[SYCL] Make handler.hpp independent from kernel_bundle.hpp

sycl/include/sycl/backend.hpp

@@ -23,15 +23,13 @@
 #include <sycl/event.hpp>                     // for event, get_native
 #include <sycl/exception.hpp>                 // for make_error_code
 #include <sycl/feature_test.hpp>              // for SYCL_BACKEND_OP...
-#include <sycl/handler.hpp>                   // for buffer
 #include <sycl/image.hpp>                     // for image, image_al...
 #include <sycl/kernel.hpp>                    // for kernel, get_native
-#include <sycl/kernel_bundle.hpp>             // for kernel_bundle
-#include <sycl/kernel_bundle_enums.hpp>       // for bundle_state
-#include <sycl/platform.hpp>                  // for platform, get_n...
-#include <sycl/property_list.hpp>             // for property_list
-#include <sycl/queue.hpp>                     // for queue, get_native
-#include <ur_api.h>                           // for ur_native_handle_t
+#include <sycl/kernel_bundle_enums.hpp> // for bundle_state
+#include <sycl/platform.hpp>            // for platform, get_n...
+#include <sycl/property_list.hpp>       // for property_list
+#include <sycl/queue.hpp>
+#include <ur_api.h> // for ur_native_handle_t
 
 #if SYCL_BACKEND_OPENCL
 #include <sycl/detail/backend_traits_opencl.hpp> // for interop
@@ -47,7 +45,7 @@
 #include <sycl/detail/backend_traits_hip.hpp>
 #endif
 #if SYCL_EXT_ONEAPI_BACKEND_LEVEL_ZERO
-#include <sycl/detail/backend_traits_level_zero.hpp> // for _ze_command_lis...
+// #include <sycl/detail/backend_traits_level_zero.hpp> // for _ze_command_lis...
 #endif
 
 #include <sycl/detail/ur.hpp>
@@ -60,13 +58,13 @@
 namespace sycl {
 inline namespace _V1 {
 
+template<bundle_state State>
+class kernel_bundle;
+
 namespace detail {
 // TODO each backend can have its own custom errc enumeration
 // but the details for this are not fully specified yet
 enum class backend_errc : unsigned int {};
-
-// Convert from UR backend to SYCL backend enum
-backend convertUrBackend(ur_platform_backend_t UrBackend);
 } // namespace detail
 
 template <backend Backend> class backend_traits {
@@ -154,13 +152,15 @@ auto get_native(const queue &Obj) -> backend_return_t<BackendName, queue> {
   int32_t IsImmCmdList;
   ur_native_handle_t Handle = Obj.getNative(IsImmCmdList);
   backend_return_t<BackendName, queue> RetVal;
+#if 0 && SYCL_EXT_ONEAPI_BACKEND_LEVEL_ZERO
   if constexpr (BackendName == backend::ext_oneapi_level_zero)
     RetVal = IsImmCmdList
                  ? backend_return_t<BackendName, queue>{reinterpret_cast<
                        ze_command_list_handle_t>(Handle)}
                  : backend_return_t<BackendName, queue>{
                        reinterpret_cast<ze_command_queue_handle_t>(Handle)};
   else
+#endif
     RetVal = reinterpret_cast<backend_return_t<BackendName, queue>>(Handle);
 
   return RetVal;

sycl/include/sycl/detail/backend_traits_cuda.hpp

@@ -18,7 +18,6 @@
 #include <sycl/detail/backend_traits.hpp>
 #include <sycl/device.hpp>
 #include <sycl/event.hpp>
-#include <sycl/kernel_bundle.hpp>
 #include <sycl/queue.hpp>
 
 typedef int CUdevice;

sycl/include/sycl/detail/backend_traits_hip.hpp

@@ -18,7 +18,6 @@
 #include <sycl/detail/backend_traits.hpp>
 #include <sycl/device.hpp>
 #include <sycl/event.hpp>
-#include <sycl/kernel_bundle.hpp>
 #include <sycl/queue.hpp>
 
 typedef int HIPdevice;

sycl/include/sycl/detail/backend_traits_level_zero.hpp

@@ -21,14 +21,12 @@
 #include <sycl/device.hpp>                                  // for device
 #include <sycl/event.hpp>                                   // for event
 #include <sycl/ext/oneapi/backend/level_zero_ownership.hpp> // for ownership
-#include <sycl/handler.hpp>                                 // for buffer
 #include <sycl/image.hpp>                                   // for image
 #include <sycl/kernel.hpp>                                  // for kernel
-#include <sycl/kernel_bundle.hpp>                           // for kernel_b...
+#include <sycl/kernel_bundle.hpp>
 #include <sycl/kernel_bundle_enums.hpp>                     // for bundle_s...
 #include <sycl/platform.hpp>                                // for platform
 #include <sycl/property_list.hpp>                           // for property...
-#include <sycl/queue.hpp>                                   // for queue
 #include <sycl/range.hpp>                                   // for range
 
 #include <variant> // for variant
@@ -46,6 +44,8 @@ typedef struct _ze_module_handle_t *ze_module_handle_t;
 
 namespace sycl {
 inline namespace _V1 {
+class queue;
+
 namespace detail {
 
 // Forward declarations

sycl/include/sycl/detail/backend_traits_opencl.hpp

@@ -22,17 +22,19 @@
 #include <sycl/detail/ur.hpp>             // for assertion and ur handles
 #include <sycl/device.hpp>                // for device
 #include <sycl/event.hpp>                 // for event
-#include <sycl/handler.hpp>               // for buffer
 #include <sycl/kernel.hpp>                // for kernel
-#include <sycl/kernel_bundle.hpp>         // for kernel_bundle
 #include <sycl/kernel_bundle_enums.hpp>   // for bundle_state
 #include <sycl/platform.hpp>              // for platform
-#include <sycl/queue.hpp>                 // for queue
 
 #include <vector> // for vector
 
 namespace sycl {
 inline namespace _V1 {
+
+template<bundle_state State>
+class kernel_bundle;
+class queue;
+
 namespace detail {
 
 // TODO the interops for context, device, event, platform and program

sycl/include/sycl/ext/oneapi/backend/level_zero.hpp

@@ -25,7 +25,6 @@
 #include <sycl/ext/oneapi/backend/level_zero_ownership.hpp> // for ownership
 #include <sycl/image.hpp>                                   // for image
 #include <sycl/kernel.hpp>                                  // for kernel
-#include <sycl/kernel_bundle.hpp>                           // for kernel_bu...
 #include <sycl/kernel_bundle_enums.hpp>                     // for bundle_state
 #include <sycl/platform.hpp>                                // for platform
 #include <sycl/properties/image_properties.hpp>             // for image
@@ -40,6 +39,10 @@
 
 namespace sycl {
 inline namespace _V1 {
+
+template <bundle_state State>
+class kernel_bundle;
+
 namespace ext::oneapi::level_zero::detail {
 __SYCL_EXPORT device make_device(const platform &Platform,
                                  ur_native_handle_t NativeHandle);

sycl/include/sycl/ext/oneapi/experimental/backend/backend_traits_cuda.hpp

@@ -19,7 +19,6 @@
 #include <sycl/detail/backend_traits.hpp>
 #include <sycl/device.hpp>
 #include <sycl/event.hpp>
-#include <sycl/kernel_bundle.hpp>
 #include <sycl/queue.hpp>
 
 #include <vector>

sycl/include/sycl/ext/oneapi/get_kernel_info.hpp

@@ -11,10 +11,21 @@
 #include <sycl/detail/export.hpp>
 #include <sycl/detail/info_desc_helpers.hpp>
 #include <sycl/device.hpp>
+#include <sycl/kernel_bundle_enums.hpp>
 #include <sycl/queue.hpp>
 
+#include <vector>
+
 namespace sycl {
 inline namespace _V1 {
+
+template<bundle_state State>
+class kernel_bundle;
+
+template <typename KernelName, bundle_state State>
+kernel_bundle<State> get_kernel_bundle(const context &,
+                                       const std::vector<device> &);
+
 namespace ext::oneapi {
 
 template <typename KernelName, typename Param>

sycl/include/sycl/ext/oneapi/owner_less.hpp

@@ -16,7 +16,6 @@
 #include <sycl/event.hpp>                       // for event
 #include <sycl/ext/oneapi/weak_object.hpp>      // for weak_object
 #include <sycl/kernel.hpp>                      // for kernel
-#include <sycl/kernel_bundle.hpp>               // for kernel_id
 #include <sycl/kernel_bundle_enums.hpp>         // for bundle_state
 #include <sycl/platform.hpp>                    // for platform
 #include <sycl/properties/image_properties.hpp> // for sampled_i...
@@ -27,6 +26,12 @@ namespace sycl {
 inline namespace _V1 {
 namespace ext::oneapi {
 
+class kernel_id;
+template<bundle_state State>
+class kernel_bundle;
+template<bundle_state State>
+class device_image;
+
 namespace detail {
 template <typename SyclObject> struct owner_less_base {
   bool operator()(const SyclObject &lhs, const SyclObject &rhs) const noexcept {

sycl/include/sycl/handler.hpp

@@ -43,7 +43,6 @@
 #include <sycl/id.hpp>
 #include <sycl/item.hpp>
 #include <sycl/kernel.hpp>
-#include <sycl/kernel_bundle.hpp>
 #include <sycl/kernel_bundle_enums.hpp>
 #include <sycl/kernel_handler.hpp>
 #include <sycl/nd_item.hpp>
@@ -152,6 +151,8 @@ inline namespace _V1 {
 
 // Forward declaration
 
+template<bundle_state State>
+class kernel_bundle;
 class handler;
 template <typename T, int Dimensions, typename AllocatorT, typename Enable>
 class buffer;
@@ -173,6 +174,7 @@ class graph_impl;
 } // namespace ext::oneapi::experimental::detail
 namespace detail {
 
+class kernel_bundle_impl;
 class work_group_memory_impl;
 class handler_impl;
 class kernel_impl;
@@ -1727,36 +1729,18 @@ class __SYCL_EXPORT handler {
   handler &operator=(const handler &) = delete;
   handler &operator=(handler &&) = delete;
 
+  // This is somewhat radical, but to make handler.hpp independtent from
+  // kernel_bundle.hpp, we define those methods within kernel_bundle.hpp
+  // header. Independence is needed in context of potential upcoming split of
+  // sycl.hpp so that users could do fine-grained include's, saving on
+  // compilation time by avoiding using headers for features they don't use.
   template <auto &SpecName>
   void set_specialization_constant(
-      typename std::remove_reference_t<decltype(SpecName)>::value_type Value) {
-
-    setStateSpecConstSet();
-
-    std::shared_ptr<detail::kernel_bundle_impl> KernelBundleImplPtr =
-        getOrInsertHandlerKernelBundle(/*Insert=*/true);
-
-    detail::createSyclObjFromImpl<kernel_bundle<bundle_state::input>>(
-        KernelBundleImplPtr)
-        .set_specialization_constant<SpecName>(Value);
-  }
+      typename std::remove_reference_t<decltype(SpecName)>::value_type Value);
 
   template <auto &SpecName>
   typename std::remove_reference_t<decltype(SpecName)>::value_type
-  get_specialization_constant() const {
-
-    if (isStateExplicitKernelBundle())
-      throw sycl::exception(make_error_code(errc::invalid),
-                            "Specialization constants cannot be read after "
-                            "explicitly setting the used kernel bundle");
-
-    std::shared_ptr<detail::kernel_bundle_impl> KernelBundleImplPtr =
-        getOrInsertHandlerKernelBundle(/*Insert=*/true);
-
-    return detail::createSyclObjFromImpl<kernel_bundle<bundle_state::input>>(
-               KernelBundleImplPtr)
-        .get_specialization_constant<SpecName>();
-  }
+  get_specialization_constant() const;
 
   void
   use_kernel_bundle(const kernel_bundle<bundle_state::executable> &ExecBundle);

sycl/include/sycl/kernel_bundle.hpp

@@ -14,8 +14,9 @@
 #include <sycl/detail/kernel_desc.hpp>     // for get_spec_constant_symboli...
 #include <sycl/detail/owner_less_base.hpp> // for OwnerLessBase
 #include <sycl/detail/string_view.hpp>
-#include <sycl/detail/ur.hpp>           // for cast
-#include <sycl/device.hpp>              // for device
+#include <sycl/detail/ur.hpp> // for cast
+#include <sycl/device.hpp>    // for device
+#include <sycl/handler.hpp>
 #include <sycl/kernel.hpp>              // for kernel, kernel_bundle
 #include <sycl/kernel_bundle_enums.hpp> // for bundle_state
 #include <sycl/property_list.hpp>       // for property_list
@@ -1135,6 +1136,37 @@ build(kernel_bundle<bundle_state::ext_oneapi_source> &SourceKB,
 
 } // namespace ext::oneapi::experimental
 
+template <auto &SpecName>
+void handler::set_specialization_constant(
+    typename std::remove_reference_t<decltype(SpecName)>::value_type Value) {
+
+  setStateSpecConstSet();
+
+  std::shared_ptr<detail::kernel_bundle_impl> KernelBundleImplPtr =
+      getOrInsertHandlerKernelBundle(/*Insert=*/true);
+
+  detail::createSyclObjFromImpl<kernel_bundle<bundle_state::input>>(
+      KernelBundleImplPtr)
+      .set_specialization_constant<SpecName>(Value);
+}
+
+template <auto &SpecName>
+typename std::remove_reference_t<decltype(SpecName)>::value_type
+handler::get_specialization_constant() const {
+
+  if (isStateExplicitKernelBundle())
+    throw sycl::exception(make_error_code(errc::invalid),
+                          "Specialization constants cannot be read after "
+                          "explicitly setting the used kernel bundle");
+
+  std::shared_ptr<detail::kernel_bundle_impl> KernelBundleImplPtr =
+      getOrInsertHandlerKernelBundle(/*Insert=*/true);
+
+  return detail::createSyclObjFromImpl<kernel_bundle<bundle_state::input>>(
+             KernelBundleImplPtr)
+      .get_specialization_constant<SpecName>();
+}
+
 } // namespace _V1
 } // namespace sycl
 

sycl/source/detail/allowlist.cpp

@@ -10,7 +10,7 @@
 #include <detail/device_impl.hpp>
 #include <detail/device_info.hpp>
 #include <detail/platform_info.hpp>
-#include <sycl/backend.hpp>
+#include <sycl/backend_types.hpp>
 
 #include <algorithm>
 #include <regex>

sycl/source/detail/kernel_impl.hpp

@@ -17,6 +17,7 @@
 #include <sycl/device.hpp>
 #include <sycl/ext/oneapi/experimental/root_group.hpp>
 #include <sycl/info/info_desc.hpp>
+#include <sycl/queue.hpp>
 
 #include <cassert>
 #include <memory>

sycl/source/detail/platform_impl.cpp

@@ -13,7 +13,7 @@
 #include <detail/platform_impl.hpp>
 #include <detail/platform_info.hpp>
 #include <detail/ur_info_code.hpp>
-#include <sycl/backend.hpp>
+#include <sycl/backend_types.hpp>
 #include <sycl/detail/iostream_proxy.hpp>
 #include <sycl/detail/ur.hpp>
 #include <sycl/detail/util.hpp>

sycl/source/detail/platform_impl.hpp

@@ -10,8 +10,8 @@
 
 #include <detail/adapter.hpp>
 #include <detail/platform_info.hpp>
+#include <detail/ur.hpp>
 #include <detail/ur_info_code.hpp>
-#include <sycl/backend.hpp>
 #include <sycl/backend_types.hpp>
 #include <sycl/detail/cl.h>
 #include <sycl/detail/common.hpp>
@@ -46,7 +46,7 @@ class platform_impl {
     AAdapter->call_nocheck<UrApiKind::urPlatformGetInfo>(
         APlatform, UR_PLATFORM_INFO_BACKEND, sizeof(ur_platform_backend_t),
         &UrBackend, nullptr);
-    MBackend = convertUrBackend(UrBackend);
+    MBackend = detail::convertUrBackend(UrBackend);
   }
 
   ~platform_impl() = default;

sycl/source/detail/sampler_impl.cpp

@@ -9,6 +9,7 @@
 #include <detail/context_impl.hpp>
 #include <detail/sampler_impl.hpp>
 #include <sycl/property_list.hpp>
+#include <sycl/sampler.hpp>
 
 namespace sycl {
 inline namespace _V1 {

sycl/source/detail/ur.hpp

@@ -36,6 +36,10 @@ initializeUr(ur_loader_config_handle_t LoaderConfig = nullptr);
 // Get the adapter serving given backend.
 template <backend BE> const AdapterPtr &getAdapter();
 } // namespace ur
+
+// Convert from UR backend to SYCL backend enum
+backend convertUrBackend(ur_platform_backend_t UrBackend);
+
 } // namespace detail
 } // namespace _V1
 } // namespace sycl

sycl/test-e2e/WeakObject/weak_object_owner_less.cpp

@@ -5,6 +5,7 @@
 
 #include "weak_object_utils.hpp"
 
+#include <sycl/kernel_bundle.hpp>
 #include <sycl/ext/oneapi/owner_less.hpp>
 
 #include <map>

sycl/test/include_deps/sycl_detail_core.hpp.cpp

@@ -159,8 +159,6 @@
 // CHECK-NEXT: ext/oneapi/experimental/virtual_functions.hpp
 // CHECK-NEXT: ext/oneapi/kernel_properties/properties.hpp
 // CHECK-NEXT: kernel.hpp
-// CHECK-NEXT: kernel_bundle.hpp
-// CHECK-NEXT: ext/oneapi/experimental/free_function_traits.hpp
 // CHECK-NEXT: sampler.hpp
 // CHECK-NEXT: feature_test.hpp
 // CHECK-EMPTY: