 #include <mutex>
 #include <optional>
+#include <string_view>
 
 namespace sycl {
 inline namespace _V1 {
@@ -89,18 +89,12 @@ struct FastKernelSubcacheT {
 // into this structure and get rid of the other KernelName -> * maps.
 class DeviceKernelInfo : public CompileTimeKernelInfoTy {
 public:
-#ifndef __INTEL_PREVIEW_BREAKING_CHANGES
-  // Needs to own the kernel name string in non-preview builds since we pass it
-  // using a temporary string instead of a string view there.
-  std::string Name;
-#endif
-
 #ifndef __INTEL_PREVIEW_BREAKING_CHANGES
   DeviceKernelInfo() = default;
 #endif
   DeviceKernelInfo(const CompileTimeKernelInfoTy &Info);
 
-  void init(KernelNameStrRefT KernelName);
+  void init(std::string_view KernelName);
 #ifndef __INTEL_PREVIEW_BREAKING_CHANGES
   // Initialize default-created entry that has no data recorded:
   void initIfEmpty(const CompileTimeKernelInfoTy &Info);

sycl/source/detail/device_kernel_info.hpp

@@ -61,9 +61,7 @@ class handler_impl {
            HandlerSubmissionState::EXPLICIT_KERNEL_BUNDLE_STATE;
   }
 
-  KernelNameStrRefT getKernelName() const {
-    return MKernelData.getKernelName();
-  }
+  std::string_view getKernelName() const { return MKernelData.getKernelName(); }
 
   /// Registers mutually exclusive submission states.
   HandlerSubmissionState MSubmissionState = HandlerSubmissionState::NO_STATE;

sycl/source/detail/handler_impl.hpp

@@ -35,7 +35,7 @@ __SYCL_EXPORT void waitEvents(std::vector<sycl::event> DepEvents) {
 #endif
 
 const RTDeviceBinaryImage *retrieveKernelBinary(queue_impl &Queue,
-                                                KernelNameStrRefT KernelName,
+                                                std::string_view KernelName,
                                                 CGExecKernel *KernelCG) {
   device_impl &Dev = Queue.getDeviceImpl();
   bool isNvidia = Dev.getBackend() == backend::ext_oneapi_cuda;

sycl/source/detail/helpers.cpp

@@ -9,7 +9,6 @@
 #pragma once
 
 #include <sycl/detail/impl_utils.hpp>
-#include <sycl/detail/kernel_name_str_t.hpp>
 #include <sycl/detail/type_traits.hpp>
 
 #include <ur_api.h>
@@ -18,6 +17,7 @@
 #include <iterator>
 #include <memory>
 #include <queue>
+#include <string_view>
 #include <tuple>
 #include <variant>
 #include <vector>
@@ -32,7 +32,7 @@ class queue_impl;
 class RTDeviceBinaryImage;
 
 const RTDeviceBinaryImage *
-retrieveKernelBinary(queue_impl &Queue, KernelNameStrRefT KernelName,
+retrieveKernelBinary(queue_impl &Queue, std::string_view KernelName,
                      CGExecKernel *CGKernel = nullptr);
 
 template <typename SyclTy, typename... Iterators> class variadic_iterator {

sycl/source/detail/helpers.hpp

@@ -144,7 +144,7 @@ static ::jit_compiler::BinaryFormat getTargetFormat(queue_impl &Queue) {
 
 ur_kernel_handle_t jit_compiler::materializeSpecConstants(
     queue_impl &Queue, const RTDeviceBinaryImage *BinImage,
-    KernelNameStrRefT KernelName,
+    std::string_view KernelName,
     const std::vector<unsigned char> &SpecConstBlob) {
 #ifndef _WIN32
   if (!BinImage) {